Ilia Shipitsin [Mon, 8 Jul 2024 21:08:18 +0000 (23:08 +0200)]
src/openvpn/init.c: handle strdup failures
Signed-off-by: Ilia Shipitsin <chipitsine@gmail.com> Acked-by: Antonio Quartulli <a@unstable.cc> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240708210912.566-2-chipitsine@gmail.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28884.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Extend "--static-challenge" option to take a third
argument (= scrv1 or concat) to specify that the password and
response should be concatenated instead of using the
SCRV1 protocol. If unspecified, it defaults to "scrv1"
meaning that the SCRV1 protocol should be used.
v2: use scrv1|concat instead of 0|1 as option argument
fix typos
v3: improve and correct documentation in management-notes.txt
Change-Id: I59a90446bfe73d8856516025a58a6f62cc98ab0d Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240719131407.75746-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28943.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Add a test for loading certificate and key to ssl context
The test certificate used in test_ssl.c is updated to use 2048 bit
RSA and the matching key is added.
Tests include loading certificate and key as inlined pem as well as
from files. Note that loading the key also checks that it matches
the certificate, providing an indirect test that the latter was loaded
correctly.
Change-Id: Ic6f089896191145f68ce9a11023587d05dcec4d8 Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240906103814.36839-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg29074.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
OpenSSL 3 has providers which can load keys and certificates
from various key stores and HSMs using a provider-specific URI.
While certificates are generally exportable, and some providers
support a PEM file that acts as a proxy for non-exportable private
keys, not all providers are expected to do so. A generic capability
to read keys and certificates from URIs appears useful.
This patch does this by extending the scope of the argument for
"--key" and "--cert" options to include URIs. Many of OpenSSL 3
utilities also work the same way: e.g., the "-in" option for
"openssl pkey" or "openssl x509" could be a filename or URI.
Other applications have started emulating this behaviour:
e.g., pkcs11: URI works as an alternative to a file name for
certificates and keys in apache. Even for files, this has a nice
side effect that non-PEM files get transparently parsed. E.g., a
pkcs12 file could be used in place of a PEM file without needing
any extra options.
This is backward compatible as OpenSSL falls back to treating URIs
with no scheme or unrecognized scheme as file names.
Parsing of inlined keys and certificates is unchanged (those
should be in PEM format).
Specification of URIs that OpenSSL accepts depends on the
providers that support them. Some are standard URIs such as
"file:/path", but providers may support non-standard URIs
with arbitrary scheme names. OpenSSL by itself recognizes
only file URI. However, the implementation is agnostic to the
URI specification as parsing is done by the provider that supports
the URI. A new URI gets automatically recognized when the provider
that supports it is loaded.
Below are some usage examples:
Relative or absolute path to a file or as a URI "file:/absolute/path":
--key mykey.pem (same as what is currently supported)
--key file:/path/to/mykey.pem
--cert file:/path/to/mycert.pem
Other file types supported by OpenSSL would also work:
--key client.p12
--cert client.p12
pkcs11-provider supports "pkcs11:" URI (RFC 7512):
Protect cached username, password and token on client
Keep the memory segment containing username and password in
"struct user_pass" encrypted. Works only on Windows.
Username and auth-token cached by the server are not covered
here.
v2: Encrypt username and password separately as it looks more
robust. We continue to depend on the username and password buffer
sizes to be a multiple of CRYPTPROTECTMEMORY_BLOCK_SIZE = 16,
which is the case now. An error is logged if this is not the case.
v3: move up ASSERT in auth_token.c
Change-Id: I42e17e09a02f01aedadc2b03f9527967f6e1e8ff Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240906112908.1009-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg29079.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Usage of credentials is a bit odd in this file.
Actually the copy of "struct user_pass" kept in p->up is not
required at all. It just defeats the purpose of auth-nocahe
as it never gets cleared.
Removing it is beyond the scope of this patch -- we just ensure
it's purged after use.
Change-Id: Ic6d63a319d272a56ac0e278f1356bc5241b56a34 Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240905100724.4105-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg29061.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 14 Feb 2024 13:27:19 +0000 (14:27 +0100)]
Implement support for AEAD tag at the end
Using the AEAD tag at the end is the standard way of doing AEAD. Several
APIs even only support the tag at the end (e.g. mbed TLS). Having the tag at
the front or end makes no difference for security but allows streaming HW
implementations like NICs to be much more efficient as they do not need to
buffer a whole packet content and encrypt it to finally write the tag but
instead just add the calculated tag at the end of processing.
