#include "log.h"
#include "macro.h"
#include "manager.h"
+#include "memory-util.h"
#include "missing.h"
#include "mkdir.h"
#include "namespace.h"
#include "umask-util.h"
#include "unit.h"
#include "user-util.h"
-#include "util.h"
#include "utmp-wtmp.h"
#define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
}
static int chown_terminal(int fd, uid_t uid) {
- struct stat st;
+ int r;
assert(fd >= 0);
/* Before we chown/chmod the TTY, let's ensure this is actually a tty */
- if (isatty(fd) < 1)
- return 0;
-
- /* This might fail. What matters are the results. */
- (void) fchown(fd, uid, -1);
- (void) fchmod(fd, TTY_MODE);
+ if (isatty(fd) < 1) {
+ if (IN_SET(errno, EINVAL, ENOTTY))
+ return 0; /* not a tty */
- if (fstat(fd, &st) < 0)
return -errno;
+ }
- if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
- return -EPERM;
+ /* This might fail. What matters are the results. */
+ r = fchmod_and_chown(fd, TTY_MODE, uid, -1);
+ if (r < 0)
+ return r;
- return 0;
+ return 1;
}
static int setup_confirm_stdio(const char *vc, int *_saved_stdin, int *_saved_stdout) {
return context_has_address_families(c) ||
c->memory_deny_write_execute ||
c->restrict_realtime ||
+ c->restrict_suid_sgid ||
exec_context_restrict_namespaces_set(c) ||
c->protect_kernel_tunables ||
c->protect_kernel_modules ||
if (skip_seccomp_unavailable(u, "SystemCallFilter="))
return 0;
- negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
+ negative_action = c->syscall_errno == 0 ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c->syscall_errno);
if (c->syscall_whitelist) {
default_action = negative_action;
return seccomp_restrict_realtime();
}
+static int apply_restrict_suid_sgid(const Unit* u, const ExecContext *c) {
+ assert(u);
+ assert(c);
+
+ if (!c->restrict_suid_sgid)
+ return 0;
+
+ if (skip_seccomp_unavailable(u, "RestrictSUIDSGID="))
+ return 0;
+
+ return seccomp_restrict_suid_sgid();
+}
+
static int apply_protect_sysctl(const Unit *u, const ExecContext *c) {
assert(u);
assert(c);
x = strappend("HOME=", home);
if (!x)
return -ENOMEM;
+
+ path_simplify(x + 5, true);
our_env[n_env++] = x;
}
x = strappend("SHELL=", shell);
if (!x)
return -ENOMEM;
+
+ path_simplify(x + 6, true);
our_env[n_env++] = x;
}
STRV_FOREACH(rt, context->directories[type].paths) {
_cleanup_free_ char *p = NULL, *pp = NULL;
- p = strjoin(params->prefix[type], "/", *rt);
+ p = path_join(params->prefix[type], *rt);
if (!p) {
r = -ENOMEM;
goto fail;
goto fail;
if (context->dynamic_user &&
- !IN_SET(type, EXEC_DIRECTORY_RUNTIME, EXEC_DIRECTORY_CONFIGURATION)) {
+ (!IN_SET(type, EXEC_DIRECTORY_RUNTIME, EXEC_DIRECTORY_CONFIGURATION) ||
+ (type == EXEC_DIRECTORY_RUNTIME && context->runtime_directory_preserve_mode != EXEC_PRESERVE_NO))) {
_cleanup_free_ char *private_root = NULL;
/* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
* Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
* files or sockets with other services. */
- private_root = strjoin(params->prefix[type], "/private");
+ private_root = path_join(params->prefix[type], "private");
if (!private_root) {
r = -ENOMEM;
goto fail;
if (r < 0)
goto fail;
- pp = strjoin(private_root, "/", *rt);
+ pp = path_join(private_root, *rt);
if (!pp) {
r = -ENOMEM;
goto fail;
if (r < 0)
goto fail;
- /* Lock down the access mode */
- if (chmod(pp, context->directories[type].mode) < 0) {
- r = -errno;
- goto fail;
- }
} else {
r = mkdir_label(p, context->directories[type].mode);
- if (r < 0 && r != -EEXIST)
- goto fail;
