You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@tomcat.apache.org by Michael Osipov <mi...@apache.org> on 2020/04/23 23:45:03 UTC
Position on failing tests with vendor-modified OpenSSL packages
Folks,
I run test from Tomcat master and libtcnative master on FreeBSD, RHEL 7
and HP-UX 11.31 on a regular basis and noticed that the OpenSSL 1.0.2
packages provided by Red Hat and HPE are modified which make several
tests fail. See an excerpt here [1].
To verify this I have compiled OpenSSL from OpenSSL_1_0_2-stable on
FreeBSD w/o any modification and all tests pass, so other must have
modified OpenSSL.
What is our position in terms of support/testing for modified OpenSSL
packages from OS vendors? Should we add a big fat warning somewhere? Add
this to tcnative README that we test/recommend upstream only?
[1] http://home.apache.org/~michaelo/issues/openssl-tests/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Michael Osipov <mi...@apache.org>.
Am 2020-04-24 um 01:45 schrieb Michael Osipov:
> Folks,
>
> I run test from Tomcat master and libtcnative master on FreeBSD, RHEL 7
> and HP-UX 11.31 on a regular basis and noticed that the OpenSSL 1.0.2
> packages provided by Red Hat and HPE are modified which make several
> tests fail. See an excerpt here [1].
> To verify this I have compiled OpenSSL from OpenSSL_1_0_2-stable on
> FreeBSD w/o any modification and all tests pass, so other must have
> modified OpenSSL.
>
> What is our position in terms of support/testing for modified OpenSSL
> packages from OS vendors? Should we add a big fat warning somewhere? Add
> this to tcnative README that we test/recommend upstream only?
>
> [1] http://home.apache.org/~michaelo/issues/openssl-tests/
I have now managed to installed OpenSSL 1.1.1 from OS vendor. It
contains modifications too:
http://home.apache.org/~michaelo/issues/openssl-1.1.1-tests/hpux/
M
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Mark Thomas <ma...@apache.org>.
On 24/04/2020 12:24, Michael Osipov wrote:
> Am 2020-04-24 um 08:57 schrieb Mark Thomas:
<snip/>
>> I think we have several options:
>> - document the expectation more clearly so folks can more easily assess
>> these failures
>> - support using the vendor supported versions with the unit tests
>> - provide a configuration option to skip these tests
>> - detect vendor supplied OpenSSL and automatically skip the tests
>
> We need to do at least the documentation. As for detectection and
> support: I consider this to be hardly solvable because there is no
> identifier in "openssl version -a" which says I have been modified by X.
That is unhelpful to say the least.
Looks like better docs and an option to skip is the way to go.
Mark
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Mark Thomas <ma...@apache.org>.
On 26/04/2020 10:36, Michael Osipov wrote:
> Am 2020-04-24 um 18:30 schrieb Christopher Schultz:
<snip/>
>> This comes down to algorithms which have been compiled-out of the
>> library, right? So we just need to automatically skip tests which
>> attempt to use those algorithms.
Not entirely. We are affected both by algorithms being removed and
algorithms being added.
>> Unfortunately, the whole point of the unit tests is to make sure we
>> haven't missed anything. In order to both remove unsupported
>> algorithms AND test whether the remaining algorithms are
>> properly-mapped, we need ANOTHER implementation of the mapping, or
>> something similar to cross-check the two.
I think I agree with this. If you mean that somewhere we would need to
maintain a list of algorithms there were supported by the vendor
modified package then, yes I agree. The tricky part being that we can't
always identify what is a vendor modified package.
>> Somewhere, we have a /complete/ list of OpenSSL-specified names that
>> we support.
Ciphers is an enumeration of all the Ciphers OpenSSL supports or has
supported in the past.
In TesterOpenSSL we modify that list to remove Ciphers we know are not
supported in the version we are testing.
>> That can easily be dumped somewhere and sorted
>> alphabetically. Then, "openssl ciphers 'ALL'" can be run, sorted, and
>> compared to the sorted list we have.
"ALL" does not return all supported ciphers. Sigh. You need to use
"ALL:eNULL"
>> Anything which isn't in the
>> "openssl ciphers 'ALL'" list can be removed from the list of cipher
>> suites we test.
The problem is, that is how we detected that the list of supported
Cipher suites has changed for a given OpenSSL branch (yes it can change
between point releases).
>> AIUI, it's possible to individually-remove a cipher suite from the
>> unit tests if it's known to be missing on the platform. We just need
>> to make that process automated, right?
