Af-packet/ips: Suricata process exits with segment fault (coredumped)

1

Suricata Version: 6.0.10

The config file we use:

%YAML 1.1
---
af-packet:
  - interface: antrea-l7-tap0
    threads: auto
    cluster-id: 80
    cluster-type: cluster_flow
    defrag: no
    use-mmap: yes
    tpacket-v2: yes
    checksum-checks: no
    copy-mode: ips
    copy-iface: antrea-l7-tap1
  - interface:  antrea-l7-tap1
    threads: auto
    cluster-id: 81
    cluster-type: cluster_flow
    defrag: no
    use-mmap: yes
    tpacket-v2: yes
    checksum-checks: no
    copy-mode: ips
    copy-iface: antrea-l7-tap0
multi-detect:
  enabled: yes
  selector: vlan

The extra config above which is included in /etc/suricata/suricata.yaml, and Suricata is started with command:

suricata -c /etc/suricata/suricata.yaml --af-packet

How to reproduce the issue:

  1. There is a client (assuming its IP is 10.10.0.1) and server (assuming its IP is 10.10.0.2), and the connections between the client the server are enforced to pass Suricata. Note that, the client and the server are in VLAN 1.
  2. Open a terminal the on client, run the command as following. The connections are expected to be passed after Suricata rules are added in subsequent steps.
for ((i=0;i<1000000;i++)) do curl http://10.10.0.2/api/v2/x;  done
  1. Open another terminal the on client, run the command as following. The connections are expected to be rejected after Suricata rules are added in subsequent steps.
for ((i=0;i<1000000;i++)) do curl http://10.10.0.2/api/v1/x;  done
  1. Add a tenant. Note that, DO NOT stop the command in Step 2 and 3.
  2. Add a config file /etc/suricata/antrea-tenant-1.yaml for the tenant as following:
%YAML 1.1

---
default-rule-path: /etc/suricata/rules
rule-files:
  - /etc/suricata/rules/antrea-l7-networkpolicy-1.rules
  1. Add a rule file /etc/suricata/rules/antrea-l7-networkpolicy-1.rules for the tenant as following:
reject ip any any -> any any (msg: "Reject by AntreaClusterNetworkPolicy:ingress-allow-http-request-to-api-v2"; flow: to_server, established; sid: 1;)
pass http any any -> any any (msg: "Allow http by AntreaClusterNetworkPolicy:ingress-allow-http-request-to-api-v2"; http.uri; content:"/api/v2/"; startswith; http.method; content:"GET"; sid: 2;)
  1. Register the tenant with the command as following:
suricatasc -c "register-tenant 1 /etc/suricata/antrea-tenant-1.yaml"
  1. Register the tenant handler with the command as following:
suricatasc -c "register-tenant-handler 1 vlan 1"
  1. After a few seconds, delete the tenant. Note that, DO NOT stop the command in Step 2 and 3.
  2. Unregister the tenant handler with the command as following:
suricatasc -c "unregister-tenant-handler 1 vlan 1"
  1. Unregister the tenant with the command as following:
suricatasc -c "register-tenant 1"
  1. Delete file /etc/suricata/antrea-tenant-1.yaml.
  2. Delete file /etc/suricata/rules/antrea-l7-networkpolicy-1.rules.
  3. Repeat Step 4 and Step 5 several times, stop at Step 4 finally, which means that the tenant is still there and corresponding rules take effect. Generally, the Suricata process will get Segment fault(coredumped) during repeating Step 4 and Step 5, or after stoping repeating for a while.
  4. If the Suricata process is still in good shape, stop the command in Step 2 and run it again for a while, the Suricata process might get Segment fault(coredumped) too.

Coredumped files.

I got two coredumped files and open it with gdb. We can see the the proccess is broken at this line: suricata/detect.c at 49713ebaa0b8edb057d60f1cfe9126946645a848 · OISF/suricata · GitHub

The value of det_ctx->non_pf_store_cnt should be modified unexpectedlly.

1892×1006 272 KB

1893×1022 277 KB

1 Like
2

Hi Hongliang Liu,

Thanks for this very detailed report! We noticed that you’ve also submitted a bug report. In time, we’ll look into it. :slight_smile:

3

Thanks a lot! Do you guys have any updates about this issue?

4

Not yet, but you’ll be able to track task progress from the redmine ticket you’ve reported :wink: