Merge pull request #10747 from poettering/machinectl-list-fix

properly acquire os-release file from containers

6 files changed, 86 insertions, 71 deletions

diff --git a/src/basic/process-util.c b/src/basic/process-util.c
index b2aab853e2..572d7eb38d 100644
--- a/src/basic/process-util.c
+++ b/src/basic/process-util.c
@@ -1409,6 +1409,60 @@ int safe_fork_full(
         return 0;
 }
 
+int namespace_fork(
+                const char *outer_name,
+                const char *inner_name,
+                const int except_fds[],
+                size_t n_except_fds,
+                ForkFlags flags,
+                int pidns_fd,
+                int mntns_fd,
+                int netns_fd,
+                int userns_fd,
+                int root_fd,
+                pid_t *ret_pid) {
+
+        int r;
+
+        /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle
+         * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that
+         * /proc/self/fd works correctly. */
+
+        r = safe_fork_full(outer_name, except_fds, n_except_fds, (flags|FORK_DEATHSIG) & ~(FORK_REOPEN_LOG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE), ret_pid);
+        if (r < 0)
+                return r;
+        if (r == 0) {
+                pid_t pid;
+
+                /* Child */
+
+                r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd);
+                if (r < 0) {
+                        log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m");
+                        _exit(EXIT_FAILURE);
+                }
+
+                /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */
+                r = safe_fork_full(inner_name, except_fds, n_except_fds, flags & ~(FORK_WAIT|FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_NULL_STDIO), &pid);
+                if (r < 0)
+                        _exit(EXIT_FAILURE);
+                if (r == 0) {
+                        /* Child */
+                        if (ret_pid)
+                                *ret_pid = pid;
+                        return 0;
+                }
+
+                r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0);
+                if (r < 0)
+                        _exit(EXIT_FAILURE);
+
+                _exit(r);
+        }
+
+        return 1;
+}
+
 int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) {
         bool stdout_is_tty, stderr_is_tty;
         size_t n, i;
diff --git a/src/basic/process-util.h b/src/basic/process-util.h
index ca4e4401a9..af47513fab 100644
--- a/src/basic/process-util.h
+++ b/src/basic/process-util.h
@@ -159,6 +159,8 @@ static inline int safe_fork(const char *name, ForkFlags flags, pid_t *ret_pid) {
         return safe_fork_full(name, NULL, 0, flags, ret_pid);
 }
 
+int namespace_fork(const char *outer_name, const char *inner_name, const int except_fds[], size_t n_except_fds, ForkFlags flags, int pidns_fd, int mntns_fd, int netns_fd, int userns_fd, int root_fd, pid_t *ret_pid);
+
 int fork_agent(const char *name, const int except[], size_t n_except, pid_t *pid, const char *path, ...) _sentinel_;
 
 int set_oom_score_adjust(int value);
diff --git a/src/basic/terminal-util.c b/src/basic/terminal-util.c
index ea0ba442ff..41789598f7 100644
--- a/src/basic/terminal-util.c
+++ b/src/basic/terminal-util.c
@@ -1095,7 +1095,8 @@ int openpt_in_namespace(pid_t pid, int flags) {
         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, pair) < 0)
                 return -errno;
 
-        r = safe_fork("(sd-openpt)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+        r = namespace_fork("(sd-openptns)", "(sd-openpt)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                           pidnsfd, mntnsfd, -1, usernsfd, rootfd, &child);
         if (r < 0)
                 return r;
         if (r == 0) {
@@ -1103,10 +1104,6 @@ int openpt_in_namespace(pid_t pid, int flags) {
 
                 pair[0] = safe_close(pair[0]);
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
-                if (r < 0)
-                        _exit(EXIT_FAILURE);
-
                 master = posix_openpt(flags|O_NOCTTY|O_CLOEXEC);
                 if (master < 0)
                         _exit(EXIT_FAILURE);
@@ -1122,7 +1119,7 @@ int openpt_in_namespace(pid_t pid, int flags) {
 
         pair[1] = safe_close(pair[1]);
 
-        r = wait_for_terminate_and_check("(sd-openpt)", child, 0);
+        r = wait_for_terminate_and_check("(sd-openptns)", child, 0);
         if (r < 0)
                 return r;
         if (r != EXIT_SUCCESS)
@@ -1144,7 +1141,8 @@ int open_terminal_in_namespace(pid_t pid, const char *name, int mode) {
         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, pair) < 0)
                 return -errno;
 
-        r = safe_fork("(sd-terminal)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+        r = namespace_fork("(sd-terminalns)", "(sd-terminal)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                           pidnsfd, mntnsfd, -1, usernsfd, rootfd, &child);
         if (r < 0)
                 return r;
         if (r == 0) {
@@ -1152,10 +1150,6 @@ int open_terminal_in_namespace(pid_t pid, const char *name, int mode) {
 
                 pair[0] = safe_close(pair[0]);
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
-                if (r < 0)
-                        _exit(EXIT_FAILURE);
-
                 master = open_terminal(name, mode|O_NOCTTY|O_CLOEXEC);
                 if (master < 0)
                         _exit(EXIT_FAILURE);
@@ -1168,7 +1162,7 @@ int open_terminal_in_namespace(pid_t pid, const char *name, int mode) {
 
         pair[1] = safe_close(pair[1]);
 
