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

nspawn: support pivot_root()

Browse Source 
In order to support pivot_root() we need to move mount propagation
changes after the pivot_root(). While MS_MOVE requires the source mount
to not be a shared mount pivot_root() also requires the target mount to
not be a shared mount. This guarantees that pivot_root() doesn't leak
any mounts.

Signed-off-by: Christian Brauner (Microsoft) <brauner@kernel.org>
This commit is contained in:
Christian Brauner
2022-12-05 18:34:25 +01:00
committed by Christian Brauner (Microsoft)
parent 2e776ed6c8
commit 57c10a5650
4 changed files with 96 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/core/namespace.c
+2 -2
View File
@@ -2486,7 +2486,7 @@ int setup_namespace(
goto finish;
/* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
r = mount_pivot_root(root);
r = mount_switch_root(root, MOUNT_ATTR_PROPAGATION_INHERIT);
if (r == -EINVAL && root_directory) {
/* If we are using root_directory and we don't have privileges (ie: user manager in a user
* namespace) and the root_directory is already a mount point in the parent namespace,
@@ -2496,7 +2496,7 @@ int setup_namespace(
r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL);
if (r < 0)
goto finish;
r = mount_pivot_root(root);
r = mount_switch_root(root, MOUNT_ATTR_PROPAGATION_INHERIT);
}
if (r < 0) {
log_debug_errno(r, "Failed to mount root with MS_MOVE: %m");
src/nspawn/nspawn.c
+10 -14
View File
@@ -3858,19 +3858,6 @@ 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;
@@ -3974,7 +3961,16 @@ 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");
src/shared/mount-util.c
+69 -28
View File
@@ -36,6 +36,7 @@
#include "set.h"
#include "stat-util.h"
#include "stdio-util.h"
#include "string-table.h"
#include "string-util.h"
#include "strv.h"
#include "tmpfile-util.h"
@@ -475,47 +476,41 @@ int bind_remount_one_with_mountinfo(
return 0;
}
int mount_move_root(const char *path) {
assert(path);
static const char *const mount_attr_propagation_type_table[_MOUNT_ATTR_PROPAGATION_TYPE_MAX] = {
[MOUNT_ATTR_PROPAGATION_INHERIT] = "inherited",
[MOUNT_ATTR_PROPAGATION_PRIVATE] = "private",
[MOUNT_ATTR_PROPAGATION_DEPENDENT] = "dependent",
[MOUNT_ATTR_PROPAGATION_SHARED] = "shared",
};
if (chdir(path) < 0)
return -errno;
DEFINE_STRING_TABLE_LOOKUP(mount_attr_propagation_type, MountAttrPropagationType);
if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
return -errno;
if (chroot(".") < 0)
return -errno;
return RET_NERRNO(chdir("/"));
unsigned int mount_attr_propagation_type_to_flag(MountAttrPropagationType t) {
switch (t) {
case MOUNT_ATTR_PROPAGATION_INHERIT:
return 0;
case MOUNT_ATTR_PROPAGATION_PRIVATE:
return MS_PRIVATE;
case MOUNT_ATTR_PROPAGATION_DEPENDENT:
return MS_SLAVE;
case MOUNT_ATTR_PROPAGATION_SHARED:
return MS_SHARED;
default:
assert_not_reached();
}
}
int mount_pivot_root(const char *path) {
_cleanup_close_ int fd_oldroot = -EBADF, fd_newroot = -EBADF;
assert(path);
/* pivot_root() isn't currently supported in the initramfs. */
if (in_initrd())
return mount_move_root(path);
static inline int mount_switch_root_pivot(const char *path, int fd_newroot) {
_cleanup_close_ int fd_oldroot = -EBADF;
fd_oldroot = open("/", O_PATH|O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW);
if (fd_oldroot < 0)
return log_debug_errno(errno, "Failed to open old rootfs");
fd_newroot = open(path, O_PATH|O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW);
if (fd_newroot < 0)
return log_debug_errno(errno, "Failed to open new rootfs '%s': %m", path);
/* Change into the new rootfs. */
if (fchdir(fd_newroot) < 0)
