user-util: rework how we validate user namesv239-42

This reworks the user validation infrastructure. There are now two
modes. In regular mode we are strict and test against a strict set of
valid chars. And in "relaxed" mode we just filter out some really
obvious, dangerous stuff. i.e. strict is whitelisting what is OK, but
"relaxed" is blacklisting what is really not OK.

The idea is that we use strict mode whenver we allocate a new user
(i.e. in sysusers.d or homed), while "relaxed" mode is when we process
users registered elsewhere, (i.e. userdb, logind, …)

The requirements on user name validity vary wildly. SSSD thinks its fine
to embedd "@" for example, while the suggested NAME_REGEX field on
Debian does not even allow uppercase chars…

This effectively liberaralizes a lot what we expect from usernames.

The code that warns about questionnable user names is now optional and
only used at places such as unit file parsing, so that it doesn't show
up on every userdb query, but only when processing configuration files
that know better.

Fixes: #15149 #15090
(cherry picked from commit 7a8867abfab10e5bbca10590ec2aa40c5b27d8fb)

Resolves: #1848373

1 files changed, 110 insertions, 75 deletions

diff --git a/src/basic/user-util.c b/src/basic/user-util.c
index 68a924770b..cd870c4361 100644
--- a/src/basic/user-util.c
+++ b/src/basic/user-util.c
@@ -14,6 +14,8 @@
 #include <unistd.h>
 #include <utmp.h>
 
+#include "sd-messages.h"
+
 #include "alloc-util.h"
 #include "fd-util.h"
 #include "fileio.h"
@@ -576,92 +578,125 @@ int take_etc_passwd_lock(const char *root) {
         return fd;
 }
 
-bool valid_user_group_name_full(const char *u, bool strict) {
+bool valid_user_group_name(const char *u, ValidUserFlags flags) {
         const char *i;
-        long sz;
-        bool warned = false;
 
-        /* Checks if the specified name is a valid user/group name. Also see POSIX IEEE Std 1003.1-2008, 2016 Edition,
-         * 3.437. We are a bit stricter here however. Specifically we deviate from POSIX rules:
-         *
-         * - We require that names fit into the appropriate utmp field
-         * - We don't allow empty user names
-         * - No dots in the first character
+        /* Checks if the specified name is a valid user/group name. There are two flavours of this call:
+         * strict mode is the default which is POSIX plus some extra rules; and relaxed mode where we accept
+         * pretty much everything except the really worst offending names.
          *
-         * If strict==true, additionally:
-         * - We don't allow any dots (this conflicts with chown syntax which permits dots as user/group name separator)
-         * - We don't allow a digit as the first character
-         *
-         * Note that other systems are even more restrictive, and don't permit underscores or uppercase characters.
-         */
+         * Whenever we synthesize users ourselves we should use the strict mode. But when we process users
+         * created by other stuff, let's be more liberal. */
 
-        if (isempty(u))
+        if (isempty(u)) /* An empty user name is never valid */
                 return false;
 