-        r = wait_for_terminate_and_check("(sd-terminal)", child, 0);
+        r = wait_for_terminate_and_check("(sd-terminalns)", child, 0);
         if (r < 0)
                 return r;
         if (r != EXIT_SUCCESS)
diff --git a/src/libsystemd/sd-bus/bus-container.c b/src/libsystemd/sd-bus/bus-container.c
index 38023087a3..2cfeefc2c3 100644
--- a/src/libsystemd/sd-bus/bus-container.c
+++ b/src/libsystemd/sd-bus/bus-container.c
@@ -45,51 +45,27 @@ int bus_container_connect_socket(sd_bus *b) {
         if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, pair) < 0)
                 return -errno;
 
-        r = safe_fork("(sd-buscntr)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+        r = namespace_fork("(sd-buscntrns)", "(sd-buscntr)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                           pidnsfd, mntnsfd, -1, usernsfd, rootfd, &child);
         if (r < 0)
                 return r;
         if (r == 0) {
-                pid_t grandchild;
-
                 pair[0] = safe_close(pair[0]);
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
-                if (r < 0)
-                        _exit(EXIT_FAILURE);
-
-                /* We just changed PID namespace, however it will only
-                 * take effect on the children we now fork. Hence,
-                 * let's fork another time, and connect from this
-                 * grandchild, so that SO_PEERCRED of our connection
-                 * comes from a process from within the container, and
-                 * not outside of it */
-
-                r = safe_fork("(sd-buscntr2)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &grandchild);
-                if (r < 0)
+                r = connect(b->input_fd, &b->sockaddr.sa, b->sockaddr_size);
+                if (r < 0) {
+                        /* Try to send error up */
+                        error_buf = errno;
+                        (void) write(pair[1], &error_buf, sizeof(error_buf));
                         _exit(EXIT_FAILURE);
-                if (r == 0) {
-
-                        r = connect(b->input_fd, &b->sockaddr.sa, b->sockaddr_size);
-                        if (r < 0) {
-                                /* Try to send error up */
-                                error_buf = errno;
-                                (void) write(pair[1], &error_buf, sizeof(error_buf));
-                                _exit(EXIT_FAILURE);
-                        }
-
-                        _exit(EXIT_SUCCESS);
                 }
 
-                r = wait_for_terminate_and_check("(sd-buscntr2)", grandchild, 0);
-                if (r < 0)
-                        _exit(EXIT_FAILURE);
-
-                _exit(r);
+                _exit(EXIT_SUCCESS);
         }
 
         pair[1] = safe_close(pair[1]);
 
-        r = wait_for_terminate_and_check("(sd-buscntr)", child, 0);
+        r = wait_for_terminate_and_check("(sd-buscntrns)", child, 0);
         if (r < 0)
                 return r;
         if (r != EXIT_SUCCESS)
diff --git a/src/machine/machine-dbus.c b/src/machine/machine-dbus.c
index 4f4d780db0..2f9f587973 100644
--- a/src/machine/machine-dbus.c
+++ b/src/machine/machine-dbus.c
@@ -207,7 +207,8 @@ int bus_machine_method_get_addresses(sd_bus_message *message, void *userdata, sd
                 if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, pair) < 0)
                         return -errno;
 
-                r = safe_fork("(sd-addr)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+                r = namespace_fork("(sd-addrns)", "(sd-addr)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                                   -1, -1, netns_fd, -1, -1, &child);
                 if (r < 0)
                         return sd_bus_error_set_errnof(error, r, "Failed to fork(): %m");
                 if (r == 0) {
@@ -217,10 +218,6 @@ int bus_machine_method_get_addresses(sd_bus_message *message, void *userdata, sd
 
                         pair[0] = safe_close(pair[0]);
 
-                        r = namespace_enter(-1, -1, netns_fd, -1, -1);
-                        if (r < 0)
-                                _exit(EXIT_FAILURE);
-
                         n = local_addresses(NULL, 0, AF_UNSPEC, &addresses);
                         if (n < 0)
                                 _exit(EXIT_FAILURE);
@@ -294,7 +291,7 @@ int bus_machine_method_get_addresses(sd_bus_message *message, void *userdata, sd
                                 return r;
                 }
 
-                r = wait_for_terminate_and_check("(sd-addr)", child, 0);
+                r = wait_for_terminate_and_check("(sd-addrns)", child, 0);
                 if (r < 0)
                         return sd_bus_error_set_errnof(error, r, "Failed to wait for child: %m");
                 if (r != EXIT_SUCCESS)
@@ -333,19 +330,21 @@ int bus_machine_method_get_os_release(sd_bus_message *message, void *userdata, s
                 break;
 
         case MACHINE_CONTAINER: {
-                _cleanup_close_ int mntns_fd = -1, root_fd = -1;
+                _cleanup_close_ int mntns_fd = -1, root_fd = -1, pidns_fd = -1;
                 _cleanup_close_pair_ int pair[2] = { -1, -1 };
                 _cleanup_fclose_ FILE *f = NULL;
                 pid_t child;
 
