When LDAP server disconnects ipa-otpd client connection due to a
timeout, ipa-otpd instance would stop and report an issue. This confuses
systemd service reporting, so for these situations we better to shut
down gracefully.
Fixes: https://pagure.io/freeipa/issue/6587
Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
Reviewed-By: Rob Crittenden <rcritten@redhat.com>
Reviewed-By: Simo Sorce <ssorce@redhat.com>
Python code does detection of the system encoding based on the locale
settings. On RHEL 8.4 development images we somehow get LANG=en_US which
defaults to iso8859-1 _inside_ the systemd-started service, even though
the whole environment defaults to LANG=en_US.UTF-8.
When instrumented with ExecStartPre=/usr/bin/locale, the following
output can be seen:
locale[45481]: LANG=en_US
locale[45481]: LC_CTYPE="en_US"
locale[45481]: LC_NUMERIC="en_US"
locale[45481]: LC_TIME="en_US"
locale[45481]: LC_COLLATE="en_US"
locale[45481]: LC_MONETARY="en_US"
locale[45481]: LC_MESSAGES="en_US"
locale[45481]: LC_PAPER="en_US"
locale[45481]: LC_NAME="en_US"
locale[45481]: LC_ADDRESS="en_US"
locale[45481]: LC_TELEPHONE="en_US"
locale[45481]: LC_MEASUREMENT="en_US"
locale[45481]: LC_IDENTIFICATION="en_US"
locale[45481]: LC_ALL=
ipactl[45483]: Unexpected error
ipactl[45483]: SystemEncodingError: System encoding must be UTF-8, 'iso8859-1' is not supported. Set LC_ALL="C.UTF-8", or LC_ALL="" and LC_CTYPE="C.UTF-8".
systemd[1]: ipa.service: Main process exited, code=exited, status=1/FAILURE
Set the environment to explicit LC_ALL=C.UTF-8 to please the Python
code. FreeIPA server side only cares about actual encoding, not the
language itself. We already use LC_ALL=C.UTF-8 in httpd service snippet.
Fixes: https://pagure.io/freeipa/issue/8617
Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
Reviewed-By: Thomas Woerner <twoerner@redhat.com>
In file included from /usr/include/string.h:519,
from /usr/include/lber.h:30,
from /usr/include/ldap.h:30,
from internal.h:27,
from main.c:31:
In function 'strncpy',
inlined from 'main' at main.c:237:5:
/usr/include/bits/string_fortified.h:91:10: warning: 'strncpy' specified bound 255 equals destination size [-Wstringop-truncation]
91 | return __builtin___strncpy_chk (__dest, __src, __len, __bos (__dest));
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Fixes: https://pagure.io/freeipa/issue/8585
Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
Reviewed-By: Rob Crittenden <rcritten@redhat.com>
ipa_gethostfqdn() now returns a pointer to a statically allocated buffer
or NULL in case of an error. The caller no longer has to supply a
correctly allocated buffer.
Rename IPA_HOST_HOST to_LEN IPA_HOST_FQDN_LEN and use IPA_HOST_FQDN_LEN
wherever code copies a hostname supplied from ipa_gethostfqdn().
Clarify that MAXHOSTNAMELEN and MAXHOSTFQDNLEN are different things.
Signed-off-by: Christian Heimes <cheimes@redhat.com>
Reviewed-By: Fraser Tweedale <ftweedal@redhat.com>
FreeIPA's Python and C code used different approaches to get the FQDN of
the host. Some places assumed that gethostname() returns a FQDN. Other
code paths used glibc's resolver to resolve the current node name to a
FQDN.
Python code now uses the ipalib.constants.FQDN where a fully qualified
domain name is expected. The variable is initialized only once and avoids
potential DNS lookups.
C code uses a new helper function ipa_gethostfqdn() in util package. The
function implements similar logic as gethostfqdn() except it uses more
modern getaddrinfo(). The result is cached as well.
Fixes: https://pagure.io/freeipa/issue/8501
Signed-off-by: Christian Heimes <cheimes@redhat.com>
Reviewed-By: Fraser Tweedale <ftweedal@redhat.com>
Use OpenSSL's HMAC API instead of NSS.
Fixes: Fixes: https://pagure.io/freeipa/issue/6857
Signed-off-by: Christian Heimes <cheimes@redhat.com>
Reviewed-By: Alexander Bokovoy <abokovoy@redhat.com>
Replace six.moves and six.StringIO/BytesIO imports with cannonical
Python 3 packages.
Note: six.moves.input behaves differently than builtin input function.
