RFC 8247 demoted it to SHOULD NOT. This might break connections with
Windows clients unless they are configured to use a stronger group or
matching weak proposals are configured explicitly on the server.
References #2427.
The MySQL client doesn't like overlapping queries on the same
connection, so we make sure to destroy the enumerator used to check for
an existing pool before deleting it when --replace is used.
This is not ideal as the call to C_Finalize() should be the last one via
the PKCS#11 API. Since the order in which jobs are canceled is undefined
we can't be sure there is no other thread still using the library (it could
even be the canceled job that still handles a previous slot event).
According to PKCS#11 the behavior of C_Finalize() is undefined while other
threads still make calls over the API.
However, canceling the thread, as done previously, could also be problematic
as PKCS#11 libraries could hold locks while in the C_WaitForSlotEvent() call,
which might not get released properly when the thread is just canceled,
and which then might cause later calls to other API functions to block.
Fixes#2437.
If enabled, add the RADIUS Class attributes received in Access-Accept messages
to RADIUS accounting messages as suggested by RFC 2865 section 5.25.
Fixes#2451.
We do something similar in reestablish() for break-before-make reauth.
If we don't abort we'd be sending an IKE_AUTH without any TS payloads.
References #2430.
The fix for gperf in 0ae19f0ced added the generated header to
EXTRA_DIST but that's already added to the distribution because it is
contained in *_SOURCES, what was not added, though, was the .h.in file.
Also fixes the reference to the header file in the .c rule here and for
stroke in out-of-tree builds.
Fixes: 0ae19f0ced ("configure: Fix gperf length parameter determination")
This can happen if a stream is used blocking exclusively (the FD is
never registered with watcher, but is removed in the stream's destructor
just in case it ever was - doing this conditionally would require an
additional flag in streams). There may be no thread reading from
the read end of the notify pipe (e.g. in starter), causing the write
to the notify pipe to block after it's full. Anyway, doing a relatively
expensive FD update is unnecessary if there were no changes.
Fixes#1453.
Just rely on the default proposals by charon if nothing is defined. The
hard-coded IKE proposal used curve25519, which depends on an optional
plugin (while enabled by default it might still not be loaded, or, like
on Debian, shipped in an optional package). With charon's default
proposal only loaded algorithms are proposed for IKE avoiding this issue.
With OpenSSL 1.1.0 the library is now named libcrypto too on Windows.
Check for libeay32 first so we don't link against the build environment's
version of OpenSSL instead of the native one that might be available.
The order of arguments in X509_CRL_get0_signature() is not the same as that
of X509_get0_signature().
Fixes: 989ba4b6cd ("openssl: Update CRL API to OpenSSL 1.1.0")
gperf is not actually a build dependency as the generated files are
shipped in the tarball. So the type depends on the gperf version on
the host that ran gperf and created the tarball, which might not be
the same as that on the actual build host, and gperf might not even
be installed there, leaving the type undetermined.
Fixes: e0e4322973 ("configure: Detect type of length parameter for gperf generated function")
It seems that there is a race, at least in 10.13, that lets
if_indextoname() fail for the new TUN device. So we delay the call a bit,
which seems to "fix" the issue. It's strange anyway that the previous
delay was only applied when an iface entry was already found.
Recent releases of glibc don't include the full stdint.h header in some
network headers included by utils.h. So uintptr_t might not be defined.
Since we use fixed width integers, including the latter, all over the place
we make sure the complete file is included.
Fixes#2425.
The value of DHCP_OPTEND is 255. When it is assigned this result in a
sign change as the positive int constant is cast to a signed char and -1
results. Clang 4.0 complains about this.
This is a follow up on the issue documented in the previous commit.