-        if (!(u[0] >= 'a' && u[0] <= 'z') &&
-            !(u[0] >= 'A' && u[0] <= 'Z') &&
-            !(u[0] >= '0' && u[0] <= '9' && !strict) &&
-            u[0] != '_')
-                return false;
-
-        bool only_digits_seen = u[0] >= '0' && u[0] <= '9';
-
-        if (only_digits_seen) {
-                log_warning("User or group name \"%s\" starts with a digit, accepting for compatibility.", u);
-                warned = true;
-        }
-
-        for (i = u+1; *i; i++) {
-                if (((*i >= 'a' && *i <= 'z') ||
-                     (*i >= 'A' && *i <= 'Z') ||
-                     (*i >= '0' && *i <= '9') ||
-                     IN_SET(*i, '_', '-'))) {
-                        if (!(*i >= '0' && *i <= '9'))
-                                only_digits_seen = false;
-                        continue;
-                        }
-
-                if (*i == '.' && !strict) {
-                        if (!warned) {
-                                log_warning("Bad user or group name \"%s\", accepting for compatibility.", u);
-                                warned = true;
-                        }
-
-                        continue;
-                }
-
-                return false;
+        if (parse_uid(u, NULL) >= 0) /* Something that parses as numeric UID string is valid exactly when the
+                                      * flag for it is set */
+                return FLAGS_SET(flags, VALID_USER_ALLOW_NUMERIC);
+
+        if (FLAGS_SET(flags, VALID_USER_RELAX)) {
+
+                /* In relaxed mode we just check very superficially. Apparently SSSD and other stuff is
+                 * extremely liberal (way too liberal if you ask me, even inserting "@" in user names, which
+                 * is bound to cause problems for example when used with an MTA), hence only filter the most
+                 * obvious cases, or where things would result in an invalid entry if such a user name would
+                 * show up in /etc/passwd (or equivalent getent output).
+                 *
+                 * Note that we stepped far out of POSIX territory here. It's not our fault though, but
+                 * SSSD's, Samba's and everybody else who ignored POSIX on this. (I mean, I am happy to step
+                 * outside of POSIX' bounds any day, but I must say in this case I probably wouldn't
+                 * have...) */
+
+                if (startswith(u, " ") || endswith(u, " ")) /* At least expect whitespace padding is removed
+                                                             * at front and back (accept in the middle, since
+                                                             * that's apparently a thing on Windows). Note
+                                                             * that this also blocks usernames consisting of
+                                                             * whitespace only. */
+                        return false;
+
+                if (!utf8_is_valid(u)) /* We want to synthesize JSON from this, hence insist on UTF-8 */
+                        return false;
+
+                if (string_has_cc(u, NULL)) /* CC characters are just dangerous (and \n in particular is the
+                                             * record separator in /etc/passwd), so we can't allow that. */
+                        return false;
+
+                if (strpbrk(u, ":/")) /* Colons are the field separator in /etc/passwd, we can't allow
+                                       * that. Slashes are special to file systems paths and user names
+                                       * typically show up in the file system as home directories, hence
+                                       * don't allow slashes. */
+                        return false;
+
+                if (in_charset(u, "0123456789")) /* Don't allow fully numeric strings, they might be confused
+                                                  * with with UIDs (note that this test is more broad than
+                                                  * the parse_uid() test above, as it will cover more than
+                                                  * the 32bit range, and it will detect 65535 (which is in
+                                                  * invalid UID, even though in the unsigned 32 bit range) */
+                        return false;
+
+                if (u[0] == '-' && in_charset(u + 1, "0123456789")) /* Don't allow negative fully numeric
+                                                                     * strings either. After all some people
+                                                                     * write 65535 as -1 (even though that's
+                                                                     * not even true on 32bit uid_t
+                                                                     * anyway) */
+                        return false;
+
+                if (dot_or_dot_dot(u)) /* User names typically become home directory names, and these two are
+                                        * special in that context, don't allow that. */
+                        return false;
+
+                /* Compare with strict result and warn if result doesn't match */
+                if (FLAGS_SET(flags, VALID_USER_WARN) && !valid_user_group_name(u, 0))
+                        log_struct(LOG_NOTICE,
+                                   "MESSAGE=Accepting user/group name '%s', which does not match strict user/group name rules.", u,
+                                   "USER_GROUP_NAME=%s", u,
+                                   "MESSAGE_ID=" SD_MESSAGE_UNSAFE_USER_NAME_STR);
+
+                /* Note that we make no restrictions on the length in relaxed mode! */
+        } else {
+                long sz;
+                size_t l;
+
+                /* Also see POSIX IEEE Std 1003.1-2008, 2016 Edition, 3.437. We are a bit stricter here
+                 * however. Specifically we deviate from POSIX rules:
+                 *
+                 * - We don't allow empty user names (see above)
+                 * - We require that names fit into the appropriate utmp field
+                 * - We don't allow any dots (this conflicts with chown syntax which permits dots as user/group name separator)
+                 * - We don't allow dashes or digit as the first character
+                 *
+                 * Note that other systems are even more restrictive, and don't permit underscores or uppercase characters.
+                 */
+
+                if (!(u[0] >= 'a' && u[0] <= 'z') &&
+                    !(u[0] >= 'A' && u[0] <= 'Z') &&
+                    u[0] != '_')
+                        return false;
+
+                for (i = u+1; *i; i++)
+                        if (!(*i >= 'a' && *i <= 'z') &&
+                            !(*i >= 'A' && *i <= 'Z') &&
+                            !(*i >= '0' && *i <= '9') &&
+                            !IN_SET(*i, '_', '-'))
+                                return false;
+
+                l = i - u;
+
+                sz = sysconf(_SC_LOGIN_NAME_MAX);
+                assert_se(sz > 0);
+
+                if (l > (size_t) sz)
+                        return false;
+                if (l > FILENAME_MAX)
+                        return false;
+                if (l > UT_NAMESIZE - 1)
+                        return false;
         }
 
-        if (only_digits_seen)
-                return false;
-
-        sz = sysconf(_SC_LOGIN_NAME_MAX);
-        assert_se(sz > 0);
-
-        if ((size_t) (i-u) > (size_t) sz)
-                return false;
-
-        if ((size_t) (i-u) > UT_NAMESIZE - 1)
-                return false;
-
         return true;
 }
 
-bool valid_user_group_name_or_id_full(const char *u, bool strict) {
-
-        /* Similar as above, but is also fine with numeric UID/GID specifications, as long as they are in the
-         * right range, and not the invalid user ids. */
-
-        if (isempty(u))
-                return false;
-
-        if (parse_uid(u, NULL) >= 0)
-                return true;
-
-        return valid_user_group_name_full(u, strict);
-}
-
 bool valid_gecos(const char *d) {
 
         if (!d)