Re: [Bug 1905790] Re: Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support for p11_child
>> Soo... Given we prefer to stay conservative and not change SSSD crypto > > I didn't say that! I know, I'm not saying that you took a decision on that but I was speaking in plural form as I recognize what you say in the sense that indeed there may be cases which we don't think of that we could break. >> backend fully (to be clear, I would have preferred it to follow >> upstream, not to provide a solution that will change in next LTS no >> matter what, and avoid having "frankensteins", but wasn't a strong >> requirement for me) I've been exploring ways to get only the component >> we care (p11_child) to use p11-kit and openssl. > > This is certainly a valuable angle to look at - thanks! > >> Robie, this would be better SRU approach? > > I think you misunderstand me. I'm not saying that your upload *has* to > be narrow. I've not formed an opinion that yet. What I'm saying is that > whatever size of scope you choose, there must be a regression analysis > that covers that scope. I understood this, reason why I thought that, given we have the chance to make it a narrower scope, then I tried to get that done. > But the analysis is still necessary and must not be skipped. Sure, not trying to do that, I'm just saying that I can't over all the cases myself. > I appreciate that sometimes it's harder or riskier to narrow the scope, > so I'm still open to widening the scope - *if* there is an appropriate > justification *and* full regression analysis of that wider scope > provided. Problem is that I'm quite sure we can't cover all the cases in a such complicated piece of software that may be configured in so many ways. Thus the reason I thought narrowing the scope was a better idea. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1905790 Title: Make SSSD in 20.04 using OpenSSL and p11-kit (instead of NSS) for p11_child To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/ca-certificates/+bug/1905790/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
Re: [Bug 1905790] Re: Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support for p11_child
On Wed, Dec 02, 2020 at 03:29:43AM -, Marco Trevisan (Treviño) wrote: > Soo... Given we prefer to stay conservative and not change SSSD crypto I didn't say that! > backend fully (to be clear, I would have preferred it to follow > upstream, not to provide a solution that will change in next LTS no > matter what, and avoid having "frankensteins", but wasn't a strong > requirement for me) I've been exploring ways to get only the component > we care (p11_child) to use p11-kit and openssl. This is certainly a valuable angle to look at - thanks! > Robie, this would be better SRU approach? I think you misunderstand me. I'm not saying that your upload *has* to be narrow. I've not formed an opinion that yet. What I'm saying is that whatever size of scope you choose, there must be a regression analysis that covers that scope. If you take a widely scope, then I expect a regression analysis to cover what I feel are the obvious possible implications of that change. By nature of it being wider, the regression analysis can be expected to be more work, of course. Because a wider scope generally correlates with increased regression risk, I'd also expect a justification of why the narrow scope is less desirable. But the analysis is still necessary and must not be skipped. If you take a narrow scope, then that's correlated with lower regression risk, and because a regression analysis would be narrower in scope to match, it might well be less work. I appreciate that sometimes it's harder or riskier to narrow the scope, so I'm still open to widening the scope - *if* there is an appropriate justification *and* full regression analysis of that wider scope provided. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1905790 Title: Make SSSD in 20.04 using OpenSSL and p11-kit (instead of NSS) for p11_child To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/sssd/+bug/1905790/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
[Bug 1905790] Re: Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support for p11_child
Soo... Given we prefer to stay conservative and not change SSSD crypto backend fully (to be clear, I would have preferred it to follow upstream, not to provide a solution that will change in next LTS no matter what, and avoid having "frankensteins", but wasn't a strong requirement for me) I've been exploring ways to get only the component we care (p11_child) to use p11-kit and openssl. As per this, I've prepared two possible approaches in two patches (I'd just squash those or pick one in case). The simplest approach [1] was to just compile the NSS version and then only the p11_child using OpenSSL and then manually install to the package... Ensuring that we always pass to it a PEM CA cert file. Works, but will not allow us to test it using upstream tests. So, I've added a further patch that acts mostly on upstream code and removes the usage of libnss ONLY from p11_child and its related operations (smartcard and ssh auth), you can see it better in this patch-queue branch (check the default one to see the debian/* changes): - https://salsa.debian.org/3v1n0-guest/sssd/-/commits/patch-queue/p11-kit-p11_child This works properly and it's tested, you can try the packages at: - https://launchpad.net/~ci-train-ppa-service/+archive/ubuntu/4361.1 Theoretically, it would be even possible to keep support for an NSS p11_child (i.e. provide two binaries, and select which one to use depending on the db defined in configuration file), but as said in the bug description I don't think that such change would actually matter for anyone, as we don't provide a system NSS database. Robie, this would be better SRU approach? [1] https://salsa.debian.org/3v1n0-guest/sssd/-/commits/p11-kit- p11_child-v1 ** Summary changed: - Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support for p11_child + Make SSSD in 20.04 using OpenSSL and p11-kit (instead of NSS) for p11_child -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1905790 Title: Make SSSD in 20.04 using OpenSSL and p11-kit (instead of NSS) for p11_child To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/sssd/+bug/1905790/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
