Arne Schwabe [Wed, 12 May 2021 13:15:09 +0000 (15:15 +0200)]
Move utility function from win32.c to win32-util.c
This done to allow to include parts win32.c when building unit tests
as win32.c itself has too many dependencies and cannot be included in
a small unit test.
Also fix a missing Windows.h include in error.h that otherwise
breaks complation when included from unit tests.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210512131511.1309914-8-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22348.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 12 May 2021 13:15:03 +0000 (15:15 +0200)]
Remove explicit struct iovec check (HAVE_IOVEC)
This macro is currently used only in 3 places in syshead.h
- EXTENDED_SOCKET_ERROR_CAPABILITY is linux specific anyway and
starts with #if defined(HAVE_LINUX_TYPES_H)
- port share and ip_pktinfo macros depends on sendmsg/recvmsg
that implicitly also require iovec
So in all three cases we can implicitly assume that iovec is present
and do not need to make this explicit check
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210512131511.1309914-2-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22343.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Add CRL extractor script for --crl-verify dir mode
When --crl-verify is enabled, specified CRL file gets reloaded on
every client connection. With huge CRL files it may take a significant
amount of time - seconds and tens of seconds, during which OpenVPN is
blocked and can't serve existing and/or incoming connections due its
singlethread nature.
In alternative mode --crl-verify option takes directory containing
files named as decimal serial numbers of the revoked certificates and
'dir' flag, revoked certificate check is being done by checking the
presence of client's certificate number in that directory.
This script allow to perform incremental extraction of revoked serial
numbers from CRL by adding absent ones and removing excess ones.
Arne Schwabe [Mon, 3 May 2021 10:58:50 +0000 (12:58 +0200)]
Remove support for blocking connect()
It is hard to imagine that we still have any supported OS that does
not support non-blocking connect while still supporting non-blocking
sockets in general.
Patch V2: remove epoll.h removal that should have been in other patch.
Patch V3: remove another instance in socket.c that I misseds in v1
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210503105850.3903732-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22291.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 22 Apr 2021 15:29:39 +0000 (17:29 +0200)]
Remove a number of platform specific checks in configure.ac
- Remove windows XP/old mingw compat code in socket.c
- Use _WIN32 instead checking for existence of windows.h, winsock2.h
and ws2tcpip.h in autconf
- Remove check for unlink. The last remaining use is a check inside
a Unix socket.
- Even Windows has umask, so remove the check for it
- Move epoll.h inclusion to event.c
Patch V2: Add epoll.h syshead.h that accidently was put into another
patch
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210422152939.2134046-6-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22201.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 22 Apr 2021 15:29:36 +0000 (17:29 +0200)]
Move direct.h header where it is used
The direct.h header provides only a handful of functions [1] of which
we only use _wchdir. Directly included the direct.h file where it is
used and remove autoconf magic around it.
A switch block in interactive.c is missing the _undo_type_max value
of the neum set, thus triggering a compiler warning.
Due to the logic, this value cannot really be assigned to the variable
being examinated, however, add the missing enum value to silence
the warning.
Fixes:
interactive.c: In function ‘Undo’:
interactive.c:1561:13: warning: enumeration value ‘_undo_type_max’ not
handled in switch [-Wswitch]
1561 | switch (type)
| ^~~~~~
Signed-off-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210501130640.9330-4-a@unstable.cc>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22267.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
warn_used is a static variable but that is used only under certain
conditions. These conditions do not apply for Windows and therefore
mingw is spitting a warning for unused variable.
Move variable declaration inside code block that uses it.
Fixes:
run_command.c: In function ‘openvpn_popen’:
run_command.c:236:17: warning: unused variable ‘warn_shown’
[-Wunused-variable]
236 | static bool warn_shown = false;
| ^~~~~~~~~~
Signed-off-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210501130640.9330-3-a@unstable.cc>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22265.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
windows: use appropriate and portable format specifier for 64bit pointer
To print a 64bit poiner, fprintf() must be provided with the appropriate
format specifier.
The most portable in this case is PRIx64, which is redefined accordingly
on each platform. It comes from inttypes.h and it's the recommended way
to print a 64bit long hex value.
Fixes various warnings of this type:
event.c: In function ‘we_ctl’:
event.c:235:24: warning: 'I' flag used with ‘%x’ gnu_printf format
[-Wformat=]
235 | dmsg(D_EVENT_WAIT, "WE_CTL n=%d ev=%p rwflags=0x%04x arg="
ptr_format,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
error.h:151:68: note: in definition of macro ‘dmsg’
151 | #define dmsg(flags, ...) do { if (msg_test(flags)) {x_msg((flags),
__VA_ARGS__);} EXIT_FATAL(flags); } while (false)
|
^~~~~~~~~~~
event.c:235:24: warning: format ‘%x’ expects argument of type ‘unsigned
int’, but argument 6 has type ‘long long unsigned int’ [-Wformat=]
235 | dmsg(D_EVENT_WAIT, "WE_CTL n=%d ev=%p rwflags=0x%04x arg="
ptr_format,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
......
