4 Copyright (C) Ronnie Sahlberg 2007
5 Copyright (C) Andrew Tridgell 2007
6 Copyright (C) Martin Schwenke 2011
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, see <http://www.gnu.org/licenses/>.
23 #include "lib/util/dlinklist.h"
24 #include "system/network.h"
25 #include "system/filesys.h"
26 #include "system/wait.h"
27 #include "../include/ctdb_private.h"
28 #include "../common/rb_tree.h"
31 #define TAKEOVER_TIMEOUT() timeval_current_ofs(ctdb->tunable.takeover_timeout,0)
33 #define CTDB_ARP_INTERVAL 1
34 #define CTDB_ARP_REPEAT 3
36 /* Flags used in IP allocation algorithms. */
43 struct ctdb_iface *prev, *next;
49 static const char *ctdb_vnn_iface_string(const struct ctdb_vnn *vnn)
52 return vnn->iface->name;
58 static int ctdb_add_local_iface(struct ctdb_context *ctdb, const char *iface)
62 /* Verify that we dont have an entry for this ip yet */
63 for (i=ctdb->ifaces;i;i=i->next) {
64 if (strcmp(i->name, iface) == 0) {
69 /* create a new structure for this interface */
70 i = talloc_zero(ctdb, struct ctdb_iface);
71 CTDB_NO_MEMORY_FATAL(ctdb, i);
72 i->name = talloc_strdup(i, iface);
73 CTDB_NO_MEMORY(ctdb, i->name);
75 * If link_up defaults to true then IPs can be allocated to a
76 * node during the first recovery. However, then an interface
77 * could have its link marked down during the startup event,
78 * causing the IP to move almost immediately. If link_up
79 * defaults to false then, during normal operation, IPs added
80 * to a new interface can't be assigned until a monitor cycle
81 * has occurred and marked the new interfaces up. This makes
82 * IP allocation unpredictable. The following is a neat
83 * compromise: early in startup link_up defaults to false, so
84 * IPs can't be assigned, and after startup IPs can be
85 * assigned immediately.
87 i->link_up = (ctdb->runstate == CTDB_RUNSTATE_RUNNING);
89 DLIST_ADD(ctdb->ifaces, i);
94 static bool vnn_has_interface_with_name(struct ctdb_vnn *vnn,
99 for (n = 0; vnn->ifaces[n] != NULL; n++) {
100 if (strcmp(name, vnn->ifaces[n]) == 0) {
108 /* If any interfaces now have no possible IPs then delete them. This
109 * implementation is naive (i.e. simple) rather than clever
110 * (i.e. complex). Given that this is run on delip and that operation
111 * is rare, this doesn't need to be efficient - it needs to be
112 * foolproof. One alternative is reference counting, where the logic
113 * is distributed and can, therefore, be broken in multiple places.
114 * Another alternative is to build a red-black tree of interfaces that
115 * can have addresses (by walking ctdb->vnn and ctdb->single_ip_vnn
116 * once) and then walking ctdb->ifaces once and deleting those not in
117 * the tree. Let's go to one of those if the naive implementation
118 * causes problems... :-)
120 static void ctdb_remove_orphaned_ifaces(struct ctdb_context *ctdb,
121 struct ctdb_vnn *vnn,
124 struct ctdb_iface *i;
126 /* For each interface, check if there's an IP using it. */
127 for(i=ctdb->ifaces; i; i=i->next) {
131 /* Only consider interfaces named in the given VNN. */
132 if (!vnn_has_interface_with_name(vnn, i->name)) {
136 /* Is the "single IP" on this interface? */
137 if ((ctdb->single_ip_vnn != NULL) &&
138 (ctdb->single_ip_vnn->ifaces[0] != NULL) &&
139 (strcmp(i->name, ctdb->single_ip_vnn->ifaces[0]) == 0)) {
140 /* Found, next interface please... */
143 /* Search for a vnn with this interface. */
145 for (tv=ctdb->vnn; tv; tv=tv->next) {
146 if (vnn_has_interface_with_name(tv, i->name)) {
153 /* None of the VNNs are using this interface. */
154 DLIST_REMOVE(ctdb->ifaces, i);
155 /* Caller will free mem_ctx when convenient. */
156 talloc_steal(mem_ctx, i);
162 static struct ctdb_iface *ctdb_find_iface(struct ctdb_context *ctdb,
165 struct ctdb_iface *i;
167 for (i=ctdb->ifaces;i;i=i->next) {
168 if (strcmp(i->name, iface) == 0) {
176 static struct ctdb_iface *ctdb_vnn_best_iface(struct ctdb_context *ctdb,
177 struct ctdb_vnn *vnn)
180 struct ctdb_iface *cur = NULL;
181 struct ctdb_iface *best = NULL;
183 for (i=0; vnn->ifaces[i]; i++) {
185 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
199 if (cur->references < best->references) {
208 static int32_t ctdb_vnn_assign_iface(struct ctdb_context *ctdb,
209 struct ctdb_vnn *vnn)
211 struct ctdb_iface *best = NULL;
214 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
215 "still assigned to iface '%s'\n",
216 ctdb_addr_to_str(&vnn->public_address),
217 ctdb_vnn_iface_string(vnn)));
221 best = ctdb_vnn_best_iface(ctdb, vnn);
223 DEBUG(DEBUG_ERR, (__location__ " public address '%s' "
224 "cannot assign to iface any iface\n",
225 ctdb_addr_to_str(&vnn->public_address)));
231 vnn->pnn = ctdb->pnn;
233 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
234 "now assigned to iface '%s' refs[%d]\n",
235 ctdb_addr_to_str(&vnn->public_address),
236 ctdb_vnn_iface_string(vnn),
241 static void ctdb_vnn_unassign_iface(struct ctdb_context *ctdb,
242 struct ctdb_vnn *vnn)
244 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
245 "now unassigned (old iface '%s' refs[%d])\n",
246 ctdb_addr_to_str(&vnn->public_address),
247 ctdb_vnn_iface_string(vnn),
248 vnn->iface?vnn->iface->references:0));
250 vnn->iface->references--;
253 if (vnn->pnn == ctdb->pnn) {
258 static bool ctdb_vnn_available(struct ctdb_context *ctdb,
259 struct ctdb_vnn *vnn)
263 if (vnn->iface && vnn->iface->link_up) {
267 for (i=0; vnn->ifaces[i]; i++) {
268 struct ctdb_iface *cur;
270 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
283 struct ctdb_takeover_arp {
284 struct ctdb_context *ctdb;
287 struct ctdb_tcp_array *tcparray;
288 struct ctdb_vnn *vnn;
293 lists of tcp endpoints
295 struct ctdb_tcp_list {
296 struct ctdb_tcp_list *prev, *next;
297 struct ctdb_tcp_connection connection;
301 list of clients to kill on IP release
303 struct ctdb_client_ip {
304 struct ctdb_client_ip *prev, *next;
305 struct ctdb_context *ctdb;
312 send a gratuitous arp
314 static void ctdb_control_send_arp(struct event_context *ev, struct timed_event *te,
315 struct timeval t, void *private_data)
317 struct ctdb_takeover_arp *arp = talloc_get_type(private_data,
318 struct ctdb_takeover_arp);
320 struct ctdb_tcp_array *tcparray;
321 const char *iface = ctdb_vnn_iface_string(arp->vnn);
323 ret = ctdb_sys_send_arp(&arp->addr, iface);
325 DEBUG(DEBUG_CRIT,(__location__ " sending of arp failed on iface '%s' (%s)\n",
326 iface, strerror(errno)));
329 tcparray = arp->tcparray;
331 for (i=0;i<tcparray->num;i++) {
332 struct ctdb_tcp_connection *tcon;
334 tcon = &tcparray->connections[i];
335 DEBUG(DEBUG_INFO,("sending tcp tickle ack for %u->%s:%u\n",
336 (unsigned)ntohs(tcon->dst_addr.ip.sin_port),
337 ctdb_addr_to_str(&tcon->src_addr),
338 (unsigned)ntohs(tcon->src_addr.ip.sin_port)));
339 ret = ctdb_sys_send_tcp(
344 DEBUG(DEBUG_CRIT,(__location__ " Failed to send tcp tickle ack for %s\n",
345 ctdb_addr_to_str(&tcon->src_addr)));
352 if (arp->count == CTDB_ARP_REPEAT) {
357 event_add_timed(arp->ctdb->ev, arp->vnn->takeover_ctx,
358 timeval_current_ofs(CTDB_ARP_INTERVAL, 100000),
359 ctdb_control_send_arp, arp);
362 static int32_t ctdb_announce_vnn_iface(struct ctdb_context *ctdb,
363 struct ctdb_vnn *vnn)
365 struct ctdb_takeover_arp *arp;
366 struct ctdb_tcp_array *tcparray;
368 if (!vnn->takeover_ctx) {
369 vnn->takeover_ctx = talloc_new(vnn);
370 if (!vnn->takeover_ctx) {
375 arp = talloc_zero(vnn->takeover_ctx, struct ctdb_takeover_arp);
381 arp->addr = vnn->public_address;
384 tcparray = vnn->tcp_array;
386 /* add all of the known tcp connections for this IP to the
387 list of tcp connections to send tickle acks for */
388 arp->tcparray = talloc_steal(arp, tcparray);
390 vnn->tcp_array = NULL;
391 vnn->tcp_update_needed = true;
394 event_add_timed(arp->ctdb->ev, vnn->takeover_ctx,
395 timeval_zero(), ctdb_control_send_arp, arp);
400 struct takeover_callback_state {
401 struct ctdb_req_control *c;
402 ctdb_sock_addr *addr;
403 struct ctdb_vnn *vnn;
406 struct ctdb_do_takeip_state {
407 struct ctdb_req_control *c;
408 struct ctdb_vnn *vnn;
412 called when takeip event finishes
414 static void ctdb_do_takeip_callback(struct ctdb_context *ctdb, int status,
417 struct ctdb_do_takeip_state *state =
418 talloc_get_type(private_data, struct ctdb_do_takeip_state);
423 struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
425 if (status == -ETIME) {
428 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
429 ctdb_addr_to_str(&state->vnn->public_address),
430 ctdb_vnn_iface_string(state->vnn)));
431 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
433 node->flags |= NODE_FLAGS_UNHEALTHY;
438 if (ctdb->do_checkpublicip) {
440 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
442 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
449 data.dptr = (uint8_t *)ctdb_addr_to_str(&state->vnn->public_address);
450 data.dsize = strlen((char *)data.dptr) + 1;
451 DEBUG(DEBUG_INFO,(__location__ " sending TAKE_IP for '%s'\n", data.dptr));
453 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_TAKE_IP, data);
456 /* the control succeeded */
457 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
462 static int ctdb_takeip_destructor(struct ctdb_do_takeip_state *state)
464 state->vnn->update_in_flight = false;
469 take over an ip address
471 static int32_t ctdb_do_takeip(struct ctdb_context *ctdb,
472 struct ctdb_req_control *c,
473 struct ctdb_vnn *vnn)
476 struct ctdb_do_takeip_state *state;
478 if (vnn->update_in_flight) {
479 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u rejected "
480 "update for this IP already in flight\n",
481 ctdb_addr_to_str(&vnn->public_address),
482 vnn->public_netmask_bits));
486 ret = ctdb_vnn_assign_iface(ctdb, vnn);
488 DEBUG(DEBUG_ERR,("Takeover of IP %s/%u failed to "
489 "assign a usable interface\n",
490 ctdb_addr_to_str(&vnn->public_address),
491 vnn->public_netmask_bits));
495 state = talloc(vnn, struct ctdb_do_takeip_state);
496 CTDB_NO_MEMORY(ctdb, state);
498 state->c = talloc_steal(ctdb, c);
501 vnn->update_in_flight = true;
502 talloc_set_destructor(state, ctdb_takeip_destructor);
504 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u on interface %s\n",
505 ctdb_addr_to_str(&vnn->public_address),
506 vnn->public_netmask_bits,
507 ctdb_vnn_iface_string(vnn)));
509 ret = ctdb_event_script_callback(ctdb,
511 ctdb_do_takeip_callback,
516 ctdb_vnn_iface_string(vnn),
517 ctdb_addr_to_str(&vnn->public_address),
518 vnn->public_netmask_bits);
521 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
522 ctdb_addr_to_str(&vnn->public_address),
523 ctdb_vnn_iface_string(vnn)));
531 struct ctdb_do_updateip_state {
532 struct ctdb_req_control *c;
533 struct ctdb_iface *old;
534 struct ctdb_vnn *vnn;
538 called when updateip event finishes
540 static void ctdb_do_updateip_callback(struct ctdb_context *ctdb, int status,
543 struct ctdb_do_updateip_state *state =
544 talloc_get_type(private_data, struct ctdb_do_updateip_state);
548 if (status == -ETIME) {
551 DEBUG(DEBUG_ERR,(__location__ " Failed to move IP %s from interface %s to %s\n",
552 ctdb_addr_to_str(&state->vnn->public_address),
554 ctdb_vnn_iface_string(state->vnn)));
557 * All we can do is reset the old interface
558 * and let the next run fix it
560 ctdb_vnn_unassign_iface(ctdb, state->vnn);
561 state->vnn->iface = state->old;
562 state->vnn->iface->references++;
564 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
569 if (ctdb->do_checkpublicip) {
571 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
573 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
580 /* the control succeeded */
581 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
586 static int ctdb_updateip_destructor(struct ctdb_do_updateip_state *state)
588 state->vnn->update_in_flight = false;
593 update (move) an ip address
595 static int32_t ctdb_do_updateip(struct ctdb_context *ctdb,
596 struct ctdb_req_control *c,
597 struct ctdb_vnn *vnn)
600 struct ctdb_do_updateip_state *state;
601 struct ctdb_iface *old = vnn->iface;
602 const char *new_name;
604 if (vnn->update_in_flight) {
605 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u rejected "
606 "update for this IP already in flight\n",
607 ctdb_addr_to_str(&vnn->public_address),
608 vnn->public_netmask_bits));
612 ctdb_vnn_unassign_iface(ctdb, vnn);
613 ret = ctdb_vnn_assign_iface(ctdb, vnn);
615 DEBUG(DEBUG_ERR,("update of IP %s/%u failed to "
616 "assin a usable interface (old iface '%s')\n",
617 ctdb_addr_to_str(&vnn->public_address),
618 vnn->public_netmask_bits,
623 new_name = ctdb_vnn_iface_string(vnn);
624 if (old->name != NULL && new_name != NULL && !strcmp(old->name, new_name)) {
625 /* A benign update from one interface onto itself.
