Maintaining the reqid when recreating a CHILD_SA from scratch night not
strictly be necessary as we usually don't have to replace any temporary
states in the kernel. However, there could be concurrent acquires that
might actually make it necessary (we use the reqid to keep track of
acquires and it's also part of the duplicate check).
Keeping a reference ensures that if the old SA expires before the new
one is installed, the previous reqid isn't reallocated to a concurrently
established CHILD_SA with different selectors.
The reference is kept until the reqid is either confirmed (i.e.
re-allocated) or replaced by a different reqid, which happens only once
we know the final traffic selectors, or the SA is destroyed without
installing it.
Every reqid is allocated once, we don't store the same reqid with e.g.
different marks or interface IDs that would make it necessary to make
them part of the key in that table (that's different in the other table).
To preserve the current behavior, that is, allocating a new reqid if e.g.
the marks are different, the additional selector values (which will
result in an additional policy in the Linux kernel) are compared after
the initial lookup.
This allows preventing peers from authenticating with certificates
that are locally trusted, in particular, our own local certificate (which
safeguards against accidental reuse of certificates on multiple peers).
On the other hand, if this option is enabled, end-entity certificates
for peers can't be configured anymore explicitly (e.g. via remote.certs
in swanctl.conf).
If we initially selected a group the peer doesn't support (e.g. because
curve25519 is the first ECDH group provided by plugins), then found
a supported curve, we previously still instantiated a DH object for the
original group and might have formatted the parameters incorrectly.
Since the referenced commit, the state switch to IKE_ESTABLISHED doesn't
happen immediately after completing the authentication but only after
the remaining tasks have finished. Due to that redirects during IKE_AUTH
were treated like those during IKE_SA_INIT causing the IKE_SA and task
manager to get reset. However, unlike the ike-init task, the ike-auth
task returned SUCCESS and, therefore, got destroyed, while a different
task was removed from the array that was modified during the reset.
This later caused a dereference of the freed ike-auth task and a crash.
Fixes: 5ce1c91b58 ("ikev2: Trigger ike_updown() event after all IKE-specific tasks ran")
There is a relatively recent NIAP requirement to reject certificates with
ECDSA keys that explicitly encode the curve parameters (TD0527, Test 8b).
Since explicit encoding is pretty rare (if used at all and e.g. wolfSSL
already rejects it, by default), we should follow that requirement and
just reject such keys/certificates completely.
This currently can be enforced in all crypto plugins except when using
older versions of OpenSSL (< 1.1.1h) and Botan (< 3.2.0).
Closesstrongswan/strongswan#1949
References strongswan/strongswan#1796
These are not allowed in X.509 certificates according to RFC 5480 and
some newer validations apparently explicitly check for this.
Note that WolfSSL rejects such keys, by default. Only when compiled with
WOLFSSL_NO_ASN_STRICT are they accepted.
This works around an issue that occurs when recreating an IKE_SA with
multiple CHILD_SAs that use dynamically allocated reqids.
We currently try to preserve the reqid when reestablishing, so the
create-child task gets the reqid of the previous CHILD_SA and will try
to reallocate that once the CHILD_SA is installed. Before that, the old
CHILD_SA is destroyed and the reqid is released and added to the array
of reqids to get reused. However, because of the reverse order used here,
the first reqid in the array is the one of the last CHILD_SA.
So it can happen that a newly created CHILD_SA gets a reqid reassigned
that a later CHILD_SA will then also claim for itself and get assigned
because an entry with that reqid exists. So multiple CHILD_SAs with
different traffic selectors could then share the same reqid.
References strongswan/strongswan#1855
socket.connect() takes a single argument with the address that depends
on the address family, for TCP it's a tuple with IP and port.
Fixes: 00a75e332f ("vici: Create default TCP socket on Windows in Python bindings")
Closesstrongswan/strongswan#1874
Note that displaying the notification for the background service is
apparently not strictly necessary. So it's fine if the user wants to
hide it. That the service is running can still be seen in the task
manager (pull down the status drawer twice, there is a bullet with a number
at the bottom if the service is running).
Simply use the system dialog. If the user denies it twice, it won't show
up again. The explanation dialog would not show up the first time (i.e.
shouldShowRequestPermissionRationale() returns false), only once the user
denied the permission once. Currently seems like a bit much work
as we don't need the user to allow notifications.
Particularly on Samsung devices, the connect() call to dissolve the
previous connection on an existing socket via AF_UNSPEC does fail in
some situations with ECONNREFUSED:
[KNL] failed to disconnect socket: Connection refused
While creating a new socket is potentially a bit more overhead, this
should avoid the issue.
Closesstrongswan/strongswan#1691
When adding a virtual IP on a TUN interface, the interface might get
activated (in terms of receiving the event) after we've already set the
virtual flag for the added address. As the activation repopulates the
addresses on the interface, this cleared the flag and the address would
no longer be treated as virtual IP when installing routes for CHILD_SAs
that reference it in their local traffic selectors.
Closesstrongswan/strongswan#1807
Fixes an infinite loop if e.g. Netlink event sockets get too many
packets queued and poll() just returns POLLERR for the socket. Also
increases the default receive buffer size for Netlink sockets to better
support systems with lots of route updates.
Closesstrongswan/strongswan#1757
We can't actually explicitly listen for errors by passing POLLERR in
`events` (the man page for poll() clearly states it's ignored). On the
other hand, POLLERR can be returned for any FD and, even worse, it might
be the only event indicated.
