Due to an unprotected incrementation, two load-tester initiators occasionally
use the same SPI under high load, and hence generate 2 IPsec SAs with the same
identifier. The responder IPsec stack will refuse to configure the second SA.
Use an atomic incrementation to avoid this race condition.
Signed-off-by: Christophe Gouault <christophe.gouault@6wind.com>
Due to an unprotected incrementation, two load-tester initiators occasionally
use the same identifier under high load. The responder typically drops one of
the connections.
Use an atomic incrementation to avoid this race condition.
Signed-off-by: Christophe Gouault <christophe.gouault@6wind.com>
This patch is based on one by Christoph Gouault.
Currently, to count the total number of half_open IKE_SAs,
get_half_open_count sums up the count of each segment in the SA hash
table (acquiring a lock for each segment). This procedure does not scale
well when the number of segments increases, as the method is called for
each new negotiation.
Instead, lets maintain a global atomic counter.
This optimization allows the use of big values for charon.ikesa_table_size
and charon.ikesa_table_segments.
These are available since GCC 4.7 and will eventually replace the __sync
operations. They support the memory model defined by C++11. For instance,
by using __ATOMIC_RELAXED for some operations on the reference counters we
can avoid memory barriers, which are required by __sync operations (whose
memory model essentially is __ATOMIC_SEQ_CST).
On many architectures it is safe to read the value directly (those
using cache coherency protocols, and with atomic loads for 32-bit
values) but it is not if that's not the case or if we ever decide to
make refcount_t 64-bit (load not atomic on x86).
So make sure the operation is actually atomic and that users do not
have to care about the size of refcount_t.
- get_cpi function was implemented to retrieve a CPI from the kernel.
- add_sa/update_sa/del_sa were updated to accommodate for IPComp SA.
- Updated add_policy_internal to update the SPD to support IPComp.
If one peer starts reauthentication by deleting the IKE_SA, while the other
starts CHILD_SA rekeying, we run in a race condition. To avoid it, temporarily
reject the rekey attempt while we are in the IKE_SA deleting state.
RFC 4306/5996 is not exactly clear about this collision, but it should be safe
to reject CHILD_SA rekeying during this stage, as the reauth will re-trigger the
CHILD_SA. For non-rekeying CHILD_SA creations, it's up to the peer to retry
establishing the CHILD_SA on the reauthenticated IKE_SA.
The extensions and conditions apply to the rekeyed IKE_SA as well, so we should
migrate them. Especially when using algorithms from private space, we need
EXT_STRONGSWAN to properly select these algorithms during IKE rekeying.
Even in Main Mode, some Sonicwall boxes seem to send ID/HASH payloads in
unencrypted form, probably to allow PSK lookup based on the ID payloads. We
by default reject that, but accept it if the
charon.accept_unencrypted_mainmode_messages option is set in strongswan.conf.
Initial patch courtesy of Paul Stewart.
Before this change a reqid set on the create_child_t task was used as
indicator of the CHILD_SA being rekeyed. Only if that was not the case
would the local traffic selector be changed to 0.0.0.0/0|::/0 (as we
don't know which virtual IP the gateway will eventually assign).
On the other hand, in case of a rekeying the VIP is expected to remain
the same, so the local TS would simply equal the VIP.
Since c949a4d501 reauthenticated CHILD_SAs also have the reqid
set. Which meant that the local TS would contain the previously
assigned VIP, basically rendering the gateway unable to assign a
different VIP to the client as the resulting TS would not match
the client's proposal anymore.
Fixes#553.
Prevents a responder peer to trick us into established state by starting
IKE_SA rekeying before the IKE_SA has been authenticated during IKE_AUTH.
Fixes CVE-2014-2338.
This is undefined behavior as per the C99 standard (sentence 1185):
"If the value of the right operand is negative or is greater or equal
to the width of the promoted left operand, the behavior is undefined."
Apparently shifts may be done modulo the width on some platforms so
a shift by 32 would not shift at all.
Christophe Gouault