get_tasks: don't hang proxy if tasks are exiting

In both GetTasks and GetTasksRecursive in the proxy, switch to a 1-second
select plus recvmsg(MSG_DONTWAIT) when waiting for pids from the cgmanager.
if the cgmanager sees a credential send failed, then it will resend the
first pid which succeeded, so that the proxy just gets a dup.  If no previous
pid has succeeded, or if that pid has now also exited, then we'll simply exit,
and the proxy will now gracefully fail.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

access_checks.c

@@ -107,8 +107,11 @@ int send_creds(int sock, struct ucred *cred)
 	msg.msg_iovlen = 1;
 
 	if (sendmsg(sock, &msg, 0) < 0) {
+		int saved_errno = errno;
 		nih_error("%s: failed at sendmsg: %s", __func__,
 			  strerror(errno));
+		if (saved_errno == 3)
+			return -3;
 		return -1;
 	}
 	return 0;
@@ -124,6 +127,8 @@ int send_creds(int sock, struct ucred *cred)
 void get_scm_creds_sync(int sock, struct ucred *cred)
 {
 	struct msghdr msg = { 0 };
+	struct timeval tv;
+	fd_set rfds;
 	struct iovec iov;
 	struct cmsghdr *cmsg;
 	char cmsgbuf[CMSG_SPACE(sizeof(*cred))];
@@ -155,10 +160,15 @@ void get_scm_creds_sync(int sock, struct ucred *cred)
 	msg.msg_iov = &iov;
 	msg.msg_iovlen = 1;
 
-	// retry logic is not ideal, especially as we are not
-	// threaded.  Sleep at most 1 second waiting for the client
-	// to send us the scm_cred
-	ret = recvmsg(sock, &msg, 0);
+	FD_ZERO(&rfds);
+	FD_SET(sock, &rfds);
+
+	tv.tv_sec = 1;
+	tv.tv_usec = 0;
+	if (select(sock+1, &rfds, NULL, NULL, &tv) < 0) {
+		return;
+	}
+	ret = recvmsg(sock, &msg, MSG_DONTWAIT);
 	if (ret < 0) {
 		nih_error("Failed to receive scm_cred: %s",
 			  strerror(errno));

cgmanager-proxy.c

@@ -867,7 +867,7 @@ int get_tasks_main (void *parent, const char *controller, const char *cgroup,
 	for (i=0; i<nrpids; i++) {
 		get_scm_creds_sync(sv[0], &tcred);
 		if (tcred.pid == -1) {
-			nih_error("%s: Failed getting pid from server",
+			nih_warn("%s: Failed getting pid from server",
 				__func__);
 			goto out;
 		}
@@ -943,7 +943,7 @@ int get_tasks_recursive_main (void *parent, const char *controller,
 	for (i=0; i<nrpids; i++) {
 		get_scm_creds_sync(sv[0], &tcred);
 		if (tcred.pid == -1) {
-			nih_error("%s: Failed getting pid from server",
+			nih_warn("%s: Failed getting pid from server",
 				__func__);
 			goto out;
 		}

frontend.c

@@ -1168,11 +1168,15 @@ int cgmanager_remove (void *data, NihDBusMessage *message, const char *controlle
 	return ret;
 }
 
-/* get_tasks - list tasks for a single cgroup */
+/*
+ * get_tasks - list tasks for a single cgroup
+ * returns -1 on error, 0 on success.
+ */
 void get_tasks_scm_complete(struct scm_sock_data *data)
 {
 	struct ucred pcred;
 	int i, ret;
+	pid_t firstvalid = -1;
 	int32_t *pids, nrpids;
 	ret = get_tasks_main(data, data->controller, data->cgroup,
 			data->pcred, data->rcred, &pids);
@@ -1187,12 +1191,25 @@ void get_tasks_scm_complete(struct scm_sock_data *data)
 		return;
 	}
 	pcred.uid = 0; pcred.gid = 0;
-	for (i=0; i<ret; i++) {
+	for (i=0; i<nrpids; i++) {
 		pcred.pid = pids[i];
-		if (send_creds(data->fd, &pcred)) {
-			nih_error("get_tasks_scm: error writing pids back to client");
+again:
+		ret = send_creds(data->fd, &pcred);
+		if (ret == -3) {
+			if (firstvalid == -1 || firstvalid == pcred.pid) {
+				nih_error("gettasks: too much pid churn.  Last valid pid was %d\n",
+						firstvalid);
+				return;
+			}
+			nih_info("gettasks: sending dup pid %d in place of exited pid %d\n",
+					firstvalid, pcred.pid);
+			pcred.pid = firstvalid;
+			goto again;
+
+		} else if (ret < 0)
 			return;
-		}
+		if (firstvalid == -1)
+			firstvalid = pids[i];
 	}
 }
 
@@ -1275,6 +1292,7 @@ void get_tasks_recursive_scm_complete(struct scm_sock_data *data)
 {
 	struct ucred pcred;
 	int i, ret;
+	pid_t firstvalid = -1;
 	int32_t *pids, nrpids;
 
 	ret = get_tasks_recursive_main(data, data->controller, data->cgroup,
@@ -1290,12 +1308,24 @@ void get_tasks_recursive_scm_complete(struct scm_sock_data *data)
 		return;
 	}
 	pcred.uid = 0; pcred.gid = 0;
-	for (i=0; i<ret; i++) {
+	for (i=0; i<nrpids; i++) {
 		pcred.pid = pids[i];
-		if (send_creds(data->fd, &pcred)) {
-			nih_error("get_tasks_recursive_scm: error writing pids back to client");
+again:
+		ret = send_creds(data->fd, &pcred);
+		if (ret == -3) {
+			if (firstvalid == -1 || firstvalid == pcred.pid) {
+				nih_error("gettasks: too much pid churn.  Last valid pid was %d\n",
+						firstvalid);
+				return;
+			}
+			nih_info("gettasks: sending dup pid %d in place of exited pid %d\n",
+					firstvalid, pcred.pid);
+			pcred.pid = firstvalid;
+			goto again;
+		} else if (ret < 0)
 			return;
-		}
+		if (firstvalid == -1)
+			firstvalid = pids[i];
 	}
 }