nspawn: make nspawn robust to container failure

nspawn and the container child use eventfd to wait and notify each other
that they are ready so the container setup can be completed.

However in its current form the wait/notify event ignore errors that
may especially affect the child (container).

On errors the child will jump to the "child_fail" label and terminate
with _exit(EXIT_FAILURE) without notifying the parent. Since the eventfd
is created without the "EFD_NONBLOCK" flag, this leaves the parent
blocking on the eventfd_read() call. The container can also be killed
at any moment before execv() and the parent will not receive
notifications.

We can fix this by using cheap mechanisms, the new high level eventfd
API and handle SIGCHLD signals:

* Keep the cheap eventfd and EFD_NONBLOCK flag.

* Introduce eventfd states for parent and child to sync.
Child notifies parent with EVENTFD_CHILD_SUCCEEDED on success or
EVENTFD_CHILD_FAILED on failure and before _exit(). This prevents the
parent from waiting on an event that will never come.

* If the child is killed before execv() or before notifying the parent,
we install a NOP handler for SIGCHLD which will interrupt blocking calls
with EINTR. This gives a chance to the parent to call wait() and
terminate in main().

* If there are no errors, parent will block SIGCHLD, restore default
handler and notify child which will do execv(), then parent will pass
control to process_pty() to do its magic.

This was exposed in part by:
https://bugs.freedesktop.org/show_bug.cgi?id=76193

Reported-by: Tobias Hunger tobias.hunger@gmail.com

4 files changed, 282 insertions, 26 deletions

diff --git a/Makefile.am b/Makefile.am
index f517f19df5..4b60735515 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -805,7 +805,9 @@ libsystemd_shared_la_SOURCES = \
 	src/shared/ring.c \
 	src/shared/ring.h \
 	src/shared/async.c \
-	src/shared/async.h
+	src/shared/async.h \
+	src/shared/eventfd-util.c \
+	src/shared/eventfd-util.h
 
 nodist_libsystemd_shared_la_SOURCES = \
 	src/shared/errno-from-name.h \
diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c
index 0cd476cd9e..a1d77244f8 100644
--- a/src/nspawn/nspawn.c
+++ b/src/nspawn/nspawn.c
@@ -84,6 +84,7 @@
 #include "def.h"
 #include "rtnl-util.h"
 #include "udev-util.h"
+#include "eventfd-util.h"
 #include "blkid-util.h"
 #include "gpt.h"
 #include "siphash24.h"
@@ -2642,6 +2643,8 @@ static int wait_for_container(pid_t pid, ContainerStatus *container) {
         return r;
 }
 
+static void nop_handler(int sig) {}
+
 int main(int argc, char *argv[]) {
 
         _cleanup_free_ char *kdbus_domain = NULL, *device_path = NULL, *root_device = NULL, *home_device = NULL, *srv_device = NULL;
@@ -2653,8 +2656,8 @@ int main(int argc, char *argv[]) {
         const char *console = NULL;
         char veth_name[IFNAMSIZ];
         bool secondary = false;
+        sigset_t mask, mask_chld;
         pid_t pid = 0;
-        sigset_t mask;
 
         log_parse_environment();
         log_open();
@@ -2816,36 +2819,44 @@ int main(int argc, char *argv[]) {
         sd_notify(0, "READY=1");
 
         assert_se(sigemptyset(&mask) == 0);
+        assert_se(sigemptyset(&mask_chld) == 0);
+        sigaddset(&mask_chld, SIGCHLD);
         sigset_add_many(&mask, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1);
         assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0);
 
         for (;;) {
                 ContainerStatus container_status;
-                int parent_ready_fd = -1, child_ready_fd = -1;
-                eventfd_t x;
-
-                parent_ready_fd = eventfd(0, EFD_CLOEXEC);
-                if (parent_ready_fd < 0) {
-                        log_error("Failed to create event fd: %m");
+                int eventfds[2] = { -1, -1 };
+                struct sigaction sa = {
+                        .sa_handler = nop_handler,
+                        .sa_flags = SA_NOCLDSTOP,
+                };
+
+                /* Child can be killed before execv(), so handle SIGCHLD
+                 * in order to interrupt parent's blocking calls and
+                 * give it a chance to call wait() and terminate. */
+                r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
+                if (r < 0) {
+                        log_error("Failed to change the signal mask: %m");
                         goto finish;
                 }
 