The latter caused an infinite loop as we didn't notify the callback nor
clear the error by calling `getsockopt(..., SOL_SOCKET, SO_ERROR, ...)`.
And while the latter would be able to reset the state to break the loop,
it seems to leave the FD in a defunct state where no further events will
be returned by poll(). Notifying the callback works better (the error
is then reported by e.g. recvfrom()) and automatically happened already
if POLLERR was returned together with e.g. POLLIN.
So we now treat POLLERR like the other error indicators we handle (POLLHUP
and POLLINVAL) and just notify the callbacks.
Also simplify how we try to exceed the system-wide maximum. We basically
just try to force the value and simply fall back to the regular call.
The kernel actually won't let the latter fail if the value is too big,
it just caps it at the internal maximum.
We track all IKE_SA_INIT requests as half-open IKE_SAs but didn't
correctly untrack them if their message ID or responder SPI was non-zero.
References strongswan/strongswan#1775
Fixes: b866ee88bf ("ike: Track unprocessed initial IKE messages like half-open IKE_SAs")
As specified in RFC 7296, section 2.25:
The SA that the initiator attempted to rekey is indicated by the SPI
field in the Notify payload, which is copied from the SPI field in
the REKEY_SA notification.
So we copy that and the protocol verbatim.
The get_spi_data() method is currently not used, so that has been
simplified so it can be used for any protocol type and any SPI length.
Same for set_spi_data(), which is currently used for IKEv1 to encode
two SPIs.
The created XFRM interface was not actually used (no interface IDs on the
SAs, no routes via interface). It was basically treated like the dummy
TUN device. To actually install the routes via XFRM interface, we have
to create it before we install the SAs and policies, signal_ip_config()
happens too late. We also have to mark the ESP packets the same as IKE
the packets to avoid a routing loop if the server's IP is included in
the remote traffic selector (in particular if it's 0.0.0.0/0 or ::/0).
Fixes: 58f278f932 ("charon-nm: Use an XFRM interface if available")
The all-zero Ed25519 public key is rejected by botan_pubkey_check_key()
when the key is loaded.
Note that Botan 3 requires GCC 11 or CLANG 14, i.e. can't easily be built
on Debian bullseye or Ubuntu 20.04.
The thread-local storage function gets flagged via various botan FFI
functions when using Botan 3, whitelist that instead of all of them.
The services running on alice seem to require a bit more memory with
Debian bookworm, so increase the memory allocation. But at the same
time reduce winnetou's allocation by the same amount as it really doesn't
require that much memory.
The unit change makes it easier to read.
Because do-tests runs the restore-defaults script, fstab would get reset
to the default version and the mount point wouldn't be available anymore
after stopping and restarting the guests (unless the guest images were
rebuilt in between).
By default, rsyslog is not installed anymore to avoid storing everything
twice (since journald is the default). If this becomes an issue, we
could delete /var/log/journal to only log via rsyslog.
System-wide installation via pip isn't easily possible anymore on Debian
bookworm, so just use the Debian package for this (is available in old
releases as well).
Because libcrypto and libssl are measured, we need a new group for Debian
versions with OpenSSL 3 (I've rather added a suffix to the old group as
that could eventually get removed, although we might need a 3.1 variant
in the future - maybe we should measure some other files?).
OOM-killer is now already triggered with `import daemon`, so set the
limit before that. Also some PEP8 fixes (including an exclusion for
the above fix as that causes imports to not be at the beginning of the
file).
Without this, Git refuses to operate on the build dirs that are mounted
with weird ownership. When running as root in the chroot, Git checks
SUDO_UID, which won't match.
This is necessary because TKM can't read PKCS#8 files and in some
scenarios we don't have the pkcs8 plugin loaded that would be required
to read/decrypt the non-traditional files.
The wrapper called the command twice for any unit but "strongswan" and
it didn't return the correct exit code. This was noticed when an
if-updown script tried to check if systemd-resolved is active and always
succeeded, which caused failing attempts to configure it.
But now that the return code is correct, trying to enable bind9 won't
fail silently anymore if the unit doesn't exist (similar on older systems
for named), so this is adapted.
With Debian bookworm, the PQC KE sntrup761x25519-sha512 is negotiated, by
default. This increases the overhead significantly, in particular, the
size of the KE message, which wouldn't get through IPsec tunnels without
MSS clamping.
Since 17fd304e60 ("resolve: Don't install individual servers via
resolvconf"), DNS servers were sorted if getting installed via resolvconf.
In some setups the order might be important (even though relying on it
isn't a good idea in general as stub resolvers are free to use all of
the servers as they please).
The previous code was problematic if a certificate request for a known
but unrelated CA was received and the local trust chain was incomplete.
Due to the received anchor, the incomplete trust chain was dismissed and
any intermediate CA certificates were, therefore, not sent to the peer.
This new approach doesn't dismiss an incomplete trust chain, but prefers
one that can be resolved to a received anchor. If no such chain is found,
the first one is used.
This adds support for sending/receiving ESP packets without UDP
encapsulation to libipsec and kernel-libipsec. Only Linux is currently
supported and the feature is disabled by default.
This is currently only supported on Linux and with the appropriate
permissions.
Since it's experimental, it's disabled by default.
The log messages for each sent and received ESP message are logged in NET
like the ones in the socket-default plugin for UDP-encapsulated messages.
Tobias Brunner