11:16 a.m.
At this moment, Keycloak internally saves all usernames in the DB in
lowercase. This means that if Keycloak DB contains user "john" and user
fills the login form with uppercase like "John", Keycloak is still able
to find the user because it lowercase the username from the login form
to "john" and then query DB for lookup of username "john" . At the
same
time, DB queries have good performance as they are not needed to be case
sensitive.
Also when we import users from the 3rd party user storage like ldap, we
first convert them to lowercase. Unfortunately this causes issues for
some users, as they have LDAP users in uppercase like "JOHN" and
usernames in keycloak tokens are returned in lowercase like "john",
which causes issues in some applications.
I see 2 possibilities to address this issue:
1) Have a switch at the realm level to support case-sensitive usernames
2) Address this issue at LDAP level and add special attribute to users
imported from LDAP
More details:
The solution (1) has some more flexibility and is not LDAP specific,
however it may either have non-trivial performance penalty due the
case-sensitive DB searches or it may mean possibility of duplicated
usernames different just by cases like "john", "John" and
"JOHN".
I can imagine that switch at realm level will have 3 options similar to
this:
* (a) Don't support case insensitive usernames (default value and
current behaviour). So usernames are lowercased before saving to
Keycloak DB and before DB searches.
* (b) Support case sensitive usernames with case-sensitive duplicated
users. If this option is chosen, it will be possible to have for
example 3 separate Keycloak users like "mposolda", "Mposolda" and
"MPOSOLDA". This will mean that users will need to fill exactly
their case-sensitive username on the login form
* (c) Support case sensitive usernames without case-sensitive
duplicated users. If this option is chosen, you will have
case-sensitive user saved in the DB, however it won't be allowed to
have duplicated usernames different just by cases. So you won't be
able to have "mposolda", "Mposolda" and "MPOSOLDA". The
third option
has price-to-pay as DB searches will need to be case-sensitive and
hence may have quite bad performance as mentioned for example in
this blog:
https://alvinalexander.com/sql/sql-select-case-insensitive-query-queries-...
. I am not even convinced whether to ever support this option, I
rather vote for no. Performance may be better once we move away from
RDBMS to Keycloak.next storage, but not sure.
2) Address this issue at LDAP level and add special attribute to users
imported from LDAP. That will allow that LDAP user "MPOSOLDA" can be
mapped to Keycloak user "mposolda". At the Keycloak level, the user
"mposolda" will have the attribute like LDAP_USERNAME with the value
"MPOSOLDA" . In this case, people may need to create protocol mapper,
which will map the LDAP_USERNAME attribute to their token. Protocol
mapper can be added to client scope, which can be added to all required
clients. Not sure if this is workaround or not
So far, I vote for (1) and maybe just add the (a) and (b) options. WDYT?
Marek