Change-Id: I00821d75342daf3f813b829812d648fe298bea81 Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240214132719.3031492-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28239.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Lev Stipakov [Fri, 9 Aug 2024 19:22:56 +0000 (21:22 +0200)]
Use a more robust way to get dco-win version
The current way doesn't work if the device is already in use.
Starting from 1.3.0, dco-win creates a non-exclusive
control device \\.\ovpn-dco-ver which can be opened by
multiple apps and supports a single IOCTL to get
a version number.
https://github.com/OpenVPN/ovpn-dco-win/pull/76
This will be expecially handy later when checking which
features driver supports.
Change-Id: Ieb6f3a9d14d76000c1caf8ee1e959c6d0de832bf Signed-off-by: Lev Stipakov <lev@openvpn.net> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240809192257.24208-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg29009.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Selva Nair [Mon, 12 Aug 2024 23:21:58 +0000 (19:21 -0400)]
test_pkcs11.c: set file offset to 0 after ftruncate
Currently key and cert file fd's are reused after ftruncate()
without setting the offset to zero. This causes subsequent
data to be written at some finite offset with the hole in
the file automatically filled by zeros. Fix it by calling
lseek() to set the offset to zero.
The test works nevertheless because p11tool seem to generously
ignore any junk before the "BEGIN" marker.
Change-Id: Ib0fe15a4ba18d89216b0288e6cd6be66ed377bd4 Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240812232158.3776869-1-selva.nair@gmail.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg29010.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Incompatible changes to the --dns server address and --dns server
exclude-domains options were introduced after the code for handling them
was released. Add and send a new IV_PROTO flag, so servers which act on
the flags set can differentiate between clients which have implemented
--dns and those which just support the new option. This enables them to
decide which variant of options to send to the client.
Change-Id: I975057c20c1457ef88111f8d142ca3fd2039d5ff Signed-off-by: Heiko Hund <heiko@ist.eigentlich.net> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240725112248.21075-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28970.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 10 Jul 2024 14:06:23 +0000 (16:06 +0200)]
Allow trailing \r and \n in control channel message
Writing a reason from a script will easily end up adding extra \r\n characters
at the end of the reason. Our current code pushes this to the peer. So be more
liberal in accepting these message.
Github: closes OpenVPN/openvpn#568
Change-Id: I47c992b6b73b1475cbff8a28f720cf50dc1fbe3e Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240710140623.172829-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28910.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
- exit after a timeout if unable to kill servers
- use sudo or equivalent only for server stop/start
- use /bin/sh directly instead of through /usr/bin/env
- simplify sudo call in the sample rc file
- remove misleading and outdated documentation
- make it work on OpenBSD 7.5
- make it work on NetBSD 10.0
- make server logs readable by normal users
Change-Id: I2cce8ad4e0d262e1404ab1eb6ff673d8590b6b3a Signed-off-by: Samuli Seppänen <samuli.seppanen@gmail.com> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240704133337.26595-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28871.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Max Fillinger [Wed, 3 Jul 2024 17:41:58 +0000 (19:41 +0200)]
mbedtls: Warn if --tls-version-min is too low
Recent versions of mbedtls only support TLS 1.2. When the minimum
version is set to TLS 1.0 or 1.1, log a warning and use 1.2 as the
actual minimum version.
Change-Id: Ibc641388d8016533c94dfef3618376f6dfa91f4e Signed-off-by: Max Fillinger <maximilian.fillinger@foxcrypto.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240703174158.7137-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28865.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
v2:
- simplify code by removing -llzo special handling
v3:
- reintroduce support for autodetection without pkg-config,
no need to break backwards compatibility right now
v7:
- Handle case correctly where lzo/lzo1x.h can not be included
at all. On most distros this works even though the .pc
file suggests to use it without. We had some partly
solution for that but it wasn't really working.
v8:
- Handle systems that do not implicitly include limits.h
in configure test builds.
lzodefs.h usually relies on lzoconf.h to include it.
