ctdb-transport: Replace ctdb_logging.h with common/logging.h
[obnox/samba/samba-obnox.git] / ctdb / tcp / tcp_connect.c
index 9d28d48a1f0d5a2c155bfd9b964b2d94965746c2..fbb17c8ab09d6c62284814c2c1f3745de62d750d 100644 (file)
    along with this program; if not, see <http://www.gnu.org/licenses/>.
 */
 
-#include "includes.h"
-#include "lib/events/events.h"
-#include "lib/tdb/include/tdb.h"
+#include "replace.h"
 #include "system/network.h"
 #include "system/filesys.h"
-#include "../include/ctdb_private.h"
+
+#include <talloc.h>
+#include <tevent.h>
+
+#include "lib/util/debug.h"
+#include "lib/util/time.h"
+
+#include "ctdb_private.h"
+
+#include "common/system.h"
+#include "common/common.h"
+#include "common/logging.h"
+
 #include "ctdb_tcp.h"
 
 /*
@@ -61,15 +71,16 @@ void ctdb_tcp_tnode_cb(uint8_t *data, size_t cnt, void *private_data)
        }
 
        ctdb_tcp_stop_connection(node);
-       tnode->connect_te = event_add_timed(node->ctdb->ev, tnode,
-                                           timeval_current_ofs(3, 0),
-                                           ctdb_tcp_node_connect, node);
+       tnode->connect_te = tevent_add_timer(node->ctdb->ev, tnode,
+                                            timeval_current_ofs(3, 0),
+                                            ctdb_tcp_node_connect, node);
 }
 
 /*
   called when socket becomes writeable on connect
 */
-static void ctdb_node_connect_write(struct event_context *ev, struct fd_event *fde, 
+static void ctdb_node_connect_write(struct tevent_context *ev,
+                                   struct tevent_fd *fde,
                                    uint16_t flags, void *private_data)
 {
        struct ctdb_node *node = talloc_get_type(private_data,
@@ -87,7 +98,7 @@ static void ctdb_node_connect_write(struct event_context *ev, struct fd_event *f
        if (getsockopt(tnode->fd, SOL_SOCKET, SO_ERROR, &error, &len) != 0 ||
            error != 0) {
                ctdb_tcp_stop_connection(node);
-               tnode->connect_te = event_add_timed(ctdb->ev, tnode, 
+               tnode->connect_te = tevent_add_timer(ctdb->ev, tnode,
                                                    timeval_current_ofs(1, 0),
                                                    ctdb_tcp_node_connect, node);
                return;
@@ -96,8 +107,14 @@ static void ctdb_node_connect_write(struct event_context *ev, struct fd_event *f
        talloc_free(tnode->connect_fde);
        tnode->connect_fde = NULL;
 
-        setsockopt(tnode->fd,IPPROTO_TCP,TCP_NODELAY,(char *)&one,sizeof(one));
-        setsockopt(tnode->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
+        if (setsockopt(tnode->fd,IPPROTO_TCP,TCP_NODELAY,(char *)&one,sizeof(one)) == -1) {
+               DEBUG(DEBUG_WARNING, ("Failed to set TCP_NODELAY on fd - %s\n",
+                                     strerror(errno)));
+       }
+        if (setsockopt(tnode->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one)) == -1) {
+               DEBUG(DEBUG_WARNING, ("Failed to set KEEPALIVE on fd - %s\n",
+                                     strerror(errno)));
+       }
 
        ctdb_queue_set_fd(tnode->out_queue, tnode->fd);
 
@@ -106,20 +123,10 @@ static void ctdb_node_connect_write(struct event_context *ev, struct fd_event *f
 }
 