>
> Up to the point! That's exactly the route one should go.
We would need to clearly separate (from memory I don't recall how clear
the separation is right now):
- Tests that compare the ciphers we think a version of OpenSSL supports
with the ciphers it reports it supports. These only need to run when
testing with an unmodified OpenSSL from openssl.org
- Tests just need to know which ciphers OpenSSL supports and, as long as
they have the correct list, will always pass
Mark
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Michael Osipov <mi...@apache.org>.
Am 2020-04-24 um 18:30 schrieb Christopher Schultz:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> Mark and Michael,
>
> On 4/24/20 07:24, Michael Osipov wrote:
>> Am 2020-04-24 um 08:57 schrieb Mark Thomas:
>>> On 24/04/2020 00:45, Michael Osipov wrote:
>>>> Folks,
>>>>
>>>> I run test from Tomcat master and libtcnative master on
>>>> FreeBSD, RHEL 7 and HP-UX 11.31 on a regular basis and noticed
>>>> that the OpenSSL 1.0.2 packages provided by Red Hat and HPE are
>>>> modified which make several tests fail. See an excerpt here
>>>> [1]. To verify this I have compiled OpenSSL from
>>>> OpenSSL_1_0_2-stable on FreeBSD w/o any modification and all
>>>> tests pass, so other must have modified OpenSSL.
>>>>
>>>> What is our position in terms of support/testing for modified
>>>> OpenSSL packages from OS vendors? Should we add a big fat
>>>> warning somewhere? Add this to tcnative README that we
>>>> test/recommend upstream only?
>>>
>>> The good news is that the tests that are failing are the ones I
>>> would expect to fail.
>>>
>>> When we added the code that lets us use OpenSSL syntax to define
>>> ciphers for JSSE (and JSSE syntax to define ciphers for OpenSSL)
>>> we added a these tests to ensure that we correctly tracked things
>>> like "ALL", "DEFAULT" as well as "ECDHE" etc. These are a moving
>>> target as support for new ciphers is added and ciphers viewed as
>>> less secure are removed.
>>>
>>> Our unit tests aim to work with the current version of all
>>> publicly supported OpenSSL branches. Currently, master (3.0.x)
>>> and 1.1.1.
>>>
>>> I expect the vendor supported 1.0.2 packages are close to current
>>> 1.1.1 but I wouldn't be surprised to see some minor differences.
>>>
>>> I think we have several options: - document the expectation more
>>> clearly so folks can more easily assess these failures - support
>>> using the vendor supported versions with the unit tests - provide
>>> a configuration option to skip these tests - detect vendor
>>> supplied OpenSSL and automatically skip the tests
>>
>> We need to do at least the documentation. As for detectection and
>> support: I consider this to be hardly solvable because there is no
>> identifier in "openssl version -a" which says I have been modified
>> by X. See:
>>
>>> $ openssl version -a OpenSSL 1.0.2r 26 Feb 2019 built on:
>>> reproducible build, date unspecified platform: hpux-ia64-cc
>>> options: bn(64,64) rc4(idx,int) des(idx,risc1,16,long)
>>> blowfish(idx) compiler: cc -I. -I.. -I../include
>>> -I/ssh_ssl_build/ssl-build/OpenSSL_1_0_FIPS_205/build/zlib +Z
>>> -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -DDSO_DLFCN
>>> -DHAVE_DLFCN_H -D_REENTRANT -Ae +DD32 +O2 +Olit=all -z -DB_ENDIAN
>>> -D_REENTRANT -DOPENSSL_BN_ASM_MONT -DRC4_ASM -DSHA1_ASM
>>> -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DGHASH_ASM
>>> OPENSSLDIR: "/opt/openssl"
>>
>> and
>>
>>> $ openssl version -a OpenSSL 1.0.2k-fips 26 Jan 2017 built on:
>>> reproducible build, date unspecified platform: linux-x86_64
>>> options: bn(64,64) md2(int) rc4(16x,int) des(idx,cisc,16,int)
>>> idea(int) blowfish(idx) compiler: gcc -I. -I.. -I../include
>>> -fPIC -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -D_REENTRANT
>>> -DDSO_DLFCN -DHAVE_DLFCN_H -DKRB5_MIT -m64 -DL_ENDIAN -Wall -O2
>>> -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions
>>> -fstack-protector-strong --param=ssp-buffer-size=4
>>> -grecord-gcc-switches -m64 -mtune=generic -Wa,--noexecstack
>>> -DPURIFY -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT
>>> -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DRC4_ASM -DSHA1_ASM
>>> -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM
>>> -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM -DECP_NISTZ256_ASM
>>> OPENSSLDIR: "/etc/pki/tls" engines: rdrand dynamic
>>
>> The only option I see is to check for OS name/LSB release, call
>> with depdency graph with ldd and check for libssl/libcrypto in
>> default locations, but this is really really ugly.