626 * no need to run the eventscripts in this case, just return
629 ctdb_request_control_reply(ctdb, c, NULL, 0, NULL);
633 state = talloc(vnn, struct ctdb_do_updateip_state);
634 CTDB_NO_MEMORY(ctdb, state);
636 state->c = talloc_steal(ctdb, c);
640 vnn->update_in_flight = true;
641 talloc_set_destructor(state, ctdb_updateip_destructor);
643 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u from "
644 "interface %s to %s\n",
645 ctdb_addr_to_str(&vnn->public_address),
646 vnn->public_netmask_bits,
650 ret = ctdb_event_script_callback(ctdb,
652 ctdb_do_updateip_callback,
655 CTDB_EVENT_UPDATE_IP,
659 ctdb_addr_to_str(&vnn->public_address),
660 vnn->public_netmask_bits);
662 DEBUG(DEBUG_ERR,(__location__ " Failed update IP %s from interface %s to %s\n",
663 ctdb_addr_to_str(&vnn->public_address),
664 old->name, new_name));
673 Find the vnn of the node that has a public ip address
674 returns -1 if the address is not known as a public address
676 static struct ctdb_vnn *find_public_ip_vnn(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
678 struct ctdb_vnn *vnn;
680 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
681 if (ctdb_same_ip(&vnn->public_address, addr)) {
690 take over an ip address
692 int32_t ctdb_control_takeover_ip(struct ctdb_context *ctdb,
693 struct ctdb_req_control *c,
698 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
699 struct ctdb_vnn *vnn;
700 bool have_ip = false;
701 bool do_updateip = false;
702 bool do_takeip = false;
703 struct ctdb_iface *best_iface = NULL;
705 if (pip->pnn != ctdb->pnn) {
706 DEBUG(DEBUG_ERR,(__location__" takeoverip called for an ip '%s' "
707 "with pnn %d, but we're node %d\n",
708 ctdb_addr_to_str(&pip->addr),
709 pip->pnn, ctdb->pnn));
713 /* update out vnn list */
714 vnn = find_public_ip_vnn(ctdb, &pip->addr);
716 DEBUG(DEBUG_INFO,("takeoverip called for an ip '%s' that is not a public address\n",
717 ctdb_addr_to_str(&pip->addr)));
721 if (ctdb->do_checkpublicip) {
722 have_ip = ctdb_sys_have_ip(&pip->addr);
724 best_iface = ctdb_vnn_best_iface(ctdb, vnn);
725 if (best_iface == NULL) {
726 DEBUG(DEBUG_ERR,("takeoverip of IP %s/%u failed to find"
727 "a usable interface (old %s, have_ip %d)\n",
728 ctdb_addr_to_str(&vnn->public_address),
729 vnn->public_netmask_bits,
730 ctdb_vnn_iface_string(vnn),
735 if (vnn->iface == NULL && vnn->pnn == -1 && have_ip && best_iface != NULL) {
736 DEBUG(DEBUG_ERR,("Taking over newly created ip\n"));
741 if (vnn->iface == NULL && have_ip) {
742 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
743 "but we have no interface assigned, has someone manually configured it? Ignore for now.\n",
744 ctdb_addr_to_str(&vnn->public_address)));
748 if (vnn->pnn != ctdb->pnn && have_ip && vnn->pnn != -1) {
749 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
750 "and we have it on iface[%s], but it was assigned to node %d"
751 "and we are node %d, banning ourself\n",
752 ctdb_addr_to_str(&vnn->public_address),
753 ctdb_vnn_iface_string(vnn), vnn->pnn, ctdb->pnn));
758 if (vnn->pnn == -1 && have_ip) {
759 vnn->pnn = ctdb->pnn;
760 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
761 "and we already have it on iface[%s], update local daemon\n",
762 ctdb_addr_to_str(&vnn->public_address),
763 ctdb_vnn_iface_string(vnn)));
768 if (vnn->iface != best_iface) {
769 if (!vnn->iface->link_up) {
771 } else if (vnn->iface->references > (best_iface->references + 1)) {
772 /* only move when the rebalance gains something */
780 ctdb_vnn_unassign_iface(ctdb, vnn);
787 ret = ctdb_do_takeip(ctdb, c, vnn);
791 } else if (do_updateip) {
792 ret = ctdb_do_updateip(ctdb, c, vnn);
798 * The interface is up and the kernel known the ip
801 DEBUG(DEBUG_INFO,("Redundant takeover of IP %s/%u on interface %s (ip already held)\n",
802 ctdb_addr_to_str(&pip->addr),
803 vnn->public_netmask_bits,
804 ctdb_vnn_iface_string(vnn)));
808 /* tell ctdb_control.c that we will be replying asynchronously */
815 takeover an ip address old v4 style
817 int32_t ctdb_control_takeover_ipv4(struct ctdb_context *ctdb,
818 struct ctdb_req_control *c,
824 data.dsize = sizeof(struct ctdb_public_ip);
825 data.dptr = (uint8_t *)talloc_zero(c, struct ctdb_public_ip);
826 CTDB_NO_MEMORY(ctdb, data.dptr);
828 memcpy(data.dptr, indata.dptr, indata.dsize);
829 return ctdb_control_takeover_ip(ctdb, c, data, async_reply);
833 kill any clients that are registered with a IP that is being released
835 static void release_kill_clients(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
837 struct ctdb_client_ip *ip;
839 DEBUG(DEBUG_INFO,("release_kill_clients for ip %s\n",
840 ctdb_addr_to_str(addr)));
842 for (ip=ctdb->client_ip_list; ip; ip=ip->next) {
843 ctdb_sock_addr tmp_addr;
846 DEBUG(DEBUG_INFO,("checking for client %u with IP %s\n",
848 ctdb_addr_to_str(&ip->addr)));
850 if (ctdb_same_ip(&tmp_addr, addr)) {
851 struct ctdb_client *client = ctdb_reqid_find(ctdb,
854 DEBUG(DEBUG_INFO,("matched client %u with IP %s and pid %u\n",
856 ctdb_addr_to_str(&ip->addr),
859 if (client->pid != 0) {
860 DEBUG(DEBUG_INFO,(__location__ " Killing client pid %u for IP %s on client_id %u\n",
861 (unsigned)client->pid,
862 ctdb_addr_to_str(addr),
864 kill(client->pid, SIGKILL);
871 called when releaseip event finishes
873 static void release_ip_callback(struct ctdb_context *ctdb, int status,
876 struct takeover_callback_state *state =
877 talloc_get_type(private_data, struct takeover_callback_state);
880 if (status == -ETIME) {
884 if (ctdb->do_checkpublicip && ctdb_sys_have_ip(state->addr)) {
885 DEBUG(DEBUG_ERR, ("IP %s still hosted during release IP callback, failing\n",
886 ctdb_addr_to_str(state->addr)));
887 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
892 /* send a message to all clients of this node telling them
893 that the cluster has been reconfigured and they should
894 release any sockets on this IP */
895 data.dptr = (uint8_t *)talloc_strdup(state, ctdb_addr_to_str(state->addr));
896 CTDB_NO_MEMORY_VOID(ctdb, data.dptr);
897 data.dsize = strlen((char *)data.dptr)+1;
899 DEBUG(DEBUG_INFO,(__location__ " sending RELEASE_IP for '%s'\n", data.dptr));
901 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELEASE_IP, data);
903 /* kill clients that have registered with this IP */
904 release_kill_clients(ctdb, state->addr);
906 ctdb_vnn_unassign_iface(ctdb, state->vnn);
908 /* the control succeeded */
909 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
913 static int ctdb_releaseip_destructor(struct takeover_callback_state *state)
915 state->vnn->update_in_flight = false;
920 release an ip address
922 int32_t ctdb_control_release_ip(struct ctdb_context *ctdb,
923 struct ctdb_req_control *c,
928 struct takeover_callback_state *state;
929 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
930 struct ctdb_vnn *vnn;
933 /* update our vnn list */
934 vnn = find_public_ip_vnn(ctdb, &pip->addr);
936 DEBUG(DEBUG_INFO,("releaseip called for an ip '%s' that is not a public address\n",
937 ctdb_addr_to_str(&pip->addr)));
942 /* stop any previous arps */
943 talloc_free(vnn->takeover_ctx);
944 vnn->takeover_ctx = NULL;
946 /* Some ctdb tool commands (e.g. moveip, rebalanceip) send
947 * lazy multicast to drop an IP from any node that isn't the
948 * intended new node. The following causes makes ctdbd ignore
949 * a release for any address it doesn't host.
951 if (ctdb->do_checkpublicip) {
952 if (!ctdb_sys_have_ip(&pip->addr)) {
953 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u on interface %s (ip not held)\n",
954 ctdb_addr_to_str(&pip->addr),
955 vnn->public_netmask_bits,
956 ctdb_vnn_iface_string(vnn)));
957 ctdb_vnn_unassign_iface(ctdb, vnn);
961 if (vnn->iface == NULL) {
962 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u (ip not held)\n",
963 ctdb_addr_to_str(&pip->addr),
964 vnn->public_netmask_bits));
969 /* There is a potential race between take_ip and us because we
970 * update the VNN via a callback that run when the
971 * eventscripts have been run. Avoid the race by allowing one
972 * update to be in flight at a time.
974 if (vnn->update_in_flight) {
975 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u rejected "
976 "update for this IP already in flight\n",
977 ctdb_addr_to_str(&vnn->public_address),
978 vnn->public_netmask_bits));
982 if (ctdb->do_checkpublicip) {
983 iface = ctdb_sys_find_ifname(&pip->addr);
985 DEBUG(DEBUG_ERR, ("Could not find which interface the ip address is hosted on. can not release it\n"));
988 if (vnn->iface == NULL) {
990 ("Public IP %s is hosted on interface %s but we have no VNN\n",
991 ctdb_addr_to_str(&pip->addr),
993 } else if (strcmp(iface, ctdb_vnn_iface_string(vnn)) != 0) {
995 ("Public IP %s is hosted on inteterface %s but VNN says %s\n",
996 ctdb_addr_to_str(&pip->addr),
998 ctdb_vnn_iface_string(vnn)));
999 /* Should we fix vnn->iface? If we do, what
1000 * happens to reference counts?
1004 iface = strdup(ctdb_vnn_iface_string(vnn));
1007 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u on interface %s node:%d\n",
1008 ctdb_addr_to_str(&pip->addr),
1009 vnn->public_netmask_bits,
1013 state = talloc(ctdb, struct takeover_callback_state);
1014 CTDB_NO_MEMORY(ctdb, state);
1016 state->c = talloc_steal(state, c);
1017 state->addr = talloc(state, ctdb_sock_addr);
1018 CTDB_NO_MEMORY(ctdb, state->addr);
1019 *state->addr = pip->addr;
1022 vnn->update_in_flight = true;
1023 talloc_set_destructor(state, ctdb_releaseip_destructor);
1025 ret = ctdb_event_script_callback(ctdb,
1026 state, release_ip_callback, state,
1028 CTDB_EVENT_RELEASE_IP,
1031 ctdb_addr_to_str(&pip->addr),
1032 vnn->public_netmask_bits);
1035 DEBUG(DEBUG_ERR,(__location__ " Failed to release IP %s on interface %s\n",
1036 ctdb_addr_to_str(&pip->addr),
1037 ctdb_vnn_iface_string(vnn)));
1042 /* tell the control that we will be reply asynchronously */
1043 *async_reply = true;
1048 release an ip address old v4 style
1050 int32_t ctdb_control_release_ipv4(struct ctdb_context *ctdb,
1051 struct ctdb_req_control *c,
1057 data.dsize = sizeof(struct ctdb_public_ip);
1058 data.dptr = (uint8_t *)talloc_zero(c, struct ctdb_public_ip);
1059 CTDB_NO_MEMORY(ctdb, data.dptr);
1061 memcpy(data.dptr, indata.dptr, indata.dsize);
1062 return ctdb_control_release_ip(ctdb, c, data, async_reply);
1066 static int ctdb_add_public_address(struct ctdb_context *ctdb,
1067 ctdb_sock_addr *addr,
1068 unsigned mask, const char *ifaces,
1071 struct ctdb_vnn *vnn;
1078 tmp = strdup(ifaces);
1079 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1080 if (!ctdb_sys_check_iface_exists(iface)) {
1081 DEBUG(DEBUG_CRIT,("Interface %s does not exist. Can not add public-address : %s\n", iface, ctdb_addr_to_str(addr)));
1088 /* Verify that we dont have an entry for this ip yet */
1089 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1090 if (ctdb_same_sockaddr(addr, &vnn->public_address)) {
1091 DEBUG(DEBUG_CRIT,("Same ip '%s' specified multiple times in the public address list \n",
1092 ctdb_addr_to_str(addr)));
1097 /* create a new vnn structure for this ip address */
1098 vnn = talloc_zero(ctdb, struct ctdb_vnn);
1099 CTDB_NO_MEMORY_FATAL(ctdb, vnn);
1100 vnn->ifaces = talloc_array(vnn, const char *, num + 2);
1101 tmp = talloc_strdup(vnn, ifaces);
1102 CTDB_NO_MEMORY_FATAL(ctdb, tmp);
1103 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1104 vnn->ifaces = talloc_realloc(vnn, vnn->ifaces, const char *, num + 2);
1105 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces);
1106 vnn->ifaces[num] = talloc_strdup(vnn, iface);
1107 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces[num]);
1111 vnn->ifaces[num] = NULL;
1112 vnn->public_address = *addr;
1113 vnn->public_netmask_bits = mask;
1115 if (check_address) {
1116 if (ctdb_sys_have_ip(addr)) {
1117 DEBUG(DEBUG_ERR,("We are already hosting public address '%s'. setting PNN to ourself:%d\n", ctdb_addr_to_str(addr), ctdb->pnn));
1118 vnn->pnn = ctdb->pnn;
1122 for (i=0; vnn->ifaces[i]; i++) {
1123 ret = ctdb_add_local_iface(ctdb, vnn->ifaces[i]);
1125 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1126 "for public_address[%s]\n",
1127 vnn->ifaces[i], ctdb_addr_to_str(addr)));
1133 DLIST_ADD(ctdb->vnn, vnn);
1139 setup the event script directory
1141 int ctdb_set_event_script_dir(struct ctdb_context *ctdb, const char *script_dir)
1143 ctdb->event_script_dir = talloc_strdup(ctdb, script_dir);
1144 CTDB_NO_MEMORY(ctdb, ctdb->event_script_dir);
1148 static void ctdb_check_interfaces_event(struct event_context *ev, struct timed_event *te,
1149 struct timeval t, void *private_data)
1151 struct ctdb_context *ctdb = talloc_get_type(private_data,
1152 struct ctdb_context);
1153 struct ctdb_vnn *vnn;
1155 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1158 for (i=0; vnn->ifaces[i] != NULL; i++) {
1159 if (!ctdb_sys_check_iface_exists(vnn->ifaces[i])) {
1160 DEBUG(DEBUG_CRIT,("Interface %s does not exist but is used by public ip %s\n",
1162 ctdb_addr_to_str(&vnn->public_address)));
1167 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1168 timeval_current_ofs(30, 0),
1169 ctdb_check_interfaces_event, ctdb);
1173 int ctdb_start_monitoring_interfaces(struct ctdb_context *ctdb)
1175 if (ctdb->check_public_ifaces_ctx != NULL) {
1176 talloc_free(ctdb->check_public_ifaces_ctx);
1177 ctdb->check_public_ifaces_ctx = NULL;
1180 ctdb->check_public_ifaces_ctx = talloc_new(ctdb);
1181 if (ctdb->check_public_ifaces_ctx == NULL) {
1182 ctdb_fatal(ctdb, "failed to allocate context for checking interfaces");
1185 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1186 timeval_current_ofs(30, 0),
1187 ctdb_check_interfaces_event, ctdb);
1194 setup the public address lists from a file
1196 int ctdb_set_public_addresses(struct ctdb_context *ctdb, bool check_addresses)
1202 lines = file_lines_load(ctdb->public_addresses_file, &nlines, ctdb);
1203 if (lines == NULL) {
1204 ctdb_set_error(ctdb, "Failed to load public address list '%s'\n", ctdb->public_addresses_file);
1207 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
1211 for (i=0;i<nlines;i++) {
1213 ctdb_sock_addr addr;
1214 const char *addrstr;
1219 while ((*line == ' ') || (*line == '\t')) {
1225 if (strcmp(line, "") == 0) {
1228 tok = strtok(line, " \t");
1230 tok = strtok(NULL, " \t");
1232 if (NULL == ctdb->default_public_interface) {
1233 DEBUG(DEBUG_CRIT,("No default public interface and no interface specified at line %u of public address list\n",
1238 ifaces = ctdb->default_public_interface;
1243 if (!addrstr || !parse_ip_mask(addrstr, ifaces, &addr, &mask)) {
1244 DEBUG(DEBUG_CRIT,("Badly formed line %u in public address list\n", i+1));
1248 if (ctdb_add_public_address(ctdb, &addr, mask, ifaces, check_addresses)) {
1249 DEBUG(DEBUG_CRIT,("Failed to add line %u to the public address list\n", i+1));
1260 int ctdb_set_single_public_ip(struct ctdb_context *ctdb,
1264 struct ctdb_vnn *svnn;
1265 struct ctdb_iface *cur = NULL;
1269 svnn = talloc_zero(ctdb, struct ctdb_vnn);
1270 CTDB_NO_MEMORY(ctdb, svnn);
1272 svnn->ifaces = talloc_array(svnn, const char *, 2);
1273 CTDB_NO_MEMORY(ctdb, svnn->ifaces);
1274 svnn->ifaces[0] = talloc_strdup(svnn->ifaces, iface);
1275 CTDB_NO_MEMORY(ctdb, svnn->ifaces[0]);
1276 svnn->ifaces[1] = NULL;
1278 ok = parse_ip(ip, iface, 0, &svnn->public_address);
1284 ret = ctdb_add_local_iface(ctdb, svnn->ifaces[0]);
1286 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1287 "for single_ip[%s]\n",
1289 ctdb_addr_to_str(&svnn->public_address)));
1294 /* assume the single public ip interface is initially "good" */
1295 cur = ctdb_find_iface(ctdb, iface);
1297 DEBUG(DEBUG_CRIT,("Can not find public interface %s used by --single-public-ip", iface));
1300 cur->link_up = true;
1302 ret = ctdb_vnn_assign_iface(ctdb, svnn);
1308 ctdb->single_ip_vnn = svnn;
1312 struct ctdb_public_ip_list {
1313 struct ctdb_public_ip_list *next;
1315 ctdb_sock_addr addr;
1318 /* Given a physical node, return the number of
1319 public addresses that is currently assigned to this node.
