shaba/systemd
1
0
Fork 0
mirror of https://github.com/systemd/systemd.git synced 2024-12-26 03:22:00 +03:00
Code

namespace: make setup_namespace() less crazy

Browse Source 
Let's replace the ridiculous number of arguments with a structure, to
make this function less weird.

No change in behaviour, just some refactoring.
This commit is contained in:
Lennart Poettering 2023-10-10 21:36:50 +02:00 committed by Yu Watanabe
parent 22d7fb6646
commit 79d956db34
5 changed files with 332 additions and 426 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

152
src/core/execute.c
View File

@ -3138,7 +3138,6 @@ static int apply_mount_namespace(
*extension_dir = NULL, *host_os_release_stage = NULL;
const char *root_dir = NULL, *root_image = NULL, *tmp_dir = NULL, *var_tmp_dir = NULL;
char **read_write_paths;
NamespaceInfo ns_info;
bool needs_sandboxing, setup_os_release_symlink;
BindMount *bind_mounts = NULL;
size_t n_bind_mounts = 0;
@ -3180,10 +3179,9 @@ static int apply_mount_namespace(
needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command_flags & EXEC_COMMAND_FULLY_PRIVILEGED);
if (needs_sandboxing) {
/* The runtime struct only contains the parent of the private /tmp,
* which is non-accessible to world users. Inside of it there's a /tmp
* that is sticky, and that's the one we want to use here.
* This does not apply when we are using /run/systemd/empty as fallback. */
/* The runtime struct only contains the parent of the private /tmp, which is non-accessible
* to world users. Inside of it there's a /tmp that is sticky, and that's the one we want to
* use here. This does not apply when we are using /run/systemd/empty as fallback. */
if (context->private_tmp && runtime && runtime->shared) {
if (streq_ptr(runtime->shared->tmp_dir, RUN_SYSTEMD_EMPTY))
@ -3196,39 +3194,10 @@ static int apply_mount_namespace(
else if (runtime->shared->var_tmp_dir)
var_tmp_dir = strjoina(runtime->shared->var_tmp_dir, "/tmp");
}
ns_info = (NamespaceInfo) {
.ignore_protect_paths = false,
.private_dev = context->private_devices,
.protect_control_groups = context->protect_control_groups,
.protect_kernel_tunables = context->protect_kernel_tunables,
.protect_kernel_modules = context->protect_kernel_modules,
.protect_kernel_logs = context->protect_kernel_logs,
.protect_hostname = context->protect_hostname,
.mount_apivfs = exec_context_get_effective_mount_apivfs(context),
.protect_home = context->protect_home,
.protect_system = context->protect_system,
.protect_proc = context->protect_proc,
.proc_subset = context->proc_subset,
.private_network = exec_needs_network_namespace(context),
.private_ipc = exec_needs_ipc_namespace(context),
/* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
.mount_nosuid = context->no_new_privileges && !mac_selinux_use(),
};
} else if (!context->dynamic_user && root_dir)
/*
* If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
* sandbox info, otherwise enforce it, don't ignore protected paths and
* fail if we are enable to apply the sandbox inside the mount namespace.
*/
ns_info = (NamespaceInfo) {
.ignore_protect_paths = true,
};
else
ns_info = (NamespaceInfo) {};
}
/* Symlinks (exec dirs, os-release) are set up after other mounts, before they are made read-only. */
setup_os_release_symlink = ns_info.mount_apivfs && (root_dir || root_image);
setup_os_release_symlink = needs_sandboxing && exec_context_get_effective_mount_apivfs(context) && (root_dir || root_image);
r = compile_symlinks(context, params, setup_os_release_symlink, &symlinks);
if (r < 0)
return r;
@ -3287,44 +3256,81 @@ static int apply_mount_namespace(
return r;
}
r = setup_namespace(
root_dir,
root_image,
context->root_image_options,
context->root_image_policy ?: &image_policy_service,
&ns_info,
read_write_paths,
needs_sandboxing ? context->read_only_paths : NULL,
needs_sandboxing ? context->inaccessible_paths : NULL,
needs_sandboxing ? context->exec_paths : NULL,
needs_sandboxing ? context->no_exec_paths : NULL,
empty_directories,
symlinks,
bind_mounts,
n_bind_mounts,
context->temporary_filesystems,
context->n_temporary_filesystems,
context->mount_images,
context->n_mount_images,
context->mount_image_policy ?: &image_policy_service,
tmp_dir,
var_tmp_dir,
creds_path,
context->log_namespace,
context->mount_propagation_flag,
&verity,
context->extension_images,
context->n_extension_images,
context->extension_image_policy ?: &image_policy_sysext,
context->extension_directories,
propagate_dir,
incoming_dir,
extension_dir,
root_dir || root_image ? params->notify_socket : NULL,
host_os_release_stage,
params->runtime_scope,
error_path);
NamespaceParameters parameters = {
.runtime_scope = params->runtime_scope,
.root_directory = root_dir,
.root_image = root_image,
.root_image_options = context->root_image_options,
.root_image_policy = context->root_image_policy ?: &image_policy_service,
.read_write_paths = read_write_paths,
.read_only_paths = needs_sandboxing ? context->read_only_paths : NULL,
.inaccessible_paths = needs_sandboxing ? context->inaccessible_paths : NULL,
.exec_paths = needs_sandboxing ? context->exec_paths : NULL,
.no_exec_paths = needs_sandboxing ? context->no_exec_paths : NULL,
.empty_directories = empty_directories,
.symlinks = symlinks,
.bind_mounts = bind_mounts,
.n_bind_mounts = n_bind_mounts,
.temporary_filesystems = context->temporary_filesystems,
.n_temporary_filesystems = context->n_temporary_filesystems,
.mount_images = context->mount_images,
.n_mount_images = context->n_mount_images,
.mount_image_policy = context->mount_image_policy ?: &image_policy_service,
