openldap-bugs

openldap-bugs@openldap.org

1 participants
21065 discussions

(ITS#6463) dnssrv_back_referrals: ldaps not handled, search base gone
by jochen＠keutel.de 26 Jan '10

26 Jan '10

Full_Name: Jochen Keutel Version: 2.4.21 OS: Solaris 10 URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (87.159.206.14) dnssrv_back_referrals() always returns ldap URLs - even when the original request was a ldaps query. Additionally only protocol and hostname are put into the URL - the search base doesn't appear any more. Problematic code: url.bv_len = STRLENOF( "ldap://" ) + strlen( hosts[i] ); url.bv_val = ch_malloc( url.bv_len + 1 ); strcpy( url.bv_val, "ldap://" ); strcpy( &url.bv_val[STRLENOF( "ldap://" )], hosts[i] ); I'm not sure whether you want me to deliver a complete patch or not ... Thanks to my colleague Manuel Gaupp for detecting this problem. Best regards, Jochen.

1 0

(ITS#6462) DNS SRV records: ldaps ???
by jochen＠keutel.de 26 Jan '10

26 Jan '10

Full_Name: Jochen Keutel Version: 2.4.21 OS: Solaris 10 URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (87.159.206.14) The code for detecting LDAP servers via DNS SRV records seems to handle only ldap URLs - not ldaps URLs. Esp.: If you access an OpenLDAP server with ldaps://host1/..., then the DNS SRV code returns a URL like ldap://host2/... . So the LDAP server chains the original ldaps request as a LDAP query without TLS. This gets problematic when the (company / project) requirements clearly state that only ldaps has to be used ... The problematic code seems to be in libraries/libldap/dnssrv.c, function ldap_domain2hostlist(): request = LDAP_MALLOC(strlen(domain) + sizeof("_ldap._tcp.")); ... sprintf(request, "_ldap._tcp.%s", domain); So always "_ldap._tcp" is used (hard coded) - not "ldaps._tcp" in case of ldaps in the original query. I'd suggest to introduce a third parameter to that function: int ldap_domain2hostlist( LDAP_CONST char *protocol, LDAP_CONST char *domain, char **list ) protocol is either "ldap" oder "ldaps". The code calling this function (slapd/back-dnssrv/search.c and referral.c) has to be adjusted as well. I'm not sure whether you want me to deliver a complete patch or not ... Thanks to my colleague Manuel Gaupp for detecting this problem. Best regards, Jochen.

1 0

Re: (ITS#6457) hiding remote attribute with rwm-map in relay context crashes server
by masarati＠aero.polimi.it 25 Jan '10

25 Jan '10

Guillaume.Rousse(a)inria.fr wrote: > Le 18/01/2010 17:13, masarati(a)aero.polimi.it a écrit : >> Could not reproduce. Apparently, it's something else that's creating >> hassle. Please provide a stack backtrace. > > Actually, it's an assertion failure un entry.c: > slapd: entry.c:483: entry_clean: Assertion `e->e_private == ((void *)0)' > failed. > Abandon > > Here the backtrace, but compilation optimisation may hinder its usability: Well, at least it shows where the failure is. However, since I can't reproduce it, I'll need your (sanitized) slapd.conf, a LDIF to load and the operation that causes the failure. p.

1 0

Re: (ITS#6460) SASL EXTERNAL fails with long certificate serial numbers
by masarati＠aero.polimi.it 25 Jan '10

25 Jan '10

Howard Chu wrote: > masarati(a)aero.polimi.it wrote: >>> * masarati(a)aero.polimi.it [2010-01-24 16:01:23 +0100]: >>>> Funny enough, the same thing is dealt with correctly in certificate >>>> validation/normalization in slapd/schema_init.c >>> That was a result of ITS#5070 (which you filed). >> right :) >> >>> Maybe there is an >>> opportunity for refactoring, but I wouldn't be a good judge of that. >> I don't quite bother about refactoring to minimize code duplication. >> Rather, I think the libldap function x509_cert_get_dn() should first >> validate the certificate, much like slapd's certificateValidate() does. > > Since the cert was obtained thru a TLS handshake, we assume it has already > been validated by the TLS library. Further validation is not needed. What I mean is that the TLS library may handle certificates that our function does not like (as in this case). Slapd's code, while skipping fields, checks their tags. We should do the same here, IMHO. p.

1 0

Re: (ITS#6460) SASL EXTERNAL fails with long certificate serial numbers
by hyc＠symas.com 24 Jan '10

24 Jan '10

masarati(a)aero.polimi.it wrote: >> * masarati(a)aero.polimi.it [2010-01-24 16:01:23 +0100]: >>> Funny enough, the same thing is dealt with correctly in certificate >>> validation/normalization in slapd/schema_init.c >> >> That was a result of ITS#5070 (which you filed). > > right :) > >> Maybe there is an >> opportunity for refactoring, but I wouldn't be a good judge of that. > > I don't quite bother about refactoring to minimize code duplication. > Rather, I think the libldap function x509_cert_get_dn() should first > validate the certificate, much like slapd's certificateValidate() does. Since the cert was obtained thru a TLS handshake, we assume it has already been validated by the TLS library. Further validation is not needed. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/