Change-Id: I1c038dc4ec80d3499582d81eee61fee74f26e693 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240626161921.179301-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28848.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Caching proxy credentials was not working due to the
lack of handling already defined creds in get_user_pass(),
which prevented the caching from working properly.
Fix this issue by getting the value of c->first_time,
that indicates if we're at the first iteration
of the main loop and use it as second argument of the
get_user_pass_http(). Otherwise, on SIGUSR1 or SIGHUP
upon instance context restart credentials would be erased
every time.
The nocache member has been added to the struct
http_proxy_options and also a getter method to retrieve
that option from ssl has been added, by doing this
we're able to erase previous queried user credentials
to ensure correct operation.
Fixes: Trac #1187 Signed-off-by: Gianmarco De Gregori <gianmarco@mandelbit.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Change-Id: Ia3e06c0832c4ca0ab868c845279fb71c01a1a78a Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240623200551.20092-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28835.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Lev Stipakov [Wed, 19 Jun 2024 14:46:08 +0000 (17:46 +0300)]
interactive.c: Improve access control for gui<->service pipe
At the moment everyone but anonymous are permitted
to create a pipe with the same name as interactive service creates,
which makes it possible for malicious process with SeImpersonatePrivilege
impersonate as local user.
This hardens the security of the pipe, making it possible only for
processes running as SYSTEM (such as interactive service) create the
pipe with the same name.
While on it, replace EXPLICIT_ACCESS structures with SDDL string.
CVE: 2024-4877
Change-Id: I35e783b79a332d247606e05a39e41b4d35d39b5d
Reported by: Zeze with TeamT5 <zeze7w@gmail.com> Signed-off-by: Lev Stipakov <lev@openvpn.net> Acked-by: Selva Nair <selva.nair@gmail.com>
Message-Id: <20240619144629.1718-2-lev@openvpn.net>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28808.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Mon, 27 May 2024 13:02:41 +0000 (15:02 +0200)]
Properly handle null bytes and invalid characters in control messages
This makes OpenVPN more picky in accepting control message in two aspects:
- Characters are checked in the whole buffer and not until the first
NUL byte
- if the message contains invalid characters, we no longer continue
evaluating a fixed up version of the message but rather stop
processing it completely.
Previously it was possible to get invalid characters to end up in log
files or on a terminal.
This also prepares the logic a bit in the direction of having a proper
framing of control messages separated by null bytes instead of relying
on the TLS framing for that. All OpenVPN implementations write the 0
bytes between control commands.
This patch also include several improvement suggestion from Reynir
(thanks!).
CVE: 2024-5594
Reported-By: Reynir Björnsson <reynir@reynir.dk>
Change-Id: I0d926f910637dabc89bf5fa919dc6beef1eb46d9 Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <a@unstable.cc>
Message-Id: <20240619103004.56460-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28791.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
5andr0 [Fri, 15 Mar 2024 16:20:11 +0000 (17:20 +0100)]
Implement server_poll_timeout for socks
So far --server-poll-timeout was only applied
for HTTP proxies, apply it also to SOCKS proxies.
This removes the default 5 second socks connect timeout
which can be too small depending on network setup and
replaces it with the configurable overall connect timeout
(default 120 seconds).
Trac: #328
Github: fixes OpenVPN/openvpn#267
Change-Id: I2b109f8c551c23045a1be355778b08f0fd4d309f Signed-off-by: 5andr0 <sandro.trianni@gmail.com> Tested-By: ValdikSS <valdikss@gmail.com> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240315162011.1661139-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28408.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
This option is very old (from SVN days) and has been
used by Access Server for many years. I don't think it
makes sense to claim that it is "experimental" at this
point.
Change-Id: I913bb70c5e527e78e7cdb43110e23a8944f35a22 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240618120156.4836-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28772.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Max Fillinger [Tue, 18 Jun 2024 12:02:19 +0000 (14:02 +0200)]
mbedtls: Remove support for old TLS versions
Recent versions of mbedtls have dropped support for TLS 1.0 and 1.1.
Rather than checking which versions are supported, drop support for
everything before 1.2.