.tmp_dir = tmp_dir,
.var_tmp_dir = var_tmp_dir,
.creds_path = creds_path,
.log_namespace = context->log_namespace,
.mount_propagation_flag = context->mount_propagation_flag,
.verity = &verity,
.extension_images = context->extension_images,
.n_extension_images = context->n_extension_images,
.extension_image_policy = context->extension_image_policy ?: &image_policy_sysext,
.extension_directories = context->extension_directories,
.propagate_dir = propagate_dir,
.incoming_dir = incoming_dir,
.extension_dir = extension_dir,
.notify_socket = root_dir || root_image ? params->notify_socket : NULL,
.host_os_release_stage = host_os_release_stage,
/* If DynamicUser=no and RootDirectory= is set then lets pass a relaxed sandbox info,
* otherwise enforce it, don't ignore protected paths and fail if we are enable to apply the
* sandbox inside the mount namespace. */
.ignore_protect_paths = !needs_sandboxing && !context->dynamic_user && root_dir,
.protect_control_groups = needs_sandboxing && context->protect_control_groups,
.protect_kernel_tunables = needs_sandboxing && context->protect_kernel_tunables,
.protect_kernel_modules = needs_sandboxing && context->protect_kernel_modules,
.protect_kernel_logs = needs_sandboxing && context->protect_kernel_logs,
.protect_hostname = needs_sandboxing && context->protect_hostname,
.private_dev = needs_sandboxing && context->private_devices,
.private_network = needs_sandboxing && exec_needs_network_namespace(context),
.private_ipc = needs_sandboxing && exec_needs_ipc_namespace(context),
.mount_apivfs = needs_sandboxing && exec_context_get_effective_mount_apivfs(context),
/* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
.mount_nosuid = needs_sandboxing && context->no_new_privileges && !mac_selinux_use(),
.protect_home = needs_sandboxing && context->protect_home,
.protect_system = needs_sandboxing && context->protect_system,
.protect_proc = needs_sandboxing && context->protect_proc,
.proc_subset = needs_sandboxing && context->proc_subset,
};
r = setup_namespace(&parameters, error_path);
/* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
* that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
* sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a

357
src/core/namespace.c
View File

@ -1121,16 +1121,16 @@ static int mount_private_sysfs(const MountEntry *m) {
return 0;
}
static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) {
static int mount_procfs(const MountEntry *m, const NamespaceParameters *p) {
_cleanup_free_ char *opts = NULL;
const char *entry_path;
int r, n;
assert(m);
assert(ns_info);
assert(p);
if (ns_info->protect_proc != PROTECT_PROC_DEFAULT ||
ns_info->proc_subset != PROC_SUBSET_ALL) {
if (p->protect_proc != PROTECT_PROC_DEFAULT ||
p->proc_subset != PROC_SUBSET_ALL) {
/* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
* pretended to be per-instance but actually was per-namespace), hence let's make use of it
@ -1138,9 +1138,9 @@ static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) {
* per-instance, we'll exclusively use the textual value for hidepid=, since support was
* added in the same commit: if it's supported it is thus also per-instance. */
const char *hpv = ns_info->protect_proc == PROTECT_PROC_DEFAULT ?
const char *hpv = p->protect_proc == PROTECT_PROC_DEFAULT ?
"off" :
protect_proc_to_string(ns_info->protect_proc);
protect_proc_to_string(p->protect_proc);
/* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
* 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
@ -1154,7 +1154,7 @@ static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) {
return -ENOMEM;
}
if (ns_info->proc_subset == PROC_SUBSET_PID &&
if (p->proc_subset == PROC_SUBSET_PID &&
mount_option_supported("proc", "subset", "pid") != 0)
if (!strextend_with_separator(&opts, ",", "subset=pid"))
return -ENOMEM;
@ -1383,10 +1383,7 @@ static int follow_symlink(
static int apply_one_mount(
const char *root_directory,
MountEntry *m,
const ImagePolicy *mount_image_policy,
const ImagePolicy *extension_image_policy,
const NamespaceInfo *ns_info,
RuntimeScope scope) {
const NamespaceParameters *p) {
_cleanup_free_ char *inaccessible = NULL;
bool rbind = true, make = false;
@ -1394,7 +1391,7 @@ static int apply_one_mount(
int r;
assert(m);
assert(ns_info);
assert(p);
log_debug("Applying namespace mount on %s", mount_entry_path(m));
@ -1544,7 +1541,7 @@ static int apply_one_mount(
break;
case PRIVATE_DEV:
return mount_private_dev(m, scope);
return mount_private_dev(m, p->runtime_scope);
case BIND_DEV:
return mount_bind_dev(m);
@ -1556,7 +1553,7 @@ static int apply_one_mount(
return mount_bind_sysfs(m);
case PROCFS:
return mount_procfs(m, ns_info);
return mount_procfs(m, p);
case RUN:
return mount_run(m);
@ -1565,10 +1562,10 @@ static int apply_one_mount(
return mount_mqueuefs(m);
case MOUNT_IMAGES:
return mount_image(m, NULL, mount_image_policy);
return mount_image(m, NULL, p->mount_image_policy);
case EXTENSION_IMAGES:
return mount_image(m, root_directory, extension_image_policy);
return mount_image(m, root_directory, p->extension_image_policy);
case OVERLAY_MOUNT:
return mount_overlay(m);
@ -1709,8 +1706,8 @@ static int make_nosuid(const MountEntry *m, FILE *proc_self_mountinfo) {
return 0;
}
static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) {
assert(ns_info);
static bool namespace_parameters_mount_apivfs(const NamespaceParameters *p) {
assert(p);
/*
* ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
@ -1718,81 +1715,64 @@ static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) {
* first place...