 
-static int ctdb_tcp_get_address(struct ctdb_context *ctdb,
-                               const char *address, ctdb_sock_addr *addr)
-{
-       if (parse_ip(address, NULL, 0, addr) == 0) {
-               DEBUG(DEBUG_CRIT, (__location__ " Unparsable address : %s.\n", address));
-               return -1;
-       }
-       return 0;
-}
-
 /*
   called when we should try and establish a tcp connection to a node
 */
-void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te, 
+void ctdb_tcp_node_connect(struct tevent_context *ev, struct tevent_timer *te,
                           struct timeval t, void *private_data)
 {
        struct ctdb_node *node = talloc_get_type(private_data,
@@ -134,40 +141,25 @@ void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te,
 
        ctdb_tcp_stop_connection(node);
 
-       ZERO_STRUCT(sock_out);
-#ifdef HAVE_SOCK_SIN_LEN
-       sock_out.ip.sin_len = sizeof(sock_out);
-#endif
-       if (ctdb_tcp_get_address(ctdb, node->address.address, &sock_out) != 0) {
-               return;
-       }
-       switch (sock_out.sa.sa_family) {
-       case AF_INET:
-               sock_out.ip.sin_port = htons(node->address.port);
-               break;
-       case AF_INET6:
-               sock_out.ip6.sin6_port = htons(node->address.port);
-               break;
-       default:
-               DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
-                       sock_out.sa.sa_family));
-               return;
-       }
+       sock_out = node->address;
 
        tnode->fd = socket(sock_out.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
+       if (tnode->fd == -1) {
+               DEBUG(DEBUG_ERR, (__location__ "Failed to create socket\n"));
+               return;
+       }
        set_nonblocking(tnode->fd);
        set_close_on_exec(tnode->fd);
 
+       DEBUG(DEBUG_DEBUG, (__location__ " Created TCP SOCKET FD:%d\n", tnode->fd));
+
        /* Bind our side of the socketpair to the same address we use to listen
         * on incoming CTDB traffic.
         * We must specify this address to make sure that the address we expose to
         * the remote side is actually routable in case CTDB traffic will run on
         * a dedicated non-routeable network.
         */
-       ZERO_STRUCT(sock_in);
-       if (ctdb_tcp_get_address(ctdb, ctdb->address.address, &sock_in) != 0) {
-               return;
-       }
+       sock_in = *ctdb->address;
 
        /* AIX libs check to see if the socket address and length
           arguments are consistent with each other on calls like
@@ -176,43 +168,49 @@ void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te,
        */
        switch (sock_in.sa.sa_family) {
        case AF_INET:
+               sock_in.ip.sin_port = 0 /* Any port */;
                sockin_size = sizeof(sock_in.ip);
                sockout_size = sizeof(sock_out.ip);
                break;
        case AF_INET6:
+               sock_in.ip6.sin6_port = 0 /* Any port */;
                sockin_size = sizeof(sock_in.ip6);
                sockout_size = sizeof(sock_out.ip6);
                break;
        default:
                DEBUG(DEBUG_ERR, (__location__ " unknown family %u\n",
                        sock_in.sa.sa_family));
+               close(tnode->fd);
+               return;
+       }
+
+       if (bind(tnode->fd, (struct sockaddr *)&sock_in, sockin_size) == -1) {
+               DEBUG(DEBUG_ERR, (__location__ "Failed to bind socket %s(%d)\n",
+                                 strerror(errno), errno));
+               close(tnode->fd);
                return;
        }
-#ifdef HAVE_SOCK_SIN_LEN
-       sock_in.ip.sin_len = sockin_size;
-       sock_out.ip.sin_len = sockout_size;
-#endif
-       bind(tnode->fd, (struct sockaddr *)&sock_in, sockin_size);
 
        if (connect(tnode->fd, (struct sockaddr *)&sock_out, sockout_size) != 0 &&
            errno != EINPROGRESS) {
                ctdb_tcp_stop_connection(node);
-               tnode->connect_te = event_add_timed(ctdb->ev, tnode, 
-                                                   timeval_current_ofs(1, 0),
-                                                   ctdb_tcp_node_connect, node);
+               tnode->connect_te = tevent_add_timer(ctdb->ev, tnode,
+                                                    timeval_current_ofs(1, 0),
+                                                    ctdb_tcp_node_connect, node);
                return;
        }
 