-                child_ready_fd = eventfd(0, EFD_CLOEXEC);
-                if (child_ready_fd < 0) {
-                        log_error("Failed to create event fd: %m");
+                r = sigaction(SIGCHLD, &sa, NULL);
+                if (r < 0) {
+                        log_error("Failed to install SIGCHLD handler: %m");
                         goto finish;
                 }
 
-                pid = syscall(__NR_clone,
-                              SIGCHLD|CLONE_NEWNS|
-                              (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS)|
-                              (arg_private_network ? CLONE_NEWNET : 0), NULL);
+                pid = clone_with_eventfd(SIGCHLD|CLONE_NEWNS|
+                                         (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS)|
+                                         (arg_private_network ? CLONE_NEWNET : 0), eventfds);
                 if (pid < 0) {
                         if (errno == EINVAL)
                                 log_error("clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
                         else
                                 log_error("clone() failed: %m");
 
+                        r = pid;
                         goto finish;
                 }
 
@@ -2986,8 +2997,11 @@ int main(int argc, char *argv[]) {
                         /* Tell the parent that we are ready, and that
                          * it can cgroupify us to that we lack access
                          * to certain devices and resources. */
-                        eventfd_write(child_ready_fd, 1);
-                        child_ready_fd = safe_close(child_ready_fd);
+                        r = eventfd_send_state(eventfds[1],
+                                               EVENTFD_CHILD_SUCCEEDED);
+                        eventfds[1] = safe_close(eventfds[1]);
+                        if (r < 0)
+                                goto child_fail;
 
                         if (chdir(arg_directory) < 0) {
                                 log_error("chdir(%s) failed: %m", arg_directory);
@@ -3089,8 +3103,10 @@ int main(int argc, char *argv[]) {
                                 env_use = (char**) envp;
 
                         /* Wait until the parent is ready with the setup, too... */
-                        eventfd_read(parent_ready_fd, &x);
-                        parent_ready_fd = safe_close(parent_ready_fd);
+                        r = eventfd_parent_succeeded(eventfds[0]);
+                        eventfds[0] = safe_close(eventfds[0]);
+                        if (r < 0)
+                                goto child_fail;
 
                         if (arg_boot) {
                                 char **a;
@@ -3121,17 +3137,27 @@ int main(int argc, char *argv[]) {
                         log_error("execv() failed: %m");
 
                 child_fail:
+                        /* Tell the parent that the setup failed, so he
+                         * can clean up resources and terminate. */
+                        if (eventfds[1] != -1)
+                                eventfd_send_state(eventfds[1],
+                                                   EVENTFD_CHILD_FAILED);
                         _exit(EXIT_FAILURE);
                 }
 
                 fdset_free(fds);
                 fds = NULL;
 
-                /* Wait until the child reported that it is ready with
-                 * all it needs to do with privileges. After we got
-                 * the notification we can make the process join its
-                 * cgroup which might limit what it can do */
-                eventfd_read(child_ready_fd, &x);
+                /* Wait for the child event:
+                 * If EVENTFD_CHILD_FAILED, the child will terminate soon.
+                 * If EVENTFD_CHILD_SUCCEEDED, the child is reporting that
+                 * it is ready with all it needs to do with priviliges.
+                 * After we got the notification we can make the process
+                 * join its cgroup which might limit what it can do */
+                r = eventfd_child_succeeded(eventfds[1]);
+                eventfds[1] = safe_close(eventfds[1]);
+                if (r < 0)
+                        goto check_container_status;
 
                 r = register_machine(pid);
                 if (r < 0)
@@ -3153,10 +3179,25 @@ int main(int argc, char *argv[]) {
                 if (r < 0)
                         goto finish;
 
