atopacctd: don't hang when running in a linux container (it can't succeed though)

atopacct.service

@@ -5,6 +5,11 @@ Conflicts=psacct.service
 After=syslog.target
 Before=atop.service
 
+# Let's skip on containers here, at least for now.
+# Currently, Linux Containers don't support process accounting
+# (and failed services are painful for Debian-style packages).
+ConditionVirtualization=!container
+
 [Service]
 Type=forking
 PIDFile=/var/run/atopacctd.pid

atopacctd.c

@@ -119,11 +119,12 @@ int		netlink_recv(int, int);		// from netlink.c
 int
 main(int argc, char *argv[])
 {
-	int			i, nfd, afd, sfd;
-	int			parentpid;
+	int			i, rv, nfd, afd, sfd;
 	struct stat		dirstat;
 	struct rlimit		rlim;
 	FILE			*pidf;
+	int			readypipe[2];
+	int8_t			ready;
 
 	struct sembuf		semincr = {0, +1, SEM_UNDO};
 
@@ -261,37 +262,59 @@ main(int argc, char *argv[])
 		}
 	}
 
-	/*
- 	** prepare cleanup signal handler
-	*/
-	memset(&sigcleanup, 0, sizeof sigcleanup);
-	sigcleanup.sa_handler	= cleanup;
-	sigemptyset(&sigcleanup.sa_mask);
-
 	/*
 	** daemonize this process
 	** i.e. be sure that the daemon is no session leader (any more)
 	** and get rid of a possible bad context that might have been
 	** inherited from ancestors
 	*/
-	parentpid = getpid();		// to be killed when initialized
-	(void) sigaction(SIGTERM, &sigcleanup, (struct sigaction *)0);
+	if ( pipe(readypipe) == -1)
+	{
+		perror("cannot create readiness pipe");
+		exit(4);
+	}
+
+	rv = fork();
 
-	if ( fork() )			// implicitly switch to background
-   	{
+	if (rv == -1)
+	{
+		perror("cannot create process");
+		exit(4);
+	}
+
+	if (rv > 0)
+	{
 		/*
- 		** parent after forking first child:
-		** wait for signal 15 from child before terminating
-		** because systemd expects parent to terminate whenever
-		** service is up and running
+		** parent after forking first child:
+		** wait for message from child before terminating
+		** because dependent services need parent to terminate
+		** whenever service is up and running
 		*/
-		pause();		// wait for signal from child
-		exit(0);		// finish parent
+
+		(void) close(readypipe[1]);	// close write end of pipe
+
+		rv = read(readypipe[0], &ready, sizeof(ready));
+
+		if (rv == sizeof(ready))
+			exit(0);		// service is ready now
+		else
+			exit(4);		// service exited
 	}
 
+	(void) close(readypipe[0]);	// close read end of pipe
+	readypipe[0] = -1;
+
 	setsid();			// become session leader to lose ctty
 
-	if ( fork() )			// finish parent; continue in child
+	rv = fork();			// finish parent; continue in child
+
+	if (rv == -1)
+	{
+		perror("cannot create process");
+		exit(4);
+	}
+
+	if (rv > 0)
 		exit(0);		// --> no session leader, no ctty
 
 	getrlimit(RLIMIT_NOFILE, &rlim);
@@ -312,8 +335,7 @@ main(int argc, char *argv[])
 	if ( semop(semprv, &semincr, 1) == -1)
        	{
 		perror("cannot increment private semaphore");
-		kill(parentpid, SIGTERM);
-		exit(4);
+		exit(5);
 	}
 
 	/*
@@ -327,8 +349,7 @@ main(int argc, char *argv[])
 	if ( (afd = creat(accountpath, 0600)) == -1)
        	{
 		perror(accountpath);
-		kill(parentpid, SIGTERM);
-		exit(5);
+		exit(6);
 	}
 
 	(void) close(afd);
@@ -339,20 +360,18 @@ main(int argc, char *argv[])
 	if ( (afd = open(accountpath, O_RDONLY)) == -1)
        	{
 		perror(accountpath);
-		kill(parentpid, SIGTERM);
-		exit(5);
+		exit(6);
 	}
 
 	/*
- 	** create directory to store the shadow files
+	** create directory to store the shadow files
 	*/
 	snprintf(shadowdir, sizeof shadowdir, "%s/%s", pacctdir, PACCTSHADOWD);
 
-	if ( mkdir(shadowdir, 0755) == -1)
+	if ( mkdir(shadowdir, 0755) == -1 && errno != EEXIST)
        	{
 		perror(shadowdir);
-		kill(parentpid, SIGTERM);
-		exit(5);
+		exit(6);
 	}
 
 	sfd = createshadow(curshadow);
@@ -377,8 +396,7 @@ main(int argc, char *argv[])
 	if ( (nfd = netlink_open()) == -1)
 	{
 		(void) unlink(accountpath);
-		kill(parentpid, SIGTERM);
-		exit(5);
+		exit(6);
 	}
 
 	/*
@@ -387,8 +405,7 @@ main(int argc, char *argv[])
 	if ( swonpacct(afd, accountpath) == -1)
 	{
 		(void) unlink(accountpath);
-		kill(parentpid, SIGTERM);
-		exit(6);
+		exit(7);
 	}
 
 	syslog(LOG_INFO, "accounting to %s", accountpath);
@@ -398,6 +415,10 @@ main(int argc, char *argv[])
 	*/
 	(void) signal(SIGHUP, SIG_IGN);
 
+	memset(&sigcleanup, 0, sizeof sigcleanup);
+	sigcleanup.sa_handler	= cleanup;
+	sigemptyset(&sigcleanup.sa_mask);
+
 	(void) sigaction(SIGINT,  &sigcleanup, (struct sigaction *)0);
 	(void) sigaction(SIGQUIT, &sigcleanup, (struct sigaction *)0);
 	(void) sigaction(SIGTERM, &sigcleanup, (struct sigaction *)0);
@@ -412,9 +433,11 @@ main(int argc, char *argv[])
 	}
 
 	/*
-	** terminate parent: service  initialized
+	** notify parent: service  initialized
 	*/
-	kill(parentpid, SIGTERM);
+	ready = 1;
+	(void) write(readypipe[1], &ready, sizeof(ready));
+	(void) close(readypipe[1]);
 
 	/*
 	** main loop

netlink.c

@@ -68,6 +68,7 @@ netlink_open(void)
 		&cpudef, strlen(cpudef)+1) == -1)
 	{
 		fprintf(stderr, "register cpumask failed\n");
+		close(nlsock);
 		return -1;
 	}
 
@@ -108,17 +109,18 @@ nlsock_getfam(int nlsock)
 
         if ( (len = recv(nlsock, &msg, sizeof msg, 0)) == -1)
 	{
-		perror("receive NETLINK family");
+		perror("receive NETLINK family for taskstats");
 		exit(1);
 	}
 
 	if  (msg.n.nlmsg_type == NLMSG_ERROR || !NLMSG_OK(&msg.n, len))
 	{
 		struct nlmsgerr *err = NLMSG_DATA(&msg);
 
-		fprintf(stderr, "receive NETLINK family, errno %d\n",
-                                err->error);
+		/* nlmsgerr holds a negative errno */
+		errno = -err->error;
 
+		perror("get NETLINK family for taskstats");
 		exit(1);
 	}
 
@@ -152,6 +154,7 @@ nlsock_open(void)
 									== -1)
 	{
 		perror("set length receive buffer");
+		close(nlsock);
 		exit(1);
 	}