Program Load Failure When Mixing Rate-Limited and Non-Rate-Limited kprobes

[justin0u0]

What happened?

After upgrading to v1.3.0, I observed that applying kprobes with both rate-limited and non-rate-limited actions in a single kprobe TracingPolicy fails.

The tracing policy is defined as follows:

description: "Detects privileges raising operations"
kprobes:
  - call: "__sys_setuid"
    message: "Privileged operation setuid to root"
    syscall: false
    args:
    - index: 0
      type: "int"
    selectors:
    - matchArgs:
      - index: 0
        operator: "Equal"
        values:
        - "0"
      matchActions:
      - action: Post
        rateLimit: "1m"
  - call: "__sys_setgid"
    message: "Privileged operation setgid to root"
    syscall: false
    args:
    - index: 0
      type: "int"
    selectors:
    - matchArgs:
      - index: 0
        operator: "Equal"
        values:
        - "0"
      matchActions:
      - action: Post
        # rateLimit: "1m"

By either setting rateLimit for both or removing rateLimit from both matchActions, the program loads successfully.

Tetragon Version

v1.3.0

Kernel Version

Linux 5.15.0-118-generic x86_64

Kubernetes Version

v1.30.3

Bugtool

No response

Relevant log output

time="2024-12-23T03:59:05Z" level=info msg="adding tracing policy" info="privileges-raise (object:1/7fc86724-664b-437d-8580-8bb01820ec23) (type:/)" name=privileges-raise
time="2024-12-23T03:59:06Z" level=info msg="Added kprobe" function=__sys_setuid override=false return=false
time="2024-12-23T03:59:06Z" level=info msg="Added kprobe" function=__sys_setgid override=false return=false
time="2024-12-23T03:59:06Z" level=info msg="Added generic kprobe sensor: /var/lib/tetragon/bpf_generic_kprobe_v511.o -> __sys_setuid" override=false
time="2024-12-23T03:59:06Z" level=info msg="Added generic kprobe sensor: /var/lib/tetragon/bpf_generic_kprobe_v511.o -> __sys_setgid" override=false
time="2024-12-23T03:59:06Z" level=info msg="BTF file: using metadata file" metadata=/sys/kernel/btf/vmlinux
time="2024-12-23T03:59:06Z" level=info msg="Loading sensor" name=generic_kprobe
time="2024-12-23T03:59:06Z" level=info msg="Loading kernel version 5.15.143"
time="2024-12-23T03:59:09Z" level=warning msg="incompatible map found" error="expected max entries 1, got 32768: map spec is incompatible with existing map" map-name=ratelimit_map path=/sys/fs/bpf/tetragon/privileges-raise/generic_kprobe/ratelimit_map
time="2024-12-23T03:59:09Z" level=warning msg="will delete and recreate" map=ratelimit_map
time="2024-12-23T03:59:09Z" level=warning msg="adding tracing policy failed" error="sensor generic_kprobe from collection privileges-raise failed to load: failed prog /var/lib/tetragon/bpf_generic_kprobe_v511.o kern_version 331663 loadInstance: opening collection '/var/lib/tetragon/bpf_generic_kprobe_v511.o' failed: using replacement map ratelimit_map: MaxEntries: 1 changed to 32768: map spec is incompatible with existing map"

Anything else?

Upon investigation, I found that in PR #2128's commit d7a30b3, the ratelimit_map was changed to a per-sensor map. This modification causes the ratelimit_map to be shared among programs, but rate-limited and non-rate-limited programs expect different MaxEntries values. As a result, the program load fails due to map specification incompatibility.