namespace helpers: Allow entering a UID namespace

To be able to use `systemd-run` or `machinectl login` on a container
that is in a private user namespace, the sub-process must have entered
the user namespace before connecting to the container's D-Bus, otherwise
the UID and GID in the peer credentials are garbage.

So we extend namespace_open and namespace_enter to support UID namespaces,
and we enter the UID namespace in bus_container_connect_{socket,kernel}.

namespace_open will degrade to a no-op if user namespaces are not enabled
in the kernel.

Special handling is required for the setns call in namespace_enter with
a user namespace, since transitioning to your own namespace is forbidden,
as it would result in re-entering your user namespace as root.

Arguably it may be valid to check this at the call site, rather than
inside namespace_enter, but it is less code to do it inside, and if the
intention of calling namespace_enter is to *be* in the target namespace,
rather than to transition to the target namespace, it is a reasonable
approach.

The check for whether the user namespace is the same must happen before
entering namespaces, as we may not be able to access /proc during the
intermediate transition stage.

We can't instead attempt to enter the user namespace and then ignore
the failure from it being the same namespace, since the error code is
not distinct, and we can't compare namespaces while mid-transition.

6 files changed, 54 insertions, 23 deletions

diff --git a/src/basic/util.c b/src/basic/util.c
index d4d3d3c83a..3811460895 100644
--- a/src/basic/util.c
+++ b/src/basic/util.c
@@ -4950,8 +4950,8 @@ int container_get_leader(const char *machine, pid_t *pid) {
         return 0;
 }
 
-int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *root_fd) {
-        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, netnsfd = -1;
+int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *userns_fd, int *root_fd) {
+        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, netnsfd = -1, usernsfd = -1;
         int rfd = -1;
 
         assert(pid >= 0);
@@ -4983,6 +4983,15 @@ int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *
                         return -errno;
         }
 
+        if (userns_fd) {
+                const char *userns;
+
+                userns = procfs_file_alloca(pid, "ns/user");
+                usernsfd = open(userns, O_RDONLY|O_NOCTTY|O_CLOEXEC);
+                if (usernsfd < 0 && errno != ENOENT)
+                        return -errno;
+        }
+
         if (root_fd) {
                 const char *root;
 
@@ -5001,15 +5010,33 @@ int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *
         if (netns_fd)
                 *netns_fd = netnsfd;
 
+        if (userns_fd)
+                *userns_fd = usernsfd;
+
         if (root_fd)
                 *root_fd = rfd;
 
-        pidnsfd = mntnsfd = netnsfd = -1;
+        pidnsfd = mntnsfd = netnsfd = usernsfd = -1;
 
         return 0;
 }
 
-int namespace_enter(int pidns_fd, int mntns_fd, int netns_fd, int root_fd) {
+int namespace_enter(int pidns_fd, int mntns_fd, int netns_fd, int userns_fd, int root_fd) {
+        if (userns_fd >= 0) {
+                /* Can't setns to your own userns, since then you could
+                 * escalate from non-root to root in your own namespace, so
+                 * check if namespaces equal before attempting to enter. */
+                _cleanup_free_ char *userns_fd_path = NULL;
+                int r;
+                if (asprintf(&userns_fd_path, "/proc/self/fd/%d", userns_fd) < 0)
+                        return -ENOMEM;
+
+                r = files_same(userns_fd_path, "/proc/self/ns/user");
+                if (r < 0)
+                        return r;
+                if (r)
+                        userns_fd = -1;
+        }
 
         if (pidns_fd >= 0)
                 if (setns(pidns_fd, CLONE_NEWPID) < 0)
@@ -5023,6 +5050,10 @@ int namespace_enter(int pidns_fd, int mntns_fd, int netns_fd, int root_fd) {
                 if (setns(netns_fd, CLONE_NEWNET) < 0)
                         return -errno;
 
