X-Git-Url: https://git.exim.org/exim.git/blobdiff_plain/55414b25bee9f0195ccd1e47f3d3b5cba766e099..3124effbfe192d3a9f1ee5e887402343cc2dd3de:/src/src/ip.c

diff --git a/src/src/ip.c b/src/src/ip.c
index e4a43e69a..2e8968528 100644
--- a/src/src/ip.c
+++ b/src/src/ip.c
@@ -2,12 +2,12 @@
 *     Exim - an Internet mail transport agent    *
 *************************************************/
 
-/* Copyright (c) University of Cambridge 1995 - 2014 */
+/* Copyright (c) University of Cambridge 1995 - 2018 */
 /* See the file NOTICE for conditions of use and distribution. */
 
 /* Functions for doing things with sockets. With the advent of IPv6 this has
 got messier, so that it's worth pulling out the code into separate functions
-that other parts of Exim can call, expecially as there are now several
+that other parts of Exim can call, especially as there are now several
 different places in the code where sockets are used. */
 
 
@@ -97,24 +97,11 @@ ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr)
 *         Bind socket to interface and port      *
 *************************************************/
 
-/* This function binds a socket to a local interface address and port. For a
-wildcard IPv6 bind, the address is ":".
-
-Arguments:
-  sock           the socket
-  af             AF_INET or AF_INET6 - the socket type
-  address        the IP address, in text form
-  port           the IP port (host order)
-
-Returns:         the result of bind()
-*/
-
 int
-ip_bind(int sock, int af, uschar *address, int port)
+ip_addr(void * sin_, int af, const uschar * address, int port)
 {
-int s_len;
-union sockaddr_46 sin;
-memset(&sin, 0, sizeof(sin));
+union sockaddr_46 * sin = sin_;
+memset(sin, 0, sizeof(*sin));
 
 /* Setup code when using an IPv6 socket. The wildcard address is ":", to
 ensure an IPv6 socket is used. */
@@ -124,15 +111,13 @@ if (af == AF_INET6)
   {
   if (address[0] == ':' && address[1] == 0)
     {
-    sin.v6.sin6_family = AF_INET6;
-    sin.v6.sin6_addr = in6addr_any;
+    sin->v6.sin6_family = AF_INET6;
+    sin->v6.sin6_addr = in6addr_any;
     }
   else
-    {
-    ip_addrinfo(address, &sin.v6);  /* Panic-dies on error */
-    }
-  sin.v6.sin6_port = htons(port);
-  s_len = sizeof(sin.v6);
+    ip_addrinfo(address, &sin->v6);  /* Panic-dies on error */
+  sin->v6.sin6_port = htons(port);
+  return sizeof(sin->v6);
   }
 else
 #else     /* HAVE_IPv6 */
@@ -142,22 +127,59 @@ af = af;  /* Avoid compiler warning */
 /* Setup code when using IPv4 socket. The wildcard address is "". */
 
   {
-  sin.v4.sin_family = AF_INET;
-  sin.v4.sin_port = htons(port);
-  s_len = sizeof(sin.v4);
-  if (address[0] == 0)
-    sin.v4.sin_addr.s_addr = (S_ADDR_TYPE)INADDR_ANY;
-  else
-    sin.v4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CS address);
+  sin->v4.sin_family = AF_INET;
+  sin->v4.sin_port = htons(port);
+  sin->v4.sin_addr.s_addr = address[0] == 0
+    ? (S_ADDR_TYPE)INADDR_ANY
+    : (S_ADDR_TYPE)inet_addr(CS address);
+  return sizeof(sin->v4);
   }
+}
+
+
+
+/* This function binds a socket to a local interface address and port. For a
+wildcard IPv6 bind, the address is ":".
 