1321 static int node_ip_coverage(struct ctdb_context *ctdb,
1323 struct ctdb_public_ip_list *ips)
1327 for (;ips;ips=ips->next) {
1328 if (ips->pnn == pnn) {
1336 /* Can the given node host the given IP: is the public IP known to the
1337 * node and is NOIPHOST unset?
1339 static bool can_node_host_ip(struct ctdb_context *ctdb, int32_t pnn,
1340 struct ctdb_ipflags ipflags,
1341 struct ctdb_public_ip_list *ip)
1343 struct ctdb_all_public_ips *public_ips;
1346 if (ipflags.noiphost) {
1350 public_ips = ctdb->nodes[pnn]->available_public_ips;
1352 if (public_ips == NULL) {
1356 for (i=0; i<public_ips->num; i++) {
1357 if (ctdb_same_ip(&ip->addr, &public_ips->ips[i].addr)) {
1358 /* yes, this node can serve this public ip */
1366 static bool can_node_takeover_ip(struct ctdb_context *ctdb, int32_t pnn,
1367 struct ctdb_ipflags ipflags,
1368 struct ctdb_public_ip_list *ip)
1370 if (ipflags.noiptakeover) {
1374 return can_node_host_ip(ctdb, pnn, ipflags, ip);
1377 /* search the node lists list for a node to takeover this ip.
1378 pick the node that currently are serving the least number of ips
1379 so that the ips get spread out evenly.
1381 static int find_takeover_node(struct ctdb_context *ctdb,
1382 struct ctdb_ipflags *ipflags,
1383 struct ctdb_public_ip_list *ip,
1384 struct ctdb_public_ip_list *all_ips)
1386 int pnn, min=0, num;
1389 numnodes = talloc_array_length(ipflags);
1391 for (i=0; i<numnodes; i++) {
1392 /* verify that this node can serve this ip */
1393 if (!can_node_takeover_ip(ctdb, i, ipflags[i], ip)) {
1394 /* no it couldnt so skip to the next node */
1398 num = node_ip_coverage(ctdb, i, all_ips);
1399 /* was this the first node we checked ? */
1411 DEBUG(DEBUG_WARNING,(__location__ " Could not find node to take over public address '%s'\n",
1412 ctdb_addr_to_str(&ip->addr)));
1422 static uint32_t *ip_key(ctdb_sock_addr *ip)
1424 static uint32_t key[IP_KEYLEN];
1426 bzero(key, sizeof(key));
1428 switch (ip->sa.sa_family) {
1430 key[3] = htonl(ip->ip.sin_addr.s_addr);
1433 uint32_t *s6_a32 = (uint32_t *)&(ip->ip6.sin6_addr.s6_addr);
1434 key[0] = htonl(s6_a32[0]);
1435 key[1] = htonl(s6_a32[1]);
1436 key[2] = htonl(s6_a32[2]);
1437 key[3] = htonl(s6_a32[3]);
1441 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", ip->sa.sa_family));
1448 static void *add_ip_callback(void *parm, void *data)
1450 struct ctdb_public_ip_list *this_ip = parm;
1451 struct ctdb_public_ip_list *prev_ip = data;
1453 if (prev_ip == NULL) {
1456 if (this_ip->pnn == -1) {
1457 this_ip->pnn = prev_ip->pnn;
1463 static int getips_count_callback(void *param, void *data)
1465 struct ctdb_public_ip_list **ip_list = (struct ctdb_public_ip_list **)param;
1466 struct ctdb_public_ip_list *new_ip = (struct ctdb_public_ip_list *)data;
1468 new_ip->next = *ip_list;
1473 static struct ctdb_public_ip_list *
1474 create_merged_ip_list(struct ctdb_context *ctdb)
1477 struct ctdb_public_ip_list *ip_list;
1478 struct ctdb_all_public_ips *public_ips;
1480 if (ctdb->ip_tree != NULL) {
1481 talloc_free(ctdb->ip_tree);
1482 ctdb->ip_tree = NULL;
1484 ctdb->ip_tree = trbt_create(ctdb, 0);
1486 for (i=0;i<ctdb->num_nodes;i++) {
1487 public_ips = ctdb->nodes[i]->known_public_ips;
1489 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
1493 /* there were no public ips for this node */
1494 if (public_ips == NULL) {
1498 for (j=0;j<public_ips->num;j++) {
1499 struct ctdb_public_ip_list *tmp_ip;
1501 tmp_ip = talloc_zero(ctdb->ip_tree, struct ctdb_public_ip_list);
1502 CTDB_NO_MEMORY_NULL(ctdb, tmp_ip);
1503 /* Do not use information about IP addresses hosted
1504 * on other nodes, it may not be accurate */
1505 if (public_ips->ips[j].pnn == ctdb->nodes[i]->pnn) {
1506 tmp_ip->pnn = public_ips->ips[j].pnn;
1510 tmp_ip->addr = public_ips->ips[j].addr;
1511 tmp_ip->next = NULL;
1513 trbt_insertarray32_callback(ctdb->ip_tree,
1514 IP_KEYLEN, ip_key(&public_ips->ips[j].addr),
1521 trbt_traversearray32(ctdb->ip_tree, IP_KEYLEN, getips_count_callback, &ip_list);
1527 * This is the length of the longtest common prefix between the IPs.
1528 * It is calculated by XOR-ing the 2 IPs together and counting the
1529 * number of leading zeroes. The implementation means that all
1530 * addresses end up being 128 bits long.
1532 * FIXME? Should we consider IPv4 and IPv6 separately given that the
1533 * 12 bytes of 0 prefix padding will hurt the algorithm if there are
1534 * lots of nodes and IP addresses?
1536 static uint32_t ip_distance(ctdb_sock_addr *ip1, ctdb_sock_addr *ip2)
1538 uint32_t ip1_k[IP_KEYLEN];
1543 uint32_t distance = 0;
1545 memcpy(ip1_k, ip_key(ip1), sizeof(ip1_k));
1547 for (i=0; i<IP_KEYLEN; i++) {
1548 x = ip1_k[i] ^ t[i];
1552 /* Count number of leading zeroes.
1553 * FIXME? This could be optimised...
1555 while ((x & (1 << 31)) == 0) {
1565 /* Calculate the IP distance for the given IP relative to IPs on the
1566 given node. The ips argument is generally the all_ips variable
1567 used in the main part of the algorithm.
1569 static uint32_t ip_distance_2_sum(ctdb_sock_addr *ip,
1570 struct ctdb_public_ip_list *ips,
1573 struct ctdb_public_ip_list *t;
1578 for (t=ips; t != NULL; t=t->next) {
1579 if (t->pnn != pnn) {
1583 /* Optimisation: We never calculate the distance
1584 * between an address and itself. This allows us to
1585 * calculate the effect of removing an address from a
1586 * node by simply calculating the distance between
1587 * that address and all of the exitsing addresses.
1588 * Moreover, we assume that we're only ever dealing
1589 * with addresses from all_ips so we can identify an
1590 * address via a pointer rather than doing a more
1591 * expensive address comparison. */
1592 if (&(t->addr) == ip) {
1596 d = ip_distance(ip, &(t->addr));
1597 sum += d * d; /* Cheaper than pulling in math.h :-) */
1603 /* Return the LCP2 imbalance metric for addresses currently assigned
1606 static uint32_t lcp2_imbalance(struct ctdb_public_ip_list * all_ips, int pnn)
1608 struct ctdb_public_ip_list *t;
1610 uint32_t imbalance = 0;
1612 for (t=all_ips; t!=NULL; t=t->next) {
1613 if (t->pnn != pnn) {
1616 /* Pass the rest of the IPs rather than the whole
1619 imbalance += ip_distance_2_sum(&(t->addr), t->next, pnn);
1625 /* Allocate any unassigned IPs just by looping through the IPs and
1626 * finding the best node for each.
1628 static void basic_allocate_unassigned(struct ctdb_context *ctdb,
1629 struct ctdb_ipflags *ipflags,
1630 struct ctdb_public_ip_list *all_ips)
1632 struct ctdb_public_ip_list *tmp_ip;
1634 /* loop over all ip's and find a physical node to cover for
1637 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1638 if (tmp_ip->pnn == -1) {
1639 if (find_takeover_node(ctdb, ipflags, tmp_ip, all_ips)) {
1640 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1641 ctdb_addr_to_str(&tmp_ip->addr)));
1647 /* Basic non-deterministic rebalancing algorithm.
1649 static void basic_failback(struct ctdb_context *ctdb,
1650 struct ctdb_ipflags *ipflags,
1651 struct ctdb_public_ip_list *all_ips,
1655 int maxnode, maxnum, minnode, minnum, num, retries;
1656 struct ctdb_public_ip_list *tmp_ip;
1658 numnodes = talloc_array_length(ipflags);
1665 /* for each ip address, loop over all nodes that can serve
1666 this ip and make sure that the difference between the node
1667 serving the most and the node serving the least ip's are
1670 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1671 if (tmp_ip->pnn == -1) {
1675 /* Get the highest and lowest number of ips's served by any
1676 valid node which can serve this ip.
1680 for (i=0; i<numnodes; i++) {
1681 /* only check nodes that can actually serve this ip */
1682 if (!can_node_takeover_ip(ctdb, i, ipflags[i], tmp_ip)) {
1683 /* no it couldnt so skip to the next node */
1687 num = node_ip_coverage(ctdb, i, all_ips);
1688 if (maxnode == -1) {
1697 if (minnode == -1) {
1707 if (maxnode == -1) {
1708 DEBUG(DEBUG_WARNING,(__location__ " Could not find maxnode. May not be able to serve ip '%s'\n",
1709 ctdb_addr_to_str(&tmp_ip->addr)));
1714 /* if the spread between the smallest and largest coverage by
1715 a node is >=2 we steal one of the ips from the node with
1716 most coverage to even things out a bit.
1717 try to do this a limited number of times since we dont
1718 want to spend too much time balancing the ip coverage.
1720 if ( (maxnum > minnum+1)
1721 && (retries < (num_ips + 5)) ){
1722 struct ctdb_public_ip_list *tmp;
1724 /* Reassign one of maxnode's VNNs */
1725 for (tmp=all_ips;tmp;tmp=tmp->next) {
1726 if (tmp->pnn == maxnode) {
1727 (void)find_takeover_node(ctdb, ipflags, tmp, all_ips);
1736 struct ctdb_rebalancenodes {
1737 struct ctdb_rebalancenodes *next;
1740 static struct ctdb_rebalancenodes *force_rebalance_list = NULL;
1743 /* set this flag to force the node to be rebalanced even if it just didnt
1744 become healthy again.
1746 void lcp2_forcerebalance(struct ctdb_context *ctdb, uint32_t pnn)
1748 struct ctdb_rebalancenodes *rebalance;
1750 for (rebalance = force_rebalance_list; rebalance; rebalance = rebalance->next) {
1751 if (rebalance->pnn == pnn) {
1756 rebalance = talloc(ctdb, struct ctdb_rebalancenodes);
1757 rebalance->pnn = pnn;
1758 rebalance->next = force_rebalance_list;
1759 force_rebalance_list = rebalance;
1762 /* Do necessary LCP2 initialisation. Bury it in a function here so
1763 * that we can unit test it.
1765 static void lcp2_init(struct ctdb_context *tmp_ctx,
1766 struct ctdb_ipflags *ipflags,
1767 struct ctdb_public_ip_list *all_ips,
1768 uint32_t **lcp2_imbalances,
1769 bool **rebalance_candidates)
1772 struct ctdb_public_ip_list *tmp_ip;
1774 numnodes = talloc_array_length(ipflags);
1776 *rebalance_candidates = talloc_array(tmp_ctx, bool, numnodes);
1777 CTDB_NO_MEMORY_FATAL(tmp_ctx, *rebalance_candidates);
1778 *lcp2_imbalances = talloc_array(tmp_ctx, uint32_t, numnodes);
1779 CTDB_NO_MEMORY_FATAL(tmp_ctx, *lcp2_imbalances);
1781 for (i=0; i<numnodes; i++) {
1782 (*lcp2_imbalances)[i] = lcp2_imbalance(all_ips, i);
1783 /* First step: assume all nodes are candidates */
1784 (*rebalance_candidates)[i] = true;
1787 /* 2nd step: if a node has IPs assigned then it must have been
1788 * healthy before, so we remove it from consideration. This
1789 * is overkill but is all we have because we don't maintain
1790 * state between takeover runs. An alternative would be to
1791 * keep state and invalidate it every time the recovery master
1794 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1795 if (tmp_ip->pnn != -1) {
1796 (*rebalance_candidates)[tmp_ip->pnn] = false;
1800 /* 3rd step: if a node is forced to re-balance then
1801 we allow failback onto the node */
1802 while (force_rebalance_list != NULL) {
1803 struct ctdb_rebalancenodes *next = force_rebalance_list->next;
1805 if (force_rebalance_list->pnn <= numnodes) {
1806 (*rebalance_candidates)[force_rebalance_list->pnn] = true;
1809 DEBUG(DEBUG_ERR,("During ipreallocation, forced rebalance of node %d\n", force_rebalance_list->pnn));
1810 talloc_free(force_rebalance_list);
1811 force_rebalance_list = next;
1815 /* Allocate any unassigned addresses using the LCP2 algorithm to find
1816 * the IP/node combination that will cost the least.