239 | (ptr_type)arg);
| ~~~~~~~~~~~~~
| |
| long long unsigned int
Signed-off-by: Antonio Quartulli <a@unstable.cc> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210501130640.9330-2-a@unstable.cc>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22268.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Add documentation on EVENT_READ/EVENT_WRITE constants
Changes from v4:
- get rid of the overly complex EVENT_SHIFT() macro
Changes from v3:
- re-introduce READ/WRITE_SHIFT because they are different from
EVENT_READ/WRITE
- define also EVENT_READ/WRITE using READ/WRITE_SHIFT
Changes from v2:
- moved event definitions to event.h
- removed READ/WRITE_SHIFT and use EVENT_READ/WRITE
- removed ifdefs around *_SHIFTS definitions in event.h
Changes from v1:
- fixed typ0s
- extended comment
- moved *_SHIFT definition to openvpn.h
- made READ/WRITE events dependant on _SHIFT definition with a macro
Signed-off-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210427191314.21317-1-a@unstable.cc>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22247.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Crypto backends are implementing the cipher_des_encrypt_ecb()
function without fully respecting the type of the argumentis as described
in the function prototype.
All ECB arguments (key, input block and output block) are expected to
be 8 bytes long, for this reason the prototype specifies the arguments
as 3 arrays of 8 bytes in size.
Convert the implementations to also explicitly mention the size of the
array they expect to receive in input.
Fixes these warnings:
crypto_openssl.c:866:39: warning: argument 2 of type ‘unsigned char *’
declared as a pointer [-Warray-parameter=]
866 | unsigned char *src,
| ~~~~~~~~~~~~~~~^~~
In file included from crypto.h:125,
from crypto_openssl.c:42:
crypto_backend.h:202:43: note: previously declared as an array ‘unsigned
char[8]’
202 | unsigned char src[DES_KEY_LENGTH],
| ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~
crypto_openssl.c:867:39: warning: argument 3 of type ‘unsigned char *’
declared as a pointer [-Warray-parameter=]
867 | unsigned char *dst)
| ~~~~~~~~~~~~~~~^~~
In file included from crypto.h:125,
from crypto_openssl.c:42:
crypto_backend.h:203:43: note: previously declared as an array ‘unsigned
char[8]’
203 | unsigned char dst[DES_KEY_LENGTH]);
| ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~
Signed-off-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210428081054.29081-1-a@unstable.cc>
URL: https://www.mail-archive.com/search?l=mid&q=20210428081054.29081-1-a@unstable.cc Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 7 Apr 2021 15:49:51 +0000 (17:49 +0200)]
Implement deferred auth for scripts
This patch also refactors the if condition that checks the result of
the authentication since that has become quite unreadable. It renames
s1/s2 and extracts some parts of the condition into individual variables
to make the condition better understandle
Patch v2: add refactoring of the if condition
Patch v4: fix documentation not mentioning method as 2nd line
Patch v5: fix deferred auth used by both plugin and script not working
Patch v6: Add missing async inotify for script deferred auth
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Tested-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210407154951.13330-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22072.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Gert Doering [Tue, 27 Apr 2021 11:03:00 +0000 (13:03 +0200)]
rewrite parse_hash_fingerprint()
The existing code was doing far too much work for too little
gain - copying the string segment for scanf(), checking extra
for spaces, making the result quite unreadable.
Verify each segment with (short-circuited) isxdigit() checks,
then feed directly to scanf(), which will stop parsing on ':'
or end-of-string.
Rewrite error message to differentiate "hash too short" (including
number of bytes read) and "hash too long" (it did not terminate when
we had enough bytes).
While at it, add an option printer for the resulting o->verify_hash
list to show_settings().
v2:
fix typo in commit message
appease whitespace dragon
add printing of verify_hash_algo and verify_hash_depth
print correct hash length for SHA1 certs
fix incorrect assignment to options->verify_hash_algo in c3a7065d5
Signed-off-by: Gert Doering <gert@greenie.muc.de> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210427110300.6911-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22241.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 8 Apr 2021 14:02:26 +0000 (16:02 +0200)]
Change options->data_channel_use_ekm to flags
Instead maintaining two different representation of the data channel
options in struct options and struct tls_options, use the same
flags variable that tls_options uses.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210408140229.31824-2-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22084.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 21 Apr 2021 12:34:15 +0000 (14:34 +0200)]
Allow all GCM ciphers
OpenSSL also allows ARIA-GCM and that works well with our implementation
While the handpicked list was needed for earlier OpenSSL versions (and
is still needed for Chacha20-Poly1305), the API nowadays with OpenSSL
1.0.2 and 1.1.x works as expected.
Patch V2: Remove special cases for AES-GCM ciphers.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210421123415.1942917-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22168.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 22 Apr 2021 15:17:18 +0000 (17:17 +0200)]
Move tls_select_primary_key into its own function
tls_pre_encrypt mainly performs the task of selecting the primary
encryption key but also performs other minor tasks. To allow only
querying for the key that should be used for encryption extract this
part of the function into its own function.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210422151724.2132573-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22198.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
options: check for blanks in fingerprints and reject string if found
A fingerprint is not expected to contains any blank (white space),
however, the parser routine will still attempt parsing the octect
and ignore the space.