*/
return ns_info->mount_apivfs ||
ns_info->protect_control_groups ||
ns_info->protect_kernel_tunables ||
ns_info->protect_proc != PROTECT_PROC_DEFAULT ||
ns_info->proc_subset != PROC_SUBSET_ALL;
return p->mount_apivfs ||
p->protect_control_groups ||
p->protect_kernel_tunables ||
p->protect_proc != PROTECT_PROC_DEFAULT ||
p->proc_subset != PROC_SUBSET_ALL;
}
static size_t namespace_calculate_mounts(
const NamespaceInfo *ns_info,
char** read_write_paths,
char** read_only_paths,
char** inaccessible_paths,
char** exec_paths,
char** no_exec_paths,
char** empty_directories,
size_t n_bind_mounts,
size_t n_temporary_filesystems,
size_t n_mount_images,
size_t n_extension_images,
size_t n_extension_directories,
size_t n_hierarchies,
const char* tmp_dir,
const char* var_tmp_dir,
const char *creds_path,
const char* log_namespace,
bool setup_propagate,
const char* notify_socket,
const char* host_os_release) {
const NamespaceParameters *p,
char **hierarchies,
bool setup_propagate) {
size_t protect_home_cnt;
size_t protect_system_cnt =
(ns_info->protect_system == PROTECT_SYSTEM_STRICT ?
(p->protect_system == PROTECT_SYSTEM_STRICT ?
ELEMENTSOF(protect_system_strict_table) :
((ns_info->protect_system == PROTECT_SYSTEM_FULL) ?
((p->protect_system == PROTECT_SYSTEM_FULL) ?
ELEMENTSOF(protect_system_full_table) :
((ns_info->protect_system == PROTECT_SYSTEM_YES) ?
((p->protect_system == PROTECT_SYSTEM_YES) ?
ELEMENTSOF(protect_system_yes_table) : 0)));
protect_home_cnt =
(ns_info->protect_home == PROTECT_HOME_YES ?
(p->protect_home == PROTECT_HOME_YES ?
ELEMENTSOF(protect_home_yes_table) :
((ns_info->protect_home == PROTECT_HOME_READ_ONLY) ?
((p->protect_home == PROTECT_HOME_READ_ONLY) ?
ELEMENTSOF(protect_home_read_only_table) :
((ns_info->protect_home == PROTECT_HOME_TMPFS) ?
((p->protect_home == PROTECT_HOME_TMPFS) ?
ELEMENTSOF(protect_home_tmpfs_table) : 0)));
return !!tmp_dir + !!var_tmp_dir +
strv_length(read_write_paths) +
strv_length(read_only_paths) +
strv_length(inaccessible_paths) +
strv_length(exec_paths) +
strv_length(no_exec_paths) +
strv_length(empty_directories) +
n_bind_mounts +
n_mount_images +
(n_extension_images > 0 || n_extension_directories > 0 ? /* Mount each image and directory plus an overlay per hierarchy */
n_hierarchies + n_extension_images + n_extension_directories: 0) +
n_temporary_filesystems +
ns_info->private_dev +
(ns_info->protect_kernel_tunables ?
return !!p->tmp_dir + !!p->var_tmp_dir +
strv_length(p->read_write_paths) +
strv_length(p->read_only_paths) +
strv_length(p->inaccessible_paths) +
strv_length(p->exec_paths) +
strv_length(p->no_exec_paths) +
strv_length(p->empty_directories) +
p->n_bind_mounts +
p->n_mount_images +
(p->n_extension_images > 0 || !strv_isempty(p->extension_directories) ? /* Mount each image and directory plus an overlay per hierarchy */
strv_length(hierarchies) + p->n_extension_images + strv_length(p->extension_directories) : 0) +
p->n_temporary_filesystems +
p->private_dev +
(p->protect_kernel_tunables ?
ELEMENTSOF(protect_kernel_tunables_proc_table) + ELEMENTSOF(protect_kernel_tunables_sys_table) : 0) +
(ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) +
(ns_info->protect_kernel_logs ?