1818 static void lcp2_allocate_unassigned(struct ctdb_context *ctdb,
1819 struct ctdb_ipflags *ipflags,
1820 struct ctdb_public_ip_list *all_ips,
1821 uint32_t *lcp2_imbalances)
1823 struct ctdb_public_ip_list *tmp_ip;
1824 int dstnode, numnodes;
1827 uint32_t mindsum, dstdsum, dstimbl, minimbl;
1828 struct ctdb_public_ip_list *minip;
1830 bool should_loop = true;
1831 bool have_unassigned = true;
1833 numnodes = talloc_array_length(ipflags);
1835 while (have_unassigned && should_loop) {
1836 should_loop = false;
1838 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1839 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES (UNASSIGNED)\n"));
1845 /* loop over each unassigned ip. */
1846 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1847 if (tmp_ip->pnn != -1) {
1851 for (dstnode=0; dstnode<numnodes; dstnode++) {
1852 /* only check nodes that can actually takeover this ip */
1853 if (!can_node_takeover_ip(ctdb, dstnode,
1856 /* no it couldnt so skip to the next node */
1860 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1861 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1862 DEBUG(DEBUG_DEBUG,(" %s -> %d [+%d]\n",
1863 ctdb_addr_to_str(&(tmp_ip->addr)),
1865 dstimbl - lcp2_imbalances[dstnode]));
1868 if ((minnode == -1) || (dstdsum < mindsum)) {
1878 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1880 /* If we found one then assign it to the given node. */
1881 if (minnode != -1) {
1882 minip->pnn = minnode;
1883 lcp2_imbalances[minnode] = minimbl;
1884 DEBUG(DEBUG_INFO,(" %s -> %d [+%d]\n",
1885 ctdb_addr_to_str(&(minip->addr)),
1890 /* There might be a better way but at least this is clear. */
1891 have_unassigned = false;
1892 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1893 if (tmp_ip->pnn == -1) {
1894 have_unassigned = true;
1899 /* We know if we have an unassigned addresses so we might as
1902 if (have_unassigned) {
1903 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1904 if (tmp_ip->pnn == -1) {
1905 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1906 ctdb_addr_to_str(&tmp_ip->addr)));
1912 /* LCP2 algorithm for rebalancing the cluster. Given a candidate node
1913 * to move IPs from, determines the best IP/destination node
1914 * combination to move from the source node.
1916 static bool lcp2_failback_candidate(struct ctdb_context *ctdb,
1917 struct ctdb_ipflags *ipflags,
1918 struct ctdb_public_ip_list *all_ips,
1921 uint32_t *lcp2_imbalances,
1922 bool *rebalance_candidates)
1924 int dstnode, mindstnode, numnodes;
1925 uint32_t srcimbl, srcdsum, dstimbl, dstdsum;
1926 uint32_t minsrcimbl, mindstimbl;
1927 struct ctdb_public_ip_list *minip;
1928 struct ctdb_public_ip_list *tmp_ip;
1930 /* Find an IP and destination node that best reduces imbalance. */
1937 numnodes = talloc_array_length(ipflags);
1939 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1940 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES FROM %d [%d]\n", srcnode, candimbl));
1942 for (tmp_ip=all_ips; tmp_ip; tmp_ip=tmp_ip->next) {
1943 /* Only consider addresses on srcnode. */
1944 if (tmp_ip->pnn != srcnode) {
1948 /* What is this IP address costing the source node? */
1949 srcdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, srcnode);
1950 srcimbl = candimbl - srcdsum;
1952 /* Consider this IP address would cost each potential
1953 * destination node. Destination nodes are limited to
1954 * those that are newly healthy, since we don't want
1955 * to do gratuitous failover of IPs just to make minor
1956 * balance improvements.
1958 for (dstnode=0; dstnode<numnodes; dstnode++) {
1959 if (!rebalance_candidates[dstnode]) {
1963 /* only check nodes that can actually takeover this ip */
1964 if (!can_node_takeover_ip(ctdb, dstnode,
1965 ipflags[dstnode], tmp_ip)) {
1966 /* no it couldnt so skip to the next node */
1970 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1971 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1972 DEBUG(DEBUG_DEBUG,(" %d [%d] -> %s -> %d [+%d]\n",
1973 srcnode, srcimbl - lcp2_imbalances[srcnode],
1974 ctdb_addr_to_str(&(tmp_ip->addr)),
1975 dstnode, dstimbl - lcp2_imbalances[dstnode]));
1977 if ((dstimbl < candimbl) && (dstdsum < srcdsum) && \
1978 ((mindstnode == -1) || \
1979 ((srcimbl + dstimbl) < (minsrcimbl + mindstimbl)))) {
1982 minsrcimbl = srcimbl;
1983 mindstnode = dstnode;
1984 mindstimbl = dstimbl;
1988 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1990 if (mindstnode != -1) {
1991 /* We found a move that makes things better... */
1992 DEBUG(DEBUG_INFO,("%d [%d] -> %s -> %d [+%d]\n",
1993 srcnode, minsrcimbl - lcp2_imbalances[srcnode],
1994 ctdb_addr_to_str(&(minip->addr)),
1995 mindstnode, mindstimbl - lcp2_imbalances[mindstnode]));
1998 lcp2_imbalances[srcnode] = srcimbl;
1999 lcp2_imbalances[mindstnode] = mindstimbl;
2000 minip->pnn = mindstnode;
2009 struct lcp2_imbalance_pnn {
2014 static int lcp2_cmp_imbalance_pnn(const void * a, const void * b)
2016 const struct lcp2_imbalance_pnn * lipa = (const struct lcp2_imbalance_pnn *) a;
2017 const struct lcp2_imbalance_pnn * lipb = (const struct lcp2_imbalance_pnn *) b;
2019 if (lipa->imbalance > lipb->imbalance) {
2021 } else if (lipa->imbalance == lipb->imbalance) {
2028 /* LCP2 algorithm for rebalancing the cluster. This finds the source
2029 * node with the highest LCP2 imbalance, and then determines the best
2030 * IP/destination node combination to move from the source node.
2032 static void lcp2_failback(struct ctdb_context *ctdb,
2033 struct ctdb_ipflags *ipflags,
2034 struct ctdb_public_ip_list *all_ips,
2035 uint32_t *lcp2_imbalances,
2036 bool *rebalance_candidates)
2038 int i, num_rebalance_candidates, numnodes;
2039 struct lcp2_imbalance_pnn * lips;
2042 numnodes = talloc_array_length(ipflags);
2046 /* It is only worth continuing if we have suitable target
2047 * nodes to transfer IPs to. This check is much cheaper than
2050 num_rebalance_candidates = 0;
2051 for (i=0; i<numnodes; i++) {
2052 if (rebalance_candidates[i]) {
2053 num_rebalance_candidates++;
2056 if (num_rebalance_candidates == 0) {
2060 /* Put the imbalances and nodes into an array, sort them and
2061 * iterate through candidates. Usually the 1st one will be
2062 * used, so this doesn't cost much...
2064 lips = talloc_array(ctdb, struct lcp2_imbalance_pnn, numnodes);
2065 for (i=0; i<numnodes; i++) {
2066 lips[i].imbalance = lcp2_imbalances[i];
2069 qsort(lips, numnodes, sizeof(struct lcp2_imbalance_pnn),
2070 lcp2_cmp_imbalance_pnn);
2073 for (i=0; i<numnodes; i++) {
2074 /* This means that all nodes had 0 or 1 addresses, so
2075 * can't be imbalanced.
2077 if (lips[i].imbalance == 0) {
2081 if (lcp2_failback_candidate(ctdb,
2087 rebalance_candidates)) {
2099 static void unassign_unsuitable_ips(struct ctdb_context *ctdb,
2100 struct ctdb_ipflags *ipflags,
2101 struct ctdb_public_ip_list *all_ips)
2103 struct ctdb_public_ip_list *tmp_ip;
2105 /* verify that the assigned nodes can serve that public ip
2106 and set it to -1 if not
2108 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2109 if (tmp_ip->pnn == -1) {
2112 if (!can_node_host_ip(ctdb, tmp_ip->pnn,
2113 ipflags[tmp_ip->pnn], tmp_ip) != 0) {
2114 /* this node can not serve this ip. */
2115 DEBUG(DEBUG_DEBUG,("Unassign IP: %s from %d\n",
2116 ctdb_addr_to_str(&(tmp_ip->addr)),
2123 static void ip_alloc_deterministic_ips(struct ctdb_context *ctdb,
2124 struct ctdb_ipflags *ipflags,
2125 struct ctdb_public_ip_list *all_ips)
2127 struct ctdb_public_ip_list *tmp_ip;
2130 numnodes = talloc_array_length(ipflags);
2132 DEBUG(DEBUG_NOTICE,("Deterministic IPs enabled. Resetting all ip allocations\n"));
2133 /* Allocate IPs to nodes in a modulo fashion so that IPs will
2134 * always be allocated the same way for a specific set of
2135 * available/unavailable nodes.
2138 for (i=0,tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next,i++) {
2139 tmp_ip->pnn = i % numnodes;
2142 /* IP failback doesn't make sense with deterministic
2143 * IPs, since the modulo step above implicitly fails
2144 * back IPs to their "home" node.
2146 if (1 == ctdb->tunable.no_ip_failback) {
2147 DEBUG(DEBUG_WARNING, ("WARNING: 'NoIPFailback' set but ignored - incompatible with 'DeterministicIPs\n"));
2150 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2152 basic_allocate_unassigned(ctdb, ipflags, all_ips);
2154 /* No failback here! */
2157 static void ip_alloc_nondeterministic_ips(struct ctdb_context *ctdb,
2158 struct ctdb_ipflags *ipflags,
2159 struct ctdb_public_ip_list *all_ips)
2161 /* This should be pushed down into basic_failback. */
2162 struct ctdb_public_ip_list *tmp_ip;
2164 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2168 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2170 basic_allocate_unassigned(ctdb, ipflags, all_ips);
2172 /* If we don't want IPs to fail back then don't rebalance IPs. */
2173 if (1 == ctdb->tunable.no_ip_failback) {
2177 /* Now, try to make sure the ip adresses are evenly distributed
2180 basic_failback(ctdb, ipflags, all_ips, num_ips);
2183 static void ip_alloc_lcp2(struct ctdb_context *ctdb,
2184 struct ctdb_ipflags *ipflags,
2185 struct ctdb_public_ip_list *all_ips)
2187 uint32_t *lcp2_imbalances;
2188 bool *rebalance_candidates;
2190 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2192 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2194 lcp2_init(tmp_ctx, ipflags, all_ips,
2195 &lcp2_imbalances, &rebalance_candidates);
2197 lcp2_allocate_unassigned(ctdb, ipflags, all_ips, lcp2_imbalances);
2199 /* If we don't want IPs to fail back then don't rebalance IPs. */
2200 if (1 == ctdb->tunable.no_ip_failback) {
2204 /* Now, try to make sure the ip adresses are evenly distributed
2207 lcp2_failback(ctdb, ipflags, all_ips,
2208 lcp2_imbalances, rebalance_candidates);
2211 talloc_free(tmp_ctx);
2214 static bool all_nodes_are_disabled(struct ctdb_node_map *nodemap)
2218 /* Count how many completely healthy nodes we have */
2220 for (i=0;i<nodemap->num;i++) {
2221 if (!(nodemap->nodes[i].flags & (NODE_FLAGS_INACTIVE|NODE_FLAGS_DISABLED))) {
2226 return num_healthy == 0;
2229 /* The calculation part of the IP allocation algorithm. */
2230 static void ctdb_takeover_run_core(struct ctdb_context *ctdb,
2231 struct ctdb_ipflags *ipflags,
2232 struct ctdb_public_ip_list **all_ips_p)
2234 /* since nodes only know about those public addresses that
2235 can be served by that particular node, no single node has
2236 a full list of all public addresses that exist in the cluster.
2237 Walk over all node structures and create a merged list of
2238 all public addresses that exist in the cluster.