Change-Id: Ia3883a26ac26df6bbb5353fb074a2e0f814737be Signed-off-by: Max Fillinger <maximilian.fillinger@foxcrypto.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240618120219.5053-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28773.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Heiko Wundram [Thu, 6 Jun 2024 10:34:41 +0000 (12:34 +0200)]
Implement Windows CA template match for Crypto-API selector
The certificate selection process for the Crypto API certificates
is currently fixed to match on subject or identifier. Especially
if certificates that are used for OpenVPN are managed by a Windows CA,
it is appropriate to select the certificate to use by the template
that it is generated from, especially on domain-joined clients which
automatically acquire/renew the corresponding certificate.
The attached match implements the match on TMPL: with either a template
name (which is looked up through CryptFindOIDInfo) or by specifying the
OID of the template directly, which then is matched against the
corresponding X509 extensions specifying the template that the certificate
was generated from.
The logic requires to walk all certificates in the underlying store and
to match the certificate extensions directly. The hook which is
implemented in the certificate selection logic is generic to allow
other Crypto-API certificate matches to also be implemented at some
point in the future.
The logic to match the certificate template is taken from the
implementation in the .NET core runtime, see Pal.Windows/FindPal.cs in
in the implementation of System.Security.Cryptography.X509Certificates.
Heiko Hund [Wed, 5 Jun 2024 12:38:56 +0000 (14:38 +0200)]
Windows: enforce 'block-local' with WFP filters
In an attempt to better defend against the TunnelCrack attacks, enforce
that no traffic can pass to anything else than the VPN interface when
the 'block-local' flags is given with either --redirect-gateway or
--redirect-private.
Reuse much of the existing --block-outside-dns code, but make it more
general, so that it can also block any traffic, not just port 53.
Uses the Windows Filtering Platform for enforcement in addition to the
routes redirecting the networks into the tunnel.
Change-Id: Ic9bf797bfc7e2d471998a84cb0f071db3e4832ba Signed-off-by: Heiko Hund <heiko@ist.eigentlich.net> Acked-by: Lev Stipakov <lstipakov@gmail.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240605123856.26267-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28717.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Franco Fichtner [Tue, 28 May 2024 17:42:52 +0000 (17:42 +0000)]
Allow to set ifmode for existing DCO interfaces in FreeBSD
While prexisting devices work well TUN/TAP the DCO interfaces require
setting the ifmode which cannot be done by FreeBSD base tooling. In
peer-to-peer mode this is not a problem because that is the default mode.
Subnet mode, however, will fail to be set and the resulting connection does
not start:
Failed to create interface ovpns2 (SIOCSIFNAME): File exists (errno=17)
DCO device ovpns2 already exists, won't be destroyed at shutdown
/sbin/ifconfig ovpns2 10.1.8.1/24 mtu 1500 up
ifconfig: in_exec_nl(): Empty IFA_LOCAL/IFA_ADDRESS
ifconfig: ioctl (SIOCAIFADDR): Invalid argument
FreeBSD ifconfig failed: external program exited with error status: 1
Exiting due to fatal error
Slightly restructure the code to catch the specific error
condition and execute dco_set_ifmode() in this case as well.
Signed-off-by: Franco Fichtner <franco@opnsense.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <AE20A784-506C-488B-9302-2D3AE775B168@opnsense.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28688.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Mon, 8 Apr 2024 12:49:33 +0000 (14:49 +0200)]
Allow the TLS session to send out TLS alerts
Previous OpenVPN versions shut down the TLS control channel immediately
when encountering an error. This also meant that we would not send out
TLS alerts to notify a client about potential problems like mismatching
TLS versions or having no common cipher.
This commit adds a new key_state S_ERROR_PRE which still allows to
send out the remaining TLS packets of the control session which are
typically the alert message and then going to S_ERROR. We do not
wait for retries. So this is more a one-shot notify but that is
acceptable in this situation.
Sending out alerts is a slight compromise in security as alerts give
out a bit of information that otherwise is not given
out. But since all other consumers TLS implementations are already doing this
and TLS implementations (nowadays) are very careful not to leak (sensitive)
information by alerts and since the user experience is much better with
alerts, this compromise is worth it.
Change-Id: I0ad48915004ddee587e97c8ed190ba8ee989e48d Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240408124933.243991-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28540.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Reynir Björnsson [Thu, 16 May 2024 11:58:08 +0000 (13:58 +0200)]
Only schedule_exit() once
If an exit has already been scheduled we should not schedule it again.
