OpenSSH 8.3 released (and ssh-rsa deprecation notice)
([Security] May 27, 2020 15:38 UTC (Wed) (corbet))
- Reference: 0000821544
- News link: https://lwn.net/Articles/821544
- Source link:
The OpenSSH 8.3 release is out. This primarily a bug-fix release with a handful of minor new features. It does, however, carry a prominent notice that ssh-rsa signature algorithm will be disabled in " a near-future release ". The announcement includes information on how to determine whether hosts you care about are affected.
From :
Damien Miller <djm-AT-openbsd.org>
To :
oss-security-AT-lists.openwall.com
Subject :
[oss-security] Announce: OpenSSH 8.3 released
Date :
Wed, 27 May 2020 00:12:59 -0600 (MDT)
Message-ID :
<9739d4a6516e888a@openbsd.org>
Archive-link :
[1]Article
OpenSSH 8.3 has just been released. It will be available from the
mirrors listed at https://www.openssh.com/ shortly.
OpenSSH is a 100% complete SSH protocol 2.0 implementation and
includes sftp client and server support.
Once again, we would like to thank the OpenSSH community for their
continued support of the project, especially those who contributed
code or patches, reported bugs, tested snapshots or donated to the
project. More information on donations may be found at:
https://www.openssh.com/donations.html
Future deprecation notice
=========================
It is now possible[1] to perform chosen-prefix attacks against the
SHA-1 algorithm for less than USD$50K. For this reason, we will be
disabling the "ssh-rsa" public key signature algorithm by default in a
near-future release.
This algorithm is unfortunately still used widely despite the
existence of better alternatives, being the only remaining public key
signature algorithm specified by the original SSH RFCs.
The better alternatives include:
* The RFC8332 RSA SHA-2 signature algorithms rsa-sha2-256/512. These
algorithms have the advantage of using the same key type as
"ssh-rsa" but use the safe SHA-2 hash algorithms. These have been
supported since OpenSSH 7.2 and are already used by default if the
client and server support them.
* The ssh-ed25519 signature algorithm. It has been supported in
OpenSSH since release 6.5.
* The RFC5656 ECDSA algorithms: ecdsa-sha2-nistp256/384/521. These
have been supported by OpenSSH since release 5.7.
To check whether a server is using the weak ssh-rsa public key
algorithm, for host authentication, try to connect to it after
removing the ssh-rsa algorithm from ssh(1)'s allowed list:
ssh -oHostKeyAlgorithms=-ssh-rsa user@host
If the host key verification fails and no other supported host key
types are available, the server software on that host should be
upgraded.
A future release of OpenSSH will enable UpdateHostKeys by default
to allow the client to automatically migrate to better algorithms.
Users may consider enabling this option manually. Vendors of devices
that implement the SSH protocol should ensure that they support the
new signature algorithms for RSA keys.
[1] "SHA-1 is a Shambles: First Chosen-Prefix Collision on SHA-1 and
Application to the PGP Web of Trust" Leurent, G and Peyrin, T
(2020) https://eprint.iacr.org/2020/014.pdf
Security
========
* scp(1): when receiving files, scp(1) could be become desynchronised
if a utimes(2) system call failed. This could allow file contents
to be interpreted as file metadata and thereby permit an adversary
to craft a file system that, when copied with scp(1) in a
configuration that caused utimes(2) to fail (e.g. under a SELinux
policy or syscall sandbox), transferred different file names and
contents to the actual file system layout.
Exploitation of this is not likely as utimes(2) does not fail under
normal circumstances. Successful exploitation is not silent - the
output of scp(1) would show transfer errors followed by the actual
file(s) that were received.
Finally, filenames returned from the peer are (since openssh-8.0)
matched against the user's requested destination, thereby
disallowing a successful exploit from writing files outside the
user's selected target glob (or directory, in the case of a
recursive transfer). This ensures that this attack can achieve no
more than a hostile peer is already able to achieve within the scp
protocol.
Potentially-incompatible changes
================================
This release includes a number of changes that may affect existing
configurations:
* sftp(1): reject an argument of "-1" in the same way as ssh(1) and
scp(1) do instead of accepting and silently ignoring it.
