4 Copyright (C) Andrew Tridgell 2006
5 Copyright (C) Ronnie Sahlberg 2008
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, see <http://www.gnu.org/licenses/>.
22 #include "lib/tevent/tevent.h"
23 #include "lib/tdb/include/tdb.h"
24 #include "system/network.h"
25 #include "system/filesys.h"
26 #include "../include/ctdb_private.h"
30 stop any connecting (established or pending) to a node
32 void ctdb_tcp_stop_connection(struct ctdb_node *node)
34 struct ctdb_tcp_node *tnode = talloc_get_type(
35 node->private_data, struct ctdb_tcp_node);
37 ctdb_queue_set_fd(tnode->out_queue, -1);
38 talloc_free(tnode->connect_te);
39 talloc_free(tnode->connect_fde);
40 tnode->connect_fde = NULL;
41 tnode->connect_te = NULL;
42 if (tnode->fd != -1) {
50 called when a complete packet has come in - should not happen on this socket
51 unless the other side closes the connection with RST or FIN
53 void ctdb_tcp_tnode_cb(uint8_t *data, size_t cnt, void *private_data)
55 struct ctdb_node *node = talloc_get_type(private_data, struct ctdb_node);
56 struct ctdb_tcp_node *tnode = talloc_get_type(
57 node->private_data, struct ctdb_tcp_node);
60 node->ctdb->upcalls->node_dead(node);
63 ctdb_tcp_stop_connection(node);
64 tnode->connect_te = event_add_timed(node->ctdb->ev, tnode,
65 timeval_current_ofs(3, 0),
66 ctdb_tcp_node_connect, node);
70 called when socket becomes writeable on connect
72 static void ctdb_node_connect_write(struct event_context *ev, struct fd_event *fde,
73 uint16_t flags, void *private_data)
75 struct ctdb_node *node = talloc_get_type(private_data,
77 struct ctdb_tcp_node *tnode = talloc_get_type(node->private_data,
78 struct ctdb_tcp_node);
79 struct ctdb_context *ctdb = node->ctdb;
81 socklen_t len = sizeof(error);
84 talloc_free(tnode->connect_te);
85 tnode->connect_te = NULL;
87 if (getsockopt(tnode->fd, SOL_SOCKET, SO_ERROR, &error, &len) != 0 ||
89 ctdb_tcp_stop_connection(node);
90 tnode->connect_te = event_add_timed(ctdb->ev, tnode,
91 timeval_current_ofs(1, 0),
92 ctdb_tcp_node_connect, node);
96 talloc_free(tnode->connect_fde);
97 tnode->connect_fde = NULL;
99 setsockopt(tnode->fd,IPPROTO_TCP,TCP_NODELAY,(char *)&one,sizeof(one));
100 setsockopt(tnode->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
102 ctdb_queue_set_fd(tnode->out_queue, tnode->fd);
104 /* the queue subsystem now owns this fd */
109 static int ctdb_tcp_get_address(struct ctdb_context *ctdb,
110 const char *address, ctdb_sock_addr *addr)
112 if (parse_ip(address, NULL, 0, addr) == 0) {
113 DEBUG(DEBUG_CRIT, (__location__ " Unparsable address : %s.\n", address));
120 called when we should try and establish a tcp connection to a node
122 void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te,
123 struct timeval t, void *private_data)
125 struct ctdb_node *node = talloc_get_type(private_data,
127 struct ctdb_tcp_node *tnode = talloc_get_type(node->private_data,
128 struct ctdb_tcp_node);
129 struct ctdb_context *ctdb = node->ctdb;
130 ctdb_sock_addr sock_in;
133 ctdb_sock_addr sock_out;
135 ctdb_tcp_stop_connection(node);
137 ZERO_STRUCT(sock_out);
138 #ifdef HAVE_SOCK_SIN_LEN
139 sock_out.ip.sin_len = sizeof(sock_out);
141 if (ctdb_tcp_get_address(ctdb, node->address.address, &sock_out) != 0) {
144 switch (sock_out.sa.sa_family) {
146 sock_out.ip.sin_port = htons(node->address.port);
149 sock_out.ip6.sin6_port = htons(node->address.port);
152 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
153 sock_out.sa.sa_family));
157 tnode->fd = socket(sock_out.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
158 set_nonblocking(tnode->fd);
159 set_close_on_exec(tnode->fd);
161 DEBUG(DEBUG_DEBUG, (__location__ " Created TCP SOCKET FD:%d\n", tnode->fd));
163 /* Bind our side of the socketpair to the same address we use to listen
164 * on incoming CTDB traffic.
