[Bug] Exporter panics on empty MQTT topic filter

[Megabyte135]

Hi! I've been reading through the telemetry pipeline as part of getting familiar with the codebase, and I noticed a parsing edge case that can crash the exporter.
Since the issue affects multiple parts of the repo, I’m opening an issue first. It seems to me that there’s room to improve the overall effectiveness of the parsing logic.

Description

mqtt_pit.c doesn't validate topicLength before copying the topic string. When a packet arrives with topicLength = 0, the C code sets topic[0] = '\0' and snprintf produces a message with a double space:

MQTT SUBSCRIBE  1

On the Go side, strings.Fields() collapses the double space, so the resulting slice has only 3 elements instead of the expected 4. The handler then tries to access fields[3] and this index is out of range.

runtime error: index out of range [3] with length 3

The panic kills listenForMetrics, and since there's no recover() wrapper, metric ingestion stops for all tarpits. The /metrics endpoint keeps serving though, so Prometheus continues scraping data with no alerts.

Steps to reproduce

Prerequisites: build the project with make all.

1. Start the exporter and MQTT tarpit

rm -f /tmp/tarpit_exporter.sock
GEO_DB=./prometheus/GeoLite2-Country.mmdb ./bin/prometheus_exporter &
./bin/mqtt_pit 11883 4096 5000 10000 50 4096 &

2. Verify metrics are alive

curl -s http://localhost:9101/metrics | grep "^mqtt_pit"

Expected: counters at 0.

3. Send a normal CONNECT followed by a SUBSCRIBE with topicLength=0

import socket, time

sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.connect(('127.0.0.1', 11883))

client_id = b'attacker'
vh = b'\x00\x04MQTT\x04\x00\x00\x3c'
payload = len(client_id).to_bytes(2, 'big') + client_id
remaining = vh + payload
sock.send(bytes([0x10, len(remaining)]) + remaining)
time.sleep(0.5)
sock.recv(1024) 

sock.send(bytes([0x82, 0x05, 0x00, 0x01, 0x00, 0x00, 0x01]))
time.sleep(1)
sock.close()

4. Observe the crash

curl -s http://localhost:9101/metrics | grep "^mqtt_pit"

The exporter log shows:

panic: runtime error: index out of range [3] with length 3

goroutine 25 [running]:
main.handleMetric(...)
    prometheus/main.go:220
main.listenForMetrics(...)
    prometheus/main.go:164

5. Verify the exporter process state

  ps aux | grep prometheus_exporter | grep -v grep                                               
  curl -s http://localhost:9101/metrics | head -5                                                

The prometheus_exporter process is dead. ps returns nothing, curl returns empty response (connection refused).

Impact

A single malformed packet kills all telemetry collection with no error indication.

Suggested fix

I'd be happy to submit a PR with following steps:

1. Add a len(fields) guard to the SUBSCRIBE/PUBLISH branches in handleMetric
2. Wrap handleMetric calls in recover() so one bad message can't take down the whole pipeline
3. Validate topicLength > 0 on the C side before building the message (since it's aligned with the protocol specification)

Is this approach okay?