        /* non-blocking connect - wait for write event */
-       tnode->connect_fde = event_add_fd(node->ctdb->ev, tnode, tnode->fd, 
-                                         EVENT_FD_WRITE|EVENT_FD_READ, 
-                                         ctdb_node_connect_write, node);
+       tnode->connect_fde = tevent_add_fd(node->ctdb->ev, tnode, tnode->fd,
+                                          TEVENT_FD_WRITE|TEVENT_FD_READ,
+                                          ctdb_node_connect_write, node);
 
        /* don't give it long to connect - retry in one second. This ensures
           that we find a node is up quickly (tcp normally backs off a syn reply
           delay by quite a lot) */
-       tnode->connect_te = event_add_timed(ctdb->ev, tnode, timeval_current_ofs(1, 0), 
-                                           ctdb_tcp_node_connect, node);
+       tnode->connect_te = tevent_add_timer(ctdb->ev, tnode,
+                                            timeval_current_ofs(1, 0),
+                                            ctdb_tcp_node_connect, node);
 }
 
 /*
@@ -220,7 +218,7 @@ void ctdb_tcp_node_connect(struct event_context *ev, struct timed_event *te,
   currently makes no attempt to check if the connection is really from a ctdb
   node in our cluster
 */
-static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde, 
+static void ctdb_listen_event(struct tevent_context *ev, struct tevent_fd *fde,
                              uint16_t flags, void *private_data)
 {
        struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
@@ -230,18 +228,16 @@ static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde,
        int fd, nodeid;
        struct ctdb_incoming *in;
        int one = 1;
-       const char *incoming_node;
 
        memset(&addr, 0, sizeof(addr));
        len = sizeof(addr);
        fd = accept(ctcp->listen_fd, (struct sockaddr *)&addr, &len);
        if (fd == -1) return;
 
-       incoming_node = ctdb_addr_to_str(&addr);
-       nodeid = ctdb_ip_to_nodeid(ctdb, incoming_node);
+       nodeid = ctdb_ip_to_nodeid(ctdb, &addr);
 
        if (nodeid == -1) {
-               DEBUG(DEBUG_ERR, ("Refused connection from unknown node %s\n", incoming_node));
+               DEBUG(DEBUG_ERR, ("Refused connection from unknown node %s\n", ctdb_addr_to_str(&addr)));
                close(fd);
                return;
        }
@@ -253,10 +249,15 @@ static void ctdb_listen_event(struct event_context *ev, struct fd_event *fde,
        set_nonblocking(in->fd);
        set_close_on_exec(in->fd);
 
-        setsockopt(in->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one));
+       DEBUG(DEBUG_DEBUG, (__location__ " Created SOCKET FD:%d to incoming ctdb connection\n", fd));
 
-       in->queue = ctdb_queue_setup(ctdb, in, in->fd, CTDB_TCP_ALIGNMENT, 
-                                    ctdb_tcp_read_cb, in);
+        if (setsockopt(in->fd,SOL_SOCKET,SO_KEEPALIVE,(char *)&one,sizeof(one)) == -1) {
+               DEBUG(DEBUG_WARNING, ("Failed to set KEEPALIVE on fd - %s\n",
+                                     strerror(errno)));
+       }
+
+       in->queue = ctdb_queue_setup(ctdb, in, in->fd, CTDB_TCP_ALIGNMENT,
+                                    ctdb_tcp_read_cb, in, "ctdbd-%s", ctdb_addr_to_str(&addr));
 }
 
 
@@ -269,10 +270,20 @@ static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
                                                struct ctdb_tcp);
         ctdb_sock_addr sock;
        int lock_fd, i;
-       const char *lock_path = "/tmp/.ctdb_socket_lock";
+       const char *lock_path = CTDB_RUNDIR "/.socket_lock";
        struct flock lock;
        int one = 1;
        int sock_size;
+       struct tevent_fd *fde;
+
+       /* If there are no nodes, then it won't be possible to find
+        * the first one.  Log a failure and short circuit the whole
+        * process.
+        */
+       if (ctdb->num_nodes == 0) {
+               DEBUG(DEBUG_CRIT,("No nodes available to attempt bind to - is the nodes file empty?\n"));
+               return -1;
+       }
 