(p->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) +
(p->protect_kernel_logs ?
ELEMENTSOF(protect_kernel_logs_proc_table) + ELEMENTSOF(protect_kernel_logs_dev_table) : 0) +
(ns_info->protect_control_groups ? 1 : 0) +
(p->protect_control_groups ? 1 : 0) +
protect_home_cnt + protect_system_cnt +
(ns_info->protect_hostname ? 2 : 0) +
(namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0) +
(creds_path ? 2 : 1) +
!!log_namespace +
(p->protect_hostname ? 2 : 0) +
(namespace_parameters_mount_apivfs(p) ? ELEMENTSOF(apivfs_table) : 0) +
(p->creds_path ? 2 : 1) +
!!p->log_namespace +
setup_propagate + /* /run/systemd/incoming */
!!notify_socket +
!!host_os_release +
ns_info->private_network + /* /sys */
ns_info->private_ipc; /* /dev/mqueue */
!!p->notify_socket +
!!p->host_os_release_stage +
p->private_network + /* /sys */
p->private_ipc; /* /dev/mqueue */
}
/* Walk all mount entries and dropping any unused mounts. This affects all
@ -1880,13 +1860,9 @@ static void mount_entry_path_debug_string(const char *root, MountEntry *m, char
static int apply_mounts(
const char *root,
const ImagePolicy *mount_image_policy,
const ImagePolicy *extension_image_policy,
const NamespaceInfo *ns_info,
const NamespaceParameters *p,
MountEntry *mounts,
size_t *n_mounts,
RuntimeScope scope,
char **symlinks,
char **error_path) {
_cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
@ -1936,7 +1912,7 @@ static int apply_mounts(
break;
}
r = apply_one_mount(root, m, mount_image_policy, extension_image_policy, ns_info, scope);
r = apply_one_mount(root, m, p);
if (r < 0) {
mount_entry_path_debug_string(root, m, error_path);
return r;
@ -1955,7 +1931,7 @@ static int apply_mounts(
* read-only switches are flipped, create the exec dirs and other symlinks.
* Note that when /var/lib is not empty/tmpfs, these symlinks will already
* exist, which means this will be a no-op. */
r = create_symlinks_from_tuples(root, symlinks);
r = create_symlinks_from_tuples(root, p->symlinks);
if (r < 0)
return log_debug_errno(r, "Failed to set up symlinks inside mount namespace: %m");
@ -1991,7 +1967,7 @@ static int apply_mounts(
}
/* Fourth round, flip the nosuid bits without a deny list. */
if (ns_info->mount_nosuid)
if (p->mount_nosuid)
for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) {
r = make_nosuid(m, proc_self_mountinfo);
if (r < 0) {
@ -2052,49 +2028,13 @@ static bool home_read_only(
return false;
}
int setup_namespace(
const char* root_directory,
const char* root_image,
const MountOptions *root_image_mount_options,
const ImagePolicy *root_image_policy,
const NamespaceInfo *ns_info,
char** read_write_paths,
char** read_only_paths,
char** inaccessible_paths,
char** exec_paths,
char** no_exec_paths,
char** empty_directories,
char** symlinks,
const BindMount *bind_mounts,
size_t n_bind_mounts,
const TemporaryFileSystem *temporary_filesystems,
size_t n_temporary_filesystems,
const MountImage *mount_images,
size_t n_mount_images,
const ImagePolicy *mount_image_policy,
const char* tmp_dir,
const char* var_tmp_dir,
const char *creds_path,
const char *log_namespace,
unsigned long mount_propagation_flag,
VeritySettings *verity,
const MountImage *extension_images,
size_t n_extension_images,
const ImagePolicy *extension_image_policy,
char **extension_directories,
const char *propagate_dir,
const char *incoming_dir,
const char *extension_dir,
const char *notify_socket,
const char *host_os_release_stage,
RuntimeScope scope,
char **error_path) {
int setup_namespace(const NamespaceParameters *p, char **error_path) {
_cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
_cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
_cleanup_strv_free_ char **hierarchies = NULL;
MountEntry *m = NULL, *mounts = NULL;
bool require_prefix = false, setup_propagate = false;
bool require_prefix = false;
const char *root;
DissectImageFlags dissect_image_flags =
DISSECT_IMAGE_GENERIC_ROOT |
@ -2109,32 +2049,29 @@ int setup_namespace(
size_t n_mounts;
int r;
assert(ns_info);
assert(p);
/* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
* we configure take effect */
BLOCK_WITH_UMASK(0000);
if (!isempty(propagate_dir) && !isempty(incoming_dir))
setup_propagate = true;
bool setup_propagate = !isempty(p->propagate_dir) && !isempty(p->incoming_dir);
unsigned long mount_propagation_flag = p->mount_propagation_flag != 0 ? p->mount_propagation_flag : MS_SHARED;
if (mount_propagation_flag == 0)
mount_propagation_flag = MS_SHARED;
if (root_image) {
if (p->root_image) {
/* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