>>
>> Compared to OpenSSL 1.1.1 from FreeBSD base and OpenSSL 1.0.2
>> compiled myself:
>>> $ openssl version -a OpenSSL 1.1.1e-freebsd 17 Mar 2020 built
>>> on: reproducible build, date unspecified platform: FreeBSD-amd64
>>> options: bn(64,64) rc4(16x,int) des(int) idea(int)
>>> blowfish(ptr) compiler: clang OPENSSLDIR: "/etc/ssl" ENGINESDIR:
>>> "/usr/lib/engines" Seeding source: os-specific
>>
>> Granted, this one says in the version string it is from FreeBSD.
>>
>>> $ /tmp/openssl-1.0.2/bin/openssl version -a OpenSSL 1.0.2v-dev
>>> xx XXX xxxx built on: reproducible build, date unspecified
>>> platform: BSD-x86_64 options: bn(64,64) rc4(16x,int)
>>> des(idx,cisc,16,int) idea(int) blowfish(idx) compiler: cc -I.
>>> -I.. -I../include -fPIC -DOPENSSL_PIC -DOPENSSL_THREADS -pthread
>>> -D_THREAD_SAFE -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DL_ENDIAN
>>> -O3 -Wall -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT
>>> -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DRC4_ASM -DSHA1_ASM
>>> -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM
>>> -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM -DECP_NISTZ256_ASM
>>> OPENSSLDIR: "/tmp/openssl-1.0.2/ssl"
>>
>>> There are probably a few options I've missed.
>>>
>>> I will add at this point that debugging a failure and figuring
>>> out the right fix hasn't always easy.
>>>
>>> I'm currently wondering if some sort of combination of the above
>>> might work. Detect vendor variations, handle them where we can
>>> recognise them and skip them where we don't.
>>>
>>> I think this needs some further thought.
>>
>> Please let me know if you need further assistance here!
>
> This comes down to algorithms which have been compiled-out of the
> library, right? So we just need to automatically skip tests which
> attempt to use those algorithms.
>
> Unfortunately, the whole point of the unit tests is to make sure we
> haven't missed anything. In order to both remove unsupported
> algorithms AND test whether the remaining algorithms are
> properly-mapped, we need ANOTHER implementation of the mapping, or
> something similar to cross-check the two.
>
> Somewhere, we have a /complete/ list of OpenSSL-specified names that
> we support. That can easily be dumped somewhere and sorted
> alphabetically. Then, "openssl ciphers 'ALL'" can be run, sorted, and
> compared to the sorted list we have. Anything which isn't in the
> "openssl ciphers 'ALL'" list can be removed from the list of cipher
> suites we test.
>
> AIUI, it's possible to individually-remove a cipher suite from the
> unit tests if it's known to be missing on the platform. We just need
> to make that process automated, right?
Up to the point! That's exactly the route one should go.
M
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Christopher Schultz <ch...@christopherschultz.net>.
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256
Mark and Michael,
On 4/24/20 07:24, Michael Osipov wrote:
> Am 2020-04-24 um 08:57 schrieb Mark Thomas:
>> On 24/04/2020 00:45, Michael Osipov wrote:
>>> Folks,
>>>
>>> I run test from Tomcat master and libtcnative master on
>>> FreeBSD, RHEL 7 and HP-UX 11.31 on a regular basis and noticed
>>> that the OpenSSL 1.0.2 packages provided by Red Hat and HPE are
>>> modified which make several tests fail. See an excerpt here
>>> [1]. To verify this I have compiled OpenSSL from
>>> OpenSSL_1_0_2-stable on FreeBSD w/o any modification and all
>>> tests pass, so other must have modified OpenSSL.
>>>
>>> What is our position in terms of support/testing for modified
>>> OpenSSL packages from OS vendors? Should we add a big fat
>>> warning somewhere? Add this to tcnative README that we
>>> test/recommend upstream only?
