SymbX/systemd
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #25513 from brauner/pivot_root.nspawn

Browse Source 
nspawn: support pivot_root()
This commit is contained in:
Luca Boccassi
2022-12-06 01:51:51 +01:00
committed by GitHub
parent 379864f890 e79581ddfe
commit 9cd4881d47
6 changed files with 227 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/nspawn/nspawn.c
+76 -29
View File
@@ -114,6 +114,7 @@
/* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
#define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
#define NSPAWN_MOUNT_TUNNEL "/run/host/incoming"
#define EXIT_FORCE_RESTART 133
@@ -2776,7 +2777,7 @@ static int reset_audit_loginuid(void) {
return 0;
}
static int setup_propagate(const char *root) {
static int mount_tunnel_dig(const char *root) {
const char *p, *q;
int r;
@@ -2789,11 +2790,11 @@ static int setup_propagate(const char *root) {
if (r < 0)
return log_error_errno(r, "Failed to create /run/host: %m");
r = userns_mkdir(root, "/run/host/incoming", 0600, 0, 0);
r = userns_mkdir(root, NSPAWN_MOUNT_TUNNEL, 0600, 0, 0);
if (r < 0)
return log_error_errno(r, "Failed to create /run/host/incoming: %m");
return log_error_errno(r, "Failed to create "NSPAWN_MOUNT_TUNNEL": %m");
q = prefix_roota(root, "/run/host/incoming");
q = prefix_roota(root, NSPAWN_MOUNT_TUNNEL);
r = mount_nofollow_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
if (r < 0)
return r;
@@ -2802,8 +2803,17 @@ static int setup_propagate(const char *root) {
if (r < 0)
return r;
/* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
return 0;
}
static int mount_tunnel_open(void) {
int r;
r = mount_follow_verbose(LOG_ERR, NULL, NSPAWN_MOUNT_TUNNEL, NULL, MS_SLAVE, NULL);
if (r < 0)
return r;
return 0;
}
static int setup_machine_id(const char *directory) {
@@ -3632,7 +3642,7 @@ static int outer_child(
_cleanup_(bind_user_context_freep) BindUserContext *bind_user_context = NULL;
_cleanup_strv_free_ char **os_release_pairs = NULL;
_cleanup_close_ int fd = -1;
_cleanup_close_ int fd = -1, mntns_fd = -EBADF;
bool idmap = false;
const char *p;
pid_t pid;
@@ -3697,6 +3707,15 @@ static int outer_child(
return r;
if (arg_userns_mode != USER_NAMESPACE_NO) {
r = namespace_open(0, NULL, &mntns_fd, NULL, NULL, NULL);
if (r < 0)
return log_error_errno(r, "Failed to pin outer mount namespace: %m");
l = send_one_fd(notify_socket, mntns_fd, 0);
if (l < 0)
return log_error_errno(l, "Failed to send outer mount namespace fd: %m");
mntns_fd = safe_close(mntns_fd);
/* Let the parent know which UID shift we read from the image */
l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
if (l < 0)
@@ -3858,19 +3877,6 @@ static int outer_child(
unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
}
/* Mark everything as shared so our mounts get propagated down. This is required to make new bind
* mounts available in systemd services inside the container that create a new mount namespace. See
* https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
* will inherit the shared propagation mode.
*
* IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
* directory mount to root later on.
* https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
*/
r = mount_nofollow_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
if (r < 0)
return r;
r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
if (r < 0)
return r;
@@ -3910,7 +3916,7 @@ static int outer_child(
if (r < 0)
return r;
r = setup_propagate(directory);
r = mount_tunnel_dig(directory);
if (r < 0)
return r;
@@ -3974,10 +3980,40 @@ static int outer_child(
return r;
}
r = mount_move_root(directory);
/* Mark everything as shared so our mounts get propagated down. This is required to make new bind
* mounts available in systemd services inside the container that create a new mount namespace. See
* https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
* will inherit the shared propagation mode.