This means that a fingerprint like
"5 :F0: 8:75:70:46:6E:(...)"
will be parsed successfully.
Explicitly check for spaces in the various octets, before conversion,
and error out if any is found.
Signed-off-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210421234908.12817-1-a@unstable.cc>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22182.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Max Fillinger [Thu, 15 Apr 2021 09:34:54 +0000 (11:34 +0200)]
Abort if CRL file can't be stat-ed in ssl_init
Now that the path for the CRL file is handled correctly when using
chroot, there's no good reason for the file to be inaccessible during
ssl_init().
This commit ensures that the CRL file is accessed successfully at least
once, which fixes a bug where the mbedtls version of OpenVPN wouldn't
use a reloaded CRL if it initially failed to access the file.
Signed-off-by: Max Fillinger <maximilian.fillinger@foxcrypto.com> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210415093454.18324-1-maximilian.fillinger@foxcrypto.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22118.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Max Fillinger [Thu, 15 Apr 2021 09:12:48 +0000 (11:12 +0200)]
In init_ssl, open the correct CRL path pre-chroot
When using the chroot option, the init_ssl function can be called before
entering the chroot or, when OpenVPN receives a SIGHUP, afterwards. This
commit ensures that OpenVPN tries to open the correct path for the CRL
file in either situation.
This commit does not address key and certificate files. For these, the
--persist-key option should be used.
Signed-off-by: Max Fillinger <maximilian.fillinger@foxcrypto.com> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210415091248.18149-1-maximilian.fillinger@foxcrypto.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22117.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 28 Mar 2021 12:02:41 +0000 (14:02 +0200)]
Fix condition to generate session keys
When OpenVPN sees a new (SSL) connection via HARD_RESET or SOFT_RESET with
the same port/ip as an existing session, it will give it the slot of the
renegotiation session (TM_UNTRUSTED). And when the authentication
succeeds it will replace the current session. In the case of a SOFT_RESET
this a renegotiation and we will generated data channel keys at the of
key_method_2_write function as key-id > 0.
For a HARD RESET the key-id is 0. Since we already have gone through
connect stages and set context_auth to CAS_SUCCEEDED, we don't
call all the connect stages again, and therefore also never call
multi_client_generate_tls_keys for this session.
This commit changes postponing the key generation to be done only if
the multi_connect has not yet been finished.
Patch V2: Explain better in the commit message why this change is done.
Trac: #1316
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210328120241.27605-2-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21873.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 18 Apr 2021 16:01:11 +0000 (18:01 +0200)]
Move context_auth from context_2 to tls_multi and name it multi_state
context_2 and tls_multi have the same life cycle for TLS connections
but so this move does not affect behaviour of the variable.
OpenVPN TLS multi code has a grown a lot more complex and code that
handles multi objects needs to know the state that the object is in.
Since not all code has access to the context_2 struct, the code that
does not have access is often not checking the state directly but
checks other parts of multi that have been affected from a state
change.
This patch also renames it to multi_state as this variable represents
the multi state machine status rather than just the state of the connect
authentication (more upcoming patches will move other states
into this variable).
Patch V2: also rename context_auth to multi_state, explain a bit why this
change is done.
Patch V3: Add comments for c2->multi NULL check forwarding. Fix compile
with ENABLE_ASYNC_PUSH.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210418160111.1494779-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22155.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Fri, 16 Apr 2021 11:09:55 +0000 (13:09 +0200)]
Add parsing of dhcp-option PROXY_HTTP
This adds support for setting a HTTP proxy that should be used after
connecting to a VPN.
The syntax has been picked to have compatibility with OpenVPN3.
Otherwise I would have used HTTP-PROXY instead.
Since this option requires an additional argument compared to the
existing dhcp-option keywords, move checking the number of arguments
to the individual keywords.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210416110955.1162574-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22129.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Fix IPv4 default gateway with multiple route tables
Current default gateway selection for zero destination address just
dumps and parses all the routing tables. If any of non-main table
with default route comes first, wrong default gateway can be picked.
Since adding/removing routes currently handles only main table,
let's stick to RT_TABLE_MAIN while selecting default route too.
v2: keep gateway address unchanged on lookup error
v3: reduce ammout of gateway address copying
Reported-by: Donald Sharp <donaldsharp72@gmail.com> Signed-off-by: Vladislav Grishenko <themiron@yandex-team.ru> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210416120708.1532-1-themiron@yandex-team.ru>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22130.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 15 Apr 2021 11:45:41 +0000 (13:45 +0200)]
Remove LibreSSL specific defines not needed for modern LibreSSL
Most of the functions remove were either already have a version check
against LibreSSL 2.9.0 or are also now deprecated in LibreSSL as well
according to the man pages in OpenBSD 6.8 like SSL_CTX_set_ecdh_auto
and SSL_library_init.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210415114541.1001644-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22119.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 8 Apr 2021 12:00:27 +0000 (14:00 +0200)]
Always save/restore pull options
The makes the code path for pull and non-pull more aligned and even
though this might do extra work for non-pull scenarios, saving the
few bytes of memory is not a worthwhile optimisation here.