[Bug 1905790] Re: Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support for p11_child
** Summary changed: - Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support + Recompile SSSD in 20.04 using OpenSSL (instead of NSS) support for p11_child ** Description changed: [ Impact ] SSSD supports in 20.04 two security backends: NSS and OpenSSL (speaking in past tense as upstream dropped NSS support completely). Those two backends are used for various generic crypto features (so they are interchangeable), but also for the management of the PKCS#11 modules for smart cards. In this case, the main problem is that by using NSS it also relies on the presence of a "system NSS" database [1] that is something present in Fedora and RHEL, but not in ubuntu or generic Linux distributions. In order to make SSSD to find a smart card module, we would then need to create a such database that mentions a p11kit proxy that will eventually load the p11-kit module and then add the card CA certificate to the same DB (see more details in [2]). And even in such case... It will not work at login phase. This is making support for Smart-card based authentication in 20.04 quite complicated, and hard to implement in professional environments (see bug #1865226). - As per this, just recompiling SSSD to use OpenSSL (as it already happens + As per this, recompiling SSSD to use OpenSSL (as it already happens starting from 20.10) would be enough to make the p11_child tool (the one in charge for smartcard authentications) to be able to get the smartcard devices from p11-kit allowed modules and to check their certificate using CA certificates in the ubuntu system ca certificate files (or other configured file). One more mayor reason to do this, is also that if we fix 20.04 now to use the "proper" method, people who will configure smartcard access there via SSSD (not easily possible right now) won't be affected by future migrations. [ Test case ] With a smartcard reader available (and with a card in its slot) launch: - sudo /usr/libexec/sssd/p11_child --pre -d 10 --debug-fd=2 \ --nssdb=/etc/ssl/certs/ca-certificates.crt The tool should find your card: (2020-11-26 21:34:22:020395): [p11_child[100729]] [do_card] (0x4000): Module List: (2020-11-26 21:34:22:020481): [p11_child[100729]] [do_card] (0x4000): common name: [p11-kit-trust]. (2020-11-26 21:34:22:020497): [p11_child[100729]] [do_card] (0x4000): dll name: [/usr/lib/x86_64-linux-gnu/pkcs11/p11-kit-trust.so]. (2020-11-26 21:34:22:020569): [p11_child[100729]] [do_card] (0x4000): Description [/etc/ssl/certs/ca-certificates.crt PKCS#11 Kit ] Manufacturer [PKCS#11 Kit ] flags [1] removable [false] token present [true]. (2020-11-26 21:34:22:020611): [p11_child[100729]] [do_card] (0x4000): common name: [opensc-pkcs11]. (2020-11-26 21:34:22:020646): [p11_child[100729]] [do_card] (0x4000): dll name: [/usr/lib/x86_64-linux-gnu/pkcs11/opensc-pkcs11.so]. (2020-11-26 21:34:22:025443): [p11_child[100729]] [do_card] (0x4000): Description [VMware Virtual USB CCID 00 00 VMware ] Manufacturer [VMware ] flags [7] removable [true] token present [true]. (2020-11-26 21:34:22:025725): [p11_child[100729]] [do_card] (0x4000): Found [MARCO TREVISAN (PIN CNS0)] in slot [VMware Virtual USB CCID 00 00][0] of module [1][/usr/lib/x86_64-linux-gnu/pkcs11/opensc-pkcs11.so]. Then the tool might fail if the card certificate is not added to the ca- certificates.crt, but this is outside the scope of the test case. What it matters is that the card is found. [ Regression potential ] While the change may involve quite different code paths when it comes to security features, I think we trust OpenSSL enough to be an acceptable crypto backend. And behavior should not change. Also assuming that upstream dropped NSS support completely in latest release, keeping the same functionalities. The only binary that is really affected in its behavior is p11_child. And I'm confident that even changing this a lot, it will break only those setup (if there are any, given that smartcard access is currently not supported by ubuntu) that have been manually configured using an unsupported system NSS db. In the remote case there are such configurations, though, the fix will be as easy as adding the CA certificates to the new PAM cert DB (by default /etc/sssd/pki/sssd_auth_ca_db.pem), while the p11-kit modules will continue to work as before. It's also technically easy to do a postinst script that will just export all the certificates from the old nss db (/etc/pki/nssdb) into the new file, if we want to avoid any unlikely breakage. [1] https://github.com/SSSD/sssd/blob/sssd-2_3_1/src/responder/pam/pamsrv.c#L53 [2] https://hackmd.io/@3v1n0/ubuntu-smartcard-login#NSS-Database-to-be-deprecated