>>
>> The good news is that the tests that are failing are the ones I
>> would expect to fail.
>>
>> When we added the code that lets us use OpenSSL syntax to define
>> ciphers for JSSE (and JSSE syntax to define ciphers for OpenSSL)
>> we added a these tests to ensure that we correctly tracked things
>> like "ALL", "DEFAULT" as well as "ECDHE" etc. These are a moving
>> target as support for new ciphers is added and ciphers viewed as
>> less secure are removed.
>>
>> Our unit tests aim to work with the current version of all
>> publicly supported OpenSSL branches. Currently, master (3.0.x)
>> and 1.1.1.
>>
>> I expect the vendor supported 1.0.2 packages are close to current
>> 1.1.1 but I wouldn't be surprised to see some minor differences.
>>
>> I think we have several options: - document the expectation more
>> clearly so folks can more easily assess these failures - support
>> using the vendor supported versions with the unit tests - provide
>> a configuration option to skip these tests - detect vendor
>> supplied OpenSSL and automatically skip the tests
>
> We need to do at least the documentation. As for detectection and
> support: I consider this to be hardly solvable because there is no
> identifier in "openssl version -a" which says I have been modified
> by X. See:
>
>> $ openssl version -a OpenSSL 1.0.2r 26 Feb 2019 built on:
>> reproducible build, date unspecified platform: hpux-ia64-cc
>> options: bn(64,64) rc4(idx,int) des(idx,risc1,16,long)
>> blowfish(idx) compiler: cc -I. -I.. -I../include
>> -I/ssh_ssl_build/ssl-build/OpenSSL_1_0_FIPS_205/build/zlib +Z
>> -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -DDSO_DLFCN
>> -DHAVE_DLFCN_H -D_REENTRANT -Ae +DD32 +O2 +Olit=all -z -DB_ENDIAN
>> -D_REENTRANT -DOPENSSL_BN_ASM_MONT -DRC4_ASM -DSHA1_ASM
>> -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DGHASH_ASM
>> OPENSSLDIR: "/opt/openssl"
>
> and
>
>> $ openssl version -a OpenSSL 1.0.2k-fips 26 Jan 2017 built on:
>> reproducible build, date unspecified platform: linux-x86_64
>> options: bn(64,64) md2(int) rc4(16x,int) des(idx,cisc,16,int)
>> idea(int) blowfish(idx) compiler: gcc -I. -I.. -I../include
>> -fPIC -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -D_REENTRANT
>> -DDSO_DLFCN -DHAVE_DLFCN_H -DKRB5_MIT -m64 -DL_ENDIAN -Wall -O2
>> -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions
>> -fstack-protector-strong --param=ssp-buffer-size=4
>> -grecord-gcc-switches -m64 -mtune=generic -Wa,--noexecstack
>> -DPURIFY -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT
>> -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DRC4_ASM -DSHA1_ASM
>> -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM
>> -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM -DECP_NISTZ256_ASM
>> OPENSSLDIR: "/etc/pki/tls" engines: rdrand dynamic
>
> The only option I see is to check for OS name/LSB release, call
> with depdency graph with ldd and check for libssl/libcrypto in
> default locations, but this is really really ugly.
>
> Compared to OpenSSL 1.1.1 from FreeBSD base and OpenSSL 1.0.2
> compiled myself:
>> $ openssl version -a OpenSSL 1.1.1e-freebsd 17 Mar 2020 built
>> on: reproducible build, date unspecified platform: FreeBSD-amd64
>> options: bn(64,64) rc4(16x,int) des(int) idea(int)
>> blowfish(ptr) compiler: clang OPENSSLDIR: "/etc/ssl" ENGINESDIR:
>> "/usr/lib/engines" Seeding source: os-specific
>
> Granted, this one says in the version string it is from FreeBSD.
>
>> $ /tmp/openssl-1.0.2/bin/openssl version -a OpenSSL 1.0.2v-dev
>> xx XXX xxxx built on: reproducible build, date unspecified
>> platform: BSD-x86_64 options: bn(64,64) rc4(16x,int)
>> des(idx,cisc,16,int) idea(int) blowfish(idx) compiler: cc -I.
>> -I.. -I../include -fPIC -DOPENSSL_PIC -DOPENSSL_THREADS -pthread
>> -D_THREAD_SAFE -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DL_ENDIAN
>> -O3 -Wall -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT
>> -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DRC4_ASM -DSHA1_ASM
>> -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM
>> -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM -DECP_NISTZ256_ASM
>> OPENSSLDIR: "/tmp/openssl-1.0.2/ssl"
>
>> There are probably a few options I've missed.
