[CRIU] [PATCH] pid_ns: Destroy helpers via kill()

Kirill Tkhai ktkhai at virtuozzo.com
Mon May 15 04:13:09 PDT 2017


When INIT_PID of a pid_ns emergency exits, kernel
kills all processes belonging to the namespace.
So, it's hopelessly to wait helper answer to destroy
request. Use kill() to destroy instead of that.
It will be noop in case of a handler is already
killed, and we won't stuck.

Signed-off-by: Kirill Tkhai <ktkhai at virtuozzo.com>
---
 criu/namespaces.c |   76 +++++++++++++++++++++++++++--------------------------
 1 file changed, 39 insertions(+), 37 deletions(-)

diff --git a/criu/namespaces.c b/criu/namespaces.c
index cfde683b9..103235c74 100644
--- a/criu/namespaces.c
+++ b/criu/namespaces.c
@@ -55,8 +55,6 @@ static struct ns_desc *ns_desc_array[] = {
 };
 
 static unsigned int join_ns_flags;
-/* Creation of every helper are synchronized by userns_sync_lock */
-static int nr_pid_ns_helper_created = 0;
 
 int check_namespace_opts(void)
 {
@@ -1521,21 +1519,29 @@ static void unsc_msg_pid_fd(struct unsc_msg *um, pid_t *pid, int *fd)
 static void usernsd_handler(int signal, siginfo_t *siginfo, void *data)
 {
 	pid_t pid = siginfo->si_pid;
+	struct ns_id *ns;
+	struct pid *tpid;
 	int status;
-	int exit;
 
 	while (pid) {
 		pid = waitpid(-1, &status, WNOHANG);
 		if (pid <= 0)
 			return;
+		for (ns = ns_ids; ns; ns = ns->next) {
+			if (ns->nd != &pid_ns_desc)
+				continue;
+			tpid = __pstree_pid_by_virt(ns, ns->ns_pid);
+			if (tpid->real == pid) {
+				tpid->real = -1;
+				break;
+			}
+		}
+
+		if (!ns)
+			pr_err("Spurious pid ns helper: pid=%d\n", pid);
 
-		exit = WIFEXITED(status);
-		status = exit ? WEXITSTATUS(status) : WTERMSIG(status);
-		if (status) {
-			futex_abort_and_wake(&task_entries->nr_in_progress);
-			pr_err("%d finished abnormal\n", pid);
-		} else
-			pr_info("%d exited normally\n", pid);
+		pr_err("%d finished unexpected: status=%d\n", pid, status);
+		futex_abort_and_wake(&task_entries->nr_in_progress);
 	}
 }
 
@@ -2597,11 +2603,9 @@ static int pid_ns_helper(struct ns_id *ns, int sk)
 			break;
 		}
 
-		if (pid != 0) {
-			if (__set_next_pid(pid) < 0) {
-				pr_err("Can't set next pid\n");
-				answer = -1;
-			}
+		if (__set_next_pid(pid) < 0) {
+			pr_err("Can't set next pid\n");
+			answer = -1;
 		}
 
 		iov.iov_base = &answer;
@@ -2610,9 +2614,6 @@ static int pid_ns_helper(struct ns_id *ns, int sk)
 			pr_perror("Can't send answer");
 			break;
 		}
-
-		if (pid == 0)
-			return 0;
 	}
 
 	return -1;
@@ -2704,7 +2705,7 @@ static int do_create_pid_ns_helper(void *arg, int sk, pid_t unused_pid)
 	futex_set_and_wake(&ns->pid.helper_created, 1);
 	flock(lock_fd, LOCK_UN);
 	close(lock_fd);
-	nr_pid_ns_helper_created++;
+	pid->real = child;
 
 	if (setns(pid_ns_fd, CLONE_NEWPID) < 0) {
 		pr_perror("Restore ns");
@@ -2755,42 +2756,43 @@ int create_pid_ns_helper(struct ns_id *ns)
 	return 0;
 }
 
-static int do_destroy_pid_ns_helper(void *arg, int fd, pid_t pid)
+static int do_destroy_pid_ns_helper(void *arg, int fd, pid_t unused)
 {
-	int i, sk, status, sig_blocked = true, nr_ok = 0, ret = 0;
+	int status, sig_blocked = true, ret = 0;
 	sigset_t sig_mask;
 	struct ns_id *ns;
-
-	if (!nr_pid_ns_helper_created)
-		return 0;
+	struct pid *pid;
 
 	if (block_sigmask(&sig_mask, SIGCHLD)) {
 		sig_blocked = false;
 		ret = -1;
 	}
 
-	sk = get_service_fd(TRANSPORT_FD_OFF);
-	BUG_ON(sk < 0);
-
 	for (ns = ns_ids; ns; ns = ns->next) {
 		if (ns->nd != &pid_ns_desc)
 			continue;
-		if (request_set_next_pid(ns->id, 0, sk) == 0)
-			nr_ok++;
-	}
+		pid = __pstree_pid_by_virt(ns, ns->ns_pid);
+		if (!pid) {
+			pr_err("Can't find pid_ns helper\n");
+			ret = -1;
+			continue;
+		}
+		if (pid->real <= 0)
+			continue;
 
-	if (nr_ok != nr_pid_ns_helper_created) {
-		pr_err("Not all pid_ns helpers killed\n");
-		ret = -1;
-	}
+		if (kill(pid->real, SIGKILL) < 0) {
+			pr_perror("Can't kill\n");
+			ret = -1;
+			continue;
+		}
 
-	for (i = 0; i < nr_ok; i++) {
-		if (waitpid(-1, &status, 0) < 0) {
+		if (waitpid(pid->real, &status, 0) != pid->real) {
 			pr_perror("Error during waiting pid_ns helper");
 			ret = -1;
+			continue;
 		}
+		pid->real = -1;
 	}
-	nr_pid_ns_helper_created = 0;
 
 	if (sig_blocked && restore_sigmask(&sig_mask))
 		ret = -1;



More information about the CRIU mailing list