Changes since OpenSSH 8.2
=========================
The focus of this release is bug fixing.
New Features
------------
* sshd(8): make IgnoreRhosts a tri-state option: "yes" to ignore
rhosts/shosts, "no" allow rhosts/shosts or (new) "shosts-only"
to allow .shosts files but not .rhosts.
* sshd(8): allow the IgnoreRhosts directive to appear anywhere in a
sshd_config, not just before any Match blocks; bz3148
* ssh(1): add %TOKEN percent expansion for the LocalFoward and
RemoteForward keywords when used for Unix domain socket forwarding.
bz#3014
* all: allow loading public keys from the unencrypted envelope of a
private key file if no corresponding public key file is present.
* ssh(1), sshd(8): prefer to use chacha20 from libcrypto where
possible instead of the (slower) portable C implementation included
in OpenSSH.
* ssh-keygen(1): add ability to dump the contents of a binary key
revocation list via "ssh-keygen -lQf /path" bz#3132
Bugfixes
--------
* ssh(1): fix IdentitiesOnly=yes to also apply to keys loaded from
a PKCS11Provider; bz#3141
* ssh-keygen(1): avoid NULL dereference when trying to convert an
invalid RFC4716 private key.
* scp(1): when performing remote-to-remote copies using "scp -3",
start the second ssh(1) channel with BatchMode=yes enabled to
avoid confusing and non-deterministic ordering of prompts.
* ssh(1), ssh-keygen(1): when signing a challenge using a FIDO token,
perform hashing of the message to be signed in the middleware layer
rather than in OpenSSH code. This permits the use of security key
middlewares that perform the hashing implicitly, such as Windows
Hello.
* ssh(1): fix incorrect error message for "too many known hosts
files." bz#3149
* ssh(1): make failures when establishing "Tunnel" forwarding
terminate the connection when ExitOnForwardFailure is enabled;
bz#3116
* ssh-keygen(1): fix printing of fingerprints on private keys and add
a regression test for same.
* sshd(8): document order of checking AuthorizedKeysFile (first) and
AuthorizedKeysCommand (subsequently, if the file doesn't match);
bz#3134
* sshd(8): document that /etc/hosts.equiv and /etc/shosts.equiv are
not considered for HostbasedAuthentication when the target user is
root; bz#3148
* ssh(1), ssh-keygen(1): fix NULL dereference in private certificate
key parsing (oss-fuzz #20074).
* ssh(1), sshd(8): more consistency between sets of %TOKENS are
accepted in various configuration options.
* ssh(1), ssh-keygen(1): improve error messages for some common
PKCS#11 C_Login failure cases; bz#3130
* ssh(1), sshd(8): make error messages for problems during SSH banner
exchange consistent with other SSH transport-layer error messages
and ensure they include the relevant IP addresses bz#3129
* various: fix a number of spelling errors in comments and debug/error
messages
* ssh-keygen(1), ssh-add(1): when downloading FIDO2 resident keys
from a token, don't prompt for a PIN until the token has told us
that it needs one. Avoids double-prompting on devices that
implement on-device authentication.
* sshd(8), ssh-keygen(1): no-touch-required FIDO certificate option
should be an extension, not a critical option.
* ssh(1), ssh-keygen(1), ssh-add(1): offer a better error message
when trying to use a FIDO key function and SecurityKeyProvider is
empty.
* ssh-add(1), ssh-agent(8): ensure that a key lifetime fits within
the values allowed by the wire format (u32). Prevents integer
wraparound of the timeout values. bz#3119
* ssh(1): detect and prevent trivial configuration loops when using
ProxyJump. bz#3057.
Portability
-----------
* Detect systems where signals flagged with SA_RESTART will interrupt
select(2). POSIX permits implementations to choose whether
select(2) will return when interrupted with a SA_RESTART-flagged
signal, but OpenSSH requires interrupting behaviour.
* Several compilation fixes for HP/UX and AIX.
* On platforms that do not support setting process-wide routing
domains (all excepting OpenBSD at present), fail to accept a
configuration attempts to set one at process start time rather than
fatally erroring at run time. bz#3126
* Improve detection of egrep (used in regression tests) on platforms
that offer a poor default one (e.g. Solaris).
