Merge pull request #4280 from giuseppe/unprivileged-user

[RFC] run systemd in an unprivileged container

src/basic/audit-util.c

@@ -92,8 +92,11 @@ bool use_audit(void) {
                 int fd;
 
                 fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_AUDIT);
-                if (fd < 0)
+                if (fd < 0) {
-                        cached_use = errno != EAFNOSUPPORT && errno != EPROTONOSUPPORT;
+                        cached_use = !IN_SET(errno, EAFNOSUPPORT, EPROTONOSUPPORT, EPERM);
+                        if (errno == EPERM)
+                                log_debug_errno(errno, "Audit access prohibited, won't talk to audit");
+                }
                 else {
                         cached_use = true;
                         safe_close(fd);

src/basic/capability-util.c

@@ -31,6 +31,7 @@
 #include "log.h"
 #include "macro.h"
 #include "parse-util.h"
+#include "user-util.h"
 #include "util.h"
 
 int have_effective_cap(int value) {
@@ -295,7 +296,7 @@ int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
         if (setresgid(gid, gid, gid) < 0)
                 return log_error_errno(errno, "Failed to change group ID: %m");
 
-        if (setgroups(0, NULL) < 0)
+        if (maybe_setgroups(0, NULL) < 0)
                 return log_error_errno(errno, "Failed to drop auxiliary groups list: %m");
 
         /* Ensure we keep the permitted caps across the setresuid() */

src/basic/user-util.c

@@ -33,6 +33,7 @@
 
 #include "alloc-util.h"
 #include "fd-util.h"
+#include "fileio.h"
 #include "formats-util.h"
 #include "macro.h"
 #include "missing.h"
@@ -460,7 +461,7 @@ int get_shell(char **_s) {
 
 int reset_uid_gid(void) {
 
-        if (setgroups(0, NULL) < 0)
+        if (maybe_setgroups(0, NULL) < 0)
                 return -errno;
 
         if (setresgid(0, 0, 0) < 0)
@@ -602,3 +603,27 @@ bool valid_home(const char *p) {
 
         return true;
 }
+
+int maybe_setgroups(size_t size, const gid_t *list) {
+        static int cached_can_setgroups = -1;
+        /* check if setgroups is allowed before we try to drop all the auxiliary groups */
+        if (size == 0) {
+                if (cached_can_setgroups < 0) {
+                        _cleanup_free_ char *setgroups_content = NULL;
+                        int r = read_one_line_file("/proc/self/setgroups", &setgroups_content);
+                        if (r < 0 && errno != ENOENT)
+                                return r;
+                        if (r < 0) {
+                                /* old kernels don't have /proc/self/setgroups, so assume we can use setgroups */
+                                cached_can_setgroups = true;
+                        } else {
+                                cached_can_setgroups = streq(setgroups_content, "allow");
+                                if (!cached_can_setgroups)
+                                        log_debug("skip setgroups, /proc/self/setgroups is set to 'deny'");
+                        }
+                }
+                if (!cached_can_setgroups)
+                        return 0;
+        }
+        return setgroups(size, list);
+}

src/basic/user-util.h

@@ -86,3 +86,5 @@ bool valid_user_group_name(const char *u);
 bool valid_user_group_name_or_id(const char *u);
 bool valid_gecos(const char *d);
 bool valid_home(const char *p);
+
+int maybe_setgroups(size_t size, const gid_t *list);

src/core/execute.c

@@ -781,7 +781,7 @@ static int enforce_groups(const ExecContext *context, const char *username, gid_
                         k++;
                 }
 
-                if (setgroups(k, gids) < 0) {
+                if (maybe_setgroups(k, gids) < 0) {
                         free(gids);
                         return -errno;
                 }

src/core/umount.c

@@ -375,7 +375,7 @@ static int mount_points_list_umount(MountPoint **head, bool *changed, bool log_e
                 /* If we are in a container, don't attempt to
                    read-only mount anything as that brings no real
                    benefits, but might confuse the host, as we remount
-                   the superblock here, not the bind mound. */
+                   the superblock here, not the bind mount. */
                 if (detect_container() <= 0)  {
                         _cleanup_free_ char *options = NULL;
                         /* MS_REMOUNT requires that the data parameter