return log_debug_errno(errno, "Failed to change into new rootfs '%s': %m", path);
/* Let the kernel tuck the new root under the old one. */
if (pivot_root(".", ".") < 0)
return log_debug_errno(errno, "Failed to pivot root to new rootfs '%s': %m", path);
/* At this point the new root is tucked under the old root. If we want
* to unmount it we cannot be fchdir()ed into it. So escape back to the
* old root. */
@@ -535,6 +530,52 @@ int mount_pivot_root(const char *path) {
return 0;
}
static inline int mount_switch_root_move(const char *path) {
if (mount(path, "/", NULL, MS_MOVE, NULL) < 0)
return log_debug_errno(errno, "Failed to move new rootfs '%s': %m", path);
if (chroot(".") < 0)
return log_debug_errno(errno, "Failed to chroot to new rootfs '%s': %m", path);
if (chdir("/"))
return log_debug_errno(errno, "Failed to chdir to new rootfs '%s': %m", path);
return 0;
}
int mount_switch_root(const char *path, MountAttrPropagationType type) {
int r;
_cleanup_close_ int fd_newroot = -EBADF;
unsigned int flags;
assert(path);
fd_newroot = open(path, O_PATH|O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW);
if (fd_newroot < 0)
return log_debug_errno(errno, "Failed to open new rootfs '%s': %m", path);
/* Change into the new rootfs. */
if (fchdir(fd_newroot) < 0)
return log_debug_errno(errno, "Failed to change into new rootfs '%s': %m", path);
r = mount_switch_root_pivot(path, fd_newroot);
if (r < 0) {
/* Failed to pivot_root() fallback to MS_MOVE. For example, this may happen if the
* rootfs is an initramfs in which case pivot_root() isn't supported. */
log_debug_errno(r, "Failed to pivot into new rootfs '%s': %m", path);
r = mount_switch_root_move(path);
}
if (r < 0)
return log_debug_errno(r, "Failed to switch to new rootfs '%s': %m", path);
/* Finally, let's establish the requested propagation type. */
flags = mount_attr_propagation_type_to_flag(type);
if ((flags != 0) && mount(NULL, ".", NULL, flags|MS_REC, 0) < 0)
return log_debug_errno(errno, "Failed to turn new rootfs '%s' into %s mount: %m",
mount_attr_propagation_type_to_string(type), path);
return 0;
}
int repeat_unmount(const char *path, int flags) {
bool done = false;
src/shared/mount-util.h
+15 -2
View File
@@ -11,6 +11,20 @@
#include "errno-util.h"
#include "macro.h"
typedef enum MountAttrPropagationType {
MOUNT_ATTR_PROPAGATION_INHERIT, /* no special MS_* propagation flags */
MOUNT_ATTR_PROPAGATION_PRIVATE, /* MS_PRIVATE */
MOUNT_ATTR_PROPAGATION_DEPENDENT, /* MS_SLAVE */
MOUNT_ATTR_PROPAGATION_SHARED, /* MS_SHARE */
_MOUNT_ATTR_PROPAGATION_TYPE_MAX,
_MOUNT_ATTR_PROPAGATION_TYPE_INVALID = -EINVAL,
} MountAttrPropagationType;
const char* mount_attr_propagation_type_to_string(MountAttrPropagationType t) _const_;
MountAttrPropagationType mount_attr_propagation_type_from_string(const char *s) _pure_;
unsigned int mount_attr_propagation_type_to_flag(MountAttrPropagationType t);
/* The limit used for /dev itself. 4MB should be enough since device nodes and symlinks don't
* consume any space and udev isn't supposed to create regular file either. There's no limit on the
* max number of inodes since such limit is hard to guess especially on large storage array
@@ -54,8 +68,7 @@ static inline int bind_remount_recursive(const char *prefix, unsigned long new_f
int bind_remount_one_with_mountinfo(const char *path, unsigned long new_flags, unsigned long flags_mask, FILE *proc_self_mountinfo);
int mount_move_root(const char *path);
int mount_pivot_root(const char *path);
int mount_switch_root(const char *path, MountAttrPropagationType type);
DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(FILE*, endmntent, NULL);
#define _cleanup_endmntent_ _cleanup_(endmntentp)