- if (r == -EEXIST) {
- struct stat st;
-
- if (stat(p, &st) < 0) {
- r = -errno;
+ if (r < 0) {
+ if (r != -EEXIST)
goto fail;
- }
- if (((st.st_mode ^ context->directories[type].mode) & 07777) != 0)
- log_warning("%s \'%s\' already exists but the mode is different. "
- "(filesystem: %o %sMode: %o)",
- exec_directory_type_to_string(type), *rt,
- st.st_mode & 07777, exec_directory_type_to_string(type), context->directories[type].mode & 07777);
- if (!context->dynamic_user)
+
+ if (type == EXEC_DIRECTORY_CONFIGURATION) {
+ struct stat st;
+
+ /* Don't change the owner/access mode of the configuration directory,
+ * as in the common case it is not written to by a service, and shall
+ * not be writable. */
+
+ if (stat(p, &st) < 0) {
+ r = -errno;
+ goto fail;
+ }
+
+ /* Still complain if the access mode doesn't match */
+ if (((st.st_mode ^ context->directories[type].mode) & 07777) != 0)
+ log_warning("%s \'%s\' already exists but the mode is different. "
+ "(File system: %o %sMode: %o)",
+ exec_directory_type_to_string(type), *rt,
+ st.st_mode & 07777, exec_directory_type_to_string(type), context->directories[type].mode & 07777);
+
continue;
+ }
}
}
- /* Don't change the owner of the configuration directory, as in the common case it is not written to by
- * a service, and shall not be writable. */
- if (type == EXEC_DIRECTORY_CONFIGURATION)
- continue;
+ /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
+ * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
+ * current UID/GID ownership.) */
+ r = chmod_and_chown(pp ?: p, context->directories[type].mode, UID_INVALID, GID_INVALID);
+ if (r < 0)
+ goto fail;
- /* Then, change the ownership of the whole tree, if necessary */
- r = path_chown_recursive(pp ?: p, uid, gid);
+ /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
+ * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
+ * assignments to exist.*/
+ r = path_chown_recursive(pp ?: p, uid, gid, context->dynamic_user ? 01777 : 07777);
if (r < 0)
goto fail;
}
.source = s,
.destination = d,
.read_only = false,
+ .nosuid = context->dynamic_user, /* don't allow suid/sgid when DynamicUser= is on */
.recursive = true,
.ignore_enoent = false,
};
const ExecCommand *command,
const ExecContext *context,
const ExecParameters *params,
- const ExecRuntime *runtime) {
+ const ExecRuntime *runtime,
+ char **error_path) {
_cleanup_strv_free_ char **empty_directories = NULL;
char *tmp = NULL, *var = NULL;
needs_sandboxing ? context->protect_home : PROTECT_HOME_NO,
needs_sandboxing ? context->protect_system : PROTECT_SYSTEM_NO,
context->mount_flags,
- DISSECT_IMAGE_DISCARD_ON_LOOP);
+ DISSECT_IMAGE_DISCARD_ON_LOOP,
+ error_path);
bind_mount_free_many(bind_mounts, n_bind_mounts);
/* 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 proceeed, but only if exclusively
+ * 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
* completely different execution environment. */
if (r == -ENOANO) {
if (!c->working_directory_home)
return 0;
- if (uid == 0) {
- /* Hardcode /root as home directory for UID 0 */
- *home = "/root";
- return 1;
- }
-
r = get_home_dir(buf);
if (r < 0)
return r;
}
}
+ if (context->network_namespace_path && runtime && runtime->netns_storage_socket[0] >= 0) {
+ r = open_netns_path(runtime->netns_storage_socket, context->network_namespace_path);
+ if (r < 0) {
+ *exit_status = EXIT_NETWORK;
+ return log_unit_error_errno(unit, r, "Failed to open network namespace path %s: %m", context->network_namespace_path);