+                /* Block SIGCHLD here, before notifying child.
+                 * process_pty() will handle it with the other signals. */
+                r = sigprocmask(SIG_BLOCK, &mask_chld, NULL);
+                if (r < 0)
+                        goto finish;
+
+                /* Reset signal to default */
+                r = default_signals(SIGCHLD, -1);
+                if (r < 0)
+                        goto finish;
+
                 /* Notify the child that the parent is ready with all
-                 * its setup, and thtat the child can now hand over
+                 * its setup, and that the child can now hand over
                  * control to the code to run inside the container. */
-                eventfd_write(parent_ready_fd, 1);
+                r = eventfd_send_state(eventfds[0],
+                                       EVENTFD_PARENT_SUCCEEDED);
+                eventfds[0] = safe_close(eventfds[0]);
+                if (r < 0)
+                        goto finish;
 
                 k = process_pty(master, &mask, arg_boot ? pid : 0, SIGRTMIN+3);
                 if (k < 0) {
@@ -3170,6 +3211,7 @@ int main(int argc, char *argv[]) {
                 /* Kill if it is not dead yet anyway */
                 terminate_machine(pid);
 
+check_container_status:
                 /* Redundant, but better safe than sorry */
                 kill(pid, SIGKILL);
 
diff --git a/src/shared/eventfd-util.c b/src/shared/eventfd-util.c
new file mode 100644
index 0000000000..27b7cf788f
--- /dev/null
+++ b/src/shared/eventfd-util.c
@@ -0,0 +1,169 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2014 Djalal Harouni
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU Lesser General Public License as published by
+  the Free Software Foundation; either version 2.1 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <assert.h>
+#include <errno.h>
+#include <unistd.h>
+#include <sys/eventfd.h>
+#include <sys/syscall.h>
+
+#include "eventfd-util.h"
+#include "util.h"
+
+
+/*
+ * Use this to create processes that need to setup a full context
+ * and sync it with their parents using cheap mechanisms.
+ *
+ * This will create two blocking eventfd(s). A pair for the parent and
+ * the other for the child so they can be used as a notify mechanism.
+ * Each process will gets its copy of the parent and child eventfds.
+ *
+ * This is useful in case:
+ * 1) If the parent fails or dies, the child must die.
+ * 2) Child will install PR_SET_PDEATHSIG as soon as possible.
+ * 3) Parent and child need to sync using less resources.
+ * 4) If parent is not able to install a SIGCHLD handler:
+ *    parent will wait using a blocking eventfd_read() or
+ *    eventfd_child_succeeded() call on the child eventfd.
+ *
+ *    * If the child setup succeeded, child should notify with an
+ *      EVENTFD_CHILD_SUCCEEDED, parent will continue.
+ *    * If the child setup failed, child should notify with an
+ *      EVENTFD_CHILD_FAILED before any _exit(). This avoids blocking
+ *      the parent.
+ *
+ * 5) If parent is able to install a SIGCHLD handler:
+ *    An empty signal handler without SA_RESTART will do it, since the
+ *    blocking eventfd_read() or eventfd_parent_succeeded() of the
+ *    parent will be interrupted by SIGCHLD and the call will fail with
+ *    EINTR. This is useful in case the child dies abnormaly and did
+ *    not have a chance to notify its parent using EVENTFD_CHILD_FAILED.
+ *
+ * 6) Call wait*() in the main instead of the signal handler in order
+ *    to: 1) reduce side effects and 2) have a better handling for
+ *    child termination in order to reduce various race conditions.
+ *
+ *
+ * The return value of clone_with_eventfd() is the same of clone().
+ * On success the eventfds[] will contain the two eventfd(s). These
+ * file descriptors can be closed later with safe_close(). On failure,