Additionally with the upcoming P2P mode NCP, the client needs to
save/restore a subset of these options anyway.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210408120029.19438-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22079.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Tue, 6 Apr 2021 16:25:18 +0000 (18:25 +0200)]
Remove OpenSSL configure checks
These checks for the functions take a lot of time in configure call and
also having these checks make it more blurry for which of the supported
OpenSSL versions (and libraries claiming to be OpenSSL) are actually
needed.
Tested with OpenSSL 1.1.1(Ubuntu 20, macOS), 1.0.2 (CentOS7),
1.1.0 (Debian stretch), LibreSSL (OpenBSD 6.8) and wolfSSL
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210406162518.4075-5-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22051.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Max Fillinger [Mon, 12 Apr 2021 17:46:17 +0000 (19:46 +0200)]
Fix build with mbedtls w/o SSL renegotiation support
In mbedtls, support for SSL renegotiation can be disabled at
compile-time. However, OpenVPN cannot be built with such a library
because it calls mbedtls_ssl_conf_renegotiation() to disable this
feature at runtime. This function doesn't exist when mbedtls was built
without support for SSL renegotiation.
This commit fixes the build by ifdef'ing out the function call when
mbedtls was built without support for SSL renegotiation.
Signed-off-by: Max Fillinger <maximilian.fillinger@foxcrypto.com> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <E1lW0eX-00012w-9n@sfs-ml-1.v29.lw.sourceforge.com>
URL: https://www.mail-archive.com/search?l=mid&q=E1lW0eX-00012w-9n@sfs-ml-1.v29.lw.sourceforge.com Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Tue, 6 Apr 2021 16:25:16 +0000 (18:25 +0200)]
Remove a number of checks for functions/headers that are always present
For the unlink function we actually have code that just ignores
the unlink call if the unlink function is not present. But all
platforms should have an unlink function.
This also removes all conditionals check for the headers that
belong to the C99 standard library header list
(https://en.cppreference.com/w/c/header).
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210406162518.4075-3-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22053.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 4 Apr 2021 11:06:02 +0000 (13:06 +0200)]
Remove conditionals compilation for P2MP, ENABLE_SHAPER and TIME_BACKTRACK_PROTECTION
Using OpenVPN without P2MP support (pull, TLS) is unrealistic and
building a binary without it is not something we realistically want
to support anyway. Building P2MP support currently only depended
on HAVE_GETTIMEOFDAY or _WIN32, which has a compat function for it.
So we basically can assume that gettimeofday is always availabe,
either natively or through our compat function.
Remove all the #ifdef P2MP logic, simplify code and reduce maintenance
effort.
This also removes the ENABLE_SHAPER and TIME_BACKTRACK_PROTECTION
defines, which also depended only on the HAVE_GETTIMEOFDAY or _WIN32.
I kept the configure.ac check and ifdef in compat since mingw actually
provides a gettimeofday and we will use that instead of our own compat
function.
Patch V2: Remove dco parts that slipped into the patch, mention the
other removed defines that are always enabled.
Patch V3: Also remove the TIME_BACKTRACK_PROTECTION defines from otime.h
Message-Id: <20210403184626.23067-1-arne@rfc2549.org> Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210404110602.20374-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22030.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Fixes:
tun.c: In function ‘do_ifconfig_ipv4’:
tun.c:1217:17: warning: variable ‘ifconfig_remote_netmask’ set but not
used [-Wunused-but-set-variable]
const char *ifconfig_remote_netmask = NULL;
tun.c:1213:10: warning: unused variable ‘tun’ [-Wunused-variable]
bool tun = is_tun_p2p(tt);
Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210403172403.9452-1-selva.nair@gmail.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22019.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Gert Doering [Fri, 2 Apr 2021 17:34:14 +0000 (19:34 +0200)]
Fix potential NULL ptr crash if compiled with DMALLOC
In the unlikely case that we are compiled with -DDMALLOC *and*
malloc() returns NULL, there is an uncaught memset() which would
crash then. Remove the memset(), as the right the next operation
after check_malloc_return() is a mempcy() which will overwrite
the whole memory block anyway.
Trac: #586
Signed-off-by: Gert Doering <gert@greenie.muc.de> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210402173414.14216-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21981.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sat, 3 Apr 2021 12:24:44 +0000 (14:24 +0200)]
Fix async push broken after auth deferred refactor
Commit c5fec838e moved the auth control file related
states into its own struct. Unfortunately I forgot
to also do the part inside #if defined(ENABLE_ASYNC_PUSH)
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210403122444.17090-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22007.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sat, 3 Apr 2021 12:30:00 +0000 (14:30 +0200)]
log file descriptor in more socket related error messages
This add the fd to the epoll event error message and the x_check_status
message. This helps debugging when thing go wrong with event handling.
Also add logging when ep_del fails to remove a socket from the structure.
In constract to ep_ctl that has this as a FATAL message (M_ERR), we only
log here since the code has been ignoring the status forever there might
be corner cases where a FATAL message could trigger an unintened
regression.