+        if (userns_fd >= 0)
+                if (setns(userns_fd, CLONE_NEWUSER) < 0)
+                        return -errno;
+
         if (root_fd >= 0) {
                 if (fchdir(root_fd) < 0)
                         return -errno;
@@ -6012,7 +6043,7 @@ int ptsname_malloc(int fd, char **ret) {
 }
 
 int openpt_in_namespace(pid_t pid, int flags) {
-        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, rootfd = -1;
+        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, usernsfd = -1, rootfd = -1;
         _cleanup_close_pair_ int pair[2] = { -1, -1 };
         union {
                 struct cmsghdr cmsghdr;
@@ -6029,7 +6060,7 @@ int openpt_in_namespace(pid_t pid, int flags) {
 
         assert(pid > 0);
 
-        r = namespace_open(pid, &pidnsfd, &mntnsfd, NULL, &rootfd);
+        r = namespace_open(pid, &pidnsfd, &mntnsfd, NULL, &usernsfd, &rootfd);
         if (r < 0)
                 return r;
 
@@ -6045,7 +6076,7 @@ int openpt_in_namespace(pid_t pid, int flags) {
 
                 pair[0] = safe_close(pair[0]);
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, rootfd);
+                r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
                 if (r < 0)
                         _exit(EXIT_FAILURE);
 
diff --git a/src/basic/util.h b/src/basic/util.h
index 426b7f7d16..1667b35f00 100644
--- a/src/basic/util.h
+++ b/src/basic/util.h
@@ -803,8 +803,8 @@ int get_proc_cmdline_key(const char *parameter, char **value);
 
 int container_get_leader(const char *machine, pid_t *pid);
 
-int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *root_fd);
-int namespace_enter(int pidns_fd, int mntns_fd, int netns_fd, int root_fd);
+int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *userns_fd, int *root_fd);
+int namespace_enter(int pidns_fd, int mntns_fd, int netns_fd, int userns_fd, int root_fd);
 
 int getpeercred(int fd, struct ucred *ucred);
 int getpeersec(int fd, char **ret);
diff --git a/src/core/machine-id-setup.c b/src/core/machine-id-setup.c
index 8e26362546..2d5ae3b3b9 100644
--- a/src/core/machine-id-setup.c
+++ b/src/core/machine-id-setup.c
@@ -325,7 +325,7 @@ int machine_id_commit(const char *root) {
         fd = safe_close(fd);
 
         /* Store current mount namespace */
-        r = namespace_open(0, NULL, &initial_mntns_fd, NULL, NULL);
+        r = namespace_open(0, NULL, &initial_mntns_fd, NULL, NULL, NULL);
         if (r < 0)
                 return log_error_errno(r, "Can't fetch current mount namespace: %m");
 
@@ -351,7 +351,7 @@ int machine_id_commit(const char *root) {
         fd = safe_close(fd);
 
         /* Return to initial namespace and proceed a lazy tmpfs unmount */
-        r = namespace_enter(-1, initial_mntns_fd, -1, -1);
+        r = namespace_enter(-1, initial_mntns_fd, -1, -1, -1);
         if (r < 0)
                 return log_warning_errno(r, "Failed to switch back to initial mount namespace: %m.\nWe'll keep transient %s file until next reboot.", etc_machine_id);
 
diff --git a/src/libsystemd/sd-bus/bus-container.c b/src/libsystemd/sd-bus/bus-container.c
index fa7a207448..101e4af18d 100644
--- a/src/libsystemd/sd-bus/bus-container.c
+++ b/src/libsystemd/sd-bus/bus-container.c
@@ -29,7 +29,7 @@
 #include "bus-container.h"
 
 int bus_container_connect_socket(sd_bus *b) {
-        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, rootfd = -1;
+        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, usernsfd = -1, rootfd = -1;
         pid_t child;
         siginfo_t si;
         int r;
@@ -45,7 +45,7 @@ int bus_container_connect_socket(sd_bus *b) {
                         return r;
         }
 
-        r = namespace_open(b->nspid, &pidnsfd, &mntnsfd, NULL, &rootfd);
+        r = namespace_open(b->nspid, &pidnsfd, &mntnsfd, NULL, &usernsfd, &rootfd);
         if (r < 0)
                 return r;
 