if (root_read_only(read_only_paths,
ns_info->protect_system) &&
home_read_only(read_only_paths, inaccessible_paths, empty_directories,
bind_mounts, n_bind_mounts, temporary_filesystems, n_temporary_filesystems,
ns_info->protect_home) &&
strv_isempty(read_write_paths))
if (root_read_only(p->read_only_paths,
p->protect_system) &&
home_read_only(p->read_only_paths, p->inaccessible_paths, p->empty_directories,
p->bind_mounts, p->n_bind_mounts, p->temporary_filesystems, p->n_temporary_filesystems,
p->protect_home) &&
strv_isempty(p->read_write_paths))
dissect_image_flags |= DISSECT_IMAGE_READ_ONLY;
SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, verity && verity->data_path);
SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, p->verity && p->verity->data_path);
r = loop_device_make_by_path(
root_image,
p->root_image,
FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */,
/* sector_size= */ UINT32_MAX,
FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
@ -2145,9 +2082,9 @@ int setup_namespace(
r = dissect_loop_device(
loop_device,
verity,
root_image_mount_options,
root_image_policy,
p->verity,
p->root_image_options,
p->root_image_policy,
dissect_image_flags,
&dissected_image);
if (r < 0)
@ -2156,21 +2093,21 @@ int setup_namespace(
r = dissected_image_load_verity_sig_partition(
dissected_image,
loop_device->fd,
verity);
p->verity);
if (r < 0)
return r;
r = dissected_image_decrypt(
dissected_image,
NULL,
verity,
p->verity,
dissect_image_flags);
if (r < 0)
return log_debug_errno(r, "Failed to decrypt dissected image: %m");
}
if (root_directory)
root = root_directory;
if (p->root_directory)
root = p->root_directory;
else {
/* /run/systemd should have been created by PID 1 early on already, but in some cases, like
* when running tests (test-execute), it might not have been created yet so let's make sure
@ -2189,7 +2126,7 @@ int setup_namespace(
require_prefix = true;
}
if (n_extension_images > 0 || !strv_isempty(extension_directories)) {
if (p->n_extension_images > 0 || !strv_isempty(p->extension_directories)) {
/* Hierarchy population needs to be done for sysext and confext extension images */
r = parse_env_extension_hierarchies(&hierarchies, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES");
if (r < 0)
@ -2197,92 +2134,76 @@ int setup_namespace(
}
n_mounts = namespace_calculate_mounts(
ns_info,
read_write_paths,
read_only_paths,
inaccessible_paths,
exec_paths,
no_exec_paths,
empty_directories,
n_bind_mounts,
n_temporary_filesystems,
n_mount_images,
n_extension_images,
strv_length(extension_directories),
strv_length(hierarchies),
tmp_dir, var_tmp_dir,
creds_path,
log_namespace,
setup_propagate,
notify_socket,
host_os_release_stage);
p,
hierarchies,
setup_propagate);
if (n_mounts > 0) {
m = mounts = new0(MountEntry, n_mounts);
if (!mounts)
return -ENOMEM;
r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix);
r = append_access_mounts(&m, p->read_write_paths, READWRITE, require_prefix);
if (r < 0)
goto finish;
r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix);
r = append_access_mounts(&m, p->read_only_paths, READONLY, require_prefix);
if (r < 0)
goto finish;
r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix);
r = append_access_mounts(&m, p->inaccessible_paths, INACCESSIBLE, require_prefix);
if (r < 0)
goto finish;
r = append_access_mounts(&m, exec_paths, EXEC, require_prefix);
r = append_access_mounts(&m, p->exec_paths, EXEC, require_prefix);
if (r < 0)
goto finish;
r = append_access_mounts(&m, no_exec_paths, NOEXEC, require_prefix);
r = append_access_mounts(&m, p->no_exec_paths, NOEXEC, require_prefix);
if (r < 0)
goto finish;
r = append_empty_dir_mounts(&m, empty_directories);
r = append_empty_dir_mounts(&m, p->empty_directories);
if (r < 0)
goto finish;
r = append_bind_mounts(&m, bind_mounts, n_bind_mounts);
r = append_bind_mounts(&m, p->bind_mounts, p->n_bind_mounts);
if (r < 0)
goto finish;
r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems);
r = append_tmpfs_mounts(&m, p->temporary_filesystems, p->n_temporary_filesystems);
if (r < 0)
goto finish;
if (tmp_dir) {
bool ro = streq(tmp_dir, RUN_SYSTEMD_EMPTY);
if (p->tmp_dir) {
bool ro = streq(p->tmp_dir, RUN_SYSTEMD_EMPTY);
*(m++) = (MountEntry) {
.path_const = "/tmp",
.mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP,
.source_const = tmp_dir,
.source_const = p->tmp_dir,
};
}
if (var_tmp_dir) {
bool ro = streq(var_tmp_dir, RUN_SYSTEMD_EMPTY);
if (p->var_tmp_dir) {
bool ro = streq(p->var_tmp_dir, RUN_SYSTEMD_EMPTY);
*(m++) = (MountEntry) {