PATCH v2: Fix wrong order of fd,code in printed message.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210403123000.17688-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg22008.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 1 Apr 2021 12:37:51 +0000 (14:37 +0200)]
Remove deprecated option '--keysize'
This option has been deprecated in OpenVPN 2.4 and the ciphers that allow
using this option fall all into the SWEET32 category of ciphers with
64 bit block size.
Patch V2: Remove superflous check in OpenSSL codepath to check keysize
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210401123751.31756-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21943.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 28 Mar 2021 09:05:30 +0000 (11:05 +0200)]
Deprecate non TLS mode in OpenVPN
The non-TLS mode is a relict from OpenVPN 1.x or 2.0. When TLS mode was
introduced the advantages of TLS over non-TLS were small but TLS mode
evolved to include a lot more features (NCP, multipeer, AEAD ciphers to
name a few).
Today VPN setups that use --secret are mainly used because this mode is
easier to setup and does not require setting up a PKI. This shortcoming
of TLS mode should be addressed now with the peer-fingerprint option.
The primary reason to deprecate --secret is that it is not secure enough
anymore for modern environments. This mode uses a fixed pre-shared key and
no session keys. Thus, no forward secrecy is possible, which means that
any captured VPN traffic can be decrypted later should the --secret key
get into the wrong hands. The cryptography overall used here was okay
when --secret was introduced but is not acceptable by today's standard
anymore.
Finally, modern hardware-accelerated crypto modes like AES-GCM can only
be used in TLS mode (due to IV requirements).
Patch V2: Improve commit message
Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210328090530.10653-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21868.html
Arne Schwabe [Fri, 19 Feb 2021 16:52:52 +0000 (17:52 +0100)]
Allow running a default configuration with TLS libraries without BF-CBC
Modern TLS libraries might drop Blowfish by default or distributions
might disable Blowfish in OpenSSL/mbed TLS. We still signal OCC
options with BF-CBC compatible strings. To avoid requiring BF-CBC
for this, special this one usage of BF-CBC enough to avoid a hard
requirement on Blowfish in the default configuration.
Signed-off-by: Arne Schwabe <arne@rfc2549.org>
Patch v2: add more clarifying comment, do not warn about OCC only insecure
ciphers, code improvements
Patch V3: Put ciphername resolution via ciper_kt_name in the right branch
Patch V4: Fix cornercase of BF-CBC in data-ciphers not itialising cipher.
Patch v5: I accidently resend v3 as v4. So v5 is just a resend of the real
v4 Acked-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210219165252.4562-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21577.html
Arne Schwabe [Thu, 1 Apr 2021 11:00:03 +0000 (13:00 +0200)]
Always disable TLS renegotiations
Renegotiations have been troublesome in the past and also the recent
OpenSSL security problem (CVE-2021-3449) is only exploitable if
TLS renegotiation is enabled.
mbed TLS disables it by default and says in the documentation:
Warning: It is recommended to always disable renegotation unless you
know you need it and you know what you're doing. In the past, there
have been several issues associated with renegotiation or a poor
understanding of its properties.
TLS renegotiation can be used to restart a session with different
parameters (e.g. now with client certs). This something that OpenVPN does
not use.
For OpenSSL 1.0.2 the workaround to disable renegotiation is rather
cumbersome. So we keep this to 1.1.1 only since 1.0.2 is on its way to
deprecation anyway.
Furthermore because of all these problems, also TLS 1.3 completely
drops support for renegotiations.
Patch V2: Improve comments and commit message
Patch V3: Only disable renegotiation where the SSL_OP_NO_RENEGOTIATION
define is available. LibreSSL, wolfSSL and OpenSSL 1.0.2 are
lacking this macro. Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210401110003.19689-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21939.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Steffan Karger [Wed, 31 Mar 2021 18:03:23 +0000 (20:03 +0200)]
reliable: retransmit if 3 follow-up ACKs are received
To improve the control channel performance under packet loss conditions,
add a more aggressive retransmit policy similar to what many TCP
implementations do: retransmit a packet if the ACK timeout expires (like
we already do), *or* if three ACKs for follow-up packets are received.
The rationale behind this is that if follow-up packets *are* received, the
connection is apparently functional and we should be able to retransmit
immediately. This significantly improves performance for connections with
low (up to a few percent) packet loss. Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <E1lRfW3-0001sy-VM@sfs-ml-4.v29.lw.sourceforge.com>
URL: https://www.mail-archive.com/search?l=mid&q=E1lRfW3-0001sy-VM@sfs-ml-4.v29.lw.sourceforge.com
Arne Schwabe [Thu, 1 Apr 2021 13:13:37 +0000 (15:13 +0200)]
Remove do_init_socket_2 and do_init_socket_1 wrapper function
These two function basically just pass a number of fields of context to
the linit_socket_init1/2 functions. This wrapper add little to no value
in understanding the code, especially since the linit_socket_init1 will
just copy them to yet another structure.
Remove these wrapper functions and pass context directly to the called
function.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210401131337.3684-15-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21954.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Thu, 1 Apr 2021 13:13:35 +0000 (15:13 +0200)]
Extract multi_assign_peer_id into its own function
This makes multi_get_create_instance_udp a bit shorter and better
structured and also prepares this method to be called from the
mutlti TCP context with DCO which will also need to assign unique peer
ids to instances.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210401131337.3684-13-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21959.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Gert Doering [Fri, 2 Apr 2021 13:45:29 +0000 (15:45 +0200)]
Fix 'compress migrate' for 2.2 clients.