* A number of shell portability fixes for the regression tests.
* Fix theoretical infinite loop in the glob(3) replacement
implementation.
* Fix seccomp sandbox compilation problems for some Linux
configurations bz#3085
* Improved detection of libfido2 and some compilation fixes for some
configurations when --with-security-key-builtin is selected.
Checksums:
==========
- SHA1 (openssh-8.3.tar.gz) = 46c63b7ddbe46a0666222f7988c993866c31fcca
- SHA256 (openssh-8.3.tar.gz) = M6CnZ+duGs4bzDio8hQNLwyLQChV+3wkUEO8HWLV35c=
- SHA1 (/openssh-8.3p1.tar.gz) = 04c7adb9986f16746588db8988b910530c589819
- SHA256 (openssh-8.3p1.tar.gz) = 8r774Ecv5+t10jNA6xdTHLazqsJAdeIGa0H4FOEjh7I=
Please note that the SHA256 signatures are base64 encoded and not
hexadecimal (which is the default for most checksum tools). The PGP
key used to sign the releases is available as RELEASE_KEY.asc from
the mirror sites.
Reporting Bugs:
===============
- Please read https://www.openssh.com/report.html
Security bugs should be reported directly to openssh@openssh.com
[1] https://lwn.net/ml/oss-security/9739d4a6516e888a@openbsd.org/
From :
Damien Miller <djm-AT-openbsd.org>
To :
oss-security-AT-lists.openwall.com
Subject :
[oss-security] Announce: OpenSSH 8.3 released
Date :
Wed, 27 May 2020 00:12:59 -0600 (MDT)
Message-ID :
<9739d4a6516e888a@openbsd.org>
Archive-link :
[1]Article
OpenSSH 8.3 has just been released. It will be available from the
mirrors listed at https://www.openssh.com/ shortly.
OpenSSH is a 100% complete SSH protocol 2.0 implementation and
includes sftp client and server support.
Once again, we would like to thank the OpenSSH community for their
continued support of the project, especially those who contributed
code or patches, reported bugs, tested snapshots or donated to the
project. More information on donations may be found at:
https://www.openssh.com/donations.html
Future deprecation notice
=========================
It is now possible[1] to perform chosen-prefix attacks against the
SHA-1 algorithm for less than USD$50K. For this reason, we will be
disabling the "ssh-rsa" public key signature algorithm by default in a
near-future release.
This algorithm is unfortunately still used widely despite the
existence of better alternatives, being the only remaining public key
signature algorithm specified by the original SSH RFCs.
The better alternatives include:
* The RFC8332 RSA SHA-2 signature algorithms rsa-sha2-256/512. These
algorithms have the advantage of using the same key type as
"ssh-rsa" but use the safe SHA-2 hash algorithms. These have been
supported since OpenSSH 7.2 and are already used by default if the
client and server support them.
* The ssh-ed25519 signature algorithm. It has been supported in
OpenSSH since release 6.5.
* The RFC5656 ECDSA algorithms: ecdsa-sha2-nistp256/384/521. These
have been supported by OpenSSH since release 5.7.
To check whether a server is using the weak ssh-rsa public key
algorithm, for host authentication, try to connect to it after
removing the ssh-rsa algorithm from ssh(1)'s allowed list:
ssh -oHostKeyAlgorithms=-ssh-rsa user@host
If the host key verification fails and no other supported host key
types are available, the server software on that host should be
upgraded.
A future release of OpenSSH will enable UpdateHostKeys by default
to allow the client to automatically migrate to better algorithms.
Users may consider enabling this option manually. Vendors of devices
that implement the SSH protocol should ensure that they support the
new signature algorithms for RSA keys.
[1] "SHA-1 is a Shambles: First Chosen-Prefix Collision on SHA-1 and
Application to the PGP Web of Trust" Leurent, G and Peyrin, T
(2020) https://eprint.iacr.org/2020/014.pdf
Security
========
* scp(1): when receiving files, scp(1) could be become desynchronised
if a utimes(2) system call failed. This could allow file contents
to be interpreted as file metadata and thereby permit an adversary
to craft a file system that, when copied with scp(1) in a
configuration that caused utimes(2) to fail (e.g. under a SELinux
policy or syscall sandbox), transferred different file names and
contents to the actual file system layout.