1 0

(ITS#6461) back-sql quote characters in query
by tolich.arz＠gmail.com 24 Jan '10

24 Jan '10

Full_Name: Anatoly Version: 2.4.19 OS: GNU/Linux URL: Submission from: (NULL) (89.169.85.181) I'm using openldap 2.4.19 with sql backend. I have a troubles with queries that contains single-quote ( ' ) character. For example, if I searching for (cn=Zool'man): <==backsql_srch_query() returns SELECT DISTINCT ldap_entries.id,phpbb_users.user_id,varchar_ci('phpbbUser') AS objectClass,ldap_entries.dn AS dn FROM ldap_entries,phpbb_users WHERE phpbb_users.user_id=ldap_entries.keyval AND ldap_entries.oc_map_id=? AND 9=9 AND (varchar_ci(phpbb_users.username)='ZOOL'MAN') Constructed query: SELECT DISTINCT ldap_entries.id,phpbb_users.user_id,varchar_ci('phpbbUser') AS objectClass,ldap_entries.dn AS dn FROM ldap_entries,phpbb_users WHERE phpbb_users.user_id=ldap_entries.keyval AND ldap_entries.oc_map_id=? AND 9=9 AND (varchar_ci(phpbb_users.username)='ZOOL'MAN') id: '2' backsql_oc_get_candidates(): error executing query Return code: -1 nativeErrCode=7 SQLengineState=S1000 msg="[unixODBC]ERROR: syntax error at or near "MAN" at character 271; In this case query should be like varchar_ci(phpbb_users.username)='ZOOL\'MAN' instead of 'ZOOL'MAN' Additionally, I fear this opens a possibility of sql injection, depending on RDBMS.

1 0

Re: (ITS#6460) SASL EXTERNAL fails with long certificate serial numbers
by Sergio.Gelato＠astro.su.se 24 Jan '10

24 Jan '10

* masarati(a)aero.polimi.it [2010-01-24 19:14:02 +0100]: > I'm applying the fix, please test. I'm in production with the fix (on 2.4.11; I don't have the time to deploy HEAD right now). syncrepl (which is what I'm using SASL EXTERNAL for) now works again. As for the need to validate the certificate: shouldn't this have been taken care of during the preceding TLS handshake?

1 0

Re: (ITS#6460) SASL EXTERNAL fails with long certificate serial numbers
by masarati＠aero.polimi.it 24 Jan '10

24 Jan '10

> * masarati(a)aero.polimi.it [2010-01-24 16:01:23 +0100]: >> Funny enough, the same thing is dealt with correctly in certificate >> validation/normalization in slapd/schema_init.c > > That was a result of ITS#5070 (which you filed). right :) > Maybe there is an > opportunity for refactoring, but I wouldn't be a good judge of that. I don't quite bother about refactoring to minimize code duplication. Rather, I think the libldap function x509_cert_get_dn() should first validate the certificate, much like slapd's certificateValidate() does. I'm applying the fix, please test. p.

1 0

Re: (ITS#6460) SASL EXTERNAL fails with long certificate serial numbers
by Sergio.Gelato＠astro.su.se 24 Jan '10

24 Jan '10

* masarati(a)aero.polimi.it [2010-01-24 16:01:23 +0100]: > Funny enough, the same thing is dealt with correctly in certificate > validation/normalization in slapd/schema_init.c That was a result of ITS#5070 (which you filed). Maybe there is an opportunity for refactoring, but I wouldn't be a good judge of that.

1 0

Re: (ITS#6460) SASL EXTERNAL fails with long certificate serial numbers
by masarati＠aero.polimi.it 24 Jan '10

24 Jan '10

> Full_Name: Sergio Gelato > Version: 2.4.11 (problem also in HEAD) > OS: Debian GNU/Linux > URL: > Submission from: (NULL) (85.225.123.197) > > > Using (X.509 certificate-based) SASL EXTERNAL authentication fails when > the > serial number of the certificate is longer than 4 octets. > > I have traced the problem to the use, by > libraries/libldap/tls.c:x509_cert_get_dn(), > of ber_get_int() to consume the serial number. This fails when the value > to be > decoded > doesn't fit in a ber_int_t. > > I discovered this bug with Debian's OpenLDAP 2.4.11-1+lenny1 but a look at > the > OpenLDAP CVS indicates that the problem is still present in the current > HEAD > (1.167). > > I've successfully tested the following patch: > ==================================================================================== > RFC5280 §4.1.2.2 says "[c]ertificate users MUST be able to handle > serialNumber > values up to 20 octets". ber_get_int() fails beyond 4 octets. Since we > don't > actually care about the certificate serial number, let's just skip it. > > --- a/libraries/libldap/tls.c > +++ b/libraries/libldap/tls.c > @@ -1673,7 +1673,8 @@ > tag = ber_skip_tag( ber, &len ); /* Context + Constructed (version) */ > if ( tag == 0xa0 ) /* Version is optional */ > tag = ber_get_int( ber, &i ); /* Int: Version */ > - tag = ber_get_int( ber, &i ); /* Int: Serial */ > + tag = ber_skip_tag( ber, &len ); /* *Big* integer: Serial */ > + ber_skip_data( ber, len ); > tag = ber_skip_tag( ber, &len ); /* Sequence: Signature */ > ber_skip_data( ber, len ); > if ( !get_subject ) { Funny enough, the same thing is dealt with correctly in certificate validation/normalization in slapd/schema_init.c p.

1 0

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

openldap-bugs