automatically clean up left-over Unix domain socket files when trying to bind (fixes #3723)

3 files changed, 84 insertions, 14 deletions

diff --git a/src/arm/gnunet-service-arm.c b/src/arm/gnunet-service-arm.c
index 0dffafeeb..a411546d7 100644
--- a/src/arm/gnunet-service-arm.c
+++ b/src/arm/gnunet-service-arm.c
@@ -603,7 +603,8 @@ accept_connection (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
  * @param sl service entry for the service in question
  */
 static void
-create_listen_socket (struct sockaddr *sa, socklen_t addr_len,
+create_listen_socket (struct sockaddr *sa,
+                      socklen_t addr_len,
                       struct ServiceList *sl)
 {
   static int on = 1;
@@ -652,14 +653,18 @@ create_listen_socket (struct sockaddr *sa, socklen_t addr_len,
     GNUNET_log_strerror (GNUNET_ERROR_TYPE_ERROR | GNUNET_ERROR_TYPE_BULK,
                          "setsockopt");
 #endif
-
+#ifndef WINDOWS
+  if (AF_UNIX == sa->sa_family)
+    GNUNET_NETWORK_unix_precheck ((struct sockaddr_un *) sa);
+#endif
   if (GNUNET_OK !=
       GNUNET_NETWORK_socket_bind (sock, (const struct sockaddr *) sa, addr_len))
   {
     GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
-                _
-                ("Unable to bind listening socket for service `%s' to address `%s': %s\n"),
-                sl->name, GNUNET_a2s (sa, addr_len), STRERROR (errno));
+                _("Unable to bind listening socket for service `%s' to address `%s': %s\n"),
+                sl->name,
+                GNUNET_a2s (sa, addr_len),
+                STRERROR (errno));
     GNUNET_break (GNUNET_OK == GNUNET_NETWORK_socket_close (sock));
     GNUNET_free (sa);
     return;
@@ -1394,11 +1399,11 @@ setup_service (void *cls,
                                             "FORCESTART"))
   {
     sl->force_start = GNUNET_YES;
-    /* FIXME: we might like the pre-binding even for
-       _certain_ services that have force_start set,
-       otherwise interdependencies may again force
-       client's to retry connections during startup. */
-    return;
+    if (GNUNET_YES ==
+        GNUNET_CONFIGURATION_get_value_yesno (cfg,
+                                              section,
+                                              "NOARMBIND"))
+      return;
   }
   else
   {
diff --git a/src/include/gnunet_network_lib.h b/src/include/gnunet_network_lib.h
index f70509830..83d820a5b 100644
--- a/src/include/gnunet_network_lib.h
+++ b/src/include/gnunet_network_lib.h
@@ -106,6 +106,21 @@ char *
 GNUNET_NETWORK_shorten_unixpath (char *unixpath);
 
 
+#ifndef WINDOWS
+/**
+ * If services crash, they can leave a unix domain socket file on the
+ * disk. This needs to be manually removed, because otherwise both
+ * bind() and connect() for the respective address will fail.  In this
+ * function, we test if such a left-over file exists, and if so,
+ * remove it (unless there is a listening service at the address).
+ *
+ * @param un unix domain socket address to check
+ */
+void
+GNUNET_NETWORK_unix_precheck (const struct sockaddr_un *un);
+#endif
+
+
 /**
  * Accept a new connection on a socket.  Configure it for non-blocking
  * IO and mark it as non-inheritable to child processes (set the
diff --git a/src/util/network.c b/src/util/network.c
index d141732c2..489966426 100644
--- a/src/util/network.c
+++ b/src/util/network.c
@@ -147,6 +147,53 @@ GNUNET_NETWORK_shorten_unixpath (char *unixpath)
 }
 
 
+#ifndef WINDOWS
+/**
+ * If services crash, they can leave a unix domain socket file on the
+ * disk. This needs to be manually removed, because otherwise both
+ * bind() and connect() for the respective address will fail.  In this
+ * function, we test if such a left-over file exists, and if so,
+ * remove it (unless there is a listening service at the address).
+ *
+ * @param un unix domain socket address to check
+ */
+void
+GNUNET_NETWORK_unix_precheck (const struct sockaddr_un *un)
+{
+  int s;
+  int eno;
+  struct stat sbuf;
+  int ret;
+
+  s = socket (AF_UNIX, SOCK_STREAM, 0);
+  ret = connect (s,
+                 (struct sockaddr *) un,
+                 sizeof (struct sockaddr_un));
+  eno = errno;
+  GNUNET_break (0 == close (s));
+  if (0 == ret)
+    return; /* another process is listening, do not remove! */
+  if (ECONNREFUSED != eno)
+    return; /* some other error, likely "no such file or directory" -- all well */
+  /* should unlink, but sanity checks first */
+  if (0 != stat (un->sun_path,
+                 &sbuf))
+    return; /* failed to 'stat', likely does not exist after all */
+  if (S_IFSOCK != (S_IFMT & sbuf.st_mode))
+    return; /* refuse to unlink anything except sockets */
+  /* finally, really unlink */
+  GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+              "Removing left-over `%s' from previous exeuction\n",
+              un->sun_path);
+  if (0 != unlink (un->sun_path))
+    GNUNET_log_strerror_file (GNUNET_ERROR_TYPE_WARNING,
+                              "unlink",
+                              un->sun_path);
+}
+#endif
+
+
+
 #ifndef FD_COPY
 #define FD_COPY(s, d) (memcpy ((d), (s), sizeof (fd_set)))
 #endif
@@ -447,6 +494,8 @@ GNUNET_NETWORK_socket_bind (struct GNUNET_NETWORK_Handle *desc,
 #endif
 #endif
 #ifndef WINDOWS
+  if (AF_UNIX == address->sa_family)
+    GNUNET_NETWORK_unix_precheck ((const struct sockaddr_un *) address);
   {
     const int on = 1;
 
@@ -459,8 +508,6 @@ GNUNET_NETWORK_socket_bind (struct GNUNET_NETWORK_Handle *desc,
       LOG_STRERROR (GNUNET_ERROR_TYPE_DEBUG,
                     "setsockopt");
   }
-#endif
-#ifndef WINDOWS
   {
     /* set permissions of newly created non-abstract UNIX domain socket to
        "user-only"; applications can choose to relax this later */
@@ -476,7 +523,10 @@ GNUNET_NETWORK_socket_bind (struct GNUNET_NETWORK_Handle *desc,
       old_mask = umask (S_IWGRP | S_IRGRP | S_IXGRP | S_IWOTH | S_IROTH | S_IXOTH);
 #endif
 
-    ret = bind (desc->fd, address, address_len);
+    ret = bind (desc->fd,
+                address,
+                address_len);
+
 #ifndef WINDOWS
     if (not_abstract)
       (void) umask (old_mask);
@@ -486,7 +536,7 @@ GNUNET_NETWORK_socket_bind (struct GNUNET_NETWORK_Handle *desc,
   if (SOCKET_ERROR == ret)
     SetErrnoFromWinsockError (WSAGetLastError ());
 #endif
-  if (ret != 0)
+  if (0 != ret)
     return GNUNET_SYSERR;
 #ifndef MINGW
   desc->addr = GNUNET_malloc (address_len);