Exploitation of this is not likely as utimes(2) does not fail under
normal circumstances. Successful exploitation is not silent - the
output of scp(1) would show transfer errors followed by the actual
file(s) that were received.
Finally, filenames returned from the peer are (since openssh-8.0)
matched against the user's requested destination, thereby
disallowing a successful exploit from writing files outside the
user's selected target glob (or directory, in the case of a
recursive transfer). This ensures that this attack can achieve no
more than a hostile peer is already able to achieve within the scp
protocol.
Potentially-incompatible changes
================================
This release includes a number of changes that may affect existing
configurations:
* sftp(1): reject an argument of "-1" in the same way as ssh(1) and
scp(1) do instead of accepting and silently ignoring it.
Changes since OpenSSH 8.2
=========================
The focus of this release is bug fixing.
New Features
------------
* sshd(8): make IgnoreRhosts a tri-state option: "yes" to ignore
rhosts/shosts, "no" allow rhosts/shosts or (new) "shosts-only"
to allow .shosts files but not .rhosts.
* sshd(8): allow the IgnoreRhosts directive to appear anywhere in a
sshd_config, not just before any Match blocks; bz3148
* ssh(1): add %TOKEN percent expansion for the LocalFoward and
RemoteForward keywords when used for Unix domain socket forwarding.
bz#3014
* all: allow loading public keys from the unencrypted envelope of a
private key file if no corresponding public key file is present.
* ssh(1), sshd(8): prefer to use chacha20 from libcrypto where
possible instead of the (slower) portable C implementation included
in OpenSSH.
* ssh-keygen(1): add ability to dump the contents of a binary key
revocation list via "ssh-keygen -lQf /path" bz#3132
Bugfixes
--------
* ssh(1): fix IdentitiesOnly=yes to also apply to keys loaded from
a PKCS11Provider; bz#3141
* ssh-keygen(1): avoid NULL dereference when trying to convert an
invalid RFC4716 private key.
* scp(1): when performing remote-to-remote copies using "scp -3",
start the second ssh(1) channel with BatchMode=yes enabled to
avoid confusing and non-deterministic ordering of prompts.
* ssh(1), ssh-keygen(1): when signing a challenge using a FIDO token,
perform hashing of the message to be signed in the middleware layer
rather than in OpenSSH code. This permits the use of security key
middlewares that perform the hashing implicitly, such as Windows
Hello.
* ssh(1): fix incorrect error message for "too many known hosts
files." bz#3149
* ssh(1): make failures when establishing "Tunnel" forwarding
terminate the connection when ExitOnForwardFailure is enabled;
bz#3116
* ssh-keygen(1): fix printing of fingerprints on private keys and add
a regression test for same.
* sshd(8): document order of checking AuthorizedKeysFile (first) and
AuthorizedKeysCommand (subsequently, if the file doesn't match);
bz#3134
* sshd(8): document that /etc/hosts.equiv and /etc/shosts.equiv are
not considered for HostbasedAuthentication when the target user is
root; bz#3148
* ssh(1), ssh-keygen(1): fix NULL dereference in private certificate
key parsing (oss-fuzz #20074).
* ssh(1), sshd(8): more consistency between sets of %TOKENS are
accepted in various configuration options.
* ssh(1), ssh-keygen(1): improve error messages for some common
PKCS#11 C_Login failure cases; bz#3130
* ssh(1), sshd(8): make error messages for problems during SSH banner
exchange consistent with other SSH transport-layer error messages
and ensure they include the relevant IP addresses bz#3129
* various: fix a number of spelling errors in comments and debug/error
messages
* ssh-keygen(1), ssh-add(1): when downloading FIDO2 resident keys
from a token, don't prompt for a PIN until the token has told us
that it needs one. Avoids double-prompting on devices that
implement on-device authentication.
* sshd(8), ssh-keygen(1): no-touch-required FIDO certificate option
should be an extension, not a critical option.