@@ -64,7 +64,7 @@ int bus_container_connect_socket(sd_bus *b) {
         if (child == 0) {
                 pid_t grandchild;
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, rootfd);
+                r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
                 if (r < 0)
                         _exit(255);
 
@@ -120,7 +120,7 @@ int bus_container_connect_socket(sd_bus *b) {
 
 int bus_container_connect_kernel(sd_bus *b) {
         _cleanup_close_pair_ int pair[2] = { -1, -1 };
-        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, rootfd = -1;
+        _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, usernsfd = -1, rootfd = -1;
         union {
                 struct cmsghdr cmsghdr;
                 uint8_t buf[CMSG_SPACE(sizeof(int))];
@@ -146,7 +146,7 @@ int bus_container_connect_kernel(sd_bus *b) {
                         return r;
         }
 
-        r = namespace_open(b->nspid, &pidnsfd, &mntnsfd, NULL, &rootfd);
+        r = namespace_open(b->nspid, &pidnsfd, &mntnsfd, NULL, &usernsfd, &rootfd);
         if (r < 0)
                 return r;
 
@@ -162,7 +162,7 @@ int bus_container_connect_kernel(sd_bus *b) {
 
                 pair[0] = safe_close(pair[0]);
 
-                r = namespace_enter(pidnsfd, mntnsfd, -1, rootfd);
+                r = namespace_enter(pidnsfd, mntnsfd, -1, usernsfd, rootfd);
                 if (r < 0)
                         _exit(EXIT_FAILURE);
 
diff --git a/src/machine/machine-dbus.c b/src/machine/machine-dbus.c
index 7658d7146d..b62a9bd813 100644
--- a/src/machine/machine-dbus.c
+++ b/src/machine/machine-dbus.c
@@ -212,7 +212,7 @@ int bus_machine_method_get_addresses(sd_bus_message *message, void *userdata, sd
         if (streq(us, them))
                 return sd_bus_error_setf(error, BUS_ERROR_NO_PRIVATE_NETWORKING, "Machine %s does not use private networking", m->name);
 
-        r = namespace_open(m->leader, NULL, NULL, &netns_fd, NULL);
+        r = namespace_open(m->leader, NULL, NULL, &netns_fd, NULL, NULL);
         if (r < 0)
                 return r;
 
@@ -230,7 +230,7 @@ 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);
+                r = namespace_enter(-1, -1, netns_fd, -1, -1);
                 if (r < 0)
                         _exit(EXIT_FAILURE);
 
@@ -346,7 +346,7 @@ int bus_machine_method_get_os_release(sd_bus_message *message, void *userdata, s
         if (m->class != MACHINE_CONTAINER)
                 return sd_bus_error_setf(error, SD_BUS_ERROR_NOT_SUPPORTED, "Requesting OS release data is only supported on container machines.");
 
-        r = namespace_open(m->leader, NULL, &mntns_fd, NULL, &root_fd);
+        r = namespace_open(m->leader, NULL, &mntns_fd, NULL, NULL, &root_fd);
         if (r < 0)
                 return r;
 
@@ -362,7 +362,7 @@ 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, root_fd);
+                r = namespace_enter(-1, mntns_fd, -1, -1, root_fd);
                 if (r < 0)
                         _exit(EXIT_FAILURE);
 
diff --git a/src/shared/logs-show.c b/src/shared/logs-show.c
index 068da465d9..b78cb7678c 100644
--- a/src/shared/logs-show.c
+++ b/src/shared/logs-show.c
@@ -1141,7 +1141,7 @@ static int get_boot_id_for_machine(const char *machine, sd_id128_t *boot_id) {
         if (r < 0)
                 return r;
 
-        r = namespace_open(pid, &pidnsfd, &mntnsfd, NULL, &rootfd);
+        r = namespace_open(pid, &pidnsfd, &mntnsfd, NULL, NULL, &rootfd);
         if (r < 0)
                 return r;
 
@@ -1157,7 +1157,7 @@ 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, rootfd);
+                r = namespace_enter(pidnsfd, mntnsfd, -1, -1, rootfd);
                 if (r < 0)
                         _exit(EXIT_FAILURE);