Otherwise, the exit signal is never emitted if the peer reschedules the
exit before the timeout occurs.
schedule_exit() now only takes the context as argument. The signal is
hard coded to SIGTERM, and the interval is read directly from the
context options.
Furthermore, schedule_exit() now returns a bool signifying whether an
exit was scheduled; false if exit is already scheduled. The call sites
are updated accordingly. A notable difference is that management is only
notified *once* when an exit is scheduled - we no longer notify
management on redundant exit.
This patch was assigned a CVE number after already reviewed and ACKed,
because it was discovered that a misbehaving client can use the (now
fixed) server behaviour to avoid being disconnected by means of a
managment interface "client-kill" command - the security issue here is
"client can circumvent security policy set by management interface".
This only affects previously authenticated clients, and only management
client-kill, so normal renegotion / AUTH_FAIL ("your session ends") is not
affected.
CVE: 2024-28882
Change-Id: I9457f005f4ba970502e6b667d9dc4299a588d661 Signed-off-by: Reynir Björnsson <reynir@reynir.dk> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240516120434.23499-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28679.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 15 May 2024 10:01:15 +0000 (12:01 +0200)]
Remove custom TLS 1.0 PRF implementation only used by LibreSSL/wolfSSL
After the removal of the OpenSSL 1.0.2 support, LibreSSL/wolfSSL are the
only libraries that still needs the custom implementation.
Since our LibreSSL/wolfSSL support is always best effort, we can afford to
limit LibreSSL support in this way. If they want to support this, they
should expose the functionality as well.
Change-Id: I5bfa3630ad4dff2807705658bc877c4a429a39ce Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240515100115.11056-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28672.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Tue, 14 May 2024 14:15:50 +0000 (16:15 +0200)]
Remove OpenSSL 1.0.2 support
With Centos 7/Red Hat Enterprise Linux 7 being EOL this June, the last
distributions that still support OpenSSL 1.0.2 are finally EOL. This
means we no longer need to support OpenSSL 1.0.2
Change-Id: I90875311a4e4c403e77e30b609c1878cbaaaad45 Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240514141550.17544-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28665.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 8 May 2024 22:05:40 +0000 (00:05 +0200)]
Workaround issue in LibreSSL crashing when enumerating digests/ciphers
OpenBSD/LibreSSL reimplemented EVP_get_cipherbyname/EVP_get_digestbyname
and broke calling EVP_get_cipherbynid/EVP_get_digestbyname with an
invalid nid in the process so that it would segfault.
Workaround but doing that NULL check in OpenVPN instead of leaving it
to the library.
Github: see also https://github.com/libressl/openbsd/issues/150
Change-Id: Ia08a9697d0ff41721fb0acf17ccb4cfa23cb3934 Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240508220540.12554-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28649.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
If we copy this code, let's at least make sure we update
it every decade ;)
I also considered removing it. However, then autoconf
can't be run on systems without pkg-config installed
anymore. While that is very unusual, didn't see a good
reason to break that.
Change-Id: I34e96a225446693f401549d86d872c02427ef7d5 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240506160413.7189-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28631.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Mon, 6 May 2024 10:27:10 +0000 (12:27 +0200)]
Remove openvpn_snprintf and similar functions
Old Microsoft versions did strange behaviour but according to the
newly added unit test and
https://stackoverflow.com/questions/7706936/is-snprintf-always-null-terminating
this is now standard conforming and we can use the normal snprintf
method.
Microsoft own documentation to swprintf also says you nowadays need to
define _CRT_NON_CONFORMING_SWPRINTFS to get to non-standard behaviour.
Change-Id: I07096977e3b562bcb5d2c6f11673a4175b8e12ac Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240506102710.8976-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28617.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 2 May 2024 12:22:31 +0000 (14:22 +0200)]
Replace macos11 with macos14 in github runners
Github's documentation states: macos-11 label has been deprecated and
will no longer be available after 6/28/2024. Add macos14 which is nowadays
supported instead.
The github macos-14 runner is using the M1 platform with ARM, so this
requires a bit more adjustment of paths.
Change-Id: Ia70f230b2e9a78939d1875395205c8f48c4944b7 Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240502122231.672-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/search?l=mid&q=20240502122231.672-1-gert@greenie.muc.de Signed-off-by: Gert Doering <gert@greenie.muc.de>
Use topology default of "subnet" only for server mode
The setting of --topology changes the syntax of --ifconfig.