Therefore I left six.moves.input for now.
See: https://pagure.io/freeipa/issue/7715
Signed-off-by: Christian Heimes <cheimes@redhat.com>
Reviewed-By: Fraser Tweedale <ftweedal@redhat.com>
The ipa-otpd code occasionally removes elements from one queue,
inspects and modifies them, and then inserts them into
another (possibly identical, possibly different) queue. When the next
pointer isn't cleared, this can result in element membership in both
queues, leading to double frees, or even self-referential elements,
causing infinite loops at traversal time.
Rather than eliminating the pattern, make it safe by clearing the next
field any time an element enters or exits a queue.
Related https://pagure.io/freeipa/issue/7262
Reviewed-By: Florence Blanc-Renaud <frenaud@redhat.com>
- Add missing executable bits to all scripts
- Remove executable bits from all files that are not scripts,
e.g. js, html, and Python libraries.
- Remove Python shebang from all Python library files.
It's frown upon to have executable library files in site-packages.
Signed-off-by: Christian Heimes <cheimes@redhat.com>
Reviewed-By: Armando Neto <abiagion@redhat.com>
If the element being removed were not the queue head,
otpd_queue_pop_msgid() would not actually remove the element, leading
to potential double frees and request replays.
Reviewed-By: Alexander Bokovoy <abokovoy@redhat.com>
Reviewed-By: Stanislav Laznicka <slaznick@redhat.com>
The Python 3 refactoring effort is finishing, it should be safe
to turn all scripts to run in Python 3 by default.
https://pagure.io/freeipa/issue/4985
Reviewed-By: Christian Heimes <cheimes@redhat.com>
It was working accidentally because krb5 libs are part of OPENLDAP_LIBS.
Reviewed-By: Martin Basti <mbasti@redhat.com>
Reviewed-By: Alexander Bokovoy <abokovoy@redhat.com>
In at least one case, when an LDAP socket closes, a read event is fired
rather than an error event. Without this patch, ipa-otpd silently
ignores this event and enters a state where all bind auths fail.
To remedy this problem, we pass error events along the same path as read
events. Should the actual read fail, we exit.
https://bugzilla.redhat.com/show_bug.cgi?id=1377858https://fedorahosted.org/freeipa/ticket/6368
Reviewed-By: Alexander Bokovoy <abokovoy@redhat.com>
Using a pragma instead of guards is easier to write, less error prone
and avoids name clashes (a source of very subtle bugs). This pragma
is supported on almost all compilers, including all the compilers we
care about: https://en.wikipedia.org/wiki/Pragma_once#Portability.
This patch does not change the autogenerated files: asn1/asn1c/*.h.
Reviewed-By: Martin Basti <mbasti@redhat.com>
Reviewed-By: Christian Heimes <cheimes@redhat.com>
Before this patch, if the user was configured for either OTP or password
it was possible to do a 1FA authentication through ipa-otpd. Because this
correctly respected the configuration, it is not a security error.
However, once we begin to insert authentication indicators into the
Kerberos tickets, we cannot allow 1FA authentications through this
code path. Otherwise the ticket would contain a 2FA indicator when
only 1FA was actually performed.
To solve this problem, we have ipa-otpd send a critical control during
the bind operation which informs the LDAP server that it *MUST* validate
an OTP token for authentication to be successful. Next, we implement
support for this control in the ipa-pwd-extop plugin. The end result is
that the bind operation will always fail if the control is present and
no OTP is validated.
https://fedorahosted.org/freeipa/ticket/433
Reviewed-By: Sumit Bose <sbose@redhat.com>
We always have to call find_base() in order to force libldap to open
the socket. However, if no base is actually required then there is
no reason to error out if find_base() fails. This condition can arise
when anonymous binds are disabled.
Reviewed-By: Martin Basti <mbasti@redhat.com>
The StringIO class was moved to the io module.
(In Python 2, io.StringIO is available, but is Unicode-only.)
Reviewed-By: David Kupka <dkupka@redhat.com>
Reviewed-By: Jan Cholasta <jcholast@redhat.com>
Reviewed-By: Martin Basti <mbasti@redhat.com>
This daemon listens for RADIUS packets on a well known
UNIX domain socket. When a packet is received, it queries
LDAP to see if the user is configured for RADIUS authentication.
If so, then the packet is forwarded to the 3rd party RADIUS server.
Otherwise, a bind is attempted against the LDAP server.
https://fedorahosted.org/freeipa/ticket/3366http://freeipa.org/page/V3/OTP