2240 keep the tree of ips around as ctdb->ip_tree
2242 *all_ips_p = create_merged_ip_list(ctdb);
2244 if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
2245 ip_alloc_lcp2(ctdb, ipflags, *all_ips_p);
2246 } else if (1 == ctdb->tunable.deterministic_public_ips) {
2247 ip_alloc_deterministic_ips(ctdb, ipflags, *all_ips_p);
2249 ip_alloc_nondeterministic_ips(ctdb, ipflags, *all_ips_p);
2252 /* at this point ->pnn is the node which will own each IP
2253 or -1 if there is no node that can cover this ip
2259 struct get_tunable_callback_data {
2260 const char *tunable;
2265 static void get_tunable_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;
2274 /* Already handled in fail callback */
2278 if (outdata.dsize != sizeof(uint32_t)) {
2279 DEBUG(DEBUG_ERR,("Wrong size of returned data when reading \"%s\" tunable from node %d. Expected %d bytes but received %d bytes\n",
2280 cd->tunable, pnn, (int)sizeof(uint32_t),
2281 (int)outdata.dsize));
2286 size = talloc_array_length(cd->out);
2288 DEBUG(DEBUG_ERR,("Got %s reply from node %d but nodemap only has %d entries\n",
2289 cd->tunable, pnn, size));
2294 cd->out[pnn] = *(uint32_t *)outdata.dptr;
2297 static void get_tunable_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2298 int32_t res, TDB_DATA outdata,
2301 struct get_tunable_callback_data *cd =
2302 (struct get_tunable_callback_data *)callback;
2307 ("Timed out getting tunable \"%s\" from node %d\n",
2313 DEBUG(DEBUG_WARNING,
2314 ("Tunable \"%s\" not implemented on node %d\n",
2319 ("Unexpected error getting tunable \"%s\" from node %d\n",
2325 static uint32_t *get_tunable_from_nodes(struct ctdb_context *ctdb,
2326 TALLOC_CTX *tmp_ctx,
2327 struct ctdb_node_map *nodemap,
2328 const char *tunable,
2329 uint32_t default_value)
2332 struct ctdb_control_get_tunable *t;
2335 struct get_tunable_callback_data callback_data;
2338 tvals = talloc_array(tmp_ctx, uint32_t, nodemap->num);
2339 CTDB_NO_MEMORY_NULL(ctdb, tvals);
2340 for (i=0; i<nodemap->num; i++) {
2341 tvals[i] = default_value;
2344 callback_data.out = tvals;
2345 callback_data.tunable = tunable;
2346 callback_data.fatal = false;
2348 data.dsize = offsetof(struct ctdb_control_get_tunable, name) + strlen(tunable) + 1;
2349 data.dptr = talloc_size(tmp_ctx, data.dsize);
2350 t = (struct ctdb_control_get_tunable *)data.dptr;
2351 t->length = strlen(tunable)+1;
2352 memcpy(t->name, tunable, t->length);
2353 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2354 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_TUNABLE,
2355 nodes, 0, TAKEOVER_TIMEOUT(),
2357 get_tunable_callback,
2358 get_tunable_fail_callback,
2359 &callback_data) != 0) {
2360 if (callback_data.fatal) {
2366 talloc_free(data.dptr);
2371 struct get_runstate_callback_data {
2372 enum ctdb_runstate *out;
2376 static void get_runstate_callback(struct ctdb_context *ctdb, uint32_t pnn,
2377 int32_t res, TDB_DATA outdata,
2378 void *callback_data)
2380 struct get_runstate_callback_data *cd =
2381 (struct get_runstate_callback_data *)callback_data;
2385 /* Already handled in fail callback */
2389 if (outdata.dsize != sizeof(uint32_t)) {
2390 DEBUG(DEBUG_ERR,("Wrong size of returned data when getting runstate from node %d. Expected %d bytes but received %d bytes\n",
2391 pnn, (int)sizeof(uint32_t),
2392 (int)outdata.dsize));
2397 size = talloc_array_length(cd->out);
2399 DEBUG(DEBUG_ERR,("Got reply from node %d but nodemap only has %d entries\n",
2404 cd->out[pnn] = (enum ctdb_runstate)*(uint32_t *)outdata.dptr;
2407 static void get_runstate_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2408 int32_t res, TDB_DATA outdata,
2411 struct get_runstate_callback_data *cd =
2412 (struct get_runstate_callback_data *)callback;
2417 ("Timed out getting runstate from node %d\n", pnn));
2421 DEBUG(DEBUG_WARNING,
2422 ("Error getting runstate from node %d - assuming runstates not supported\n",
2427 static enum ctdb_runstate * get_runstate_from_nodes(struct ctdb_context *ctdb,
2428 TALLOC_CTX *tmp_ctx,
2429 struct ctdb_node_map *nodemap,
2430 enum ctdb_runstate default_value)
2433 enum ctdb_runstate *rs;
2434 struct get_runstate_callback_data callback_data;
2437 rs = talloc_array(tmp_ctx, enum ctdb_runstate, nodemap->num);
2438 CTDB_NO_MEMORY_NULL(ctdb, rs);
2439 for (i=0; i<nodemap->num; i++) {
2440 rs[i] = default_value;
2443 callback_data.out = rs;
2444 callback_data.fatal = false;
2446 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2447 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_RUNSTATE,
2448 nodes, 0, TAKEOVER_TIMEOUT(),
2450 get_runstate_callback,
2451 get_runstate_fail_callback,
2452 &callback_data) != 0) {
2453 if (callback_data.fatal) {
2463 /* Set internal flags for IP allocation:
2465 * Set NOIPTAKOVER ip flags from per-node NoIPTakeover tunable
2466 * Set NOIPHOST ip flag for each INACTIVE node
2467 * if all nodes are disabled:
2468 * Set NOIPHOST ip flags from per-node NoIPHostOnAllDisabled tunable
2470 * Set NOIPHOST ip flags for disabled nodes
2472 static struct ctdb_ipflags *
2473 set_ipflags_internal(struct ctdb_context *ctdb,
2474 TALLOC_CTX *tmp_ctx,
2475 struct ctdb_node_map *nodemap,
2476 uint32_t *tval_noiptakeover,
2477 uint32_t *tval_noiphostonalldisabled,
2478 enum ctdb_runstate *runstate)
2481 struct ctdb_ipflags *ipflags;
2483 /* Clear IP flags - implicit due to talloc_zero */
2484 ipflags = talloc_zero_array(tmp_ctx, struct ctdb_ipflags, nodemap->num);
2485 CTDB_NO_MEMORY_NULL(ctdb, ipflags);
2487 for (i=0;i<nodemap->num;i++) {
2488 /* Can not take IPs on node with NoIPTakeover set */
2489 if (tval_noiptakeover[i] != 0) {
2490 ipflags[i].noiptakeover = true;
2493 /* Can not host IPs on node not in RUNNING state */
2494 if (runstate[i] != CTDB_RUNSTATE_RUNNING) {
2495 ipflags[i].noiphost = true;
2498 /* Can not host IPs on INACTIVE node */
2499 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
2500 ipflags[i].noiphost = true;
2504 if (all_nodes_are_disabled(nodemap)) {
2505 /* If all nodes are disabled, can not host IPs on node
2506 * with NoIPHostOnAllDisabled set
2508 for (i=0;i<nodemap->num;i++) {
2509 if (tval_noiphostonalldisabled[i] != 0) {
2510 ipflags[i].noiphost = true;
2514 /* If some nodes are not disabled, then can not host
2515 * IPs on DISABLED node
2517 for (i=0;i<nodemap->num;i++) {
2518 if (nodemap->nodes[i].flags & NODE_FLAGS_DISABLED) {
2519 ipflags[i].noiphost = true;
2527 static struct ctdb_ipflags *set_ipflags(struct ctdb_context *ctdb,
2528 TALLOC_CTX *tmp_ctx,
2529 struct ctdb_node_map *nodemap)
2531 uint32_t *tval_noiptakeover;
2532 uint32_t *tval_noiphostonalldisabled;
2533 struct ctdb_ipflags *ipflags;
2534 enum ctdb_runstate *runstate;
2537 tval_noiptakeover = get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2539 if (tval_noiptakeover == NULL) {
2543 tval_noiphostonalldisabled =
2544 get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2545 "NoIPHostOnAllDisabled", 0);
2546 if (tval_noiphostonalldisabled == NULL) {
2547 /* Caller frees tmp_ctx */
2551 /* Any nodes where CTDB_CONTROL_GET_RUNSTATE is not supported
2552 * will default to CTDB_RUNSTATE_RUNNING. This ensures
2553 * reasonable behaviour on a mixed cluster during upgrade.
2555 runstate = get_runstate_from_nodes(ctdb, tmp_ctx, nodemap,
2556 CTDB_RUNSTATE_RUNNING);
2557 if (runstate == NULL) {
2558 /* Caller frees tmp_ctx */
2562 ipflags = set_ipflags_internal(ctdb, tmp_ctx, nodemap,
2564 tval_noiphostonalldisabled,
2567 talloc_free(tval_noiptakeover);
2568 talloc_free(tval_noiphostonalldisabled);
2569 talloc_free(runstate);
2574 struct iprealloc_callback_data {
2577 client_async_callback fail_callback;
2578 void *fail_callback_data;
2579 struct ctdb_node_map *nodemap;
2582 static void iprealloc_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2583 int32_t res, TDB_DATA outdata,
2587 struct iprealloc_callback_data *cd =
2588 (struct iprealloc_callback_data *)callback;
2592 /* If the control timed out then that's a real error,
2593 * so call the real fail callback
2595 cd->fail_callback(ctdb, pnn, res, outdata,
2596 cd->fail_callback_data);
2599 /* If not a timeout then either the ipreallocated
2600 * eventscript (or some setup) failed. This might
2601 * have failed because the IPREALLOCATED control isn't
2602 * implemented - right now there is no way of knowing
2603 * because the error codes are all folded down to -1.
2604 * Consider retrying using EVENTSCRIPT control...
2607 numnodes = talloc_array_length(cd->retry_nodes);
2608 if (pnn > numnodes) {
2610 ("ipreallocated failure from node %d, but only %d nodes in nodemap\n",
2615 /* Can't run the "ipreallocated" event on a STOPPED node */
2616 if (cd->nodemap->nodes[pnn].flags & NODE_FLAGS_STOPPED) {
2618 ("ipreallocated failure from node %d, but node is stopped - not flagging a retry\n",
2623 DEBUG(DEBUG_WARNING,
2624 ("ipreallocated failure from node %d, flagging retry\n",
2626 cd->retry_nodes[pnn] = true;
2631 struct takeover_callback_data {
2633 client_async_callback fail_callback;
2634 void *fail_callback_data;
2635 struct ctdb_node_map *nodemap;
2638 static void takeover_run_fail_callback(struct ctdb_context *ctdb,
2639 uint32_t node_pnn, int32_t res,
2640 TDB_DATA outdata, void *callback_data)
2642 struct takeover_callback_data *cd =
2643 talloc_get_type_abort(callback_data,
2644 struct takeover_callback_data);
2647 for (i = 0; i < cd->nodemap->num; i++) {
2648 if (node_pnn == cd->nodemap->nodes[i].pnn) {
2653 if (i == cd->nodemap->num) {
2654 DEBUG(DEBUG_ERR, (__location__ " invalid PNN %u\n", node_pnn));
2658 if (!cd->node_failed[i]) {
2659 cd->node_failed[i] = true;
2660 cd->fail_callback(ctdb, node_pnn, res, outdata,
2661 cd->fail_callback_data);
2666 make any IP alias changes for public addresses that are necessary
2668 int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
2669 client_async_callback fail_callback, void *callback_data)
2672 struct ctdb_public_ip ip;
2673 struct ctdb_public_ipv4 ipv4;
2675 struct ctdb_public_ip_list *all_ips, *tmp_ip;
2677 struct timeval timeout;
2678 struct client_async_data *async_data;
2679 struct ctdb_client_control_state *state;
2680 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2681 uint32_t disable_timeout;
2682 struct ctdb_ipflags *ipflags;
2683 struct takeover_callback_data *takeover_data;
2684 struct iprealloc_callback_data iprealloc_data;
2688 * ip failover is completely disabled, just send out the
2689 * ipreallocated event.
2691 if (ctdb->tunable.disable_ip_failover != 0) {
2695 ipflags = set_ipflags(ctdb, tmp_ctx, nodemap);
2696 if (ipflags == NULL) {
2697 DEBUG(DEBUG_ERR,("Failed to set IP flags - aborting takeover run\n"));
2698 talloc_free(tmp_ctx);
2704 /* Do the IP reassignment calculations */
2705 ctdb_takeover_run_core(ctdb, ipflags, &all_ips);
2707 /* The recovery daemon does regular sanity checks of the IPs.
2708 * However, sometimes it is overzealous and thinks changes are
2709 * required when they're already underway. This stops the
2710 * checks for a while before we start moving IPs.
2712 disable_timeout = ctdb->tunable.takeover_timeout;
2713 data.dptr = (uint8_t*)&disable_timeout;
2714 data.dsize = sizeof(disable_timeout);
2715 if (ctdb_client_send_message(ctdb, CTDB_BROADCAST_CONNECTED,
2716 CTDB_SRVID_DISABLE_IP_CHECK, data) != 0) {
2717 DEBUG(DEBUG_INFO,("Failed to disable ip verification\n"));
2720 /* Now tell all nodes to release any public IPs should not
2721 * host. This will be a NOOP on nodes that don't currently
2722 * hold the given IP.
2724 takeover_data = talloc_zero(tmp_ctx, struct takeover_callback_data);
2725 CTDB_NO_MEMORY_FATAL(ctdb, takeover_data);
2727 takeover_data->node_failed = talloc_zero_array(tmp_ctx,
2728 bool, nodemap->num);
2729 CTDB_NO_MEMORY_FATAL(ctdb, takeover_data->node_failed);
2730 takeover_data->fail_callback = fail_callback;
2731 takeover_data->fail_callback_data = callback_data;
2732 takeover_data->nodemap = nodemap;
2734 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2735 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2737 async_data->fail_callback = takeover_run_fail_callback;
2738 async_data->callback_data = takeover_data;
2740 for (i=0;i<nodemap->num;i++) {
2741 /* don't talk to unconnected nodes, but do talk to banned nodes */
2742 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
2746 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2747 if (tmp_ip->pnn == nodemap->nodes[i].pnn) {
2748 /* This node should be serving this
2749 vnn so dont tell it to release the ip
2753 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2754 ipv4.pnn = tmp_ip->pnn;
2755 ipv4.sin = tmp_ip->addr.ip;
2757 timeout = TAKEOVER_TIMEOUT();
2758 data.dsize = sizeof(ipv4);
2759 data.dptr = (uint8_t *)&ipv4;
2760 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2761 0, CTDB_CONTROL_RELEASE_IPv4, 0,
2765 ip.pnn = tmp_ip->pnn;
2766 ip.addr = tmp_ip->addr;
2768 timeout = TAKEOVER_TIMEOUT();
2769 data.dsize = sizeof(ip);
2770 data.dptr = (uint8_t *)&ip;
2771 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2772 0, CTDB_CONTROL_RELEASE_IP, 0,
2777 if (state == NULL) {
2778 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_RELEASE_IP to node %u\n", nodemap->nodes[i].pnn));
2779 talloc_free(tmp_ctx);
2783 ctdb_client_async_add(async_data, state);
2786 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2787 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_RELEASE_IP failed\n"));
2788 talloc_free(tmp_ctx);
2791 talloc_free(async_data);
2794 /* tell all nodes to get their own IPs */
2795 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2796 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2798 async_data->fail_callback = fail_callback;
2799 async_data->callback_data = callback_data;
2801 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2802 if (tmp_ip->pnn == -1) {
2803 /* this IP won't be taken over */
2807 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2808 ipv4.pnn = tmp_ip->pnn;
2809 ipv4.sin = tmp_ip->addr.ip;
2811 timeout = TAKEOVER_TIMEOUT();
2812 data.dsize = sizeof(ipv4);
2813 data.dptr = (uint8_t *)&ipv4;
2814 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2815 0, CTDB_CONTROL_TAKEOVER_IPv4, 0,
2819 ip.pnn = tmp_ip->pnn;
2820 ip.addr = tmp_ip->addr;
2822 timeout = TAKEOVER_TIMEOUT();
2823 data.dsize = sizeof(ip);
2824 data.dptr = (uint8_t *)&ip;
2825 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2826 0, CTDB_CONTROL_TAKEOVER_IP, 0,
2830 if (state == NULL) {
2831 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_TAKEOVER_IP to node %u\n", tmp_ip->pnn));
2832 talloc_free(tmp_ctx);
2836 ctdb_client_async_add(async_data, state);
2838 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2839 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_TAKEOVER_IP failed\n"));
2840 talloc_free(tmp_ctx);
2846 * Tell all nodes to run eventscripts to process the
2847 * "ipreallocated" event. This can do a lot of things,
2848 * including restarting services to reconfigure them if public
2849 * IPs have moved. Once upon a time this event only used to
2852 retry_data = talloc_zero_array(tmp_ctx, bool, nodemap->num);
2853 CTDB_NO_MEMORY_FATAL(ctdb, retry_data);
2854 iprealloc_data.retry_nodes = retry_data;
2855 iprealloc_data.retry_count = 0;
2856 iprealloc_data.fail_callback = fail_callback;
2857 iprealloc_data.fail_callback_data = callback_data;
2858 iprealloc_data.nodemap = nodemap;
2860 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2861 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_IPREALLOCATED,
2862 nodes, 0, TAKEOVER_TIMEOUT(),
2864 NULL, iprealloc_fail_callback,
2865 &iprealloc_data) != 0) {
2867 /* If the control failed then we should retry to any
2868 * nodes flagged by iprealloc_fail_callback using the
2869 * EVENTSCRIPT control. This is a best-effort at
2870 * backward compatiblity when running a mixed cluster
2871 * where some nodes have not yet been upgraded to
2872 * support the IPREALLOCATED control.
