Terminate server process if serval.pid clobbered

commandline.c

@@ -33,36 +33,6 @@ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
 #include "serval.h"
 #include "rhizome.h"
 
-/** Return the PID of the currently running server process, return 0 if there is none.
- */
-static int servalNodeRunning()
-{
-  const char *instancepath = serval_instancepath();
-  struct stat st;
-  if (stat(instancepath, &st) == -1)
-    return setReason(
-	"Instance path '%s' non existant or not accessable: %s [errno=%d]"
-	" (Set SERVALINSTANCE_PATH to specify an alternate location)",
-	instancepath, strerror(errno), errno
-      );
-  if ((st.st_mode & S_IFMT) != S_IFDIR)
-    return setReason("Instance path '%s' is not a directory", instancepath);
-  char filename[1024];
-  if (!FORM_SERVAL_INSTANCE_PATH(filename, "serval.pid"))
-    return -1;
-  FILE *f = NULL;
-  if ((f = fopen(filename, "r"))) {
-    char buf[20];
-    fgets(buf, sizeof buf, f);
-    fclose(f);
-    int pid = atoi(buf);
-    if (pid > 0 && kill(pid, 0) != -1)
-      return pid;
-    unlink(filename);
-  }
-  return 0;
-}
-
 int cli_usage() {
   fprintf(stderr,"\nServal Mesh version <version>.\n");
   fprintf(stderr,"Usage:\n");
@@ -589,7 +559,7 @@ int app_server_start(int argc, const char *const *argv, struct command_line_opti
   if (!interfaces[0])
     WHY("No network interfaces configured (empty 'interfaces' config setting)");
   overlay_interface_args(interfaces);
-  int pid = servalNodeRunning();
+  int pid = server_pid();
   if (pid < 0)
     return -1;
   int ret = 1;
@@ -631,7 +601,7 @@ int app_server_start(int argc, const char *const *argv, struct command_line_opti
 	  happening. */
 	time_t timeout = time(NULL) + 5;
 	int rpid;
-	while (time(NULL) < timeout && (rpid = servalNodeRunning()) == 0)
+	while (time(NULL) < timeout && (rpid = server_pid()) == 0)
 	    usleep(200000); // 5 Hz
 	if (rpid == -1)
 	  return -1;
@@ -662,7 +632,7 @@ int app_server_stop(int argc, const char *const *argv, struct command_line_optio
   if (cli_arg(argc, argv, o, "instance path", &thisinstancepath, cli_absolute_path, NULL) == -1)
     return -1;
   const char *instancepath = serval_instancepath();
-  int pid = servalNodeRunning();
+  int pid = server_pid();
   if (pid < 0)
     return -1;
   cli_puts("instancepath");
@@ -705,7 +675,7 @@ int app_server_stop(int argc, const char *const *argv, struct command_line_optio
       /* Allow a few seconds for the process to die, and keep an eye on things while this is
 	 happening. */
       time_t timeout = time(NULL) + 2;
-      while (time(NULL) < timeout && servalNodeRunning() == pid)
+      while (time(NULL) < timeout && server_pid() == pid)
 	usleep(200000); // 5 Hz
     }
     cli_puts("tries");
@@ -720,7 +690,7 @@ int app_server_status(int argc, const char *const *argv, struct command_line_opt
 {
   if (cli_arg(argc, argv, o, "instance path", &thisinstancepath, cli_absolute_path, NULL) == -1)
     return -1;
-  int pid = servalNodeRunning();
+  int pid = server_pid();
   if (pid < 0)
     return -1;
   cli_puts("instancepath");

dna.c

@@ -317,18 +317,12 @@ void signal_handler( int signal ) {
       exit(0);
     case SIGHUP:
     case SIGINT:
-      /* Shut down.
-	The shutting down should be done from the main-line code rather than here,
-	so we first try to tell the mainline code to do so.  If, however, this is
-	not the first time we have been asked to shut down, then we will do it here. */
-      if (servalShutdown) {
-	/* We have been asked before, so shut down cleanly */
-	serverCleanUp();
-	exit(0);
-      } else {
-	WHY("Asking Serval process to shutdown cleanly");
-	servalShutdown=1;
-      }
+      /* Terminate the server process.  The shutting down should be done from the main-line code
+	 rather than here, so we first try to tell the mainline code to do so.  If, however, this is
+	 not the first time we have been asked to shut down, then we will do it here. */
+      server_shutdown_check();
+      WHY("Asking Serval process to shutdown cleanly");
+      servalShutdown = 1;
       return;
   }
 

overlay.c

@@ -78,6 +78,8 @@ keyring_file *keyring=NULL;
 
 int overlayServerMode()
 {
+  long long server_pid_time_ms = overlay_gettime_ms();
+
   /* In overlay mode we need to listen to all of our sockets, and also to
      send periodic traffic. This means we need to */
   fprintf(stderr,"Running in overlay mode.\n");
@@ -118,13 +120,10 @@ int overlayServerMode()
   /* Get rhizome server started BEFORE populating fd list so that
      the server's listen socket is in the list for poll() */
   if (rhizome_datastore_path) rhizome_server_poll();
-    
+
   while(1) {
 