.path_const = "/var/tmp",
.mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP,
.source_const = var_tmp_dir,
.source_const = p->var_tmp_dir,
};
}
r = append_mount_images(&m, mount_images, n_mount_images);
r = append_mount_images(&m, p->mount_images, p->n_mount_images);
if (r < 0)
goto finish;
r = append_extensions(&m, root, extension_dir, hierarchies, extension_images, n_extension_images, extension_directories);
r = append_extensions(&m, root, p->extension_dir, hierarchies, p->extension_images, p->n_extension_images, p->extension_directories);
if (r < 0)
goto finish;
if (ns_info->private_dev)
if (p->private_dev)
*(m++) = (MountEntry) {
.path_const = "/dev",
.mode = PRIVATE_DEV,
@ -2292,8 +2213,8 @@ int setup_namespace(
/* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the
protective mounts to non-pid /proc paths would fail. But the pid only option may have
failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */
bool ignore_protect_proc = ns_info->ignore_protect_paths || ns_info->proc_subset == PROC_SUBSET_PID;
if (ns_info->protect_kernel_tunables) {
bool ignore_protect_proc = p->ignore_protect_paths || p->proc_subset == PROC_SUBSET_PID;
if (p->protect_kernel_tunables) {
r = append_static_mounts(&m,
protect_kernel_tunables_proc_table,
ELEMENTSOF(protect_kernel_tunables_proc_table),
@ -2304,21 +2225,21 @@ int setup_namespace(
r = append_static_mounts(&m,
protect_kernel_tunables_sys_table,
ELEMENTSOF(protect_kernel_tunables_sys_table),
ns_info->ignore_protect_paths);
p->ignore_protect_paths);
if (r < 0)
goto finish;
}
if (ns_info->protect_kernel_modules) {
if (p->protect_kernel_modules) {
r = append_static_mounts(&m,
protect_kernel_modules_table,
ELEMENTSOF(protect_kernel_modules_table),
ns_info->ignore_protect_paths);
p->ignore_protect_paths);
if (r < 0)
goto finish;
}
if (ns_info->protect_kernel_logs) {
if (p->protect_kernel_logs) {
r = append_static_mounts(&m,
protect_kernel_logs_proc_table,
ELEMENTSOF(protect_kernel_logs_proc_table),
@ -2329,30 +2250,30 @@ int setup_namespace(
r = append_static_mounts(&m,
protect_kernel_logs_dev_table,
ELEMENTSOF(protect_kernel_logs_dev_table),
ns_info->ignore_protect_paths);
p->ignore_protect_paths);
if (r < 0)
goto finish;
}
if (ns_info->protect_control_groups)
if (p->protect_control_groups)
*(m++) = (MountEntry) {
.path_const = "/sys/fs/cgroup",
.mode = READONLY,
};
r = append_protect_home(&m, ns_info->protect_home, ns_info->ignore_protect_paths);
r = append_protect_home(&m, p->protect_home, p->ignore_protect_paths);
if (r < 0)
goto finish;
r = append_protect_system(&m, ns_info->protect_system, false);
r = append_protect_system(&m, p->protect_system, false);
if (r < 0)
goto finish;
if (namespace_info_mount_apivfs(ns_info)) {
if (namespace_parameters_mount_apivfs(p)) {
r = append_static_mounts(&m,
apivfs_table,
ELEMENTSOF(apivfs_table),
ns_info->ignore_protect_paths);
p->ignore_protect_paths);
if (r < 0)
goto finish;
}
@ -2360,7 +2281,7 @@ int setup_namespace(
/* Note, if proc is mounted with subset=pid then neither of the
* two paths will exist, i.e. they are implicitly protected by
* the mount option. */
if (ns_info->protect_hostname) {
if (p->protect_hostname) {
*(m++) = (MountEntry) {
.path_const = "/proc/sys/kernel/hostname",
.mode = READONLY,
@ -2373,20 +2294,20 @@ int setup_namespace(
};
}
if (ns_info->private_network)
if (p->private_network)
*(m++) = (MountEntry) {
.path_const = "/sys",
.mode = PRIVATE_SYSFS,
};
if (ns_info->private_ipc)
if (p->private_ipc)
*(m++) = (MountEntry) {
.path_const = "/dev/mqueue",
.mode = MQUEUEFS,
.flags = MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME,
};
if (creds_path) {
if (p->creds_path) {
/* If our service has a credentials store configured, then bind that one in, but hide
* everything else. */
@ -2399,10 +2320,10 @@ int setup_namespace(
};
*(m++) = (MountEntry) {
.path_const = creds_path,
.path_const = p->creds_path,
.mode = BIND_MOUNT,
.read_only = true,
.source_const = creds_path,
.source_const = p->creds_path,
.ignore = true,
};
} else {
@ -2416,10 +2337,10 @@ int setup_namespace(
};
}
if (log_namespace) {
if (p->log_namespace) {
_cleanup_free_ char *q = NULL;
q = strjoin("/run/systemd/journal.", log_namespace);
q = strjoin("/run/systemd/journal.", p->log_namespace);
if (!q) {
r = -ENOMEM;
goto finish;
@ -2436,24 +2357,24 @@ int setup_namespace(
/* Will be used to add bind mounts at runtime */
if (setup_propagate)
*(m++) = (MountEntry) {
.source_const = propagate_dir,
.path_const = incoming_dir,
.source_const = p->propagate_dir,
.path_const = p->incoming_dir,