2874 DEBUG(DEBUG_WARNING,
2875 ("Retry ipreallocated to some nodes using eventscript control\n"));
2877 nodes = talloc_array(tmp_ctx, uint32_t,
2878 iprealloc_data.retry_count);
2879 CTDB_NO_MEMORY_FATAL(ctdb, nodes);
2882 for (i=0; i<nodemap->num; i++) {
2883 if (iprealloc_data.retry_nodes[i]) {
2889 data.dptr = discard_const("ipreallocated");
2890 data.dsize = strlen((char *)data.dptr) + 1;
2891 if (ctdb_client_async_control(ctdb,
2892 CTDB_CONTROL_RUN_EVENTSCRIPTS,
2893 nodes, 0, TAKEOVER_TIMEOUT(),
2895 NULL, fail_callback,
2896 callback_data) != 0) {
2897 DEBUG(DEBUG_ERR, (__location__ " failed to send control to run eventscripts with \"ipreallocated\"\n"));
2901 talloc_free(tmp_ctx);
2907 destroy a ctdb_client_ip structure
2909 static int ctdb_client_ip_destructor(struct ctdb_client_ip *ip)
2911 DEBUG(DEBUG_DEBUG,("destroying client tcp for %s:%u (client_id %u)\n",
2912 ctdb_addr_to_str(&ip->addr),
2913 ntohs(ip->addr.ip.sin_port),
2916 DLIST_REMOVE(ip->ctdb->client_ip_list, ip);
2921 called by a client to inform us of a TCP connection that it is managing
2922 that should tickled with an ACK when IP takeover is done
2923 we handle both the old ipv4 style of packets as well as the new ipv4/6
2926 int32_t ctdb_control_tcp_client(struct ctdb_context *ctdb, uint32_t client_id,
2929 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
2930 struct ctdb_control_tcp *old_addr = NULL;
2931 struct ctdb_control_tcp_addr new_addr;
2932 struct ctdb_control_tcp_addr *tcp_sock = NULL;
2933 struct ctdb_tcp_list *tcp;
2934 struct ctdb_tcp_connection t;
2937 struct ctdb_client_ip *ip;
2938 struct ctdb_vnn *vnn;
2939 ctdb_sock_addr addr;
2941 switch (indata.dsize) {
2942 case sizeof(struct ctdb_control_tcp):
2943 old_addr = (struct ctdb_control_tcp *)indata.dptr;
2944 ZERO_STRUCT(new_addr);
2945 tcp_sock = &new_addr;
2946 tcp_sock->src.ip = old_addr->src;
2947 tcp_sock->dest.ip = old_addr->dest;
2949 case sizeof(struct ctdb_control_tcp_addr):
2950 tcp_sock = (struct ctdb_control_tcp_addr *)indata.dptr;
2953 DEBUG(DEBUG_ERR,(__location__ " Invalid data structure passed "
2954 "to ctdb_control_tcp_client. size was %d but "
2955 "only allowed sizes are %lu and %lu\n",
2957 (long unsigned)sizeof(struct ctdb_control_tcp),
2958 (long unsigned)sizeof(struct ctdb_control_tcp_addr)));
2962 addr = tcp_sock->src;
2963 ctdb_canonicalize_ip(&addr, &tcp_sock->src);
2964 addr = tcp_sock->dest;
2965 ctdb_canonicalize_ip(&addr, &tcp_sock->dest);
2968 memcpy(&addr, &tcp_sock->dest, sizeof(addr));
2969 vnn = find_public_ip_vnn(ctdb, &addr);
2971 switch (addr.sa.sa_family) {
2973 if (ntohl(addr.ip.sin_addr.s_addr) != INADDR_LOOPBACK) {
2974 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public address.\n",
2975 ctdb_addr_to_str(&addr)));
2979 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public ipv6 address.\n",
2980 ctdb_addr_to_str(&addr)));
2983 DEBUG(DEBUG_ERR,(__location__ " Unknown family type %d\n", addr.sa.sa_family));
2989 if (vnn->pnn != ctdb->pnn) {
2990 DEBUG(DEBUG_ERR,("Attempt to register tcp client for IP %s we don't hold - failing (client_id %u pid %u)\n",
2991 ctdb_addr_to_str(&addr),
2992 client_id, client->pid));
2993 /* failing this call will tell smbd to die */
2997 ip = talloc(client, struct ctdb_client_ip);
2998 CTDB_NO_MEMORY(ctdb, ip);
3002 ip->client_id = client_id;
3003 talloc_set_destructor(ip, ctdb_client_ip_destructor);
3004 DLIST_ADD(ctdb->client_ip_list, ip);
3006 tcp = talloc(client, struct ctdb_tcp_list);
3007 CTDB_NO_MEMORY(ctdb, tcp);
3009 tcp->connection.src_addr = tcp_sock->src;
3010 tcp->connection.dst_addr = tcp_sock->dest;
3012 DLIST_ADD(client->tcp_list, tcp);
3014 t.src_addr = tcp_sock->src;
3015 t.dst_addr = tcp_sock->dest;
3017 data.dptr = (uint8_t *)&t;
3018 data.dsize = sizeof(t);
3020 switch (addr.sa.sa_family) {
3022 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
3023 (unsigned)ntohs(tcp_sock->dest.ip.sin_port),
3024 ctdb_addr_to_str(&tcp_sock->src),
3025 (unsigned)ntohs(tcp_sock->src.ip.sin_port), client_id, client->pid));
3028 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
3029 (unsigned)ntohs(tcp_sock->dest.ip6.sin6_port),
3030 ctdb_addr_to_str(&tcp_sock->src),
3031 (unsigned)ntohs(tcp_sock->src.ip6.sin6_port), client_id, client->pid));
3034 DEBUG(DEBUG_ERR,(__location__ " Unknown family %d\n", addr.sa.sa_family));
3038 /* tell all nodes about this tcp connection */
3039 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
3040 CTDB_CONTROL_TCP_ADD,
3041 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
3043 DEBUG(DEBUG_ERR,(__location__ " Failed to send CTDB_CONTROL_TCP_ADD\n"));
3051 find a tcp address on a list
3053 static struct ctdb_tcp_connection *ctdb_tcp_find(struct ctdb_tcp_array *array,
3054 struct ctdb_tcp_connection *tcp)
3058 if (array == NULL) {
3062 for (i=0;i<array->num;i++) {
3063 if (ctdb_same_sockaddr(&array->connections[i].src_addr, &tcp->src_addr) &&
3064 ctdb_same_sockaddr(&array->connections[i].dst_addr, &tcp->dst_addr)) {
3065 return &array->connections[i];
3074 called by a daemon to inform us of a TCP connection that one of its
3075 clients managing that should tickled with an ACK when IP takeover is
3078 int32_t ctdb_control_tcp_add(struct ctdb_context *ctdb, TDB_DATA indata, bool tcp_update_needed)
3080 struct ctdb_tcp_connection *p = (struct ctdb_tcp_connection *)indata.dptr;
3081 struct ctdb_tcp_array *tcparray;
3082 struct ctdb_tcp_connection tcp;
3083 struct ctdb_vnn *vnn;
3085 vnn = find_public_ip_vnn(ctdb, &p->dst_addr);
3087 DEBUG(DEBUG_INFO,(__location__ " got TCP_ADD control for an address which is not a public address '%s'\n",
3088 ctdb_addr_to_str(&p->dst_addr)));
3094 tcparray = vnn->tcp_array;
3096 /* If this is the first tickle */
3097 if (tcparray == NULL) {
3098 tcparray = talloc_size(ctdb->nodes,
3099 offsetof(struct ctdb_tcp_array, connections) +
3100 sizeof(struct ctdb_tcp_connection) * 1);
3101 CTDB_NO_MEMORY(ctdb, tcparray);
3102 vnn->tcp_array = tcparray;
3105 tcparray->connections = talloc_size(tcparray, sizeof(struct ctdb_tcp_connection));
3106 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3108 tcparray->connections[tcparray->num].src_addr = p->src_addr;
3109 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
3112 if (tcp_update_needed) {
3113 vnn->tcp_update_needed = true;
3119 /* Do we already have this tickle ?*/
3120 tcp.src_addr = p->src_addr;
3121 tcp.dst_addr = p->dst_addr;
3122 if (ctdb_tcp_find(vnn->tcp_array, &tcp) != NULL) {
3123 DEBUG(DEBUG_DEBUG,("Already had tickle info for %s:%u for vnn:%u\n",
3124 ctdb_addr_to_str(&tcp.dst_addr),
3125 ntohs(tcp.dst_addr.ip.sin_port),
3130 /* A new tickle, we must add it to the array */
3131 tcparray->connections = talloc_realloc(tcparray, tcparray->connections,
3132 struct ctdb_tcp_connection,
3134 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3136 vnn->tcp_array = tcparray;
3137 tcparray->connections[tcparray->num].src_addr = p->src_addr;
3138 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
3141 DEBUG(DEBUG_INFO,("Added tickle info for %s:%u from vnn %u\n",
3142 ctdb_addr_to_str(&tcp.dst_addr),
3143 ntohs(tcp.dst_addr.ip.sin_port),
3146 if (tcp_update_needed) {
3147 vnn->tcp_update_needed = true;
3155 called by a daemon to inform us of a TCP connection that one of its
3156 clients managing that should tickled with an ACK when IP takeover is
3159 static void ctdb_remove_tcp_connection(struct ctdb_context *ctdb, struct ctdb_tcp_connection *conn)
3161 struct ctdb_tcp_connection *tcpp;
3162 struct ctdb_vnn *vnn = find_public_ip_vnn(ctdb, &conn->dst_addr);
3165 DEBUG(DEBUG_ERR,(__location__ " unable to find public address %s\n",
3166 ctdb_addr_to_str(&conn->dst_addr)));
3170 /* if the array is empty we cant remove it
3171 and we dont need to do anything
3173 if (vnn->tcp_array == NULL) {
3174 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist (array is empty) %s:%u\n",
3175 ctdb_addr_to_str(&conn->dst_addr),
3176 ntohs(conn->dst_addr.ip.sin_port)));
3181 /* See if we know this connection
3182 if we dont know this connection then we dont need to do anything
3184 tcpp = ctdb_tcp_find(vnn->tcp_array, conn);
3186 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist %s:%u\n",
3187 ctdb_addr_to_str(&conn->dst_addr),
3188 ntohs(conn->dst_addr.ip.sin_port)));
3193 /* We need to remove this entry from the array.
3194 Instead of allocating a new array and copying data to it
3195 we cheat and just copy the last entry in the existing array
3196 to the entry that is to be removed and just shring the
3199 *tcpp = vnn->tcp_array->connections[vnn->tcp_array->num - 1];
3200 vnn->tcp_array->num--;
3202 /* If we deleted the last entry we also need to remove the entire array
3204 if (vnn->tcp_array->num == 0) {
3205 talloc_free(vnn->tcp_array);
3206 vnn->tcp_array = NULL;
3209 vnn->tcp_update_needed = true;
3211 DEBUG(DEBUG_INFO,("Removed tickle info for %s:%u\n",
3212 ctdb_addr_to_str(&conn->src_addr),
3213 ntohs(conn->src_addr.ip.sin_port)));
3218 called by a daemon to inform us of a TCP connection that one of its
3219 clients used are no longer needed in the tickle database
3221 int32_t ctdb_control_tcp_remove(struct ctdb_context *ctdb, TDB_DATA indata)
3223 struct ctdb_tcp_connection *conn = (struct ctdb_tcp_connection *)indata.dptr;
3225 ctdb_remove_tcp_connection(ctdb, conn);
3232 called when a daemon restarts - send all tickes for all public addresses
3233 we are serving immediately to the new node.
3235 int32_t ctdb_control_startup(struct ctdb_context *ctdb, uint32_t vnn)
3237 /*XXX here we should send all tickes we are serving to the new node */
3243 called when a client structure goes away - hook to remove
3244 elements from the tcp_list in all daemons
3246 void ctdb_takeover_client_destructor_hook(struct ctdb_client *client)
3248 while (client->tcp_list) {
3249 struct ctdb_tcp_list *tcp = client->tcp_list;
3250 DLIST_REMOVE(client->tcp_list, tcp);
3251 ctdb_remove_tcp_connection(client->ctdb, &tcp->connection);
3257 release all IPs on shutdown
3259 void ctdb_release_all_ips(struct ctdb_context *ctdb)
3261 struct ctdb_vnn *vnn;
3264 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3265 if (!ctdb_sys_have_ip(&vnn->public_address)) {
3266 ctdb_vnn_unassign_iface(ctdb, vnn);
3273 DEBUG(DEBUG_INFO,("Release of IP %s/%u on interface %s node:-1\n",
3274 ctdb_addr_to_str(&vnn->public_address),
3275 vnn->public_netmask_bits,
3276 ctdb_vnn_iface_string(vnn)));
3278 ctdb_event_script_args(ctdb, CTDB_EVENT_RELEASE_IP, "%s %s %u",
3279 ctdb_vnn_iface_string(vnn),
3280 ctdb_addr_to_str(&vnn->public_address),
3281 vnn->public_netmask_bits);
3282 release_kill_clients(ctdb, &vnn->public_address);
3283 ctdb_vnn_unassign_iface(ctdb, vnn);
3287 DEBUG(DEBUG_NOTICE,(__location__ " Released %d public IPs\n", count));
3292 get list of public IPs
3294 int32_t ctdb_control_get_public_ips(struct ctdb_context *ctdb,
3295 struct ctdb_req_control *c, TDB_DATA *outdata)
3298 struct ctdb_all_public_ips *ips;
3299 struct ctdb_vnn *vnn;
3300 bool only_available = false;
3302 if (c->flags & CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE) {
3303 only_available = true;
3306 /* count how many public ip structures we have */
3308 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3312 len = offsetof(struct ctdb_all_public_ips, ips) +
3313 num*sizeof(struct ctdb_public_ip);
3314 ips = talloc_zero_size(outdata, len);
3315 CTDB_NO_MEMORY(ctdb, ips);
3318 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3319 if (only_available && !ctdb_vnn_available(ctdb, vnn)) {
3322 ips->ips[i].pnn = vnn->pnn;
3323 ips->ips[i].addr = vnn->public_address;
3327 len = offsetof(struct ctdb_all_public_ips, ips) +
3328 i*sizeof(struct ctdb_public_ip);
3330 outdata->dsize = len;
3331 outdata->dptr = (uint8_t *)ips;
3338 get list of public IPs, old ipv4 style. only returns ipv4 addresses
3340 int32_t ctdb_control_get_public_ipsv4(struct ctdb_context *ctdb,
3341 struct ctdb_req_control *c, TDB_DATA *outdata)
3344 struct ctdb_all_public_ipsv4 *ips;
3345 struct ctdb_vnn *vnn;
3347 /* count how many public ip structures we have */
3349 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3350 if (vnn->public_address.sa.sa_family != AF_INET) {
3356 len = offsetof(struct ctdb_all_public_ipsv4, ips) +
3357 num*sizeof(struct ctdb_public_ipv4);
3358 ips = talloc_zero_size(outdata, len);
3359 CTDB_NO_MEMORY(ctdb, ips);
3361 outdata->dsize = len;
3362 outdata->dptr = (uint8_t *)ips;
3366 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3367 if (vnn->public_address.sa.sa_family != AF_INET) {
3370 ips->ips[i].pnn = vnn->pnn;
3371 ips->ips[i].sin = vnn->public_address.ip;
3378 int32_t ctdb_control_get_public_ip_info(struct ctdb_context *ctdb,
3379 struct ctdb_req_control *c,
3384 ctdb_sock_addr *addr;
3385 struct ctdb_control_public_ip_info *info;
3386 struct ctdb_vnn *vnn;
3388 addr = (ctdb_sock_addr *)indata.dptr;
3390 vnn = find_public_ip_vnn(ctdb, addr);
3392 /* if it is not a public ip it could be our 'single ip' */
3393 if (ctdb->single_ip_vnn) {
3394 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, addr)) {
3395 vnn = ctdb->single_ip_vnn;
3400 DEBUG(DEBUG_ERR,(__location__ " Could not get public ip info, "
3401 "'%s'not a public address\n",
3402 ctdb_addr_to_str(addr)));
3406 /* count how many public ip structures we have */
3408 for (;vnn->ifaces[num];) {
3412 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3413 num*sizeof(struct ctdb_control_iface_info);
3414 info = talloc_zero_size(outdata, len);
3415 CTDB_NO_MEMORY(ctdb, info);
3417 info->ip.addr = vnn->public_address;
3418 info->ip.pnn = vnn->pnn;
3419 info->active_idx = 0xFFFFFFFF;
3421 for (i=0; vnn->ifaces[i]; i++) {
3422 struct ctdb_iface *cur;
3424 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
3426 DEBUG(DEBUG_CRIT, (__location__ " internal error iface[%s] unknown\n",
3430 if (vnn->iface == cur) {
3431 info->active_idx = i;
3433 strcpy(info->ifaces[i].name, cur->name);
3434 info->ifaces[i].link_state = cur->link_up;
3435 info->ifaces[i].references = cur->references;
3438 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3439 i*sizeof(struct ctdb_control_iface_info);
3441 outdata->dsize = len;
3442 outdata->dptr = (uint8_t *)info;
3447 int32_t ctdb_control_get_ifaces(struct ctdb_context *ctdb,
3448 struct ctdb_req_control *c,
3452 struct ctdb_control_get_ifaces *ifaces;
3453 struct ctdb_iface *cur;
3455 /* count how many public ip structures we have */
3457 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3461 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3462 num*sizeof(struct ctdb_control_iface_info);
3463 ifaces = talloc_zero_size(outdata, len);
3464 CTDB_NO_MEMORY(ctdb, ifaces);
3467 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3468 strcpy(ifaces->ifaces[i].name, cur->name);
3469 ifaces->ifaces[i].link_state = cur->link_up;
3470 ifaces->ifaces[i].references = cur->references;
3474 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3475 i*sizeof(struct ctdb_control_iface_info);
3477 outdata->dsize = len;
3478 outdata->dptr = (uint8_t *)ifaces;
3483 int32_t ctdb_control_set_iface_link(struct ctdb_context *ctdb,
3484 struct ctdb_req_control *c,
3487 struct ctdb_control_iface_info *info;
3488 struct ctdb_iface *iface;
3489 bool link_up = false;
3491 info = (struct ctdb_control_iface_info *)indata.dptr;
3493 if (info->name[CTDB_IFACE_SIZE] != '\0') {
3494 int len = strnlen(info->name, CTDB_IFACE_SIZE);
3495 DEBUG(DEBUG_ERR, (__location__ " name[%*.*s] not terminated\n",
3496 len, len, info->name));
3500 switch (info->link_state) {
3508 DEBUG(DEBUG_ERR, (__location__ " link_state[%u] invalid\n",
3509 (unsigned int)info->link_state));
3513 if (info->references != 0) {
3514 DEBUG(DEBUG_ERR, (__location__ " references[%u] should be 0\n",
3515 (unsigned int)info->references));
3519 iface = ctdb_find_iface(ctdb, info->name);
3520 if (iface == NULL) {
3524 if (link_up == iface->link_up) {
3528 DEBUG(iface->link_up?DEBUG_ERR:DEBUG_NOTICE,
3529 ("iface[%s] has changed it's link status %s => %s\n",
3531 iface->link_up?"up":"down",
3532 link_up?"up":"down"));
3534 iface->link_up = link_up;
3540 structure containing the listening socket and the list of tcp connections
3541 that the ctdb daemon is to kill
3543 struct ctdb_kill_tcp {
3544 struct ctdb_vnn *vnn;
3545 struct ctdb_context *ctdb;
3547 struct fd_event *fde;
3548 trbt_tree_t *connections;
3553 a tcp connection that is to be killed
3555 struct ctdb_killtcp_con {
3556 ctdb_sock_addr src_addr;
3557 ctdb_sock_addr dst_addr;
3559 struct ctdb_kill_tcp *killtcp;
3562 /* this function is used to create a key to represent this socketpair
3563 in the killtcp tree.
