The previous code had some issues because it handled each child config
separately. Not only was this quite inefficient because all IKE_SAs had
to be enumerated for every config, it also caused problems with the check
for other CHILD_SAs in order to decide whether to delete the IKE_SA or
not. Because CHILD_SAs are deleted with an INFORMATIONAL exchange, they
are not immediately gone. This caused a race condition and with more
than one child config and SAs the IKE_SA could be kept because it
could appear as if other, unrelated CHILD_SAs were still there.
Another race condition, which is fixed by the previous commit, occurred
when only changing child configs. Then it could happen that the code
deemed the IKE_SA empty and a delete for it was queued. If that happened
while the IKE_SA was deleting one of the CHILD_SAs (or was busy with some
other exchange), the IKE_SA was not switched to IKE_DELETING. So it
looked usable and create-child tasks for the updated configs might have
gotten queued. Unfortunately, once the ike-delete task is eventually
executed, these tasks would be gone and the replacement CHILD_SAs never
created. This commit additionally avoids actually deleting the IKE_SA
even if all child configs change or get removed if any new CHILD_SAs are
to be initiated.
The IKE_SA might be busy with a different task while a request to
terminate it is getting queued, we don't want to use such an IKE_SA to
initiate new CHILD_SAs as these tasks will get lost once the IKE_SA is
terminated.
The empty array of rules for `assert_message_empty()` and the resulting
size 0 triggers warnings like these:
allocation of insufficient size '0' for type 'listener_message_rule_t' with size '12'
Using calloc() with `nmemb` set to 0 triggers the same warning.
The number of elements is the first argument, their size the second.
The previous code triggered the following warning:
'calloc' sizes specified with 'sizeof' in the earlier argument and not in the later argument
Instead of just adding the offset internally, this way the reported
base address is always the first assignable address (e.g. for
192.168.0.0/24 vs. 192.168.0.1/24).
Closesstrongswan/strongswan#2264
If a base address is configured, we don't expect the pool to be empty,
so reject the creation (e.g. with the broadcast address as base).
References strongswan/strongswan#2205
This is a patch from the OpenWrt package sources necessary to adapt to
changes from 2008 that abstracted the option datatype (added a list
type).
Signed-off-by: Noel Kuntze <[email protected]>
Reviewed-by: Philip Prindeville <[email protected]>
musl's headers define a lot of networking structs. For some, the
definition in the Linux UAPI headers is then suppressed by e.g.
__UAPI_DEF_ETHHDR.
Since we included musl's net/ethernet.h, which includes netinet/if_ether.h
that defines `struct ethhdr` (and the above constant), **after** we
include linux/if_ether.h, there was a compilation error because the
struct was defined multiple times.
However, simply moving that include doesn't fix the problem because for
ARP-specific structs the Linux headers don't provide __UAPI_DEF* checks.
So instead of directly including the linux/ headers, we include those
provided by the C library. For glibc these usually just include the
Linux headers, but for musl this allows them to define the struct
directly. We also need to move if.h and add packet.h, which define
other structs (or include headers that do so) that we use.
Fixes: 187c72d1af ("dhcp: Port the plugin to FreeBSD/macOS")
The previous option caused such requests to be enabled if not explicitly
disabled, which only the vici plugin did, for all other backends requests
would have been sent.
References strongswan/strongswan#2016
Only the vici plugin previously set OCSP_SEND_REPLY explicitly, all other
backends would have defaulted to OCSP_SEND_BOTH.
References strongswan/strongswan#2016
In particular for static code analyzers. The previous nesting of case
statements inside of a while loop that's inside a switch statement and
a wrapping block with declaration was quite weird and Coverity didn't
like it (it figured that `type` was uninitialized even when it assumed
that get_type() returned a known type for which a case statement
existed).
This reverts commit 8b9b11919d.
Since ESN was negotiated via proposal, just configuring the SA without
ESN won't work as the ICV will be incorrect if the peer enabled ESN
on its SA. While the Linux kernel currently doesn't support disabling
replay protection for SAs that use ESN, this at least gets users an
explicit error not just dropped packets, and it will automatically work
if the kernel supports this combination at some point.
References strongswan/strongswan#2117
This also refactors the BPF handling so it can be shared between the
dhcp and farp plugins. The latter is adapted accordingly.
Closesstrongswan/strongswan#2047
Co-authored-by: Tobias Brunner <[email protected]>
CHILD_SA unique identifier allocation starts at 1. If the counter overflows,
a unique ID of 0 is assigned to an CHILD_SA, which may have unclear
consequences.
Overflowing the unique ID counter is theoretical for most setups, but on
a Gateway terminating 100'000 tunnels and rekeying CHILD_SAs every 60s
overflows the counter after a month uptime. So avoid a 0 unique identifier
by using ref_get_nonzero().
IKE_SA unique identifier allocation starts at 1. If the counter overflows,
a unique ID of 0 is assigned to an IKE_SA, which may have unclear consequences.
Overflowing the unique ID counter is theoretical for most setups, but on
a Gateway terminating 100'000 tunnels and rekeying the IKE_SA every 60s
overflows the counter after a month uptime. So avoid a 0 unique identifier
by using ref_get_nonzero().
A refcount variable is used to allocate sequential unique identifiers for
Netlink sequence numbers, subject to overflows. The risk of an overflow
has so far not been considered practical, as it requires 2^32 netlink
requests.
It seems that this issue is not only theoretical. A host with thousands
of tunnels doing aggressive rekeying and/or aggressive status checking
(via vici list-sas) may trigger the overflow after a few weeks uptime.
The consequences are rather devastating: Once the refcount overflows, a
Netlink request is sent with sequence number 0. This request is answered
by the kernel, but can't be matched to the request, resulting in the error:
"received unknown netlink seq 0, ignored". Without Netlink timeouts, the
thread indefinitely waits for a response while holding the Netlink mutex,
bringing all threads to a halt.
So at all costs avoid zero sequence numbers. Also, start at sequence number
1 instead of the arbitrary 201, so the same range is used on start and after
an overflow.
The kernel includes the XFRMA_REPLAY_ESN_VAL attribute when dumping
SAs since it was added with 2.6.39. So we basically added this attribute
twice to the message sent to the kernel, potentially exceeding the
message buffer if the window size is large.
The XFRMA_REPLAY_VAL attribute is only dumped since 3.19, so that might
still be relevant (Google seems to maintain a 3.18 kernel) and since we
have to query the current lifetime stats anyway, we can just avoid adding
this attribute twice.
Closesstrongswan/strongswan#1967
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.
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")
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
Thomas Egerer