Re: [OE-core][PATCH v2] openssh: upgrade 9.7p1 -> 9.8p1

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Thu, 2024-07-04 at 13:46 +0100, Jose Quaresma via
lists.openembedded.org wrote:
> Release notes at https://www.openssh.com/txt/release-9.8
> 
> Security
> ========
> 
> This release contains fixes for two security problems, one critical
> and one minor.
> 
> 1) Race condition in sshd(8)
> 
> A critical vulnerability in sshd(8) was present in Portable OpenSSH
> versions between 8.5p1 and 9.7p1 (inclusive) that may allow arbitrary
> code execution with root privileges.
> 
> Successful exploitation has been demonstrated on 32-bit Linux/glibc
> systems with ASLR. Under lab conditions, the attack requires on
> average 6-8 hours of continuous connections up to the maximum the
> server will accept. Exploitation on 64-bit systems is believed to be
> possible but has not been demonstrated at this time. It's likely that
> these attacks will be improved upon.
> 
> Exploitation on non-glibc systems is conceivable but has not been
> examined. Systems that lack ASLR or users of downstream Linux
> distributions that have modified OpenSSH to disable per-connection
> ASLR re-randomisation (yes - this is a thing, no - we don't
> understand why) may potentially have an easier path to exploitation.
> OpenBSD is not vulnerable.
> 
> We thank the Qualys Security Advisory Team for discovering, reporting
> and demonstrating exploitability of this problem, and for providing
> detailed feedback on additional mitigation measures.
> 
> 2) Logic error in ssh(1) ObscureKeystrokeTiming
> 
> In OpenSSH version 9.5 through 9.7 (inclusive), when connected to an
> OpenSSH server version 9.5 or later, a logic error in the ssh(1)
> ObscureKeystrokeTiming feature (on by default) rendered this feature
> ineffective - a passive observer could still detect which network
> packets contained real keystrokes when the countermeasure was active
> because both fake and real keystroke packets were being sent
> unconditionally.
> 
> This bug was found by Philippos Giavridis and also independently by
> Jacky Wei En Kung, Daniel Hugenroth and Alastair Beresford of the
> University of Cambridge Computer Lab.
> 
> Worse, the unconditional sending of both fake and real keystroke
> packets broke another long-standing timing attack mitigation. Since
> OpenSSH 2.9.9 sshd(8) has sent fake keystoke echo packets for
> traffic received on TTYs in echo-off mode, such as when entering a
> password into su(8) or sudo(8). This bug rendered these fake
> keystroke echoes ineffective and could allow a passive observer of
> a SSH session to once again detect when echo was off and obtain
> fairly limited timing information about keystrokes in this situation
> (20ms granularity by default).
> 
> This additional implication of the bug was identified by Jacky Wei
> En Kung, Daniel Hugenroth and Alastair Beresford and we thank them
> for their detailed analysis.
> 
> This bug does not affect connections when ObscureKeystrokeTiming
> was disabled or sessions where no TTY was requested.
> 
> Future deprecation notice
> =========================
> 
> OpenSSH plans to remove support for the DSA signature algorithm in
> early 2025. This release disables DSA by default at compile time.
> 
> DSA, as specified in the SSHv2 protocol, is inherently weak - being
> limited to a 160 bit private key and use of the SHA1 digest. Its
> estimated security level is only 80 bits symmetric equivalent.
> 
> OpenSSH has disabled DSA keys by default since 2015 but has retained
> run-time optional support for them. DSA was the only mandatory-to-
> implement algorithm in the SSHv2 RFCs, mostly because alternative
> algorithms were encumbered by patents when the SSHv2 protocol was
> specified.
> 
> This has not been the case for decades at this point and better
> algorithms are well supported by all actively-maintained SSH
> implementations. We do not consider the costs of maintaining DSA
> in OpenSSH to be justified and hope that removing it from OpenSSH
> can accelerate its wider deprecation in supporting cryptography
> libraries.
> 
> This release, and its deactivation of DSA by default at compile-time,
> marks the second step in our timeline to finally deprecate DSA. The
> final step of removing DSA support entirely is planned for the first
> OpenSSH release of 2025.
> 
> DSA support may be re-enabled in OpenBSD by setting "DSAKEY=yes"
> in Makefile.inc. To enable DSA support in portable OpenSSH, pass
> the "--enable-dsa-keys" option to configure.
> 
> Potentially-incompatible changes
> --------------------------------
> 
>  * all: as mentioned above, the DSA signature algorithm is now
>    disabled at compile time.
> 
>  * sshd(8): the server will now block client addresses that
>    repeatedly fail authentication, repeatedly connect without ever
>    completing authentication or that crash the server. See the
>    discussion of PerSourcePenalties below for more information.
>    Operators of servers that accept connections from many users, or
>    servers that accept connections from addresses behind NAT or
>    proxies may need to consider these settings.
> 
>  * sshd(8): the server has been split into a listener binary,
> sshd(8),
>    and a per-session binary "sshd-session". This allows for a much
>    smaller listener binary, as it no longer needs to support the SSH
>    protocol. As part of this work, support for disabling privilege
>    separation (which previously required code changes to disable) and
>    disabling re-execution of sshd(8) has been removed. Further
>    separation of sshd-session into additional, minimal binaries is