>>
>> I will add at this point that debugging a failure and figuring
>> out the right fix hasn't always easy.
>>
>> I'm currently wondering if some sort of combination of the above
>> might work. Detect vendor variations, handle them where we can
>> recognise them and skip them where we don't.
>>
>> I think this needs some further thought.
>
> Please let me know if you need further assistance here!
This comes down to algorithms which have been compiled-out of the
library, right? So we just need to automatically skip tests which
attempt to use those algorithms.
Unfortunately, the whole point of the unit tests is to make sure we
haven't missed anything. In order to both remove unsupported
algorithms AND test whether the remaining algorithms are
properly-mapped, we need ANOTHER implementation of the mapping, or
something similar to cross-check the two.
Somewhere, we have a /complete/ list of OpenSSL-specified names that
we support. That can easily be dumped somewhere and sorted
alphabetically. Then, "openssl ciphers 'ALL'" can be run, sorted, and
compared to the sorted list we have. Anything which isn't in the
"openssl ciphers 'ALL'" list can be removed from the list of cipher
suites we test.
AIUI, it's possible to individually-remove a cipher suite from the
unit tests if it's known to be missing on the platform. We just need
to make that process automated, right?
- -chris
-----BEGIN PGP SIGNATURE-----
Comment: Using GnuPG with Thunderbird - https://www.enigmail.net/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=NXq8
-----END PGP SIGNATURE-----
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Michael Osipov <mi...@apache.org>.
Am 2020-04-24 um 08:57 schrieb Mark Thomas:
> On 24/04/2020 00:45, Michael Osipov wrote:
>> Folks,
>>
>> I run test from Tomcat master and libtcnative master on FreeBSD, RHEL 7
>> and HP-UX 11.31 on a regular basis and noticed that the OpenSSL 1.0.2
>> packages provided by Red Hat and HPE are modified which make several
>> tests fail. See an excerpt here [1].
>> To verify this I have compiled OpenSSL from OpenSSL_1_0_2-stable on
>> FreeBSD w/o any modification and all tests pass, so other must have
>> modified OpenSSL.
>>
>> What is our position in terms of support/testing for modified OpenSSL
>> packages from OS vendors? Should we add a big fat warning somewhere? Add
>> this to tcnative README that we test/recommend upstream only?
>
> The good news is that the tests that are failing are the ones I would
> expect to fail.
>
> When we added the code that lets us use OpenSSL syntax to define ciphers
> for JSSE (and JSSE syntax to define ciphers for OpenSSL) we added a
> these tests to ensure that we correctly tracked things like "ALL",
> "DEFAULT" as well as "ECDHE" etc. These are a moving target as support
> for new ciphers is added and ciphers viewed as less secure are removed.
>
> Our unit tests aim to work with the current version of all publicly
> supported OpenSSL branches. Currently, master (3.0.x) and 1.1.1.
>
> I expect the vendor supported 1.0.2 packages are close to current 1.1.1
> but I wouldn't be surprised to see some minor differences.
>
> I think we have several options:
> - document the expectation more clearly so folks can more easily assess
> these failures
> - support using the vendor supported versions with the unit tests
> - provide a configuration option to skip these tests
> - detect vendor supplied OpenSSL and automatically skip the tests
We need to do at least the documentation. As for detectection and
support: I consider this to be hardly solvable because there is no
identifier in "openssl version -a" which says I have been modified by X.
See:
> $ openssl version -a
> OpenSSL 1.0.2r 26 Feb 2019
> built on: reproducible build, date unspecified
> platform: hpux-ia64-cc
> options: bn(64,64) rc4(idx,int) des(idx,risc1,16,long) blowfish(idx)
> compiler: cc -I. -I.. -I../include -I/ssh_ssl_build/ssl-build/OpenSSL_1_0_FIPS_205/build/zlib +Z -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -DDSO_DLFCN -DHAVE_DLFCN_H -D_REENTRANT -Ae +DD32 +O2 +Olit=all -z -DB_ENDIAN -D_REENTRANT -DOPENSSL_BN_ASM_MONT -DRC4_ASM -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DGHASH_ASM
> OPENSSLDIR: "/opt/openssl"
and
> $ openssl version -a
> OpenSSL 1.0.2k-fips 26 Jan 2017
> built on: reproducible build, date unspecified
> platform: linux-x86_64
> options: bn(64,64) md2(int) rc4(16x,int) des(idx,cisc,16,int) idea(int) blowfish(idx)
> compiler: gcc -I. -I.. -I../include -fPIC -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DKRB5_MIT -m64 -DL_ENDIAN -Wall -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -m64 -mtune=generic -Wa,--noexecstack -DPURIFY -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DRC4_ASM -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM -DECP_NISTZ256_ASM
> OPENSSLDIR: "/etc/pki/tls"
> engines: rdrand dynamic
The only option I see is to check for OS name/LSB release, call with
depdency graph with ldd and check for libssl/libcrypto in default
locations, but this is really really ugly.