* ssh(1), ssh-keygen(1), ssh-add(1): offer a better error message
when trying to use a FIDO key function and SecurityKeyProvider is
empty.
* ssh-add(1), ssh-agent(8): ensure that a key lifetime fits within
the values allowed by the wire format (u32). Prevents integer
wraparound of the timeout values. bz#3119
* ssh(1): detect and prevent trivial configuration loops when using
ProxyJump. bz#3057.
Portability
-----------
* Detect systems where signals flagged with SA_RESTART will interrupt
select(2). POSIX permits implementations to choose whether
select(2) will return when interrupted with a SA_RESTART-flagged
signal, but OpenSSH requires interrupting behaviour.
* Several compilation fixes for HP/UX and AIX.
* On platforms that do not support setting process-wide routing
domains (all excepting OpenBSD at present), fail to accept a
configuration attempts to set one at process start time rather than
fatally erroring at run time. bz#3126
* Improve detection of egrep (used in regression tests) on platforms
that offer a poor default one (e.g. Solaris).
* A number of shell portability fixes for the regression tests.
* Fix theoretical infinite loop in the glob(3) replacement
implementation.
* Fix seccomp sandbox compilation problems for some Linux
configurations bz#3085
* Improved detection of libfido2 and some compilation fixes for some
configurations when --with-security-key-builtin is selected.
Checksums:
==========
- SHA1 (openssh-8.3.tar.gz) = 46c63b7ddbe46a0666222f7988c993866c31fcca
- SHA256 (openssh-8.3.tar.gz) = M6CnZ+duGs4bzDio8hQNLwyLQChV+3wkUEO8HWLV35c=
- SHA1 (/openssh-8.3p1.tar.gz) = 04c7adb9986f16746588db8988b910530c589819
- SHA256 (openssh-8.3p1.tar.gz) = 8r774Ecv5+t10jNA6xdTHLazqsJAdeIGa0H4FOEjh7I=
Please note that the SHA256 signatures are base64 encoded and not
hexadecimal (which is the default for most checksum tools). The PGP
key used to sign the releases is available as RELEASE_KEY.asc from
the mirror sites.
Reporting Bugs:
===============
- Please read https://www.openssh.com/report.html
Security bugs should be reported directly to openssh@openssh.com
[1] https://lwn.net/ml/oss-security/9739d4a6516e888a@openbsd.org/
Serious questions: how worried should I be?
What algorithms do other SSH implementations support? (Has anybody here tried turning off ssh-rsa and have you gotten any complaints from users?)
I've been very leery of using ECDSA, since as I understand it, a less-than-perfect random number generator allows an attacker to determine your secret key. For SSH host/client public keys this seems especially dangerous, since there's effectively no key rotation. Am I being overly paranoid about this?
What specifically is the danger of chosen-prefix attacks on SHA-1 - how would those be used to attack SSH host or client authentication?
Serious questions: how worried should I be?
What algorithms do other SSH implementations support? (Has anybody here tried turning off ssh-rsa and have you gotten any complaints from users?)
I've been very leery of using ECDSA, since as I understand it, a less-than-perfect random number generator allows an attacker to determine your secret key. For SSH host/client public keys this seems especially dangerous, since there's effectively no key rotation. Am I being overly paranoid about this?
What specifically is the danger of chosen-prefix attacks on SHA-1 - how would those be used to attack SSH host or client authentication?
Serious questions: how worried should I be?
On a server, disabling ssh-rsa seems quite safe; users can easily generate a new key. It generates some minor grumbling, but with a well-communicated rationale, users can handle it.
On the flip side, though: I regularly run into servers that only support ssh-rsa keys, such as git repository hosting sites that have web interfaces to provide public keys. Many of those sites aren't using OpenSSH's sshd, they're using library implementations of ssh and not supporting full shells. Perhaps this deprecation will help motivate such servers to move forward.
Personally, I think ssh-ed25519 seems like a reasonable path forward.
Serious questions: how worried should I be?
On a server, disabling ssh-rsa seems quite safe; users can easily generate a new key. It generates some minor grumbling, but with a well-communicated rationale, users can handle it.
