packet loss prevention: a capture's packet count from its record
huntress ctf 2026, day 27 · metadata forensics · all three modes
tools used: archive and metadata readers and timed-attempt pipeline.
"packet captures, exports, and host artifacts describe the same incident from different angles. count traffic, reject malformed captures, and find the host narrative missing network evidence." day 27 opened a new file family for the lane: a pcap inside the evidence zip, and one question about it, its packet count.
the key insight is that the family was new to the questions, not to the readers: a pcap record (magic, version, snapshot length, link type, timestamp precision, packet count, first and last packet time) was already among the metadata readers' record families, and the question's words named one of its keys.
challenge shape
single 1 field the packet count of capture-001.pcap in the zip
time trial 32 fields the count on a fresh capture, the note and
roster fields, every earlier day's property, and
the sha-256 order field
mastery 352 fields the same 32 per set, over eleven sets
the field was d27-pcap-001-pcap-packet-count, "determine pcap.packet_count for capture-001.pcap", answer kind integer: "base-10 integer with no units, separators, or explanatory text".
artifact anatomy
a pcap is a 24-byte global header (magic, version, time zone, snapshot length, link type) followed by records, each a 16-byte header (seconds, microseconds, captured length, original length) and the captured bytes. the packet count is the number of records, walking from the first record to the end of the file by each record's captured length. the magic sets the byte order and the timestamp unit: d4 c3 b2 a1 little-endian microseconds, a1 b2 c3 d4 big-endian, 4d 3c b2 a1 and a1 b2 3c 4d the same two with nanoseconds. the mastery used all four, over ethernet, raw ip and 802.11 link types, and the single's capture used a user-defined link type no dissector decodes; none of the day's captures ended in a partial record.
extraction method
the pcap reader walks the records once and builds the record the metadata readers expose under the file's kind:
def pcap_record(data: bytes) -> dict:
magic = data[:4]
order = "<" if magic in (b"\xd4\xc3\xb2\xa1", b"\x4d\x3c\xb2\xa1") else ">"
major, minor, zone, _, snaplen, link = struct.unpack_from(
order + "HHiIII", data, 4
)
pos, count = 24, 0
while pos + 16 <= len(data):
_, _, incl_len, _ = struct.unpack_from(order + "IIII", data, pos)
count += 1
pos += 16 + incl_len # captured length, not original
return {"link_type": link, "packet_count": count, ...}
wireshark's capinfos -c capture-001.pcap prints the same number for any of the four magics. a question is answered from a record by its words: pcap-packet-count names the pcap record and its packet_count key, so the lane solver's record lookup returned the count with nothing written for the day.
validation and answer formatting
the count is a bare base-10 integer. the file's kind comes from its magic (the four above; pcapng has its own reader), and the walk steps by the captured length, so a frame cut short by the snapshot length is still one packet. no graded capture ended in a partial record, so the day never showed whether a record header with missing data counts; our walk counts it. the mastery's sets each carried their own capture and earlier days' files, read by the same lookup.
failed approaches and pivots
none on day 27.
automation across modes
the record lookup answered the single, the time trial and every mastery set; the mastery's solve took 1.8 seconds for eleven sets.
lessons
- a reader written before its first question pays off the morning the question arrives: the field's words are the lookup key.
- a pcap walk takes its byte order from the magic: the mastery mixed big- and little-endian captures, and a walk fixed to little-endian reads garbage lengths on the big-endian sets.
solve times
public scoreboard · day 27 finish: below top 10
- 0:03.8singlefirst blood, +0.51 s
- 0:06.4time trialfirst blood, +1.7 s
- 0:11.2masteryfirst blood, +1.1 s
- #38team day finishof 127 teams that cleared it, in 34:43.1
- #3single orderof the day's 8, by our flag time
from the public scoreboard: 09:00 et release to our flag; the table below counts from our attempt's start.
from the mode's first attempt start to its flag (server times). attempts counts our solver runs.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | 1.2s | 1 | 0 |
| time trial | 0.8s | 1 | 0 |
| mastery | 3.3s | 1 | 0 |
the clock
server timer of the winning attempt; used is attempt start to flag, left is what remained on the clock.
| mode | time available | time used | time left |
|---|---|---|---|
| time trial | 50.0s | 0.8s | 49.2s |
| mastery | 355.0s | 3.3s | 351.7s |
where we got stuck
- nowhere on day 27: the record reader existed and the question's words matched its key.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}