3564 this key is used to insert and lookup matching socketpairs that are
3565 to be tickled and RST
3567 #define KILLTCP_KEYLEN 10
3568 static uint32_t *killtcp_key(ctdb_sock_addr *src, ctdb_sock_addr *dst)
3570 static uint32_t key[KILLTCP_KEYLEN];
3572 bzero(key, sizeof(key));
3574 if (src->sa.sa_family != dst->sa.sa_family) {
3575 DEBUG(DEBUG_ERR, (__location__ " ERROR, different families passed :%u vs %u\n", src->sa.sa_family, dst->sa.sa_family));
3579 switch (src->sa.sa_family) {
3581 key[0] = dst->ip.sin_addr.s_addr;
3582 key[1] = src->ip.sin_addr.s_addr;
3583 key[2] = dst->ip.sin_port;
3584 key[3] = src->ip.sin_port;
3587 uint32_t *dst6_addr32 =
3588 (uint32_t *)&(dst->ip6.sin6_addr.s6_addr);
3589 uint32_t *src6_addr32 =
3590 (uint32_t *)&(src->ip6.sin6_addr.s6_addr);
3591 key[0] = dst6_addr32[3];
3592 key[1] = src6_addr32[3];
3593 key[2] = dst6_addr32[2];
3594 key[3] = src6_addr32[2];
3595 key[4] = dst6_addr32[1];
3596 key[5] = src6_addr32[1];
3597 key[6] = dst6_addr32[0];
3598 key[7] = src6_addr32[0];
3599 key[8] = dst->ip6.sin6_port;
3600 key[9] = src->ip6.sin6_port;
3604 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", src->sa.sa_family));
3612 called when we get a read event on the raw socket
3614 static void capture_tcp_handler(struct event_context *ev, struct fd_event *fde,
3615 uint16_t flags, void *private_data)
3617 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3618 struct ctdb_killtcp_con *con;
3619 ctdb_sock_addr src, dst;
3620 uint32_t ack_seq, seq;
3622 if (!(flags & EVENT_FD_READ)) {
3626 if (ctdb_sys_read_tcp_packet(killtcp->capture_fd,
3627 killtcp->private_data,
3629 &ack_seq, &seq) != 0) {
3630 /* probably a non-tcp ACK packet */
3634 /* check if we have this guy in our list of connections
3637 con = trbt_lookuparray32(killtcp->connections,
3638 KILLTCP_KEYLEN, killtcp_key(&src, &dst));
3640 /* no this was some other packet we can just ignore */
3644 /* This one has been tickled !
3645 now reset him and remove him from the list.
3647 DEBUG(DEBUG_INFO, ("sending a tcp reset to kill connection :%d -> %s:%d\n",
3648 ntohs(con->dst_addr.ip.sin_port),
3649 ctdb_addr_to_str(&con->src_addr),
3650 ntohs(con->src_addr.ip.sin_port)));
3652 ctdb_sys_send_tcp(&con->dst_addr, &con->src_addr, ack_seq, seq, 1);
3657 /* when traversing the list of all tcp connections to send tickle acks to
3658 (so that we can capture the ack coming back and kill the connection
3660 this callback is called for each connection we are currently trying to kill
3662 static int tickle_connection_traverse(void *param, void *data)
3664 struct ctdb_killtcp_con *con = talloc_get_type(data, struct ctdb_killtcp_con);
3666 /* have tried too many times, just give up */
3667 if (con->count >= 5) {
3668 /* can't delete in traverse: reparent to delete_cons */
3669 talloc_steal(param, con);
3673 /* othervise, try tickling it again */
3676 (ctdb_sock_addr *)&con->dst_addr,
3677 (ctdb_sock_addr *)&con->src_addr,
3684 called every second until all sentenced connections have been reset
3686 static void ctdb_tickle_sentenced_connections(struct event_context *ev, struct timed_event *te,
3687 struct timeval t, void *private_data)
3689 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3690 void *delete_cons = talloc_new(NULL);
3692 /* loop over all connections sending tickle ACKs */
3693 trbt_traversearray32(killtcp->connections, KILLTCP_KEYLEN, tickle_connection_traverse, delete_cons);
3695 /* now we've finished traverse, it's safe to do deletion. */
3696 talloc_free(delete_cons);
3698 /* If there are no more connections to kill we can remove the
3699 entire killtcp structure
3701 if ( (killtcp->connections == NULL) ||
3702 (killtcp->connections->root == NULL) ) {
3703 talloc_free(killtcp);
3707 /* try tickling them again in a seconds time
3709 event_add_timed(killtcp->ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3710 ctdb_tickle_sentenced_connections, killtcp);
3714 destroy the killtcp structure
3716 static int ctdb_killtcp_destructor(struct ctdb_kill_tcp *killtcp)
3718 struct ctdb_vnn *tmpvnn;
3720 /* verify that this vnn is still active */
3721 for (tmpvnn = killtcp->ctdb->vnn; tmpvnn; tmpvnn = tmpvnn->next) {
3722 if (tmpvnn == killtcp->vnn) {
3727 if (tmpvnn == NULL) {
3731 if (killtcp->vnn->killtcp != killtcp) {
3735 killtcp->vnn->killtcp = NULL;
3741 /* nothing fancy here, just unconditionally replace any existing
3742 connection structure with the new one.
3744 dont even free the old one if it did exist, that one is talloc_stolen
3745 by the same node in the tree anyway and will be deleted when the new data
3748 static void *add_killtcp_callback(void *parm, void *data)
3754 add a tcp socket to the list of connections we want to RST
3756 static int ctdb_killtcp_add_connection(struct ctdb_context *ctdb,
3760 ctdb_sock_addr src, dst;
3761 struct ctdb_kill_tcp *killtcp;
3762 struct ctdb_killtcp_con *con;
3763 struct ctdb_vnn *vnn;
3765 ctdb_canonicalize_ip(s, &src);
3766 ctdb_canonicalize_ip(d, &dst);
3768 vnn = find_public_ip_vnn(ctdb, &dst);
3770 vnn = find_public_ip_vnn(ctdb, &src);
3773 /* if it is not a public ip it could be our 'single ip' */
3774 if (ctdb->single_ip_vnn) {
3775 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, &dst)) {
3776 vnn = ctdb->single_ip_vnn;
3781 DEBUG(DEBUG_ERR,(__location__ " Could not killtcp, not a public address\n"));
3785 killtcp = vnn->killtcp;
3787 /* If this is the first connection to kill we must allocate
3790 if (killtcp == NULL) {
3791 killtcp = talloc_zero(vnn, struct ctdb_kill_tcp);
3792 CTDB_NO_MEMORY(ctdb, killtcp);
3795 killtcp->ctdb = ctdb;
3796 killtcp->capture_fd = -1;
3797 killtcp->connections = trbt_create(killtcp, 0);
3799 vnn->killtcp = killtcp;
3800 talloc_set_destructor(killtcp, ctdb_killtcp_destructor);
3805 /* create a structure that describes this connection we want to
3806 RST and store it in killtcp->connections
3808 con = talloc(killtcp, struct ctdb_killtcp_con);
3809 CTDB_NO_MEMORY(ctdb, con);
3810 con->src_addr = src;
3811 con->dst_addr = dst;
3813 con->killtcp = killtcp;
3816 trbt_insertarray32_callback(killtcp->connections,
3817 KILLTCP_KEYLEN, killtcp_key(&con->dst_addr, &con->src_addr),
3818 add_killtcp_callback, con);
3821 If we dont have a socket to listen on yet we must create it
3823 if (killtcp->capture_fd == -1) {
3824 const char *iface = ctdb_vnn_iface_string(vnn);
3825 killtcp->capture_fd = ctdb_sys_open_capture_socket(iface, &killtcp->private_data);
3826 if (killtcp->capture_fd == -1) {
3827 DEBUG(DEBUG_CRIT,(__location__ " Failed to open capturing "
3828 "socket on iface '%s' for killtcp (%s)\n",
3829 iface, strerror(errno)));
3835 if (killtcp->fde == NULL) {
3836 killtcp->fde = event_add_fd(ctdb->ev, killtcp, killtcp->capture_fd,
3838 capture_tcp_handler, killtcp);
3839 tevent_fd_set_auto_close(killtcp->fde);
3841 /* We also need to set up some events to tickle all these connections
3842 until they are all reset
3844 event_add_timed(ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3845 ctdb_tickle_sentenced_connections, killtcp);
3848 /* tickle him once now */
3857 talloc_free(vnn->killtcp);
3858 vnn->killtcp = NULL;
3863 kill a TCP connection.
3865 int32_t ctdb_control_kill_tcp(struct ctdb_context *ctdb, TDB_DATA indata)
3867 struct ctdb_control_killtcp *killtcp = (struct ctdb_control_killtcp *)indata.dptr;
3869 return ctdb_killtcp_add_connection(ctdb, &killtcp->src_addr, &killtcp->dst_addr);
3873 called by a daemon to inform us of the entire list of TCP tickles for
3874 a particular public address.
3875 this control should only be sent by the node that is currently serving
3876 that public address.