.mode = BIND_MOUNT,
.read_only = true,
};
if (notify_socket)
if (p->notify_socket)
*(m++) = (MountEntry) {
.path_const = notify_socket,
.source_const = notify_socket,
.path_const = p->notify_socket,
.source_const = p->notify_socket,
.mode = BIND_MOUNT,
.read_only = true,
};
if (host_os_release_stage)
if (p->host_os_release_stage)
*(m++) = (MountEntry) {
.path_const = "/run/host/.os-release-stage/",
.source_const = host_os_release_stage,
.source_const = p->host_os_release_stage,
.mode = BIND_MOUNT,
.read_only = true,
.ignore = true, /* Live copy, don't hard-fail if it goes missing */
@ -2486,12 +2407,12 @@ int setup_namespace(
/* Create the source directory to allow runtime propagation of mounts */
if (setup_propagate)
(void) mkdir_p(propagate_dir, 0600);
(void) mkdir_p(p->propagate_dir, 0600);
if (n_extension_images > 0 || !strv_isempty(extension_directories))
if (p->n_extension_images > 0 || !strv_isempty(p->extension_directories))
/* ExtensionImages/Directories mountpoint directories will be created while parsing the
* mounts to create, so have the parent ready */
(void) mkdir_p(extension_dir, 0600);
(void) mkdir_p(p->extension_dir, 0600);
/* Remount / as SLAVE so that nothing now mounted in the namespace
* shows up in the parent */
@ -2500,7 +2421,7 @@ int setup_namespace(
goto finish;
}
if (root_image) {
if (p->root_image) {
/* A root image is specified, mount it to the right place */
r = dissected_image_mount(
dissected_image,
@ -2528,7 +2449,7 @@ int setup_namespace(
goto finish;
}
} else if (root_directory) {
} else if (p->root_directory) {
/* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW);
@ -2550,24 +2471,20 @@ int setup_namespace(
}
/* Try to set up the new root directory before mounting anything else there. */
if (root_image || root_directory)
if (p->root_image || p->root_directory)
(void) base_filesystem_create(root, UID_INVALID, GID_INVALID);
/* Now make the magic happen */
r = apply_mounts(root,
mount_image_policy,
extension_image_policy,
ns_info,
p,
mounts, &n_mounts,
scope,
symlinks,
error_path);
if (r < 0)
goto finish;
/* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
r = mount_switch_root(root, /* mount_propagation_flag = */ 0);
if (r == -EINVAL && root_directory) {
if (r == -EINVAL && p->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,
* MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
@ -2593,9 +2510,9 @@ int setup_namespace(
/* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only
* supported for non-shared mounts. This needs to happen after remounting / or it will fail. */
if (setup_propagate) {
r = mount(NULL, incoming_dir, NULL, MS_SLAVE, NULL);
r = mount(NULL, p->incoming_dir, NULL, MS_SLAVE, NULL);
if (r < 0) {
log_error_errno(r, "Failed to remount %s with MS_SLAVE: %m", incoming_dir);
log_error_errno(r, "Failed to remount %s with MS_SLAVE: %m", p->incoming_dir);
goto finish;
}
}

128
src/core/namespace.h
View File

@ -5,7 +5,7 @@
Copyright © 2016 Djalal Harouni
***/
typedef struct NamespaceInfo NamespaceInfo;
typedef struct NamespaceParameters NamespaceParameters;
typedef struct BindMount BindMount;
typedef struct TemporaryFileSystem TemporaryFileSystem;
typedef struct MountImage MountImage;
@ -53,24 +53,6 @@ typedef enum ProcSubset {
_PROC_SUBSET_INVALID = -EINVAL,
} ProcSubset;
struct NamespaceInfo {
bool ignore_protect_paths;
bool private_dev;
bool protect_control_groups;
bool protect_kernel_tunables;
bool protect_kernel_modules;
bool protect_kernel_logs;
bool mount_apivfs;
bool protect_hostname;
bool private_network;
bool private_ipc;
bool mount_nosuid;
ProtectHome protect_home;
ProtectSystem protect_system;
ProtectProc protect_proc;
ProcSubset proc_subset;
};
struct BindMount {
char *source;
char *destination;
@ -100,43 +82,77 @@ struct MountImage {
MountImageType type;
};
int setup_namespace(
const char *root_directory,
const char *root_image,
const MountOptions *root_image_options,
const ImagePolicy *root_image_policy,
const NamespaceInfo *ns_info,
char **read_write_paths,
char **read_only_paths,
char **inaccessible_paths,
char **exec_paths,
char **no_exec_paths,
char **empty_directories,
char **symlinks,
const BindMount *bind_mounts,
size_t n_bind_mounts,
const TemporaryFileSystem *temporary_filesystems,
size_t n_temporary_filesystems,
const MountImage *mount_images,
size_t n_mount_images,
const ImagePolicy *mount_image_policy,
const char *tmp_dir,
const char *var_tmp_dir,
const char *creds_path,
const char *log_namespace,
unsigned long mount_propagation_flag,
VeritySettings *verity,