So changing the default of --topology breaks all existing
configs that use --ifconfig but not --topology.
For P2P setups that is probably a signification percentage.
For server setups the percentage is hopefully lower since
--ifconfig is implicitly set by --server. Also more people
might have set their topology explicitly since it makes a
much bigger difference. Clients will usually get the
topology and the IP config pushed by the server.
So we decided to not switch the default for everyone to
not affect P2P setups. What we care about is to change
the default for --mode server, so we only do that now. For
people using --server this should be transparent except
for a pool reset.
Github: Openvpn/openvpn#529
Change-Id: Iefd209c0856ef395ab74055496130de00b86ead0 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240501124254.29114-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28592.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
For some of the test we don't like the default log behavior
and there seems no easy way to change that except to fork
the driver. The license seems unproblematic since we're
GPL anyway.
v2:
- Do not use forked-test-driver for UTs. Default behavior
is fine for those.
Change-Id: I67d461afbcc9c06b1fc5ab4477141d7b8bd9ba8e Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240125110036.16070-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28132.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Tue, 2 Apr 2024 13:49:09 +0000 (15:49 +0200)]
Remove/combine redundant call of EVP_CipherInit before EVP_CipherInit_Ex
EVP_CipherInit basically is the same EVP_CipherInit_ex except that it
in some instances it resets/inits the ctx parameter first. We already
call EVP_CIPHER_CTX_reset to reset/init the ctx before. Also ensure that
EVP_CipherInit_Ex gets the cipher to actually be able to initialise the
context.
- Remove obsolete ifconfig_broadcast. Since this was
removed in 2.5.0, do not add a removal note but just
completely remove it.
- Add missing documentation of IPv6 variants for
ifconfig_pool_* variables.
Github: fixes Openvpn/openvpn#527
Change-Id: Ia8c8de6799f0291fc900628fbd06c8a414e741ca Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240321161623.2794161-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28438.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
phase2_tcp_server: fix Coverity issue 'Dereference after null check'
As Coverity says:
Either the check against null is unnecessary, or there may be a null
pointer dereference.
In phase2_tcp_server: Pointer is checked against null but then
dereferenced anyway
There is only one caller (link_socket_init_phase2) and it already has
an ASSERT(sig_info). So use that here was well.
v2:
- fix cleanly by actually asserting that sig_info is defined
Change-Id: I8ef199463d46303129a3f563fd9eace780a58b8a Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240325071448.12143-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28452.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Mon, 25 Mar 2024 12:50:52 +0000 (13:50 +0100)]
Use snprintf instead of sprintf for get_ssl_library_version
This is avoid a warning/error (when using -Werror) under current macOS
of sprintf:
__deprecated_msg("This function is provided for compatibility
reasons only. Due to security concerns inherent in the design
of sprintf(3), it is highly recommended that you use snprintf(3)
instead.")
Change-Id: I3c6fd36eb9daee9244d6dc6d9f22de1c5cf9d039 Signed-off-by: Arne Schwabe <arne-openvpn@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240325125052.14135-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28458.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Previously the sections "Encryption Options" and
"Data channel cipher negotiation" were on the same
level as "OPTIONS", which makes no sense. Instead
move them and their subsections one level down.
Use ` since that was already in use in section
"Virtual Routing and Forwarding".
Change-Id: Ib5a7f9a978bda5ad58830e43580232660401f66d Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240325071520.12513-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28453.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
- Remove compression settings. Not recommended anymore.
- Remove old cipher setting. Replaced by data-ciphers negotiation.
- Add comment how to set data-ciphers for very old clients.
- Remove/reword some old comments. e.g. no need to reference
OpenVPN 1.x anymore.
- Mention peer-fingerprint alternative.
- comment out "tls-auth" as that is not needed for a bare-bones VPN config
and needs additional setup.
Github: #511
Change-Id: I1a36651c0dea52259533ffc00bccb9b03bf82e26 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <20240325071320.11348-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28451.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Lev Stipakov [Mon, 18 Mar 2024 18:17:44 +0000 (19:17 +0100)]
Disable DCO if proxy is set via management
Commit
45a1cb2a ("Disable DCO if proxy is set via management")
attempted to disable DCO when proxy is set via management interface. However,
at least on Windows this doesn't work, since:
- setting tuntap_options->disable_dco to true is not enough to disable DCO
- at this point it is a bit too late, since we've already done DCO-specific
adjustments
Since proxy can be set via management only if --management-query-proxy is
specified, the better way is to add a check to dco_check_startup_option().
