There are two cases (see below) in sid_to_id where errors occur
mapping the uid but the rc which is set is overwritten (reset to 0
before return).
saved_cred = override_creds(root_cred);
sidkey = request_key(&cifs_idmap_key_type, sidstr, "");
if (IS_ERR(sidkey)) {
rc = -EINVAL;
cifs_dbg(FYI, "%s: Can't map SID %s to a %cid\n",
__func__, sidstr, sidtype == SIDOWNER ? 'u' : 'g');
goto out_revert_creds;
}
/*
* FIXME: Here we assume that uid_t and gid_t are same size. It's
* probably a safe assumption but might be better to check based on
* sidtype.
*/
BUILD_BUG_ON(sizeof(uid_t) != sizeof(gid_t));
if (sidkey->datalen != sizeof(uid_t)) {
rc = -EIO;
cifs_dbg(FYI, "%s: Downcall contained malformed key
(datalen=%hu)\n",
__func__, sidkey->datalen);
key_invalidate(sidkey);
goto out_key_put;
}
since later in the function we do:
out_key_put:
key_put(sidkey);
out_revert_creds:
revert_creds(saved_cred);
kfree(sidstr);
/*
* Note that we return 0 here unconditionally. If the mapping
* fails then we just fall back to using the ctx->linux_uid/linux_gid.
*/
got_valid_id:
rc = 0;
if (sidtype == SIDOWNER)
fattr->cf_uid = fuid;
else
fattr->cf_gid = fgid;
return rc;
}
Any thoughts on whether it would be better to return the errors, or
continue the current strategy of simply using the default uid/gid for
the mount and returning 0 (and removing the two places above where we
set rc to non zero values, since rc will be overwritten with 0)?
--
Thanks,
Steve