-                r = namespace_open(m->leader, NULL, &mntns_fd, NULL, NULL, &root_fd);
+                r = namespace_open(m->leader, &pidns_fd, &mntns_fd, NULL, NULL, &root_fd);
                 if (r < 0)
                         return r;
 
                 if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, pair) < 0)
                         return -errno;
 
-                r = safe_fork("(sd-osrel)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+                r = namespace_fork("(sd-osrelns)", "(sd-osrel)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                                   pidns_fd, mntns_fd, -1, -1, root_fd,
+                                   &child);
                 if (r < 0)
                         return sd_bus_error_set_errnof(error, r, "Failed to fork(): %m");
                 if (r == 0) {
@@ -353,10 +352,6 @@ int bus_machine_method_get_os_release(sd_bus_message *message, void *userdata, s
 
                         pair[0] = safe_close(pair[0]);
 
-                        r = namespace_enter(-1, mntns_fd, -1, -1, root_fd);
-                        if (r < 0)
-                                _exit(EXIT_FAILURE);
-
                         r = open_os_release(NULL, NULL, &fd);
                         if (r == -ENOENT)
                                 _exit(EXIT_NOT_FOUND);
@@ -382,7 +377,7 @@ int bus_machine_method_get_os_release(sd_bus_message *message, void *userdata, s
                 if (r < 0)
                         return r;
 
-                r = wait_for_terminate_and_check("(sd-osrel)", child, 0);
+                r = wait_for_terminate_and_check("(sd-osrelns)", child, 0);
                 if (r < 0)
                         return sd_bus_error_set_errnof(error, r, "Failed to wait for child: %m");
                 if (r == EXIT_NOT_FOUND)
@@ -1239,7 +1234,8 @@ int bus_machine_method_open_root_directory(sd_bus_message *message, void *userda
                 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, pair) < 0)
                         return -errno;
 
-                r = safe_fork("(sd-openroot)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+                r = namespace_fork("(sd-openrootns)", "(sd-openroot)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                                   -1, mntns_fd, -1, -1, root_fd, &child);
                 if (r < 0)
                         return sd_bus_error_set_errnof(error, r, "Failed to fork(): %m");
                 if (r == 0) {
@@ -1247,10 +1243,6 @@ int bus_machine_method_open_root_directory(sd_bus_message *message, void *userda
 
                         pair[0] = safe_close(pair[0]);
 
-                        r = namespace_enter(-1, mntns_fd, -1, -1, root_fd);
-                        if (r < 0)
-                                _exit(EXIT_FAILURE);
-
                         dfd = open("/", O_RDONLY|O_CLOEXEC|O_DIRECTORY);
                         if (dfd < 0)
                                 _exit(EXIT_FAILURE);
@@ -1265,7 +1257,7 @@ int bus_machine_method_open_root_directory(sd_bus_message *message, void *userda
 
                 pair[1] = safe_close(pair[1]);
 
-                r = wait_for_terminate_and_check("(sd-openroot)", child, 0);
+                r = wait_for_terminate_and_check("(sd-openrootns)", child, 0);
                 if (r < 0)
                         return sd_bus_error_set_errnof(error, r, "Failed to wait for child: %m");
                 if (r != EXIT_SUCCESS)
diff --git a/src/shared/logs-show.c b/src/shared/logs-show.c
index 79e47b8417..2d53307203 100644
--- a/src/shared/logs-show.c
+++ b/src/shared/logs-show.c
@@ -1342,7 +1342,8 @@ static int get_boot_id_for_machine(const char *machine, sd_id128_t *boot_id) {
         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, pair) < 0)
                 return -errno;
 
-        r = safe_fork("(sd-bootid)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &child);
+        r = namespace_fork("(sd-bootidns)", "(sd-bootid)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG,
+                           pidnsfd, mntnsfd, -1, -1, rootfd, &child);
         if (r < 0)
                 return r;
         if (r == 0) {
@@ -1350,10 +1351,6 @@ static int get_boot_id_for_machine(const char *machine, sd_id128_t *boot_id) {
 
                 pair[0] = safe_close(pair[0]);
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, -1, rootfd);
-                if (r < 0)
-                        _exit(EXIT_FAILURE);
-
                 fd = open("/proc/sys/kernel/random/boot_id", O_RDONLY|O_CLOEXEC|O_NOCTTY);
                 if (fd < 0)
                         _exit(EXIT_FAILURE);
@@ -1372,7 +1369,7 @@ static int get_boot_id_for_machine(const char *machine, sd_id128_t *boot_id) {
 
         pair[1] = safe_close(pair[1]);
 
-        r = wait_for_terminate_and_check("(sd-bootid)", child, 0);
+        r = wait_for_terminate_and_check("(sd-bootidns)", child, 0);
         if (r < 0)
                 return r;
         if (r != EXIT_SUCCESS)