*
* IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
* directory mount to root later on.
* https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
*/
r = mount_switch_root(directory, MOUNT_ATTR_PROPAGATION_SHARED);
if (r < 0)
return log_error_errno(r, "Failed to move root directory: %m");
/* We finished setting up the rootfs which is a shared mount. The mount tunnel needs to be a
* dependent mount otherwise we can't MS_MOVE mounts that were propagated from the host into
* the container. */
r = mount_tunnel_open();
if (r < 0)
return r;
if (arg_userns_mode != USER_NAMESPACE_NO) {
/* In order to mount procfs and sysfs in an unprivileged container the kernel
* requires that a fully visible instance is already present in the target mount
* namespace. Mount one here so the inner child can mount its own instances. Later
* we umount the temporary instances created here before we actually exec the
* payload. Since the rootfs is shared the umount will propagate into the container.
* Note, the inner child wouldn't be able to unmount the instances on its own since
* it doesn't own the originating mount namespace. IOW, the outer child needs to do
* this. */
r = pin_fully_visible_fs();
if (r < 0)
return r;
}
fd = setup_notify_child();
if (fd < 0)
return fd;
@@ -4735,12 +4771,12 @@ static int run_container(
rtnl_socket_pair[2] = { -1, -1 },
pid_socket_pair[2] = { -1, -1 },
uuid_socket_pair[2] = { -1, -1 },
notify_socket_pair[2] = { -1, -1 },
fd_socket_pair[2] = { -EBADF, -EBADF },
uid_shift_socket_pair[2] = { -1, -1 },
master_pty_socket_pair[2] = { -1, -1 },
unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
_cleanup_close_ int notify_socket = -1;
_cleanup_close_ int notify_socket = -1, mntns_fd = -EBADF;
_cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
_cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
_cleanup_(sd_event_unrefp) sd_event *event = NULL;
@@ -4787,7 +4823,7 @@ static int run_container(
if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
return log_error_errno(errno, "Failed to create id socket pair: %m");
if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, fd_socket_pair) < 0)
return log_error_errno(errno, "Failed to create notify socket pair: %m");
if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
@@ -4840,7 +4876,7 @@ static int run_container(
rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
fd_socket_pair[0] = safe_close(fd_socket_pair[0]);
master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
@@ -4854,7 +4890,7 @@ static int run_container(
secondary,
pid_socket_pair[1],
uuid_socket_pair[1],
notify_socket_pair[1],
fd_socket_pair[1],
kmsg_socket_pair[1],
rtnl_socket_pair[1],
uid_shift_socket_pair[1],
@@ -4876,12 +4912,16 @@ static int run_container(
rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
fd_socket_pair[1] = safe_close(fd_socket_pair[1]);
master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
if (arg_userns_mode != USER_NAMESPACE_NO) {
mntns_fd = receive_one_fd(fd_socket_pair[0], 0);
if (mntns_fd < 0)
return log_error_errno(mntns_fd, "Failed to receive mount namespace fd from outer child: %m");
/* The child just let us know the UID shift it might have read from the image. */
l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
if (l < 0)
@@ -4958,7 +4998,7 @@ static int run_container(
return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
/* We also retrieve the socket used for notifications generated by outer child */
notify_socket = receive_one_fd(notify_socket_pair[0], 0);
notify_socket = receive_one_fd(fd_socket_pair[0], 0);
if (notify_socket < 0)
return log_error_errno(notify_socket,
"Failed to receive notification socket from the outer child: %m");
@@ -5143,6 +5183,13 @@ static int run_container(
if (r < 0)
return r;
if (arg_userns_mode != USER_NAMESPACE_NO) {
r = wipe_fully_visible_fs(mntns_fd);
if (r < 0)
return r;
mntns_fd = safe_close(mntns_fd);
}
/* Let the child know that we are ready and wait that the child is completely ready now. */
if (!barrier_place_and_sync(&barrier)) /* #5 */
return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");