165 * We must specify this address to make sure that the address we expose to
166 * the remote side is actually routable in case CTDB traffic will run on
167 * a dedicated non-routeable network.
169 ZERO_STRUCT(sock_in);
170 if (ctdb_tcp_get_address(ctdb, ctdb->address.address, &sock_in) != 0) {
171 DEBUG(DEBUG_ERR, (__location__ " Failed to find our address. Failing bind.\n"));
176 /* AIX libs check to see if the socket address and length
177 arguments are consistent with each other on calls like
178 connect(). Can not get by with just sizeof(sock_in),
179 need sizeof(sock_in.ip).
181 switch (sock_in.sa.sa_family) {
183 sockin_size = sizeof(sock_in.ip);
184 sockout_size = sizeof(sock_out.ip);
187 sockin_size = sizeof(sock_in.ip6);
188 sockout_size = sizeof(sock_out.ip6);
191 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
192 sock_in.sa.sa_family));
196 #ifdef HAVE_SOCK_SIN_LEN
197 sock_in.ip.sin_len = sockin_size;
198 sock_out.ip.sin_len = sockout_size;
200 bind(tnode->fd, (struct sockaddr *)&sock_in, sockin_size);
202 if (connect(tnode->fd, (struct sockaddr *)&sock_out, sockout_size) != 0 &&
203 errno != EINPROGRESS) {
204 ctdb_tcp_stop_connection(node);
205 tnode->connect_te = event_add_timed(ctdb->ev, tnode,
206 timeval_current_ofs(1, 0),
207 ctdb_tcp_node_connect, node);
211 /* non-blocking connect - wait for write event */
212 tnode->connect_fde = event_add_fd(node->ctdb->ev, tnode, tnode->fd,
213 EVENT_FD_WRITE|EVENT_FD_READ,
214 ctdb_node_connect_write, node);
216 /* don't give it long to connect - retry in one second. This ensures
217 that we find a node is up quickly (tcp normally backs off a syn reply
218 delay by quite a lot) */
219 tnode->connect_te = event_add_timed(ctdb->ev, tnode, timeval_current_ofs(1, 0),
220 ctdb_tcp_node_connect, node);
224 called when we get contacted by another node
225 currently makes no attempt to check if the connection is really from a ctdb
228 static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde,
229 uint16_t flags, void *private_data)
231 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
232 struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data, struct ctdb_tcp);
236 struct ctdb_incoming *in;
238 const char *incoming_node;
240 memset(&addr, 0, sizeof(addr));
242 fd = accept(ctcp->listen_fd, (struct sockaddr *)&addr, &len);
243 if (fd == -1) return;
245 incoming_node = ctdb_addr_to_str(&addr);
246 nodeid = ctdb_ip_to_nodeid(ctdb, incoming_node);
249 DEBUG(DEBUG_ERR, ("Refused connection from unknown node %s\n", incoming_node));
254 in = talloc_zero(ctcp, struct ctdb_incoming);
258 set_nonblocking(in->fd);
259 set_close_on_exec(in->fd);
261 DEBUG(DEBUG_DEBUG, (__location__ " Created SOCKET FD:%d to incoming ctdb connection\n", fd));
263 setsockopt(in->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
265 in->queue = ctdb_queue_setup(ctdb, in, in->fd, CTDB_TCP_ALIGNMENT,
266 ctdb_tcp_read_cb, in, "ctdbd-%s", incoming_node);
271 automatically find which address to listen on
273 static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
275 struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
279 const char *lock_path = "/tmp/.ctdb_socket_lock";
283 struct tevent_fd *fde;
285 /* in order to ensure that we don't get two nodes with the
286 same adddress, we must make the bind() and listen() calls
287 atomic. The SO_REUSEADDR setsockopt only prevents double
288 binds if the first socket is in LISTEN state */
289 lock_fd = open(lock_path, O_RDWR|O_CREAT, 0666);
291 DEBUG(DEBUG_CRIT,("Unable to open %s\n", lock_path));
295 lock.l_type = F_WRLCK;
296 lock.l_whence = SEEK_SET;
301 if (fcntl(lock_fd, F_SETLKW, &lock) != 0) {
302 DEBUG(DEBUG_CRIT,("Unable to lock %s\n", lock_path));
307 for (i=0; i < ctdb->num_nodes; i++) {