-    if (servalShutdown) {
-      serverCleanUp();
-      exit(0);
-    }
+    server_shutdown_check();
 
     /* Work out how long we can wait before we need to tick */
     long long ms=overlay_time_until_next_tick();

serval.h

@@ -397,7 +397,9 @@ int respondSimple(keyring_identity *id,
 int requestItem(char *did,char *sid,char *item,int instance,unsigned char *buffer,int buffer_length,int *len,
 		unsigned char *transaction_id);
 int requestNewHLR(char *did,char *pin,char *sid,int recvttl,struct sockaddr *recvaddr);
+int server_pid();
 int server(char *backing_file);
+void server_shutdown_check();
 int isTransactionInCache(unsigned char *transaction_id);
 void insertTransactionInCache(unsigned char *transaction_id);
 

server.c

@@ -23,6 +23,7 @@ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
 
 #include "serval.h"
 
+static int server_getpid = 0;
 unsigned char *hlr=NULL;
 int hlr_size=0;
 
@@ -97,6 +98,35 @@ int recvwithttl(int sock,unsigned char *buffer,int bufferlen,int *ttl,
   return len;
 }
 
+/** Return the PID of the currently running server process, return 0 if there is none.
+ */
+int server_pid()
+{
+  const char *instancepath = serval_instancepath();
+  struct stat st;
+  if (stat(instancepath, &st) == -1)
+    return setReason(
+	"Instance path '%s' non existant or not accessable: %s [errno=%d]"
+	" (Set SERVALINSTANCE_PATH to specify an alternate location)",
+	instancepath, strerror(errno), errno
+      );
+  if ((st.st_mode & S_IFMT) != S_IFDIR)
+    return setReason("Instance path '%s' is not a directory", instancepath);
+  char filename[1024];
+  if (!FORM_SERVAL_INSTANCE_PATH(filename, "serval.pid"))
+    return -1;
+  FILE *f = NULL;
+  if ((f = fopen(filename, "r"))) {
+    char buf[20];
+    fgets(buf, sizeof buf, f);
+    fclose(f);
+    int pid = atoi(buf);
+    if (pid > 0 && kill(pid, 0) != -1)
+      return pid;
+    unlink(filename);
+  }
+  return 0;
+}
 
 int server(char *backing_file)
 {
@@ -122,7 +152,7 @@ int server(char *backing_file)
       getKeyring(backing_file);
     }
   
-  /* Record PID */
+  /* Record PID to advertise that the server is now running */
   char filename[1024];
   if (!FORM_SERVAL_INSTANCE_PATH(filename, "serval.pid"))
     return -1;
@@ -132,7 +162,8 @@ int server(char *backing_file)
     perror("fopen");
     return -1;
   }
-  fprintf(f,"%d\n",getpid());
+  server_getpid = getpid();
+  fprintf(f,"%d\n", server_getpid);
   fclose(f);
 
   if (!overlayMode) simpleServerMode();
@@ -141,6 +172,28 @@ int server(char *backing_file)
   return 0;
 }
 
+/* Called periodically by the server process in its main loop.
+ */
+void server_shutdown_check()
+{
+  static long long server_pid_time_ms = 0;
+  if (servalShutdown) {
+    serverCleanUp();
+    exit(0);
+  }
+  /* If this server has been supplanted with another or Serval has been uninstalled, then its PID
+      file will change or be unaccessible.  In this case, shut down without all the cleanup.
+      Perform this check at most once per second.  */
+  long long time_ms = overlay_gettime_ms();
+  if (server_pid_time_ms == 0 || time_ms - server_pid_time_ms > 1000) {
+    server_pid_time_ms = time_ms;
+    if (server_pid() != server_getpid) {
+      WHYF("Server pid file no longer contains my pid=%d -- shutting down without cleanup", server_pid);
+      exit(1);
+    }
+  }
+}
+
 int getKeyring(char *backing_file)
 {
  if (!backing_file)