-/* Now we can call the bind() function */
+Arguments:
+  sock           the socket
+  af             AF_INET or AF_INET6 - the socket type
+  address        the IP address, in text form
+  port           the IP port (host order)
 
+Returns:         the result of bind()
+*/
+
+int
+ip_bind(int sock, int af, uschar *address, int port)
+{
+union sockaddr_46 sin;
+int s_len = ip_addr(&sin, af, address, port);
 return bind(sock, (struct sockaddr *)&sin, s_len);
 }
 
 
 
+/*************************************************
+*************************************************/
+
+#ifdef EXIM_TFO_PROBE
+void
+tfo_probe(void)
+{
+# ifdef TCP_FASTOPEN
+int sock, backlog = 5;
+
+if (  (sock = socket(SOCK_STREAM, AF_INET, 0)) < 0
+   && setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN, &backlog, sizeof(backlog))
+   )
+  tcp_fastopen_ok = TRUE;
+close(sock);
+# endif
+}
+#endif
+
+
 /*************************************************
 *        Connect socket to remote host           *
 *************************************************/
@@ -173,12 +195,15 @@ Arguments:
   address     the remote address, in text form
   port        the remote port
   timeout     a timeout (zero for indefinite timeout)
+  fastopen_blob    non-null iff TCP_FASTOPEN can be used; may indicate early-data to
+		be sent in SYN segment
 
 Returns:      0 on success; -1 on failure, with errno set
 */
 
 int
-ip_connect(int sock, int af, const uschar *address, int port, int timeout)
+ip_connect(int sock, int af, const uschar *address, int port, int timeout,
+  const blob * fastopen_blob)
 {
 struct sockaddr_in s_in4;
 struct sockaddr *s_ptr;
@@ -216,9 +241,66 @@ IPv6 support. */
 /* If no connection timeout is set, just call connect() without setting a
 timer, thereby allowing the inbuilt OS timeout to operate. */
 
+callout_address = string_sprintf("[%s]:%d", address, port);
 sigalrm_seen = FALSE;
 if (timeout > 0) alarm(timeout);
-rc = connect(sock, s_ptr, s_len);
+
+#if defined(TCP_FASTOPEN) && defined(MSG_FASTOPEN)
+/* TCP Fast Open, if the system has a cookie from a previous call to
+this peer, can send data in the SYN packet.  The peer can send data
+before it gets our ACK of its SYN,ACK - the latter is useful for
+the SMTP banner.  Other (than SMTP) cases of TCP connections can
+possibly use the data-on-syn, so support that too.  */
+
+if (fastopen_blob && tcp_fastopen_ok)
+  {
+  if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len,
+		    MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0)
+	/* seen for with-data, experimental TFO option, with-cookie case */
+	/* seen for with-data, proper TFO opt, with-cookie case */
+    {
+    DEBUG(D_transport|D_v)
+      debug_printf("non-TFO mode connection attempt to %s, %lu data\n",
+	address, (unsigned long)fastopen_blob->len);
+    tcp_out_fastopen = fastopen_blob->len > 0 ?  2 : 1;
+    }
+  else if (errno == EINPROGRESS)	/* expected if we had no cookie for peer */
+	/* seen for no-data, proper TFO option, both cookie-request and with-cookie cases */
+	/*  apparently no visibility of the diffference at this point */
+	/* seen for with-data, proper TFO opt, cookie-req */
+	/*   with netwk delay, post-conn tcp_info sees unacked 1 for R, 2 for C; code in smtp_out.c */
+	/* ? older Experimental TFO option behaviour ? */
+    {					/* queue unsent data */
+    DEBUG(D_transport|D_v) debug_printf("TFO mode sendto, %s data: EINPROGRESS\n",
+      fastopen_blob->len > 0 ? "with"  : "no");
+    if (!fastopen_blob->data)
+      {
+      tcp_out_fastopen = 1;		/* we tried; unknown if useful yet */
+      rc = 0;
+      }
+    else
+      rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0);
+    }
+  else if(errno == EOPNOTSUPP)
+    {
+    DEBUG(D_transport)
+      debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n");
+    goto legacy_connect;
+    }
+  }
+else
+#endif
+  {
+legacy_connect:
+  DEBUG(D_transport|D_v) if (fastopen_blob)
+    debug_printf("non-TFO mode connection attempt to %s, %lu data\n",
+      address, (unsigned long)fastopen_blob->len);
+  if ((rc = connect(sock, s_ptr, s_len)) >= 0)
+    if (  fastopen_blob && fastopen_blob->data && fastopen_blob->len
+       && send(sock, fastopen_blob->data, fastopen_blob->len, 0) < 0)
+	rc = -1;
+  }
+
 save_errno = errno;
 alarm(0);
 
@@ -226,24 +308,22 @@ alarm(0);
 can't think of any other way of doing this. It converts a connection refused
 into a timeout if the timeout is set to 999999. */
 