+ }
+ }
+
r = setup_input(context, params, socket_fd, named_iofds);
if (r < 0) {
*exit_status = EXIT_STDIN;
USER_PROCESS,
username);
- if (context->user) {
+ if (uid_is_valid(uid)) {
r = chown_terminal(STDIN_FILENO, uid);
if (r < 0) {
*exit_status = EXIT_STDIN;
}
}
- if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
+ if ((context->private_network || context->network_namespace_path) && runtime && runtime->netns_storage_socket[0] >= 0) {
+
if (ns_type_supported(NAMESPACE_NET)) {
r = setup_netns(runtime->netns_storage_socket);
if (r < 0) {
*exit_status = EXIT_NETWORK;
return log_unit_error_errno(unit, r, "Failed to set up network namespacing: %m");
}
+ } else if (context->network_namespace_path) {
+ *exit_status = EXIT_NETWORK;
+ return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP), "NetworkNamespacePath= is not supported, refusing.");
} else
log_unit_warning(unit, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
}
needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime);
if (needs_mount_namespace) {
- r = apply_mount_namespace(unit, command, context, params, runtime);
+ _cleanup_free_ char *error_path = NULL;
+
+ r = apply_mount_namespace(unit, command, context, params, runtime, &error_path);
if (r < 0) {
*exit_status = EXIT_NAMESPACE;
- return log_unit_error_errno(unit, r, "Failed to set up mount namespacing: %m");
+ return log_unit_error_errno(unit, r, "Failed to set up mount namespacing%s%s: %m",
+ error_path ? ": " : "", strempty(error_path));
}
}
}
if (needs_setuid) {
- if (context->user) {
+ if (uid_is_valid(uid)) {
r = enforce_user(context, uid);
if (r < 0) {
*exit_status = EXIT_USER;
return log_unit_error_errno(unit, r, "Failed to apply realtime restrictions: %m");
}
+ r = apply_restrict_suid_sgid(unit, context);
+ if (r < 0) {
+ *exit_status = EXIT_SECCOMP;
+ return log_unit_error_errno(unit, r, "Failed to apply SUID/SGID restrictions: %m");
+ }
+
r = apply_restrict_namespaces(unit, context);
if (r < 0) {
*exit_status = EXIT_SECCOMP;
c->stdin_data = mfree(c->stdin_data);
c->stdin_data_size = 0;
+
+ c->network_namespace_path = mfree(c->network_namespace_path);
}
int exec_context_destroy_runtime_directory(const ExecContext *c, const char *runtime_prefix) {
STRV_FOREACH(i, c->directories[EXEC_DIRECTORY_RUNTIME].paths) {
_cleanup_free_ char *p;
- p = strjoin(runtime_prefix, "/", *i);
+ p = path_join(runtime_prefix, *i);
if (!p)
return -ENOMEM;
- /* We execute this synchronously, since we need to be sure this is gone when we start the service
- * next. */
+ /* We execute this synchronously, since we need to be sure this is gone when we start the
+ * service next. */
(void) rm_rf(p, REMOVE_ROOT);
}
return true; /* if we could not resolve, assume it may */
/* "tty0" means the active VC, so it may be the same sometimes */
- return streq(resolved, tty) || (streq(resolved, "tty0") && tty_is_vc(tty));
+ return path_equal(resolved, tty) || (streq(resolved, "tty0") && tty_is_vc(tty));
}
-bool exec_context_may_touch_console(const ExecContext *ec) {
+static bool exec_context_may_touch_tty(const ExecContext *ec) {
+ assert(ec);
- return (ec->tty_reset ||
+ return ec->tty_reset ||
ec->tty_vhangup ||
ec->tty_vt_disallocate ||
is_terminal_input(ec->std_input) ||
is_terminal_output(ec->std_output) ||
- is_terminal_output(ec->std_error)) &&
+ is_terminal_output(ec->std_error);
+}
+
+bool exec_context_may_touch_console(const ExecContext *ec) {
+
+ return exec_context_may_touch_tty(ec) &&