3878 int32_t ctdb_control_set_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata)
3880 struct ctdb_control_tcp_tickle_list *list = (struct ctdb_control_tcp_tickle_list *)indata.dptr;
3881 struct ctdb_tcp_array *tcparray;
3882 struct ctdb_vnn *vnn;
3884 /* We must at least have tickles.num or else we cant verify the size
3885 of the received data blob
3887 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3888 tickles.connections)) {
3889 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list. Not enough data for the tickle.num field\n"));
3893 /* verify that the size of data matches what we expect */
3894 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3895 tickles.connections)
3896 + sizeof(struct ctdb_tcp_connection)
3897 * list->tickles.num) {
3898 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list\n"));
3902 vnn = find_public_ip_vnn(ctdb, &list->addr);
3904 DEBUG(DEBUG_INFO,(__location__ " Could not set tcp tickle list, '%s' is not a public address\n",
3905 ctdb_addr_to_str(&list->addr)));
3910 /* remove any old ticklelist we might have */
3911 talloc_free(vnn->tcp_array);
3912 vnn->tcp_array = NULL;
3914 tcparray = talloc(ctdb->nodes, struct ctdb_tcp_array);
3915 CTDB_NO_MEMORY(ctdb, tcparray);
3917 tcparray->num = list->tickles.num;
3919 tcparray->connections = talloc_array(tcparray, struct ctdb_tcp_connection, tcparray->num);
3920 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3922 memcpy(tcparray->connections, &list->tickles.connections[0],
3923 sizeof(struct ctdb_tcp_connection)*tcparray->num);
3925 /* We now have a new fresh tickle list array for this vnn */
3926 vnn->tcp_array = talloc_steal(vnn, tcparray);
3932 called to return the full list of tickles for the puclic address associated
3933 with the provided vnn
3935 int32_t ctdb_control_get_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
3937 ctdb_sock_addr *addr = (ctdb_sock_addr *)indata.dptr;
3938 struct ctdb_control_tcp_tickle_list *list;
3939 struct ctdb_tcp_array *tcparray;
3941 struct ctdb_vnn *vnn;
3943 vnn = find_public_ip_vnn(ctdb, addr);
3945 DEBUG(DEBUG_ERR,(__location__ " Could not get tcp tickle list, '%s' is not a public address\n",
3946 ctdb_addr_to_str(addr)));
3951 tcparray = vnn->tcp_array;
3953 num = tcparray->num;
3958 outdata->dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3959 tickles.connections)
3960 + sizeof(struct ctdb_tcp_connection) * num;
3962 outdata->dptr = talloc_size(outdata, outdata->dsize);
3963 CTDB_NO_MEMORY(ctdb, outdata->dptr);
3964 list = (struct ctdb_control_tcp_tickle_list *)outdata->dptr;
3967 list->tickles.num = num;
3969 memcpy(&list->tickles.connections[0], tcparray->connections,
3970 sizeof(struct ctdb_tcp_connection) * num);
3978 set the list of all tcp tickles for a public address
3980 static int ctdb_ctrl_set_tcp_tickles(struct ctdb_context *ctdb,
3981 struct timeval timeout, uint32_t destnode,
3982 ctdb_sock_addr *addr,
3983 struct ctdb_tcp_array *tcparray)
3987 struct ctdb_control_tcp_tickle_list *list;
3990 num = tcparray->num;
3995 data.dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3996 tickles.connections) +
3997 sizeof(struct ctdb_tcp_connection) * num;
3998 data.dptr = talloc_size(ctdb, data.dsize);
3999 CTDB_NO_MEMORY(ctdb, data.dptr);
4001 list = (struct ctdb_control_tcp_tickle_list *)data.dptr;
4003 list->tickles.num = num;
4005 memcpy(&list->tickles.connections[0], tcparray->connections, sizeof(struct ctdb_tcp_connection) * num);
4008 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
4009 CTDB_CONTROL_SET_TCP_TICKLE_LIST,
4010 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
4012 DEBUG(DEBUG_ERR,(__location__ " ctdb_control for set tcp tickles failed\n"));
4016 talloc_free(data.dptr);
4023 perform tickle updates if required
4025 static void ctdb_update_tcp_tickles(struct event_context *ev,
4026 struct timed_event *te,
4027 struct timeval t, void *private_data)
4029 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
4031 struct ctdb_vnn *vnn;
4033 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
4034 /* we only send out updates for public addresses that
4037 if (ctdb->pnn != vnn->pnn) {
4040 /* We only send out the updates if we need to */
4041 if (!vnn->tcp_update_needed) {
4044 ret = ctdb_ctrl_set_tcp_tickles(ctdb,
4046 CTDB_BROADCAST_CONNECTED,
4047 &vnn->public_address,
4050 DEBUG(DEBUG_ERR,("Failed to send the tickle update for public address %s\n",
4051 ctdb_addr_to_str(&vnn->public_address)));
4055 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
4056 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
4057 ctdb_update_tcp_tickles, ctdb);
4062 start periodic update of tcp tickles
4064 void ctdb_start_tcp_tickle_update(struct ctdb_context *ctdb)
4066 ctdb->tickle_update_context = talloc_new(ctdb);
4068 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
4069 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
4070 ctdb_update_tcp_tickles, ctdb);
4076 struct control_gratious_arp {
4077 struct ctdb_context *ctdb;
4078 ctdb_sock_addr addr;
4084 send a control_gratuitous arp
4086 static void send_gratious_arp(struct event_context *ev, struct timed_event *te,
4087 struct timeval t, void *private_data)
4090 struct control_gratious_arp *arp = talloc_get_type(private_data,
4091 struct control_gratious_arp);
4093 ret = ctdb_sys_send_arp(&arp->addr, arp->iface);
4095 DEBUG(DEBUG_ERR,(__location__ " sending of gratious arp on iface '%s' failed (%s)\n",
4096 arp->iface, strerror(errno)));
4101 if (arp->count == CTDB_ARP_REPEAT) {
4106 event_add_timed(arp->ctdb->ev, arp,
4107 timeval_current_ofs(CTDB_ARP_INTERVAL, 0),
4108 send_gratious_arp, arp);
4115 int32_t ctdb_control_send_gratious_arp(struct ctdb_context *ctdb, TDB_DATA indata)
4117 struct ctdb_control_gratious_arp *gratious_arp = (struct ctdb_control_gratious_arp *)indata.dptr;
4118 struct control_gratious_arp *arp;
4120 /* verify the size of indata */
4121 if (indata.dsize < offsetof(struct ctdb_control_gratious_arp, iface)) {
4122 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_gratious_arp structure. Got %u require %u bytes\n",
4123 (unsigned)indata.dsize,
4124 (unsigned)offsetof(struct ctdb_control_gratious_arp, iface)));
4128 ( offsetof(struct ctdb_control_gratious_arp, iface)
4129 + gratious_arp->len ) ){
4131 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4132 "but should be %u bytes\n",
4133 (unsigned)indata.dsize,
4134 (unsigned)(offsetof(struct ctdb_control_gratious_arp, iface)+gratious_arp->len)));
4139 arp = talloc(ctdb, struct control_gratious_arp);
4140 CTDB_NO_MEMORY(ctdb, arp);
4143 arp->addr = gratious_arp->addr;
4144 arp->iface = talloc_strdup(arp, gratious_arp->iface);
4145 CTDB_NO_MEMORY(ctdb, arp->iface);
4148 event_add_timed(arp->ctdb->ev, arp,
4149 timeval_zero(), send_gratious_arp, arp);
4154 int32_t ctdb_control_add_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
4156 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
4159 /* verify the size of indata */
4160 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
4161 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
4165 ( offsetof(struct ctdb_control_ip_iface, iface)
4168 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4169 "but should be %u bytes\n",
4170 (unsigned)indata.dsize,
4171 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
4175 DEBUG(DEBUG_NOTICE,("Add IP %s\n", ctdb_addr_to_str(&pub->addr)));
4177 ret = ctdb_add_public_address(ctdb, &pub->addr, pub->mask, &pub->iface[0], true);
4180 DEBUG(DEBUG_ERR,(__location__ " Failed to add public address\n"));
4188 called when releaseip event finishes for del_public_address
4190 static void delete_ip_callback(struct ctdb_context *ctdb, int status,
4193 talloc_free(private_data);
4196 int32_t ctdb_control_del_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
4198 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
4199 struct ctdb_vnn *vnn;
4202 /* verify the size of indata */
4203 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
4204 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
4208 ( offsetof(struct ctdb_control_ip_iface, iface)
4211 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4212 "but should be %u bytes\n",
4213 (unsigned)indata.dsize,
4214 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
4218 DEBUG(DEBUG_NOTICE,("Delete IP %s\n", ctdb_addr_to_str(&pub->addr)));
4220 /* walk over all public addresses until we find a match */
4221 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
4222 if (ctdb_same_ip(&vnn->public_address, &pub->addr)) {
4223 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
4225 DLIST_REMOVE(ctdb->vnn, vnn);
4226 talloc_steal(mem_ctx, vnn);
4227 ctdb_remove_orphaned_ifaces(ctdb, vnn, mem_ctx);
4228 if (vnn->pnn != ctdb->pnn) {
4229 if (vnn->iface != NULL) {
4230 ctdb_vnn_unassign_iface(ctdb, vnn);
4232 talloc_free(mem_ctx);
4237 ret = ctdb_event_script_callback(ctdb,
4238 mem_ctx, delete_ip_callback, mem_ctx,
4240 CTDB_EVENT_RELEASE_IP,
4242 ctdb_vnn_iface_string(vnn),
4243 ctdb_addr_to_str(&vnn->public_address),
4244 vnn->public_netmask_bits);
4245 if (vnn->iface != NULL) {
4246 ctdb_vnn_unassign_iface(ctdb, vnn);
4259 struct ipreallocated_callback_state {
4260 struct ctdb_req_control *c;
4263 static void ctdb_ipreallocated_callback(struct ctdb_context *ctdb,
4264 int status, void *p)
4266 struct ipreallocated_callback_state *state =
4267 talloc_get_type(p, struct ipreallocated_callback_state);
4271 (" \"ipreallocated\" event script failed (status %d)\n",
4273 if (status == -ETIME) {
4274 ctdb_ban_self(ctdb);
4278 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
4282 /* A control to run the ipreallocated event */
4283 int32_t ctdb_control_ipreallocated(struct ctdb_context *ctdb,
4284 struct ctdb_req_control *c,
4288 struct ipreallocated_callback_state *state;
4290 state = talloc(ctdb, struct ipreallocated_callback_state);
4291 CTDB_NO_MEMORY(ctdb, state);
4293 DEBUG(DEBUG_INFO,(__location__ " Running \"ipreallocated\" event\n"));
4295 ret = ctdb_event_script_callback(ctdb, state,
4296 ctdb_ipreallocated_callback, state,
4297 false, CTDB_EVENT_IPREALLOCATED,
4301 DEBUG(DEBUG_ERR,("Failed to run \"ipreallocated\" event \n"));
4306 /* tell the control that we will be reply asynchronously */
4307 state->c = talloc_steal(state, c);
4308 *async_reply = true;
4314 /* This function is called from the recovery daemon to verify that a remote
4315 node has the expected ip allocation.
4316 This is verified against ctdb->ip_tree
4318 int verify_remote_ip_allocation(struct ctdb_context *ctdb,
4319 struct ctdb_all_public_ips *ips,
4322 struct ctdb_public_ip_list *tmp_ip;
4325 if (ctdb->ip_tree == NULL) {
4326 /* dont know the expected allocation yet, assume remote node
4335 for (i=0; i<ips->num; i++) {
4336 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ips->ips[i].addr));
4337 if (tmp_ip == NULL) {
4338 DEBUG(DEBUG_ERR,("Node %u has new or unknown public IP %s\n", pnn, ctdb_addr_to_str(&ips->ips[i].addr)));
4342 if (tmp_ip->pnn == -1 || ips->ips[i].pnn == -1) {
4346 if (tmp_ip->pnn != ips->ips[i].pnn) {
4348 ("Inconsistent IP allocation - node %u thinks %s is held by node %u while it is assigned to node %u\n",
4350 ctdb_addr_to_str(&ips->ips[i].addr),
4351 ips->ips[i].pnn, tmp_ip->pnn));
4359 int update_ip_assignment_tree(struct ctdb_context *ctdb, struct ctdb_public_ip *ip)
4361 struct ctdb_public_ip_list *tmp_ip;
4363 if (ctdb->ip_tree == NULL) {
4364 DEBUG(DEBUG_ERR,("No ctdb->ip_tree yet. Failed to update ip assignment\n"));
4368 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ip->addr));
4369 if (tmp_ip == NULL) {
4370 DEBUG(DEBUG_ERR,(__location__ " Could not find record for address %s, update ip\n", ctdb_addr_to_str(&ip->addr)));
4374 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));
4375 tmp_ip->pnn = ip->pnn;
4381 struct ctdb_reloadips_handle {
4382 struct ctdb_context *ctdb;
4383 struct ctdb_req_control *c;
4387 struct fd_event *fde;
4390 static int ctdb_reloadips_destructor(struct ctdb_reloadips_handle *h)
4392 if (h == h->ctdb->reload_ips) {
4393 h->ctdb->reload_ips = NULL;
4396 ctdb_request_control_reply(h->ctdb, h->c, NULL, h->status, NULL);
4399 ctdb_kill(h->ctdb, h->child, SIGKILL);
4403 static void ctdb_reloadips_timeout_event(struct event_context *ev,
4404 struct timed_event *te,
4405 struct timeval t, void *private_data)
4407 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4412 static void ctdb_reloadips_child_handler(struct event_context *ev, struct fd_event *fde,
4413 uint16_t flags, void *private_data)
4415 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4420 ret = read(h->fd[0], &res, 1);
4421 if (ret < 1 || res != 0) {
4422 DEBUG(DEBUG_ERR, (__location__ " Reloadips child process returned error\n"));
4430 static int ctdb_reloadips_child(struct ctdb_context *ctdb)
4432 TALLOC_CTX *mem_ctx = talloc_new(NULL);
4433 struct ctdb_all_public_ips *ips;
4434 struct ctdb_vnn *vnn;
4437 CTDB_NO_MEMORY(ctdb, mem_ctx);
4439 /* read the ip allocation from the local node */
4440 ret = ctdb_ctrl_get_public_ips(ctdb, TAKEOVER_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ips);
4442 DEBUG(DEBUG_ERR, ("Unable to get public ips from local node\n"));
4443 talloc_free(mem_ctx);
4447 /* re-read the public ips file */
4449 if (ctdb_set_public_addresses(ctdb, false) != 0) {
4450 DEBUG(DEBUG_ERR,("Failed to re-read public addresses file\n"));
4451 talloc_free(mem_ctx);
4456 /* check the previous list of ips and scan for ips that have been
4459 for (i = 0; i < ips->num; i++) {
4460 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4461 if (ctdb_same_ip(&vnn->public_address, &ips->ips[i].addr)) {
4466 /* we need to delete this ip, no longer available on this node */
4468 struct ctdb_control_ip_iface pub;
4470 DEBUG(DEBUG_NOTICE,("RELOADIPS: IP%s is no longer available on this node. Deleting it.\n", ctdb_addr_to_str(&ips->ips[i].addr)));
4471 pub.addr = ips->ips[i].addr;
4475 ret = ctdb_ctrl_del_public_ip(ctdb, TAKEOVER_TIMEOUT(), CTDB_CURRENT_NODE, &pub);
4477 talloc_free(mem_ctx);
4478 DEBUG(DEBUG_ERR, ("RELOADIPS: Unable to del public ip:%s from local node\n", ctdb_addr_to_str(&ips->ips[i].addr)));
4485 /* loop over all new ones and check the ones we need to add */
4486 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4487 for (i = 0; i < ips->num; i++) {
4488 if (ctdb_same_ip(&vnn->public_address, &ips->ips[i].addr)) {
4492 if (i == ips->num) {
4493 struct ctdb_control_ip_iface *pub;
4494 const char *ifaces = NULL;
4497 DEBUG(DEBUG_NOTICE,("RELOADIPS: New ip:%s found, adding it.\n", ctdb_addr_to_str(&vnn->public_address)));
4499 pub = talloc_zero(mem_ctx, struct ctdb_control_ip_iface);
4500 pub->addr = vnn->public_address;
4501 pub->mask = vnn->public_netmask_bits;
4503 ifaces = vnn->ifaces[0];
4505 while (vnn->ifaces[iface] != NULL) {
4506 ifaces = talloc_asprintf(vnn, "%s,%s", ifaces, vnn->ifaces[iface]);
4509 pub->len = strlen(ifaces)+1;
4510 pub = talloc_realloc_size(mem_ctx, pub,
4511 offsetof(struct ctdb_control_ip_iface, iface) + pub->len);
4513 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate memory\n"));
4514 talloc_free(mem_ctx);
4517 memcpy(&pub->iface[0], ifaces, pub->len);
4519 ret = ctdb_ctrl_add_public_ip(ctdb, TAKEOVER_TIMEOUT(),
4520 CTDB_CURRENT_NODE, pub);
4522 DEBUG(DEBUG_ERR, ("RELOADIPS: Unable to add public ip:%s to local node\n", ctdb_addr_to_str(&vnn->public_address)));
4523 talloc_free(mem_ctx);
4529 talloc_free(mem_ctx);
4533 /* This control is sent to force the node to re-read the public addresses file
4534 and drop any addresses we should nnot longer host, and add new addresses
4535 that we are now able to host
4537 int32_t ctdb_control_reload_public_ips(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
4539 struct ctdb_reloadips_handle *h;
4540 pid_t parent = getpid();
4542 if (ctdb->reload_ips != NULL) {
4543 talloc_free(ctdb->reload_ips);
4544 ctdb->reload_ips = NULL;
4547 h = talloc(ctdb, struct ctdb_reloadips_handle);
4548 CTDB_NO_MEMORY(ctdb, h);
4553 if (pipe(h->fd) == -1) {
4554 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
4559 h->child = ctdb_fork(ctdb);
4560 if (h->child == (pid_t)-1) {
4561 DEBUG(DEBUG_ERR, ("Failed to fork a child for reloadips\n"));
4569 if (h->child == 0) {
4570 signed char res = 0;
4573 debug_extra = talloc_asprintf(NULL, "reloadips:");
4575 ctdb_set_process_name("ctdb_reloadips");
4576 if (switch_from_server_to_client(ctdb, "reloadips-child") != 0) {
4577 DEBUG(DEBUG_CRIT,("ERROR: Failed to switch reloadips child into client mode\n"));
4580 res = ctdb_reloadips_child(ctdb);
4582 DEBUG(DEBUG_ERR,("Failed to reload ips on local node\n"));
4586 write(h->fd[1], &res, 1);
4587 /* make sure we die when our parent dies */
4588 while (ctdb_kill(ctdb, parent, 0) == 0 || errno != ESRCH) {
4594 h->c = talloc_steal(h, c);
4597 set_close_on_exec(h->fd[0]);
4599 talloc_set_destructor(h, ctdb_reloadips_destructor);
4602 h->fde = event_add_fd(ctdb->ev, h, h->fd[0],
4603 EVENT_FD_READ, ctdb_reloadips_child_handler,
4605 tevent_fd_set_auto_close(h->fde);
4607 event_add_timed(ctdb->ev, h,
4608 timeval_current_ofs(120, 0),
4609 ctdb_reloadips_timeout_event, h);
4611 /* we reply later */
4612 *async_reply = true;