Commit 8fa8a17528c001a introduces "compress migrate" to move old clients
that have "compress" or "comp-lzo" in their config towards a connection
without compression. This is done by looking at incoming OCC strings
to see if the client has compression enabled, and at incoming IV_
strings to see whether it can do "compress stub-v2" or needs to be sent
"comp-lzo no".
That check fails for 2.2 clients that do not send *any* peer-info by
default, so the server will not push back any "disable compression"
command. It works if the client connects with "--push-peer-info".
Fix: turn around the order of checks, treat "no peer_info" the same
as "peer_info does not contain IV_COMP_STUBv2".
Signed-off-by: Gert Doering <gert@greenie.muc.de> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210402134529.27866-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21974.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 24 Mar 2021 22:08:53 +0000 (23:08 +0100)]
Implement '--compress migrate' to migrate to non-compression setup
This option allow migration to a non compression server config while
still retraining compatibility with client that have a compression
setting in their config.
For existing setups that used to have comp-lzo no or another
compression setting in their configs it is a difficult to migrate to
a setup without compression without replacing all client configs at
once especially if OpenVPN 2.3 or earlier clients are in the mix that
do not support pushing stub-v2. Even with OpenVPN 2.4 and later clients
that support pushing this is not a satisfying solution as the clients
log occ mismatches and the "push stub-v2" needs to be in the server
config "forever".
If the new migrate option to compress is set and a client is detected
that indicates that compression is used (via OCC), the server will
automatically add ``--push compress stub-v2`` to the client specific
configuration if stub-v2 is supported by the client and otherwise
switch to ``comp-lzo no`` and add ``--push comp-lzo`` to the client
specific configuration.
Gert Doering [Thu, 1 Apr 2021 08:29:34 +0000 (10:29 +0200)]
Get rid of last PLUGIN_DEF_AUTH #ifdef
Commit 99d217b200 attempted to get rid of all #ifdef related to
--disable-def-auth but one of them managed to hide. Remove.
The effect of this is that the "openvpn_acf_...tmp" files get not
removed after when an async auth plugin is in use. This is can
get very annoying on a busy server.
Trac: #1186
Signed-off-by: Gert Doering <gert@greenie.muc.de> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210401082934.29922-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21933.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Tõivo Leedjärv [Sun, 28 Mar 2021 17:11:51 +0000 (17:11 +0000)]
Stop using deprecated getpass()
The getpass() function is present in SUSv2, but marked LEGACY. It is
removed in POSIX.1-2001. Additionally, on Solaris getpass() returns
maximum 9 bytes. This will make longer passwords fail with no
possibility for user to know what is happening.
This patch removes usage of getpass() completely and replaces it with
direct implementation of what getpass() does: opens tty (existing code),
outputs the prompt (existing code), turns off echoing (new code), reads
one line (existing code shared with echoed mode), restores tty state
(new code) and closes tty (existing code).
Patch v2: incorporate review feedback, incl. style fixes, merge
termios.h check in configure.ac with an existing
AC_CHECK_HEADERS, add error check and logging after
tcsettattr() when restoring tty settings
Signed-off-by: Tõivo Leedjärv <toivol@gmail.com> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210328171151.12056-1-toivol@gmail.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21889.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Selva Nair [Mon, 29 Mar 2021 04:23:18 +0000 (00:23 -0400)]
Remove automatic service
This has been replaced by openvpnserv2 since 2.4.0 and we have
stopped setting up this service in the installer since 2.5.0.
Get rid of the unused code. The mechanics of supporting multiple
services with the same executable is retained for possible future use.
For backwards compatibility, the command line option -instance
is unchanged as "-instance <name> id" although <name>="interactive"
is the only supported value now.
Signed-off-by: Selva Nair <selva.nair@gmail.com> Acked-by: Lev Stipakov <lstipakov@gmail.com>
Message-Id: <1616991798-7179-1-git-send-email-selva.nair@gmail.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21890.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 28 Mar 2021 14:20:38 +0000 (16:20 +0200)]
Remove support for non ISO C99 vararg support
We require ISO C99 as minimum support for our source code and all compilers
should support the ISO C99 macros. Especially gcc does not need
the gcc extensions anymore. Also MSVC has support for it (as defined
in the config-msvc.h but also double checked)
LCLINT seems to be a C analyzer that history has forgotten about. I could
only find https://splint.org/release1.3.html and an similarly old research
paper.
Patch V2: Also remove AX_ macros from configure.ac
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Steffan Karger <steffan.karger@foxcrypto.com> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210328142038.8826-2-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21883.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 28 Mar 2021 14:20:37 +0000 (16:20 +0200)]
Remove flexible array member autoconf check
This is configure macro that tries out how to declare a variable array
at the end of struct. This has been standardised in C99, so there is
no more need for non C99 magic. See also this stackoverflow discussion:
Arne Schwabe [Fri, 26 Mar 2021 17:57:50 +0000 (18:57 +0100)]
Cleanup print_details and add signature/ED certificate print
This commit cleans up the logic in the function a bit. It also makes it
more clear the the details printed in the second part of the message are
details about the peer certificate and not the TLS connection as such.