const MountImage *extension_images,
size_t n_extension_images,
const ImagePolicy *extension_image_policy,
char **extension_directories,
const char *propagate_dir,
const char *incoming_dir,
const char *extension_dir,
const char *notify_socket,
const char *host_os_release_stage,
RuntimeScope scope,
char **error_path);
struct NamespaceParameters {
RuntimeScope runtime_scope;
const char *root_directory;
const char *root_image;
const MountOptions *root_image_options;
const ImagePolicy *root_image_policy;
char **read_write_paths;
char **read_only_paths;
char **inaccessible_paths;
char **exec_paths;
char **no_exec_paths;
char **empty_directories;
char **symlinks;
const BindMount *bind_mounts;
size_t n_bind_mounts;
const TemporaryFileSystem *temporary_filesystems;
size_t n_temporary_filesystems;
const MountImage *mount_images;
size_t n_mount_images;
const ImagePolicy *mount_image_policy;
const char *tmp_dir;
const char *var_tmp_dir;
const char *creds_path;
const char *log_namespace;
unsigned long mount_propagation_flag;
VeritySettings *verity;
const MountImage *extension_images;
size_t n_extension_images;
const ImagePolicy *extension_image_policy;
char **extension_directories;
const char *propagate_dir;
const char *incoming_dir;
const char *extension_dir;
const char *notify_socket;
const char *host_os_release_stage;
bool ignore_protect_paths;
bool protect_control_groups;
bool protect_kernel_tunables;
bool protect_kernel_modules;
bool protect_kernel_logs;
bool protect_hostname;
bool private_dev;
bool private_network;
bool private_ipc;
bool mount_apivfs;
bool mount_nosuid;
ProtectHome protect_home;
ProtectSystem protect_system;
ProtectProc protect_proc;
ProcSubset proc_subset;
};
int setup_namespace(const NamespaceParameters *p, char **error_path);
#define RUN_SYSTEMD_EMPTY "/run/systemd/empty"

41
src/test/test-namespace.c
View File

@ -149,11 +149,12 @@ TEST(ipcns) {
}
TEST(protect_kernel_logs) {
int r;
pid_t pid;
static const NamespaceInfo ns_info = {
static const NamespaceParameters p = {
.runtime_scope = RUNTIME_SCOPE_SYSTEM,
.protect_kernel_logs = true,
};
pid_t pid;
int r;
if (geteuid() > 0) {
(void) log_tests_skipped("not root");
@ -175,39 +176,7 @@ TEST(protect_kernel_logs) {
fd = open("/dev/kmsg", O_RDONLY | O_CLOEXEC);
assert_se(fd > 0);
r = setup_namespace(NULL,
NULL,
NULL,
NULL,
&ns_info,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL, 0,
NULL, 0,
NULL, 0,
NULL,
NULL,
NULL,
NULL,
NULL,
0,
NULL,
NULL,
0,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
RUNTIME_SCOPE_SYSTEM,
NULL);
r = setup_namespace(&p, NULL);
assert_se(r == 0);
assert_se(setresuid(UID_NOBODY, UID_NOBODY, UID_NOBODY) >= 0);

80
src/test/test-ns.c
View File

@ -44,13 +44,15 @@ int main(int argc, char *argv[]) {
NULL
};
static const NamespaceInfo ns_info = {
.private_dev = true,
.protect_control_groups = true,
.protect_kernel_tunables = true,
.protect_kernel_modules = true,
.protect_proc = PROTECT_PROC_NOACCESS,
.proc_subset = PROC_SUBSET_PID,
static const BindMount bind_mount = {
.source = (char*) "/usr/bin",
.destination = (char*) "/etc/systemd",
.read_only = true,
};
static const TemporaryFileSystem tmpfs = {
.path = (char*) "/var",
.options = (char*) "ro",
};
char *root_directory;
@ -76,40 +78,36 @@ int main(int argc, char *argv[]) {
else
log_info("Not chrooted");
r = setup_namespace(root_directory,
NULL,
NULL,
NULL,
&ns_info,
(char **) writable,
(char **) readonly,
(char **) inaccessible,
NULL,
(char **) exec,
(char **) no_exec,
NULL,
&(BindMount) { .source = (char*) "/usr/bin", .destination = (char*) "/etc/systemd", .read_only = true }, 1,
&(TemporaryFileSystem) { .path = (char*) "/var", .options = (char*) "ro" }, 1,
NULL,
0,
NULL,
tmp_dir,
var_tmp_dir,
NULL,
NULL,
0,
NULL,
NULL,
0,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
RUNTIME_SCOPE_SYSTEM,
NULL);
NamespaceParameters p = {
.runtime_scope = RUNTIME_SCOPE_SYSTEM,
.root_directory = root_directory,
.read_write_paths = (char**) writable,
.read_only_paths = (char**) readonly,
.inaccessible_paths = (char**) inaccessible,
.exec_paths = (char**) exec,
.no_exec_paths = (char**) no_exec,
.tmp_dir = tmp_dir,
.var_tmp_dir = var_tmp_dir,
.bind_mounts = &bind_mount,
.n_bind_mounts = 1,
.temporary_filesystems = &tmpfs,
.n_temporary_filesystems = 1,
.private_dev = true,
.protect_control_groups = true,
.protect_kernel_tunables = true,
.protect_kernel_modules = true,
.protect_proc = PROTECT_PROC_NOACCESS,
.proc_subset = PROC_SUBSET_PID,
};
r = setup_namespace(&p, NULL);
if (r < 0) {
log_error_errno(r, "Failed to set up namespace: %m");