Github: fixes OpenVPN/openvpn#522
Change-Id: I16d6a9fefa317d7d4a195e786618328445bdbca8 Signed-off-by: Lev Stipakov <lev@openvpn.net> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240318181744.20625-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28415.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Lev Stipakov [Tue, 19 Mar 2024 15:27:11 +0000 (17:27 +0200)]
interactive.c: Fix potential stack overflow issue
When reading message from the pipe, we first peek the pipe to get the size
of the message waiting to be read and then read the message. A compromised
OpenVPN process could send an excessively large message, which would result
in a stack-allocated message buffer overflow.
To address this, we terminate the misbehaving process if the peeked message
size exceeds the maximum allowable size.
CVE: 2024-27459
Microsoft case number: 85932
Reported-by: Vladimir Tokarev <vtokarev@microsoft.com>
Change-Id: Ib5743cba0741ea11f9ee62c4978b2c6789b81ada Signed-off-by: Lev Stipakov <lev@openvpn.net> Acked-by: Heiko Hund <heiko@openvpn.net>
Message-Id: <20240319152803.1801-2-lev@openvpn.net>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28420.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Lev Stipakov [Tue, 19 Mar 2024 15:16:07 +0000 (17:16 +0200)]
interactive.c: disable remote access to the service pipe
Remote access to the service pipe is not needed and might
be a potential attack vector.
For example, if an attacker manages to get credentials for
a user which is the member of "OpenVPN Administrators" group
on a victim machine, an attacker might be able to communicate
with the privileged interactive service on a victim machine
and start openvpn processes remotely.
CVE: 2024-24974
Microsoft case number: 85925
Reported-by: Vladimir Tokarev <vtokarev@microsoft.com>
Change-Id: I8739c5f127e9ca0683fcdbd099dba9896ae46277 Signed-off-by: Lev Stipakov <lev@openvpn.net> Acked-by: Heiko Hund <heiko@openvpn.net>
Message-Id: <20240319151723.936-2-lev@openvpn.net>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28419.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Lev Stipakov [Tue, 19 Mar 2024 13:53:45 +0000 (15:53 +0200)]
win32: Enforce loading of plugins from a trusted directory
Currently, there's a risk associated with allowing plugins to be loaded
from any location. This update ensures plugins are only loaded from a
trusted directory, which is either:
- HKLM\SOFTWARE\OpenVPN\plugin_dir (or if the key is missing,
then HKLM\SOFTWARE\OpenVPN, which is installation directory)
- System directory
Loading from UNC paths is disallowed.
Note: This change affects only Windows environments.
CVE: 2024-27903
Change-Id: I154a4aaad9242c9253a64312a14c5fd2ea95f40d Reported-by: Vladimir Tokarev <vtokarev@microsoft.com> Signed-off-by: Lev Stipakov <lev@openvpn.net> Acked-by: Selva Nair <selva.nair@gmail.com>
Message-Id: <20240319135355.1279-2-lev@openvpn.net>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28416.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Individual tests can define a script to run to test
whether they should be skipped.
Included in this commit is an example check which
checks whether we can do NTLM checks. This fails
e.g. on recent versions of Fedora with mbedTLS
(tested with Fedora 39) or when NTLM support is not
compiled in.
v2:
- ntlm_support:
- support OpenSSL 3
- allow to build without cmocka
v3:
- add example to t_client.rc-sample
- t_client.sh code style
- use syshead.h in error.h
v5:
- rename SKIP_x to CHECK_SKIP_x
Change-Id: I13ea6752c8d102eabcc579e391828c05d5322899 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240308102818.9249-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/search?l=mid&q=20240308102818.9249-1-gert@greenie.muc.de Signed-off-by: Gert Doering <gert@greenie.muc.de>
Removed if-guard checking if any feature is
enabled before performing per-feature check.
It doesn't save us much but instead introduces
uneeded complexity.