Also print the signature algorithm as this might help to identify
peer certificate that still use SHA1.
The new format with for TLS 1.3 and an EC certificate.
Control Channel: TLSv1.3, cipher TLSv1.3 TLS_AES_256_GCM_SHA384, peer
certificate: 384 bit EC, curve secp384r1, signature: ecdsa-with-SHA256
Using the more generic OpenSSL functions also allows use to correctly
print details about ED certificates:
Control Channel: TLSv1.3, cipher TLSv1.3 TLS_AES_256_GCM_SHA384, peer
certificate: 253 bit ED25519, signature: ED25519
Patch v2: Cleanup multiple calls to EVP_PKEY_id, minor code restructuring
Patch v3: Always initialise sig.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210326175750.4772-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21861.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 24 Mar 2021 22:23:30 +0000 (23:23 +0100)]
Use correct types for OpenSSL and Windows APIs
The error code of OpenSSL is a long. On most Unics systems
(mac, Linux...) this happens to be the same as size_t. But on Windows
as LP64, long is a 32 bit type and size_t is a 64 bit type. So use the
same type as OpenSSL.
When calling the Windows API use DWORD for the functions that want a
DWORD.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210324222330.455-4-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21803.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 24 Mar 2021 22:23:27 +0000 (23:23 +0100)]
Make buffer related function conversion explicit when narrowing
Clang and gcc do report many of the narrowing conversion that MSVC
reports, like these:
warning C4267: 'function': conversion from 'size_t' to 'int', possible
loss of data
This commit changes int to size_t where it is safe
(e.g. checked by buf_size_valid) and add casts where necessary.
In the function buffer_read_from_file the return value of fread is
size_t (at least on Linux/Windows/macOS and cppreference), so fix the
check to actually make sense.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210324222330.455-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21805.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 17 Mar 2021 16:00:38 +0000 (17:00 +0100)]
Restore also ping related options on a reconnect
This fixes the issue that if a client reconnects the next connection
entries inherits the keepalive settings that were pushed or set by
the previous entry. Since UDP+PULL entries have an implicit 120s
timeout, this timeout also got applied to a TCP session after an
UDP entry.
Patch v2: rebase on master
Reported-By: Jan Just Keijser <janjust@nikhef.nl> Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210317160038.25828-3-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21675.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 17 Mar 2021 16:00:37 +0000 (17:00 +0100)]
Move NCP saving and restore to the prepush restore code
This unifies save/restoring options that might be changed by a push
from the server. It also removes using the context_1 to store something
that is not related to a SIGHUP lifetime.
Patch v2: rebase on master.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210317160038.25828-2-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21674.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Wed, 17 Mar 2021 16:00:36 +0000 (17:00 +0100)]
Move restoring pre pull options to initialising of c2 context
We currently delay restoring these options until we actually must
restore them. Since there is no reason to do so apart from the very
minor saving to not have to execute that code when a connection fails,
move them it into the general context_2 initialisation.
Patch V2: rebase on master.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210317160038.25828-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21676.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Simon Rozman [Mon, 22 Mar 2021 10:39:57 +0000 (11:39 +0100)]
openvpnserv: Cache last error before it is overridden
FormatMessage() sets the last error according to its own success. This
looses the original error code leading to mismatched error message and
error number when sprintfted together resulting in confusing event log
message.
Signed-off-by: Simon Rozman <simon@rozman.si> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210322103957.1234-1-simon@rozman.si>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21789.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Mon, 22 Mar 2021 10:21:19 +0000 (11:21 +0100)]
Remove empty dummy functions
These functions seem to have been added to avoid MSVC compiler warnigns.
However nowadays, they trigger compiler warnings from Clang (e.g. when
using --disable-lzo and --disable-lz4):
Arne Schwabe [Mon, 22 Mar 2021 09:16:21 +0000 (10:16 +0100)]
Deprecate the --verify-hash option
Despite trying to figure out with multiple people what the use case for
this option is, we could not come up with a good one. Checking that only
a specific CA is used can be also done by only using that CA in the --ca
directive.
Although it feels a bit strange to deprecate the option after improving
it with peer-fingerprint patches, all the improvements are needed for
--peer-fingerprint and making them specify to --peer-fingerprint would
have added more (unecessary) changes.
Patch v3: rebased on v3 version of other patches.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210322091621.7864-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21779.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Support fingerprint authentication without CA certificate
OpenVPN traditionally works around CAs. However many TLS-based protocols
also
allow an alternative simpler mode in which rather than verify certificates
against CAs, the certificate itself is hashed and compared against a
pre-known set of acceptable hashes. This is usually referred to as
"fingerprint verification". It's popular across SMTP servers, IRC servers,
XMPP servers, and even in the context of HTTP with pinning.
* Allow not specifying the --ca parameter, to specify that
certificates should not be checked against a CA.
I've included some instructions on how to use all of this.
