Ok, so not having full context as some of the discussion was happening on IRC (and no mention of it was left on the bug here + the package was not rejected from the queue), I have possibly prematurely accepted the libnet-ssleay-perl SRU. After getting logs from Rik I see there's still no consensus on whether the read()/write() behavioral change is wanted/acceptable or not. When reviewing the package I got the impression that it was an approved change (per regression potential).
To my defense: with bionic-proposed being in a broken state causing build failures for both packages and images, the only two sane options were either accepting the package as-is or removing the two previous SRUs from -proposed. Since I was obviously missing context and did barge in uninvited: Steve, if you think that the SRU is not really acceptable as is, please request a follow up upload with further changes. Thanks. -- You received this bug notification because you are a member of Ubuntu Touch seeded packages, which is subscribed to openssl in Ubuntu. https://bugs.launchpad.net/bugs/1797386 Title: [SRU] OpenSSL 1.1.1 to 18.04 LTS Status in libwww-perl package in Ubuntu: Fix Released Status in openssl package in Ubuntu: Fix Released Status in python-tornado package in Ubuntu: In Progress Status in libio-socket-ssl-perl source package in Bionic: Fix Committed Status in libnet-ssleay-perl source package in Bionic: Fix Committed Status in libwww-perl source package in Bionic: Fix Committed Status in openssl source package in Bionic: Fix Committed Status in python-cryptography source package in Bionic: Fix Committed Status in python2.7 source package in Bionic: Fix Committed Status in python3.6 source package in Bionic: Fix Committed Status in python3.7 source package in Bionic: Fix Committed Status in r-cran-openssl source package in Bionic: Fix Committed Status in ruby-openssl source package in Bionic: Fix Committed Status in ruby2.5 source package in Bionic: Fix Committed Bug description: [Impact] * OpenSSL 1.1.1 is an LTS release upstream, which will continue to receive security support for much longer than 1.1.0 series will. * OpenSSL 1.1.1 comes with support for TLS v1.3 which is expected to be rapidly adopted due to increased set of supported hashes & algoes, as well as improved handshake [re-]negotiation. * OpenSSL 1.1.1 comes with improved hw-acceleration capabilities. * OpenSSL 1.1.1 is ABI/API compatible with 1.1.0, however some software is sensitive to the negotiation handshake and may either need patches/improvements or clamp-down to maximum v1.2. [Test Case] * Rebuild all reverse dependencies * Execute autopkg tests for all of them * Clamp down to TLS v1.2 software that does not support TLS v1.3 (e.g. mongodb) * Backport TLS v1.3 support patches, where applicable [Test cases for the python updates] python3.7 is a preview in bionic as a non-supported/non-default version of python3. Passing it's own autopkgtests is sufficient validation for python3.7. It includes a point release update, with OpenSSL 1.1.1 compat and features. python3.6 not only has OpenSSL 1.1.1 compat and features patches, but also includes a point release update to 3.6.8. It has been part of the full-archive rebuild and regression analysis. Autopkgtests were triggered for python3.6 and python3-defaults with regressions already fixed in the individual packages as appropriate. python2.7 has the update from .15~rc1 to .15 final, with OpenSSL 1.1.1 compat only. It has been part of the full-archive rebuild and regression analysis. Autopkgtests were triggered for python2.7 and python-defaults with regressions already fixed in the individual packages as appropriate. The archive rebuilds done, were commulative with OpenJDK 11, OpenSSL 1.1.1 and python point releases as seen in: http://people.canonical.com/~doko/ftbfs-report/test-rebuild-20181222-bionic.html http://people.canonical.com/~doko/ftbfs-report/test-rebuild-20181222-test-bionic.html And analyzed in https://docs.google.com/spreadsheets/d/1tMIwlwoHH_1h5sbvUbNac6-HIPKi3e0Xr8ebchIOU1A/edit#gid=147857652 [ Test case libwww-perl (and deps) regression ] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=914034 1. apt install liblwp-protocol-https-perl 2. enable -proposed 3. apt install libio-socket-ssl-perl libnet-ssleay-perl 4. perl -MLWP::UserAgent -e 'LWP::UserAgent->new->post("https://facebook.com";, { data => "foo" }) or die' [Regression Potential] * Connectivity interop is the biggest issues which will be unavoidable with introducing TLS v1.3. However, tests on cosmic demonstrate that curl/nginx/google-chrome/mozilla-firefox connect and negotiate TLS v1.3 without issues. * Mitigation of discovered connectivity issues will be possible by clamping down to TLS v1.2 in either server-side or client-side software or by backporting relevant support fixes * Notable changes are listed here https://wiki.openssl.org/index.php/TLS1.3 * Most common connectivity issues so far: - client verifies SNI in TLSv1.3 mode, yet client doesn't set hostname. Solution is client change to set hostname, or to clamp down the client to TLSv1.2. - session negotiation is different in TLSv1.3, existing client code may fail to create/negotiate/resume session. Clients need to learn how to use session callback. - non-application data records. TLSv1.3 sends more of these, when compared with previous versions, and some applications may not handle this correctly. Resulting in application data not being available, when previously expected. Mitigation around these involve disabling/enabling SSL_MODE_AUTO_RETRY or setting max protocol version to TLSv1.2. For example see discussion identified in the perl stack https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=914034 Similar hangs are possible with prior versions of TLS as well, it is however easier to trigger this with TLSv1.3. * Deprecated npn extenstion does not exist in TLSv1.3 implementation. * This update bundles python 3.6 and 3.7 point releases * libnet-ssleay-perl introduces two API changes which carry some risk of regression to consuming applications. The risk is considered small. - Servers implemented in perl may now raise SIGPIPE in the event of a premature client disconnection. This may be a behavior change in openssl itself but has only been noticed in the libnet-ssleay-perl tests. This may represent a DoS attack against any third-party TLS-using servers implemented in perl if they do not already handle SIGPIPE gracefully. - The behavior of SSLeay::read() and SSLeay::write() has been changed to NOT retry on short reads/short writes, leading to the perl API more closely matching the C API. There are new ssl_read_all() / ssl_write_all() calls for applications which want the previous behavior. [Other Info] * Previous FFe for OpenSSL in 18.10 is at https://bugs.launchpad.net/ubuntu/+source/openssl/+bug/1793092 * TLS v1.3 support in NSS is expected to make it to 18.04 via security updates * TLS v1.3 support in GnuTLS is expected to be available in 19.04 * Test OpenSSL is being prepared in https://launchpad.net/~ci-train-ppa-service/+archive/ubuntu/3473 [Autopkgtest Regressions] dovecot/armhf - flakey libnet-ssleay-perl - awaiting sru accept into proposed of libnet-ssleay-perl and libio-socket-ssl-perl due to fixes and versioned breaks. linux* - rebuild testcases passes (for some edge flavours the build fails in non-ssl portions of the build), ubuntu-regression-suite testcase fails for a few variants but should have been skipped (in progress to be fixed in https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1823056) openvswitch/i386 - extremely flakey, errors out or fails mostly To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/libwww-perl/+bug/1797386/+subscriptions -- Mailing list: https://launchpad.net/~touch-packages Post to : touch-packages@lists.launchpad.net Unsubscribe : https://launchpad.net/~touch-packages More help : https://help.launchpad.net/ListHelp