308 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
312 /* if node_ip is specified we will only try to bind to that
315 if (ctdb->node_ip != NULL) {
316 if (strcmp(ctdb->node_ip, ctdb->nodes[i]->address.address)) {
322 if (ctdb_tcp_get_address(ctdb,
323 ctdb->nodes[i]->address.address,
328 switch (sock.sa.sa_family) {
330 sock.ip.sin_port = htons(ctdb->nodes[i]->address.port);
331 sock_size = sizeof(sock.ip);
334 sock.ip6.sin6_port = htons(ctdb->nodes[i]->address.port);
335 sock_size = sizeof(sock.ip6);
338 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
342 #ifdef HAVE_SOCK_SIN_LEN
343 sock.ip.sin_len = sock_size;
346 ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
347 if (ctcp->listen_fd == -1) {
348 ctdb_set_error(ctdb, "socket failed\n");
352 set_close_on_exec(ctcp->listen_fd);
354 setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
356 if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) == 0) {
360 if (errno == EADDRNOTAVAIL) {
361 DEBUG(DEBUG_DEBUG,(__location__ " Failed to bind() to socket. %s(%d)\n",
362 strerror(errno), errno));
364 DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n",
365 strerror(errno), errno));
369 if (i == ctdb->num_nodes) {
370 DEBUG(DEBUG_CRIT,("Unable to bind to any of the node addresses - giving up\n"));
373 ctdb->address.address = talloc_strdup(ctdb, ctdb->nodes[i]->address.address);
374 ctdb->address.port = ctdb->nodes[i]->address.port;
375 ctdb->name = talloc_asprintf(ctdb, "%s:%u",
376 ctdb->address.address,
378 ctdb->pnn = ctdb->nodes[i]->pnn;
379 ctdb->nodes[i]->flags &= ~NODE_FLAGS_DISCONNECTED;
380 DEBUG(DEBUG_INFO,("ctdb chose network address %s:%u pnn %u\n",
381 ctdb->address.address,
384 /* do we start out in DISABLED mode? */
385 if (ctdb->start_as_disabled != 0) {
386 DEBUG(DEBUG_INFO, ("This node is configured to start in DISABLED state\n"));
387 ctdb->nodes[i]->flags |= NODE_FLAGS_DISABLED;
389 /* do we start out in STOPPED mode? */
390 if (ctdb->start_as_stopped != 0) {
391 DEBUG(DEBUG_INFO, ("This node is configured to start in STOPPED state\n"));
392 ctdb->nodes[i]->flags |= NODE_FLAGS_STOPPED;
395 if (listen(ctcp->listen_fd, 10) == -1) {
399 fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
400 ctdb_listen_event, ctdb);
401 tevent_fd_set_auto_close(fde);
408 close(ctcp->listen_fd);
409 ctcp->listen_fd = -1;
415 listen on our own address
417 int ctdb_tcp_listen(struct ctdb_context *ctdb)
419 struct ctdb_tcp *ctcp = talloc_get_type(ctdb->private_data,
424 struct tevent_fd *fde;
426 /* we can either auto-bind to the first available address, or we can
427 use a specified address */
428 if (!ctdb->address.address) {
429 return ctdb_tcp_listen_automatic(ctdb);
433 if (ctdb_tcp_get_address(ctdb, ctdb->address.address,
438 switch (sock.sa.sa_family) {
440 sock.ip.sin_port = htons(ctdb->address.port);
441 sock_size = sizeof(sock.ip);
444 sock.ip6.sin6_port = htons(ctdb->address.port);
445 sock_size = sizeof(sock.ip6);
448 DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
452 #ifdef HAVE_SOCK_SIN_LEN
453 sock.ip.sin_len = sock_size;
456 ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
457 if (ctcp->listen_fd == -1) {
458 ctdb_set_error(ctdb, "socket failed\n");
462 set_close_on_exec(ctcp->listen_fd);
464 setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
466 if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) != 0) {
467 DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n", strerror(errno), errno));
471 if (listen(ctcp->listen_fd, 10) == -1) {
475 fde = event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ,
476 ctdb_listen_event, ctdb);
477 tevent_fd_set_auto_close(fde);
482 if (ctcp->listen_fd != -1) {
483 close(ctcp->listen_fd);
485 ctcp->listen_fd = -1;