On the flip side, though: I regularly run into servers that only support ssh-rsa keys, such as git repository hosting sites that have web interfaces to provide public keys. Many of those sites aren't using OpenSSH's sshd, they're using library implementations of ssh and not supporting full shells. Perhaps this deprecation will help motivate such servers to move forward.
Personally, I think ssh-ed25519 seems like a reasonable path forward.
Serious questions: how worried should I be?
Note that even if both client and service have RSA keys whose *public key format* starts with "ssh-rsa " (in ~/.ssh/id_rsa.pub and /etc/ssh/ssh_host_rsa_key.pub respectively), they are not necessarily using the ssh-rsa *host key algorithm*.
Assuming both the client and the server are a vaguely recent version of OpenSSH, or something else that implements newer algorithms, the same public keys can be used with the rsa-sha2-512 and rsa-sha2-256 host key algorithms, which are believed to be OK. The announcement does say this ('These algorithms have the advantage of using the same key type as "ssh-rsa"'), but perhaps that isn't making it clear enough.
For example:
$ ssh -vvv -oHostKeyAlgorithms=rsa-sha2-512,rsa-sha2-256 people.debian.org
...
debug2: host key algorithms: rsa-sha2-512,rsa-sha2-256
...
debug1: kex: algorithm: curve25519-sha256
debug1: kex: host key algorithm: rsa-sha2-512
...
debug1: Server host key: ssh-rsa SHA256:[redacted1]
...
debug1: Host 'people.debian.org' is known and matches the RSA host key.
...
debug1: Will attempt key: /home/[redacted].pub RSA SHA256:[redacted2] explicit agent
...
debug3: sign_and_send_pubkey: RSA SHA256:[redacted2]
debug3: sign_and_send_pubkey: signing using rsa-sha2-512 SHA256:[redacted2]
(My only public key that people.debian.org is configured to accept is RSA, and for this example I told my ssh client not to try validating people.debian.org's non-RSA host keys.)
I think the biggest problem here is going to be with non-OpenSSH client and server implementations, which might only implement the mandatory-to-implement ssh-rsa algorithm if they are a minimum-viable-product sort of implementation.
Serious questions: how worried should I be?
Note that even if both client and service have RSA keys whose *public key format* starts with "ssh-rsa " (in ~/.ssh/id_rsa.pub and /etc/ssh/ssh_host_rsa_key.pub respectively), they are not necessarily using the ssh-rsa *host key algorithm*.
Assuming both the client and the server are a vaguely recent version of OpenSSH, or something else that implements newer algorithms, the same public keys can be used with the rsa-sha2-512 and rsa-sha2-256 host key algorithms, which are believed to be OK. The announcement does say this ('These algorithms have the advantage of using the same key type as "ssh-rsa"'), but perhaps that isn't making it clear enough.
For example:
$ ssh -vvv -oHostKeyAlgorithms=rsa-sha2-512,rsa-sha2-256 people.debian.org
...
debug2: host key algorithms: rsa-sha2-512,rsa-sha2-256
...
debug1: kex: algorithm: curve25519-sha256
debug1: kex: host key algorithm: rsa-sha2-512
...
debug1: Server host key: ssh-rsa SHA256:[redacted1]
...
debug1: Host 'people.debian.org' is known and matches the RSA host key.
...
debug1: Will attempt key: /home/[redacted].pub RSA SHA256:[redacted2] explicit agent
...
debug3: sign_and_send_pubkey: RSA SHA256:[redacted2]
debug3: sign_and_send_pubkey: signing using rsa-sha2-512 SHA256:[redacted2]
(My only public key that people.debian.org is configured to accept is RSA, and for this example I told my ssh client not to try validating people.debian.org's non-RSA host keys.)
I think the biggest problem here is going to be with non-OpenSSH client and server implementations, which might only implement the mandatory-to-implement ssh-rsa algorithm if they are a minimum-viable-product sort of implementation.
Serious questions: how worried should I be?
I have a few OpenWRT devices (which use dropbear), and I am affected by ssh-rsa deprecation, without any way to fix this. I cannot generate anything else than ssh-rsa on the router itself, because dropbearkey, as built by OpenWRT, supports only rsa keys. This is on a recent (less than 1 week) master snapshot.