        /* in order to ensure that we don't get two nodes with the
           same adddress, we must make the bind() and listen() calls
@@ -296,30 +307,17 @@ static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
                return -1;
        }
 
-       for (i=0;i<ctdb->num_nodes;i++) {
-               /* if node_ip is specified we will only try to bind to that
-                  ip.
-               */
-               if (ctdb->node_ip != NULL) {
-                       if (strcmp(ctdb->node_ip, ctdb->nodes[i]->address.address)) {
-                               continue;
-                       }
-               }
-
-               ZERO_STRUCT(sock);
-               if (ctdb_tcp_get_address(ctdb,
-                               ctdb->nodes[i]->address.address, 
-                               &sock) != 0) {
+       for (i=0; i < ctdb->num_nodes; i++) {
+               if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
                        continue;
                }
-       
+               sock = ctdb->nodes[i]->address;
+
                switch (sock.sa.sa_family) {
                case AF_INET:
-                       sock.ip.sin_port = htons(ctdb->nodes[i]->address.port);
                        sock_size = sizeof(sock.ip);
                        break;
                case AF_INET6:
-                       sock.ip6.sin6_port = htons(ctdb->nodes[i]->address.port);
                        sock_size = sizeof(sock.ip6);
                        break;
                default:
@@ -327,9 +325,6 @@ static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
                                sock.sa.sa_family));
                        continue;
                }
-#ifdef HAVE_SOCK_SIN_LEN
-               sock.ip.sin_len = sock_size;
-#endif
 
                ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
                if (ctcp->listen_fd == -1) {
@@ -339,49 +334,69 @@ static int ctdb_tcp_listen_automatic(struct ctdb_context *ctdb)
 
                set_close_on_exec(ctcp->listen_fd);
 
-               setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
+               if (setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,
+                              (char *)&one,sizeof(one)) == -1) {
+                       DEBUG(DEBUG_WARNING, ("Failed to set REUSEADDR on fd - %s\n",
+                                             strerror(errno)));
+               }
 
                if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) == 0) {
-                       DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n", strerror(errno), errno));
                        break;
                }
+
+               if (errno == EADDRNOTAVAIL) {
+                       DEBUG(DEBUG_DEBUG,(__location__ " Failed to bind() to socket. %s(%d)\n",
+                                       strerror(errno), errno));
+               } else {
+                       DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n",
+                                       strerror(errno), errno));
+               }
+
+               close(ctcp->listen_fd);
+               ctcp->listen_fd = -1;
        }
-       
+
        if (i == ctdb->num_nodes) {
                DEBUG(DEBUG_CRIT,("Unable to bind to any of the node addresses - giving up\n"));
                goto failed;
        }
-       ctdb->address.address = talloc_strdup(ctdb, ctdb->nodes[i]->address.address);
-       ctdb->address.port    = ctdb->nodes[i]->address.port;
-       ctdb->name = talloc_asprintf(ctdb, "%s:%u", 
-                                    ctdb->address.address, 
-                                    ctdb->address.port);
-       ctdb->pnn = ctdb->nodes[i]->pnn;
-       ctdb->nodes[i]->flags &= ~NODE_FLAGS_DISCONNECTED;
-       DEBUG(DEBUG_INFO,("ctdb chose network address %s:%u pnn %u\n", 
-                ctdb->address.address, 
-                ctdb->address.port, 
-                ctdb->pnn));
-       /* do we start out in DISABLED mode? */
-       if (ctdb->start_as_disabled != 0) {
-               DEBUG(DEBUG_INFO, ("This node is configured to start in DISABLED state\n"));
-               ctdb->nodes[i]->flags |= NODE_FLAGS_DISABLED;
+       ctdb->address = talloc_memdup(ctdb,
+                                     &ctdb->nodes[i]->address,
+                                     sizeof(ctdb_sock_addr));
+       if (ctdb->address == NULL) {
+               ctdb_set_error(ctdb, "Out of memory at %s:%d",
+                              __FILE__, __LINE__);
+               goto failed;
        }
-       
+
+       ctdb->name = talloc_asprintf(ctdb, "%s:%u",
+                                    ctdb_addr_to_str(ctdb->address),
+                                    ctdb_addr_to_port(ctdb->address));
+       if (ctdb->name == NULL) {
+               ctdb_set_error(ctdb, "Out of memory at %s:%d",
+                              __FILE__, __LINE__);
+               goto failed;
+       }
+       DEBUG(DEBUG_INFO,("ctdb chose network address %s\n", ctdb->name));
+
        if (listen(ctcp->listen_fd, 10) == -1) {
                goto failed;
        }
 