>    planned for the future.
> 
>  * sshd(8): several log messages have changed. In particular, some
>    log messages will be tagged with as originating from a process
>    named "sshd-session" rather than "sshd".
> 
>  * ssh-keyscan(1): this tool previously emitted comment lines
>    containing the hostname and SSH protocol banner to standard error.
>    This release now emits them to standard output, but adds a new
>    "-q" flag to silence them altogether.
> 
>  * sshd(8): (portable OpenSSH only) sshd will no longer use argv[0]
>    as the PAM service name. A new "PAMServiceName" sshd_config(5)
>    directive allows selecting the service name at runtime. This
>    defaults to "sshd". bz2101
> 
>  * (portable OpenSSH only) Automatically-generated files, such as
>    configure, config.h.in, etc will now be checked in to the portable
>    OpenSSH git release branch (e.g. V_9_8). This should ensure that
>    the contents of the signed release branch exactly match the
>    contents of the signed release tarball.
> 
> Changes since OpenSSH 9.7
> =========================
> 
> This release contains mostly bugfixes.
> 
> New features
> ------------
> 
>  * sshd(8): as described above, sshd(8) will now penalise client
>    addresses that, for various reasons, do not successfully complete
>    authentication. This feature is controlled by a new sshd_config(5)
>    PerSourcePenalties option and is on by default.
> 
>    sshd(8) will now identify situations where the session did not
>    authenticate as expected. These conditions include when the client
>    repeatedly attempted authentication unsucessfully (possibly
>    indicating an attack against one or more accounts, e.g. password
>    guessing), or when client behaviour caused sshd to crash (possibly
>    indicating attempts to exploit bugs in sshd).
> 
>    When such a condition is observed, sshd will record a penalty of
>    some duration (e.g. 30 seconds) against the client's address. If
>    this time is above a minimum configurable threshold, then all
>    connections from the client address will be refused (along with
> any
>    others in the same PerSourceNetBlockSize CIDR range) until the
>    penalty expire.
> 
>    Repeated offenses by the same client address will accrue greater
>    penalties, up to a configurable maximum. Address ranges may be
>    fully exempted from penalties, e.g. to guarantee access from a set
>    of trusted management addresses, using the new sshd_config(5)
>    PerSourcePenaltyExemptList option.
> 
>    We hope these options will make it significantly more difficult
> for
>    attackers to find accounts with weak/guessable passwords or
> exploit
>    bugs in sshd(8) itself. This option is enabled by default.
> 
>  * ssh(8): allow the HostkeyAlgorithms directive to disable the
>    implicit fallback from certificate host key to plain host keys.
> 
> Bugfixes
> --------
> 
>  * misc: fix a number of inaccuracies in the PROTOCOL.*
>    documentation files. GHPR430 GHPR487
> 
>  * all: switch to strtonum(3) for more robust integer parsing in most
>    places.
> 
>  * ssh(1), sshd(8): correctly restore sigprocmask around ppoll()
> 
>  * ssh-keysign(8): stricter validation of messaging socket fd GHPR492
> 
>  * sftp(1): flush stdout after writing "sftp>" prompt when not using
>    editline. GHPR480
> 
>  * sftp-server(8): fix home-directory extension implementation, it
>    previously always returned the current user's home directory
>    contrary to the spec. GHPR477
> 
>  * ssh-keyscan(1): do not close stdin to prevent error messages when
>    stdin is read multiple times. E.g.
>    echo localhost | ssh-keyscan -f - -f -
> 
>  * regression tests: fix rekey test that was testing the same KEX
>    algorithm repeatedly instead of testing all of them. bz3692
> 
>  * ssh_config(5), sshd_config(5): clarify the KEXAlgorithms directive
>    documentation, especially around what is supported vs available.
>    bz3701.
> 
> Portability
> -----------
> 
>  * sshd(8): expose SSH_AUTH_INFO_0 always to PAM auth modules
>    unconditionally. The previous behaviour was to expose it only when
>    particular authentication methods were in use.
> 
>  * build: fix OpenSSL ED25519 support detection. An incorrect
> function
>    signature in configure.ac previously prevented enabling the
> recently
>    added support for ED25519 private keys in PEM PKCS8 format.
> 
>  * ssh(1), ssh-agent(8): allow the presence of the WAYLAND_DISPLAY
>    environment variable to enable SSH_ASKPASS, similarly to the X11
>    DISPLAY environment variable. GHPR479
> 
>  * build: improve detection of the -fzero-call-used-regs compiler
>    flag. bz3673.
> 
>  * build: relax OpenSSL version check to accept all OpenSSL 3.x
>    versions.
> 
>  * sshd(8): add support for notifying systemd on server listen and
>    reload, using a standalone implementation that doesn't depend on
>    libsystemd. bz2641
> 
> Signed-off-by: Jose Quaresma <jose.quaresma@foundries.io>
> ---
> 
> v2:
>  - fix musl build
>  - fix sshd-session packing on openssh-sshd
>  - rebase on top of the CVE-2024-6387 fix sent

Thanks for the fixes. Unfortunately I think there is still one issue
remaining as the openssh ptests appear to be hanging on both arm and
x86:

https://autobuilder.yoctoproject.org/typhoon/#/builders/82/builds/6600
https://autobuilder.yoctoproject.org/typhoon/#/builders/81/builds/6778

On a previous build run I ended up stopping them after 24h+ but I
wasn't sure if that was related to other issues with the update or not.
It now looks like it is a separate issue :(

Cheers,

Richard