tty_may_match_dev_console(exec_context_tty_path(ec));
}
"%sIgnoreSIGPIPE: %s\n"
"%sMemoryDenyWriteExecute: %s\n"
"%sRestrictRealtime: %s\n"
+ "%sRestrictSUIDSGID: %s\n"
"%sKeyringMode: %s\n"
"%sProtectHostname: %s\n",
prefix, c->umask,
prefix, yes_no(c->ignore_sigpipe),
prefix, yes_no(c->memory_deny_write_execute),
prefix, yes_no(c->restrict_realtime),
+ prefix, yes_no(c->restrict_suid_sgid),
prefix, exec_keyring_mode_to_string(c->keyring_mode),
prefix, yes_no(c->protect_hostname));
prefix, s);
}
+ if (c->network_namespace_path)
+ fprintf(f,
+ "%sNetworkNamespacePath: %s\n",
+ prefix, c->network_namespace_path);
+
if (c->syscall_errno > 0) {
const char *errno_name;
else
fprintf(f, "%d\n", c->syscall_errno);
}
-
- if (c->apparmor_profile)
- fprintf(f,
- "%sAppArmorProfile: %s%s\n",
- prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
}
bool exec_context_maintains_privileges(const ExecContext *c) {
c->n_log_extra_fields = 0;
}
+void exec_context_revert_tty(ExecContext *c) {
+ int r;
+
+ assert(c);
+
+ /* First, reset the TTY (possibly kicking everybody else from the TTY) */
+ exec_context_tty_reset(c, NULL);
+
+ /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
+ * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
+ * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
+
+ if (exec_context_may_touch_tty(c)) {
+ const char *path;
+
+ path = exec_context_tty_path(c);
+ if (path) {
+ r = chmod_and_chown(path, TTY_MODE, 0, TTY_GID);
+ if (r < 0 && r != -ENOENT)
+ log_warning_errno(r, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path);
+ }
+ }
+}
+
void exec_status_start(ExecStatus *s, pid_t pid) {
assert(s);
s->code = code;
s->status = status;
- if (context) {
- if (context->utmp_id)
- (void) utmp_put_dead_process(context->utmp_id, pid, code, status);
-
- exec_context_tty_reset(context, NULL);
- }
+ if (context && context->utmp_id)
+ (void) utmp_put_dead_process(context->utmp_id, pid, code, status);
}
void exec_status_reset(ExecStatus *s) {
}
static void exec_runtime_freep(ExecRuntime **rt) {
- if (*rt)
- (void) exec_runtime_free(*rt, false);
+ (void) exec_runtime_free(*rt, false);
}
-static int exec_runtime_allocate(ExecRuntime **rt) {
- assert(rt);
+static int exec_runtime_allocate(ExecRuntime **ret) {
+ ExecRuntime *n;
- *rt = new0(ExecRuntime, 1);
- if (!*rt)
+ assert(ret);
+
+ n = new(ExecRuntime, 1);
+ if (!n)
return -ENOMEM;
- (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
+ *n = (ExecRuntime) {
+ .netns_storage_socket = { -1, -1 },
+ };
+
+ *ret = n;
return 0;
}
static int exec_runtime_make(Manager *m, const ExecContext *c, const char *id, ExecRuntime **ret) {
_cleanup_free_ char *tmp_dir = NULL, *var_tmp_dir = NULL;
- _cleanup_close_pair_ int netns_storage_socket[2] = {-1, -1};
+ _cleanup_close_pair_ int netns_storage_socket[2] = { -1, -1 };
int r;
assert(m);
assert(id);
/* It is not necessary to create ExecRuntime object. */
- if (!c->private_network && !c->private_tmp)
+ if (!c->private_network && !c->private_tmp && !c->network_namespace_path)
return 0;
if (c->private_tmp) {
return r;
}
- if (c->private_network) {
+ if (c->private_network || c->network_namespace_path) {
if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, netns_storage_socket) < 0)
return -errno;
}
return r;
/* Avoid cleanup */
- netns_storage_socket[0] = -1;
- netns_storage_socket[1] = -1;
+ netns_storage_socket[0] = netns_storage_socket[1] = -1;
return 1;
}
projects / thirdparty / systemd.git / blobdiff