Record our fingerprint in an environment variable for the client to use
later:
$ server_fingerprint="$(openssl x509 -in servercert.pem -noout -sha256
-fingerprint | sed 's/.*=//;s/\(.*\)/\1/')"
Record our fingerprint in an environment variable for the server to use
later:
$ client_fingerprint="$(openssl x509 -in clientcert.pem -noout -sha256
-fingerprint | sed 's/.*=//;s/\(.*\)/\1/')"
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Patch V2: Changes in V2 (by Arne Schwabe):
- Only check peer certificates, not all cert levels, if you need
multiple levels of certificate you should use a real CA
- Use peer-fingerprint instead tls-verify on server side in
example.
- rename variable ca_file_none to verify_hash_no_ca
- do no require --ca none but allow --ca simply
to be absent when --peer-fingprint is present
- adjust warnings/errors messages to also point to
peer-fingerprint as valid verification method.
- Fix mbed TLS version of not requiring CA
not working
Patch v3: Fix minor style. Remove unessary check of verify_hash_no_ca in
ssl.c.
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210322091414.7533-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/search?l=mid&q=20210322091414.7533-1-arne@rfc2549.org Signed-off-by: Gert Doering <gert@greenie.muc.de>
wcsncat() was declared unsafe in favour of wcsncat_s(). However, the
string concatenation follows the string length check, making wcsncat()
safe too. Code analysis is just not smart enough (yet) to detect this.
The code was refactored to use wcscat_s() MSVC is considering as "safe".
Signed-off-by: Simon Rozman <simon@rozman.si> Acked-by: Gert Doering <gert@greenie.muc.de>
Message-Id: <20210322074359.527-1-simon@rozman.si>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg21774.html Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 21 Mar 2021 14:33:53 +0000 (15:33 +0100)]
Implement peer-fingerprint to check fingerprint of peer certificate
This option allows to pin one or more more peer certificates. It also
prepares for doing TLS authentication without a CA and just
self-signed certificates.
Patch V2: Allow peer-fingerprint to be specified multiple times
to allow multiple peers without needing to use inline
syntax. (e.g. on command line).
Patch V3: rebase on v3 of 1/4, reword message of verify-hash and
peer-fingerpring incompatibility
Signed-off-by: Arne Schwabe <arne@rfc2549.org> Acked-by: Antonio Quartulli <antonio@openvpn.net>
Message-Id: <20210321143353.2677-1-arne@rfc2549.org>
URL: https://www.mail-archive.com/search?l=mid&q=20210321143353.2677-1-arne@rfc2549.org Signed-off-by: Gert Doering <gert@greenie.muc.de>
Arne Schwabe [Sun, 21 Mar 2021 14:25:38 +0000 (15:25 +0100)]
Extend verify-hash to allow multiple hashes
This patch introduces support for verify-hash inlining.
When inlined, this options now allows to specify multiple fingerprints,
one per line.
Since this is a new syntax, there is no backwards compatibility to take
care of, therefore we can drop support for SHA1. Inlined fingerprints
are assumed be to SHA-256 only.
Also print a warning about SHA1 hash being deprecated to verify
certificates as it is not "industry standard" anymore.
Patch v2: fix/clarify various comments, fix a few minor problems, allow
the option to be specified multiple times and have that
added to the list.
Simon Rozman [Sun, 21 Mar 2021 14:46:27 +0000 (15:46 +0100)]
iservice: Resolve MSVC C4996 warnings
Lots of string functions were declared unsafe in favor of ..._s()
counterparts. However, the code already is careful about the buffer
size. Code analysis is just not smart enough (yet) to detect this.
The code was refactored to use ..._s() variants MSVC is considering as
"safe".
Signed-off-by: Simon Rozman <simon@rozman.si> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210321144627.1621-5-simon@rozman.si>
URL: https://www.mail-archive.com/search?l=mid&q=20210321144627.1621-5-simon@rozman.si Signed-off-by: Gert Doering <gert@greenie.muc.de>
It's about using a standard recommended alias for the wcsdup():
> warning C4996: 'wcsdup': The POSIX name for this item is deprecated.
> Instead, use the ISO C and C++ conformant name: _wcsdup. See online
> help for details.
And the documentation says:
> The Microsoft-implemented POSIX function names strdup and wcsdup are
> deprecated aliases for the _strdup and _wcsdup functions. By default,
> they generate Compiler warning (level 3) C4996. The names are
> deprecated because they don't follow the Standard C rules for
> implementation-specific names. However, the functions are still
> supported.
>
> We recommend you use _strdup and _wcsdup instead. Or, you can continue
> to use these function names, and disable the warning. For more
> information, see Turn off the warning and POSIX function names.
Reference:
https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/strdup-wcs
dup Signed-off-by: Simon Rozman <simon@rozman.si> Acked-by: Arne Schwabe <arne@rfc2549.org>
Message-Id: <20210321144627.1621-3-simon@rozman.si>
URL: https://www.mail-archive.com/search?l=mid&q=20210321144627.1621-3-simon@rozman.si Signed-off-by: Gert Doering <gert@greenie.muc.de>
projects / thirdparty / openvpn.git / log