-if (running_in_test_harness)
+if (running_in_test_harness  && save_errno == ECONNREFUSED && timeout == 999999)
   {
-  if (save_errno == ECONNREFUSED && timeout == 999999)
-    {
-    rc = -1;
-    save_errno = EINTR;
-    sigalrm_seen = TRUE;
-    }
+  rc = -1;
+  save_errno = EINTR;
+  sigalrm_seen = TRUE;
   }
 
 /* Success */
 
-if (rc >= 0) return 0;
+if (rc >= 0)
+  return 0;
 
 /* A failure whose error code is "Interrupted system call" is in fact
 an externally applied timeout if the signal handler has been run. */
 
-errno = (save_errno == EINTR && sigalrm_seen)? ETIMEDOUT : save_errno;
+errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
 return -1;
 }
 
@@ -262,15 +342,17 @@ Arguments:
   address       the remote address, in text form
   portlo,porthi the remote port range
   timeout       a timeout
-  connhost	if not NULL, host_item filled in with connection details
+  connhost	if not NULL, host_item to be filled in with connection details
   errstr        pointer for allocated string on error
+  fastopen_blob	with SOCK_STREAM, if non-null, request TCP Fast Open.
+		Additionally, optional early-data to send
 
 Return:
   socket fd, or -1 on failure (having allocated an error string)
 */
 int
 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
-	int timeout, host_item * connhost, uschar ** errstr)
+      int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen_blob)
 {
 int namelen, port;
 host_item shost;
@@ -289,9 +371,7 @@ namelen = Ustrlen(hostname);
 if (hostname[0] == '[' &&
     hostname[namelen - 1] == ']')
   {
-  uschar * host = string_copy(hostname);
-  host[namelen - 1] = 0;
-  host++;
+  uschar * host = string_copyn(hostname+1, namelen-2);
   if (string_is_ip_address(host, NULL) == 0)
     {
     *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
@@ -303,15 +383,15 @@ if (hostname[0] == '[' &&
 /* Otherwise check for an unadorned IP address */
 
 else if (string_is_ip_address(hostname, NULL) != 0)
-  shost.name = shost.address = string_copy(hostname);
+  shost.name = shost.address = string_copyn(hostname, namelen);
 
 /* Otherwise lookup IP address(es) from the name */
 
 else
   {
-  shost.name = string_copy(hostname);
-  if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, NULL,
-      FALSE) != HOST_FOUND)
+  shost.name = string_copyn(hostname, namelen);
+  if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
+      NULL, FALSE) != HOST_FOUND)
     {
     *errstr = string_sprintf("no IP address found for host %s", shost.name);
     return -1;
@@ -320,11 +400,11 @@ else
 
 /* Try to connect to the server - test each IP till one works */
 
-for (h = &shost; h != NULL; h = h->next)
+for (h = &shost; h; h = h->next)
   {
-  fd = (Ustrchr(h->address, ':') != 0)
-    ? (fd6 < 0) ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
-    : (fd4 < 0) ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
+  fd = Ustrchr(h->address, ':') != 0
+    ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
+    : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
 
   if (fd < 0)
     {
@@ -333,11 +413,12 @@ for (h = &shost; h != NULL; h = h->next)
     }
 
   for(port = portlo; port <= porthi; port++)
-    if (ip_connect(fd, af, h->address, port, timeout) == 0)
+    if (ip_connect(fd, af, h->address, port, timeout, fastopen_blob) == 0)
       {
       if (fd != fd6) close(fd6);
       if (fd != fd4) close(fd4);
-      if (connhost) {
+      if (connhost)
+	{
 	h->port = port;
 	*connhost = *h;
 	connhost->next = NULL;
@@ -354,56 +435,62 @@ bad:
 }
 
 
+/*XXX TFO? */
 int
 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
 {
-  int scan;
-  uschar hostname[256];
-  unsigned int portlow, porthigh;
-
-  /* extract host and port part */
-  scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
-  if ( scan != 3 ) {
-    if ( scan != 2 ) {
-      *errstr = string_sprintf("invalid socket '%s'", hostport);
-      return -1;
+int scan;
+uschar hostname[256];
+unsigned int portlow, porthigh;
+
+/* extract host and port part */
+scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
+if (scan != 3)
+  {
+  if (scan != 2)
+    {
+    *errstr = string_sprintf("invalid socket '%s'", hostport);
+    return -1;
     }
-    porthigh = portlow;
+  porthigh = portlow;
   }
 
