Replace relevant uses of select() by poll(). poll(2) avoids the difficulties
we have with more than 1024 open file descriptors, and seems to be fairly
portable.
Fixes#757.
If multiple sockets are ready, we previously preferred the IPv4 non-NAT socket
over others. To handle all with equal priority, use a round-robin selection.
Instead of allocating MTU-sized buffers for each packet, read to a stack buffer
and copy to an allocation of the actual packet size. While it requires an
additional copy on non-Apple platforms, this should make allocation more
efficient for small packets.
Introduces options to enable concurrent Netlink queries. While this does not
make much sense on vanilla Linux, this can help on third party stacks to
increase throughput if longer latencies are to expect. Netlink message
retransmission can be optionally enabled if transmission is unreliable.
Non-socket based IKE bypass policies and other tweaks bring better compatibility
to third party stacks using Netlink.
As some backends over unreliable transport do not cache response messages,
retransmissions due the loss of responses perform the operation again. Add an
option to ignore some errors arising from such duplicate operations.
Note: This approach can't distinguish between real EXIST/NOTFOUND errors
and packet failures, and therefore is a source of race conditions and can't
detect any of these errors actually happening. Therefore that behavior is
disabled by default, and can be enabled with the ignore_retransmit_errors
strongswan.conf option.
To properly distinguish between real and retransmission errors, a Netlink
backend should implement retransmission detection using sequence numbers.
The socket based IKE bypass policies are usually superior, but not supported
on all networking stacks. The port based variant uses global policies for the
UDP ports we have IKE sockets for.
As under vanilla Linux the kernel can't handle parallel dump queries and returns
EBUSY, it makes not much sense to use them. Disable parallel queries by default
to basically restore original behavior, improving performance.
Besides that it can improve throughput, it avoids a deadlock situation. If
all threads are busy, watcher will invoke the FD notification for NEWADDR
events itself. If the lock is held, it gets locked up. As watcher is not
dispatching anymore, it can't signal Netlink socket send() completion, and
the send() operation does not return and keeps the lock.
If the kernel can't execute a Netlink query because a different query is already
active, it returns EBUSY. As this can happen now as we support parallel queries,
retry on this error condition.
Instead of locking the socket exclusively to wait for replies, use watcher
to wait for and read in responses asynchronously. This allows multiple parallel
Netlink queries, which can significantly improve performance if the kernel
Netlink layer has longer latencies and supports parallel queries.
For vanilla Linux, parallel queries don't make much sense, as it usually returns
EBUSY for the relevant dump requests. This requires a retry, and in the end
makes queries more expensive under high load.
Instead of checking the Netlink message sequence number to detect multi-part
messages, this code now relies on the NLM_F_MULTI flag to detect them. This
has previously been avoided (by 1d51abb7). It is unclear if the flag did not
work correctly on very old Linux kernels, or if the flag was not used
appropriately by strongSwan. The flag seems to work just fine back to 2.6.18,
which is a kernel still in use by RedHat/CentOS 5.
Fixes handling of invalid policies in end entity certificates by not rejecting
the full certificate, but just invalidating the affected policy. Additionally
adds a bunch of unit tests for the constraints plugin, and some minor fixes
to the nameConstraints handling.
Currently we still reject CAs that use invalid policy mapping; we should accept
such certificates and just invalid affected policies in a next iteration.
Fixes#453.
While RFC 5280 is not very specific about the matching rules of subjectAltNames,
it has some examples how to match email and FQDN constraints. We try to follow
these examples, and restrict DNS names to subdomain matching and email to
full email, host or domain matching.
Instead of rejecting the certificate completely if a certificate has a policy
OID that is actually not allowed by the issuer CA, we accept it. However, the
certificate policy itself is still considered invalid, and is not returned
in the auth config resulting from trust chain operations.
A user must make sure to rely on the returned auth config certificate policies
instead of the policies contained in the certificate; even if the certificate
is valid, the policy OID itself in the certificate are not to be trusted
anymore.
We explicitly avoided TryAcquireSRWLockExclusive() because of crashes. This
issue was caused by a MinGW-w64 bug (mingw-w64 fix 46f77afc). Using a newer
toolchain works fine.
While try_write_lock() obviously can fail, not supporting it is not really an
option, as some algorithms depend on occasionally successful calls. Certificate
caching in the certificate manager and the cred_set cache rely on successful
try_write_lock()ing.
When invoking /bin/sh, its default PATH is used. On some systems, that does
not include the PATH where the ipsec script is installed, as charon is invoked
with a custom PATH. Explicitly setting the PATH of charon should fix this
case, properly invoking the (default) updown script.
Fixes#745.
This fixes some vici test cases on OS X, where the test thread tries to cancel
the watcher thread during cleanup, but fails as select() does not honor the
pre-issued cancellation request.
As we want to avoid the libstrongswan include dependencies for libvici, avoid
the use of the bool type. Unfortunately this change may break the ABI for
vici_dump(). As this function is mostly for debugging purposes, we do it
nonetheless; my apologies if somebody already relies on the ABI stability of
that function.
Integrates the ext-auth plugin by Vyronas Tsingaras. The new child process
abstraction simplifies implementation in both the new ext-auth and the existing
updown plugin, and makes them available on the Windows platform.
Due to the expired and reissued holder certificates of carol and dave, new
attribute certificates are required to match the holder certificates serial in
the ikev2/acert-{cached,fallback,inline} tests.
Enable missing https:// support in the curl plugin by initializing libcurl
appropriately.
To initialize the SSL backend properly as required, we rely on our specific
crypto backends (openssl, gcrypt) that already provide this functionality.
Fixes#692.
On Ubuntu 12.04, there seems to be a resource leak related to pthread keys
when initializing glib or related libraries more than once. With our repeated
initialization for libstrongswan tests, we hit the following error:
Lib (gthread-posix.c): Unexpected error from C library during
'pthread_key_create': Resource temporarily unavailable.
The problem is not reproducible on a newer Gnome stack, hence we disable the
glib based soup plugin until we have a more recent Ubuntu on Travis.
To use SSL in curl, we need to initialize the SSL library in a thread-safe
manner and provide the appropriate callbacks. As we already do that in our
crypto plugins using these libraries, we depend on these features.
This implies that we need the same plugin enabled (openssl, gcrypt) as the
curl backend is configured to use to fetch from HTTPS URIs.
Some fetcher plugins (such as curl) might build upon OpenSSL to implement
HTTPS fetching. As we set (and can't unset) threading callbacks in our
openssl plugin, we must ensure that OpenSSL functions don't get called after
openssl plugin unloading.
We achieve that by loading curl and all other fetcher plugins after the base
crypto plugins, including openssl.
Martin Willi