+ * a negative value is returned in the caller's context, and errno will
+ * be set appropriately.
+ *
+ * Extra preliminary work:
+ * 1) Child can wait before starting its setup by using the
+ *    eventfd_recv_start() call on the parent eventfd, in that case the
+ *    parent must notify with EVENTFD_START, after doing any preliminary
+ *    work.
+ *
+ * Note: this function depends on systemd internal functions
+ * safe_close() and it should be used only by direct binaries, no
+ * libraries.
+ */
+pid_t clone_with_eventfd(int flags, int eventfds[2]) {
+        pid_t pid;
+
+        assert(eventfds);
+
+        eventfds[0] = eventfd(EVENTFD_INIT, EFD_CLOEXEC);
+        if (eventfds[0] < 0)
+                return -1;
+
+        eventfds[1] = eventfd(EVENTFD_INIT, EFD_CLOEXEC);
+        if (eventfds[1] < 0)
+                goto err_eventfd0;
+
+        pid = syscall(__NR_clone, flags, NULL);
+        if (pid < 0)
+                goto err_eventfd1;
+
+        return pid;
+
+err_eventfd1:
+        eventfds[1] = safe_close(eventfds[1]);
+err_eventfd0:
+        eventfds[0] = safe_close(eventfds[0]);
+        return -1;
+}
+
+int eventfd_send_state(int efd, eventfd_t s) {
+        return eventfd_write(efd, s);
+}
+
+/*
+ * Receive an eventfd state on the eventfd file descriptor.
+ *
+ * If the third argument is set to a value other than zero, then this
+ * function will compare the received value with this argument and set
+ * the return value.
+ *
+ * On success return 0. On error, -1 will be returned, and errno will
+ * be set appropriately.
+ */
+int eventfd_recv_state(int efd, eventfd_t *e, eventfd_t s) {
+        int ret;
+
+        ret = eventfd_read(efd, e);
+        if (ret < 0)
+                return ret;
+        else if (s != 0 && *e != s) {
+                errno = EINVAL;
+                return -1;
+        }
+
+        return 0;
+}
+
+/*
+ * Receive the EVENTFD_START state on the eventfd file descriptor.
+ *
+ * On Success return 0. On error, -1 will be returned, and errno will
+ * be set appropriately.
+ */
+int eventfd_recv_start(int efd) {
+        eventfd_t e = EVENTFD_INIT;
+        return eventfd_recv_state(efd, &e, EVENTFD_START);
+}
+
+/*
+ * Receive the EVENTFD_PARENT_SUCCEEDED state on the eventfd file
+ * descriptor.
+ *
+ * On Success return 0. On error, -1 will be returned, and errno will
+ * be set appropriately.
+ */
+int eventfd_parent_succeeded(int efd) {
+        eventfd_t e = EVENTFD_INIT;
+        return eventfd_recv_state(efd, &e, EVENTFD_PARENT_SUCCEEDED);
+}
+
+/*
+ * Receive the EVENTFD_CHILD_SUCCEEDED state on the eventfd file
+ * descriptor.
+ *
+ * On Success return 0. On error, -1 will be returned, and errno will
+ * be set appropriately.
+ */
+int eventfd_child_succeeded(int efd) {
+        eventfd_t e = EVENTFD_INIT;
+        return eventfd_recv_state(efd, &e, EVENTFD_CHILD_SUCCEEDED);
+}
diff --git a/src/shared/eventfd-util.h b/src/shared/eventfd-util.h
new file mode 100644
index 0000000000..0120f0409e
--- /dev/null
+++ b/src/shared/eventfd-util.h
@@ -0,0 +1,43 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+#pragma once
+
+/***
+  This file is part of systemd.
+
+  Copyright 2014 Djalal Harouni
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU Lesser General Public License as published by
+  the Free Software Foundation; either version 2.1 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <sys/types.h>
+#include <sys/eventfd.h>
+
+enum {
+        EVENTFD_INIT,
+        EVENTFD_START,
+        EVENTFD_PARENT_SUCCEEDED,
+        EVENTFD_PARENT_FAILED,
+        EVENTFD_CHILD_SUCCEEDED,
+        EVENTFD_CHILD_FAILED,
+};
+
+pid_t clone_with_eventfd(int flags, int eventfds[2]);
+
+int eventfd_send_state(int efd, eventfd_t s);
+int eventfd_recv_state(int efd, eventfd_t *e, eventfd_t s);
+
+int eventfd_recv_start(int efd);
+int eventfd_parent_succeeded(int efd);
+int eventfd_child_succeeded(int efd);