While at it, fixed a typo IMCP -> ICMP for defined
PIPV6_ICMP_NOHOST_CLIENT and PIPV6_ICMP_NOHOST_SERVER
macros.
Fixes: Trac https://community.openvpn.net/openvpn/ticket/269
Change-Id: I4b5e8357d872c920efdb64632e9bce72cebee202 Signed-off-by: Gianmarco De Gregori <gianmarco@mandelbit.com> Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240307124616.16358-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28345.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Including "ssl.h" conflicts with the wolfSSL ssl.h header file. The openvpn/src directory needs to be included before include/wolfssl. include/wolfssl needs to be included so that openvpn can pick up wolfSSL compatibility headers instead of OpenSSL headers without changing the paths.
src/openvpn/Makefile.am does not need to be modified because AM_CPPFLAGS is placed before AM_CFLAGS in the output Makefile.
Signed-off-by: Juliusz Sosinowicz <juliusz@wolfssl.com> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20240212132522.125903-1-juliusz@wolfssl.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28229.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Fri, 9 Feb 2024 11:06:29 +0000 (12:06 +0100)]
Implement generating TLS 1.0 PRF using new OpenSSL 3.0 APIs
OpenSSL 3.0 introduced a new API for doing key derivation. So this leaves
us now with three different implementation for 1.0.2, 1.1.x and 3.x.
This was initially done to maybe still have a working TLS 1.0 PRF when
using OpenSSL 3.0 in FIPS but it gives the same error as with the older API.
But since moving to a new API is always good, we use the new API when using
OpenSSL 3.0. We also print the internal OpenSSL error message when
the KDF fails.
This also allows us now to compile an OpenSSL build that has been built with
OPENSSL_NO_MD5. Which is not yet common but might be in the future.
Change-Id: Ic74195a4ed340547c5e862dc2438f95be318c286 Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240209110629.15364-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28203.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 8 Feb 2024 08:57:49 +0000 (09:57 +0100)]
Add unit test for encrypting/decrypting data channel
This test is reusing code from --test-crypto but is modified to not rely
on the static key functionality and also only tests the most common
algorithm. So it does not yet completely replace --test-crypto
Change-Id: Ifa5ae96165d17b3cae4afc53e844bb34d1610e58 Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240208085749.869-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28195.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Kristof Provost [Wed, 24 Jan 2024 15:27:39 +0000 (16:27 +0100)]
dco-freebsd: dynamically re-allocate buffer if it's too small
It's possible for the buffer we provide for OVPN_GET_PEER_STATS to be
too small. Handle the error, re-allocate a larger buffer and try again
rather than failing.
Signed-off-by: Kristof Provost <kprovost@netgate.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20240124152739.28248-1-kprovost@netgate.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28128.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
documentation: Update and fix documentation for --push-peer-info
- description of IV_PROTO was outdated, missing a lot
of flags
- complete list of compression flags, but separate them out
- various other style/grammar/typo fixes
Change-Id: I7f854a5a14d2a2a391ebb78a2a92b3e14cfd8be6 Signed-off-by: Frank Lichtenheld <frank@lichtenheld.com> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20240206141057.46249-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28178.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 1 Feb 2024 14:48:17 +0000 (15:48 +0100)]
Allow unit tests to fall back to hard coded location
Settings the environment variable required for running unit tests is
tiresome in my IDE (Clion). So allow unit tests to fall back to a hard
coded location in case the environment variable is not set.
Change-Id: Ide72b81f497088dd0fd2cdcfff83cbce5b48f145 Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240201144817.188884-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28161.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Tue, 23 Jan 2024 10:43:58 +0000 (11:43 +0100)]
Ensure that all unit tests use unbuffered stdout and stderr
stderr is normally always unbuffered but stdout can be buffered. Especially,
when stdout is redirected it will become buffered while it is normally
unbuffered when connected to a terminal. This mean that if the unit exits
prematurely, the output in the buffered output will be lost.
As the unit test x_msg mock implementation prints even fatal on stdout
we ensure with this setup method that stdout is also unbuffered.
Change-Id: I5c06dc13e9d8ab73997f79b13c30ee8949e5e993 Acked-by: Frank Lichtenheld <frank@lichtenheld.com>
Message-Id: <20240123104358.495517-1-frank@lichtenheld.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28122.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
projects / thirdparty / openvpn.git / log