-  return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
-			    tmo, NULL, errstr);
+return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
+			  tmo, NULL, errstr, NULL);
 }
 
 int
 ip_unixsocket(const uschar * path, uschar ** errstr)
 {
-  int sock;
-  struct sockaddr_un server;
+int sock;
+struct sockaddr_un server;
 
-  if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) {
-    *errstr = US"can't open UNIX socket.";
-    return -1;
+if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
+  {
+  *errstr = US"can't open UNIX socket.";
+  return -1;
   }
 
-  server.sun_family = AF_UNIX;
-  Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
-  server.sun_path[sizeof(server.sun_path)-1] = '\0';
-  if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0) {
-    int err = errno;
-    (void)close(sock);
-    *errstr =  string_sprintf("unable to connect to UNIX socket (%s): %s",
-		  path, strerror(err));
-    return -1;
-    }
-  return sock;
+callout_address = string_copy(path);
+server.sun_family = AF_UNIX;
+Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
+server.sun_path[sizeof(server.sun_path)-1] = '\0';
+if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
+  {
+  int err = errno;
+  (void)close(sock);
+  *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
+		path, strerror(err));
+  return -1;
+  }
+return sock;
 }
 
 int
 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
 {
-  return *spec == '/'
-    ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
+return *spec == '/'
+  ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
 }
 
 /*************************************************
@@ -425,7 +512,7 @@ ip_keepalive(int sock, const uschar *address, BOOL torf)
 {
 int fodder = 1;
 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
-    (uschar *)(&fodder), sizeof(fodder)) != 0)
+    US (&fodder), sizeof(fodder)) != 0)
   log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
     "failed: %s", torf? "to":"from", address, strerror(errno));
 }
@@ -447,23 +534,22 @@ BOOL
 fd_ready(int fd, int timeout)
 {
 fd_set select_inset;
-struct timeval tv;
 time_t start_recv = time(NULL);
+int time_left = timeout;
 int rc;
 
-if (timeout <= 0)
+if (time_left <= 0)
   {
   errno = ETIMEDOUT;
   return FALSE;
   }
 /* Wait until the socket is ready */
 
-for (;;)
+do
   {
+  struct timeval tv = { .tv_sec = time_left, .tv_usec = 0 };
   FD_ZERO (&select_inset);
   FD_SET (fd, &select_inset);
-  tv.tv_sec = timeout;
-  tv.tv_usec = 0;
 
   /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
   rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
@@ -475,17 +561,17 @@ for (;;)
   Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
   the interrupt not at all rare. Since the timeout is typically more than 2
   minutes, the effect was to block the timeout completely. To prevent this
-  happening again, we do an explicit time test. */
+  happening again, we do an explicit time test and adjust the timeout
+  accordingly */
 
   if (rc < 0 && errno == EINTR)
     {
     DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
-    if (time(NULL) - start_recv < timeout) continue;
-    DEBUG(D_transport) debug_printf("total wait time exceeds timeout\n");
-    }
 
-  /* Handle a timeout, and treat any other select error as a timeout, including
-  an EINTR when we have been in this loop for longer than timeout. */
+    /* Watch out, 'continue' jumps to the condition, not to the loops top */
+    time_left = timeout - (time(NULL) - start_recv);
+    if (time_left > 0) continue;
+    }
 
   if (rc <= 0)
     {
@@ -493,10 +579,10 @@ for (;;)
     return FALSE;
     }
 
-  /* If the socket is ready, break out of the loop. */
-
-  if (FD_ISSET(fd, &select_inset)) break;
+  /* Checking the FD_ISSET is not enough, if we're interrupted, the
+  select_inset may still contain the 'input'. */
   }
+while (rc < 0 || !FD_ISSET(fd, &select_inset));
 return TRUE;
 }
 
@@ -699,13 +785,9 @@ while (last > first)
     return TRUE;
     }
   else if (c > 0)
-    {
     first = middle + 1;
-    }
   else
-    {
     last = middle;
-    }
   }
 return FALSE;
 }
@@ -720,3 +802,5 @@ for (i=0; i < dscp_table_size; ++i)
 
 
 /* End of ip.c */
+/* vi: aw ai sw=2
+*/