-       event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE, 
-                    ctdb_listen_event, ctdb);  
+       fde = tevent_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, TEVENT_FD_READ,
+                           ctdb_listen_event, ctdb);
+       tevent_fd_set_auto_close(fde);
 
        close(lock_fd);
+
        return 0;
-       
+
 failed:
        close(lock_fd);
-       close(ctcp->listen_fd);
-       ctcp->listen_fd = -1;
+       if (ctcp->listen_fd != -1) {
+               close(ctcp->listen_fd);
+               ctcp->listen_fd = -1;
+       }
        return -1;
 }
 
@@ -396,26 +411,21 @@ int ctdb_tcp_listen(struct ctdb_context *ctdb)
         ctdb_sock_addr sock;
        int sock_size;
        int one = 1;
+       struct tevent_fd *fde;
 
        /* we can either auto-bind to the first available address, or we can
           use a specified address */
-       if (!ctdb->address.address) {
+       if (!ctdb->address) {
                return ctdb_tcp_listen_automatic(ctdb);
        }
 
-       ZERO_STRUCT(sock);
-       if (ctdb_tcp_get_address(ctdb, ctdb->address.address, 
-                                &sock) != 0) {
-               goto failed;
-       }
-       
+       sock = *ctdb->address;
+
        switch (sock.sa.sa_family) {
        case AF_INET:
-               sock.ip.sin_port = htons(ctdb->address.port);
                sock_size = sizeof(sock.ip);
                break;
        case AF_INET6:
-               sock.ip6.sin6_port = htons(ctdb->address.port);
                sock_size = sizeof(sock.ip6);
                break;
        default:
@@ -423,9 +433,6 @@ int ctdb_tcp_listen(struct ctdb_context *ctdb)
                        sock.sa.sa_family));
                goto failed;
        }
-#ifdef HAVE_SOCK_SIN_LEN
-       sock.ip.sin_len = sock_size;
-#endif
 
        ctcp->listen_fd = socket(sock.sa.sa_family, SOCK_STREAM, IPPROTO_TCP);
        if (ctcp->listen_fd == -1) {
@@ -435,7 +442,10 @@ int ctdb_tcp_listen(struct ctdb_context *ctdb)
 
        set_close_on_exec(ctcp->listen_fd);
 
-        setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one));
+        if (setsockopt(ctcp->listen_fd,SOL_SOCKET,SO_REUSEADDR,(char *)&one,sizeof(one)) == -1) {
+               DEBUG(DEBUG_WARNING, ("Failed to set REUSEADDR on fd - %s\n",
+                                     strerror(errno)));
+       }
 
        if (bind(ctcp->listen_fd, (struct sockaddr * )&sock, sock_size) != 0) {
                DEBUG(DEBUG_ERR,(__location__ " Failed to bind() to socket. %s(%d)\n", strerror(errno), errno));
@@ -446,8 +456,9 @@ int ctdb_tcp_listen(struct ctdb_context *ctdb)
                goto failed;
        }
 
-       event_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE, 
-                    ctdb_listen_event, ctdb);  
+       fde = tevent_add_fd(ctdb->ev, ctcp, ctcp->listen_fd, TEVENT_FD_READ,
+                           ctdb_listen_event, ctdb);
+       tevent_fd_set_auto_close(fde);
 
        return 0;