To build with -Werror and Clang 3.9 we'd have to change all enum arguments
that are used as last argument before ... to e.g. u_int, which affects
quite a lot of places (crypto-factory, MODP_CUSTOM constructors, auth-cfg,
bus, vici-builder, vici-message). Besides that it doesn't look as nice
it also seems a bit too much hassle just to cater to the whims of a
particular version of one compiler, so we just don't build with that
version on Travis and use 4.0 instead.
This fixes compilation with -Werror when using Clang 4.0 (but not 3.9)
and possibly prevents undefined behavior.
According to the C standard the following applies to the second
parameter of the va_start() macro (subclause 7.16.1.4, paragraph 4):
The parameter parmN is the identifier of the rightmost parameter
in the variable parameter list in the function definition (the
one just before the ...). If the parameter parmN is declared with
the register storage class, with a function or array type, or with
a type that is not compatible with the type that results after
application of the default argument promotions, the behavior is
undefined.
Because bool is usually just 1 byte and therefore smaller than int (i.e.
the result of default argument promotion) its use as last argument before
... might result in undefined behavior. This theoretically can also
apply to enums as a compiler may use a smaller base type than int.
Since Clang 3.9 (currently in use on Travis by default) a warning is
issued about this, however, that version did not yet compare the actual
size of the argument's type, causing warnings where they are not
warranted (basically for all cases where enum types are used for the
last argument). This was apparently fixed with Clang 4.0, which only
warns about this use of bool with va_start(), which makes sense.
They now match the dh_constructor_t signature. This is a follow up for
the changes merged with b668bf3f9e and should fix use of MODP_CUSTOM on
Apple's ARM64 platform.
This causes problems e.g. on Android where we handle the alert (and
reestablish the IKE_SA) even though it usually is no problem if the
peer retries with the requested group. We don't consider it as a
failure on the initiator either.
If an IPsec SA is actually replaced with a rekeying its entry in the
manager is freed. That means that when the hard expire is triggered a
new entry might be found at the cached pointer location. So we have
to make sure we trigger the expire only if we found the right SA.
We could use SPI and addresses for the lookup, but this here requires
a bit less memory and is just a small change. Another option would be to
somehow cancel the queued job, but our scheduler doesn't allow that at
the moment.
Fixes#2399.
This fixes "packet too short" errors when parsing fragmented IPv4
packets and correctly determines the protocol in fragmented IPv6 packets.
Protocol headers are only parsed in unfragmented IPv4 and IPv6 packets,
or IPv4 first fragments.
Closesstrongswan/strongswan#80.
We don't attempt to parse the transport headers for fragments, not even
for the initial fragment (it's not guaranteed they contain the header,
depending on the number and type of extension headers).
In case we send retransmits for an IKE_SA_INIT where we propose a DH
group the responder will reject we might later receive delayed responses
that either contain INVALID_KE_PAYLOAD notifies with the group we already
use or, if we retransmitted an IKE_SA_INIT with the requested group but
then had to restart again, a KE payload with a group different from the
one we proposed. So far we didn't change the initiator SPI when
restarting the connection, i.e. these delayed responses were processed
and might have caused fatal errors due to a failed DH negotiation or
because of the internal retry counter in the ike-init task. Changing
the initiator SPI avoids that as we won't process the delayed responses
anymore that caused this confusion.
Caches CRLs in the app directory, adds support for OCSP, adds a button
to reconnect to the "already connected" dialog, only apply/configure app
selection on Android >= 5 (older versions don't support the API), and catches
some random exceptions.
If an interface is renamed we already have an entry (based on the
ifindex) allocated but previously only set the usable state once
based on the original name.
Fixes#2403.
There is a bug in some versions of lcov that causes it to fail writing
to files via relative paths after it issued warnings (e.g. due to
negative counts in the tracefile).
The generic field of size 0 in the union that was used previously
triggered index-out-of-bounds errors with the UBSAN sanitizer that's
used on OSS-Fuzz. Since the two family specific union members don't
really provide any advantage, we can just use a single buffer for both
families to avoid the errors.
Tobias Brunner