12.3. Exception Policies

Suricata has a set of configuration variables to indicate what should the engine do when certain exception conditions, such as hitting a memcap, are reached.

They are called Exception Policies and are configurable via suricata.yaml. If enabled, the engine will call them when it reaches exception states. Stats for any applied exception policies can be found in counters related to the specific configuration setting (read more). Some configuration is available directly via the stats settings.

For developers or for researching purposes, there are also simulation options exposed in debug mode and passed via command-line. These exist to force or simulate failures or errors and understand Suricata behavior under such conditions.

12.3.1. Exception Policies

12.3.1.1. Master Switch

It is possible to set all configuration policies via what we call "master switch". This offers a quick way to define what the engine should do in case of traffic exceptions, while still allowing for the flexibility of indicating a different behavior for specific exception policies your setup/environment may have the need to.

# Define a common behavior for all exception policies.
# In IPS mode, the default is drop-flow. For cases when that's not possible, the
# engine will fall to drop-packet. To fallback to old behavior (setting each of
# them individually, or ignoring all), set this to ignore.
# All values available for exception policies can be used, and there is one
# extra option: auto - which means drop-flow or drop-packet (as explained above)
# in IPS mode, and ignore in IDS mode. Exception policy values are: drop-packet,
# drop-flow, reject, bypass, pass-packet, pass-flow, ignore (disable).
exception-policy: auto

This value will be overwritten by specific exception policies whose settings are also defined in the yaml file.

12.3.1.1.1. Auto

In IPS mode, the default behavior for most of the exception policies is to fail close. This means dropping the flow, or the packet, when the flow action is not supported. The default policy for the midstream exception will be ignore if midstream flows are accepted.

It is possible to disable this default, by setting the exception policies' "master switch" yaml config option to ignore.

In IDS mode, setting auto mode actually means disabling the master-switch, or ignoring the exception policies.

Note

If no exception policy is enabled, Suricata will not log exception policy stats.

12.3.1.2. Specific settings

Exception policies are implemented for:

Exception Policy configuration variables
Config setting	Policy variable	Affects	Expected behavior
stream.memcap	memcap-policy	Flow or packet	If a stream memcap limit is reached, apply the memcap policy to the packet and/or flow.
stream.midstream	midstream-policy	Flow	If a session is picked up midstream, apply the midstream policy to the flow.
stream.reassembly.memcap	memcap-policy	Flow or packet	If stream reassembly reaches memcap limit, apply memcap policy to the packet and/or flow.
flow.memcap	memcap-policy	Packet	Apply policy when the memcap limit for flows is reached and no flow could be freed up. Policy can only be applied to the packet.
defrag.memcap	memcap-policy	Packet	Apply policy when the memcap limit for defrag is reached and no tracker could be picked up. Policy can only be applied to the packet.
app-layer	error-policy	Flow or packet	Apply policy if a parser reaches an error state. Policy can be applied to packet and/or flow.

To change any of these, go to the specific section in the suricata.yaml file (for more configuration details, check the suricata.yaml's documentation).

The possible values for the exception policies, and the resulting behaviors, are:

drop-flow: disable inspection for the whole flow (packets, payload, application layer protocol), drop the packet and all future packets in the flow.
drop-packet: drop the packet.
reject: same as drop-flow, but reject the current packet as well (see reject action in Rule's Action).
bypass: bypass the flow. No further inspection is done. Bypass may be offloaded.
pass-flow: disable payload and packet detection; stream reassembly, app-layer parsing and logging still happen.
pass-packet: disable detection, still does stream updates and app-layer parsing (depending on which policy triggered it).
ignore: do not apply exception policies (default behavior).

The drop, pass and reject are similar to the rule actions described in rule actions.

12.3.2. Exception Policies and Midstream Pick-up Sessions

Suricata behavior can be difficult to track in case of midstream session pick-ups. Consider this matrix illustrating the different interactions for midstream pick-ups enabled or not and the various exception policy values:

**Exception Policy Behaviors - IDS Mode**
Exception Policy	Midstream pick-up sessions ENABLED (stream.midstream=true)	Midstream pick-up sessions DISABLED (stream.midstream=false)
Ignore	Session tracked and parsed, inspect and log app-layer traffic, do detection.	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly.
Drop-flow	Not valid.*	Not valid.*
Drop-packet	Not valid.*	Not valid.*
Reject	Not valid.*	Session not tracked, flow REJECTED.
Pass-flow	Track session, inspect and log app-layer traffic, no detection.	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly.
Pass-packet	Not valid.*	Not valid.*
Bypass	Not valid.*	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly.
Auto	Midstream policy applied: "ignore". Same behavior.	Midstream policy applied: "ignore". Same behavior.