Compared to OpenSSL 1.1.1 from FreeBSD base and OpenSSL 1.0.2 compiled
myself:
> $ openssl version -a
> OpenSSL 1.1.1e-freebsd 17 Mar 2020
> built on: reproducible build, date unspecified
> platform: FreeBSD-amd64
> options: bn(64,64) rc4(16x,int) des(int) idea(int) blowfish(ptr)
> compiler: clang
> OPENSSLDIR: "/etc/ssl"
> ENGINESDIR: "/usr/lib/engines"
> Seeding source: os-specific
Granted, this one says in the version string it is from FreeBSD.
> $ /tmp/openssl-1.0.2/bin/openssl version -a
> OpenSSL 1.0.2v-dev xx XXX xxxx
> built on: reproducible build, date unspecified
> platform: BSD-x86_64
> options: bn(64,64) rc4(16x,int) des(idx,cisc,16,int) idea(int) blowfish(idx)
> compiler: cc -I. -I.. -I../include -fPIC -DOPENSSL_PIC -DOPENSSL_THREADS -pthread -D_THREAD_SAFE -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DL_ENDIAN -O3 -Wall -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DRC4_ASM -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM -DECP_NISTZ256_ASM
> OPENSSLDIR: "/tmp/openssl-1.0.2/ssl"
> There are probably a few options I've missed.
>
> I will add at this point that debugging a failure and figuring out the
> right fix hasn't always easy.
>
> I'm currently wondering if some sort of combination of the above might
> work. Detect vendor variations, handle them where we can recognise them
> and skip them where we don't.
>
> I think this needs some further thought.
Please let me know if you need further assistance here!
Michael
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org
Re: Position on failing tests with vendor-modified OpenSSL packages
Posted by Mark Thomas <ma...@apache.org>.
On 24/04/2020 00:45, Michael Osipov wrote:
> Folks,
>
> I run test from Tomcat master and libtcnative master on FreeBSD, RHEL 7
> and HP-UX 11.31 on a regular basis and noticed that the OpenSSL 1.0.2
> packages provided by Red Hat and HPE are modified which make several
> tests fail. See an excerpt here [1].
> To verify this I have compiled OpenSSL from OpenSSL_1_0_2-stable on
> FreeBSD w/o any modification and all tests pass, so other must have
> modified OpenSSL.
>
> What is our position in terms of support/testing for modified OpenSSL
> packages from OS vendors? Should we add a big fat warning somewhere? Add
> this to tcnative README that we test/recommend upstream only?
The good news is that the tests that are failing are the ones I would
expect to fail.
When we added the code that lets us use OpenSSL syntax to define ciphers
for JSSE (and JSSE syntax to define ciphers for OpenSSL) we added a
these tests to ensure that we correctly tracked things like "ALL",
"DEFAULT" as well as "ECDHE" etc. These are a moving target as support
for new ciphers is added and ciphers viewed as less secure are removed.
Our unit tests aim to work with the current version of all publicly
supported OpenSSL branches. Currently, master (3.0.x) and 1.1.1.
I expect the vendor supported 1.0.2 packages are close to current 1.1.1
but I wouldn't be surprised to see some minor differences.
I think we have several options:
- document the expectation more clearly so folks can more easily assess
these failures
- support using the vendor supported versions with the unit tests
- provide a configuration option to skip these tests
- detect vendor supplied OpenSSL and automatically skip the tests
There are probably a few options I've missed.
I will add at this point that debugging a failure and figuring out the
right fix hasn't always easy.
I'm currently wondering if some sort of combination of the above might
work. Detect vendor variations, handle them where we can recognise them
and skip them where we don't.
I think this needs some further thought.
Mark
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org