The main difference between IDS and IPS scenarios is that in IPS mode flows can be allowed or blocked (as in with the PASS and DROP rule actions). Packet actions are not valid, as midstream pick-up is a configuration that affects the whole flow.

**Exception Policy Behaviors - IPS Mode**
Exception Policy	Midstream pick-up sessions ENABLED (stream.midstream=true)	Midstream pick-up sessions DISABLED (stream.midstream=false)
Ignore	Session tracked and parsed, inspect and log app-layer traffic, do detection.	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly.
Drop-flow	Not valid.*	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly. Flow DROPPED.
Drop-packet	Not valid.*	Not valid.*
Reject	Not valid.*	Session not tracked, flow DROPPED and REJECTED.
Pass-flow	Track session, inspect and log app-layer traffic, no detection.	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly.
Pass-packet	Not valid.*	Not valid.*
Bypass	Not valid.*	Session not tracked. No app-layer inspection or logging. No detection. No stream reassembly. Packets ALLOWED.
Auto	Midstream policy applied: "ignore". Same behavior.	Midstream policy applied: "drop-flow". Same behavior.

Notes:

Not valid means that Suricata will error out and won't start.

REJECT will make Suricata send a Reset-packet unreach error to the sender of the matching packet.

12.3.3. Log Output

12.3.3.1. Flow Event

When an Exception Policy is triggered, this will be indicated in the flow log event for the associated flow, also indicating which target triggered that, and what policy was applied. If no exception policy is triggered, that field won't be present in the logs.

Note that this is true even if the policy is applied only to certain packets from a flow.

In the log sample below, the flow triggered the midstream policy, leading to Suricata applying the behavior that had been configured for such scenario: to pass the flow (pass_flow). It also did trigger the app_layer_error exception policy, but that is set up to ignore:

"flow": {
  "pkts_toserver": 4,
  "pkts_toclient": 5,
  "bytes_toserver": 495,
  "bytes_toclient": 351,
  "start": "2016-07-13T22:42:07.199672+0000",
  "end": "2016-07-13T22:42:07.573174+0000",
  "age": 0,
  "state": "new",
  "reason": "shutdown",
  "alerted": false,
  "action": "pass",
  "exception_policy": [
     {
       "target": "stream_midstream",
       "policy": "pass_flow"
     },
     {
       "target": "app_layer_error",
       "policy": "ignore"
     }
   ]
}

12.3.3.2. Available Stats

There are stats counters for each supported exception policy scenario that will be logged when exception policies are enabled:

**Exception Policy Stats Counters**
Setting	Counters
stream.memcap	exception_policy.tcp.ssn_memcap
stream.reassembly.memcap	exception_policy.tcp.reassembly.memcap
stream.midstream	exception_policy.tcp.midstream
defrag.memcap	exception_policy.defrag.memcap
flow.memcap	exception_policy.flow.memcap
app-layer.error	exception_policy.app_layer.error app_layer.error.exception_policy

If a given exception policy does not apply for a setting, no related counter is logged.

Stats for application layer errors are available in summarized form or per application layer protocol. As the latter is extremely verbose, by default Suricata logs only the summary. If any further investigation is needed, it is recommended to enable per-app-proto exception policy error counters temporarily (for more, read stats configuration).

12.3.4. Command-line Options for Simulating Exceptions

It is also possible to force specific exception scenarios, to check engine behavior under failure or error conditions.

The available command-line options are:

simulate-applayer-error-at-offset-ts: force an applayer error in the to server direction at the given offset.
simulate-applayer-error-at-offset-tc: force an applayer error in the to client direction at the given offset.
simulate-packet-loss: simulate that the packet with the given number (pcap_cnt) from the session was lost.
simulate-packet-tcp-reassembly-memcap: simulate that the TCP stream reassembly reached memcap for the specified packet.
simulate-packet-tcp-ssn-memcap: simulate that the TCP session hit the memcap for the specified packet.
simulate-packet-flow-memcap: force the engine to assume that flow memcap is hit at the given packet.
simulate-packet-defrag-memcap: force Suricata to assume memcap is hit when defragmenting specified packet.
simulate-alert-queue-realloc-failure: prevent the engine from dynamically growing the temporary alert queue, during alerts processing.

12.3.5. Common abbreviations

applayer: application layer protocol
memcap: (maximum) memory capacity available
defrag: defragmentation