bridge troll: a managed module at the p/invoke crossing
huntress ctf 2026, day 22 · binary · all three modes
tools used: binary analysis and timed-attempt pipeline.
"managed code on one side, native code on the other, and a bridge of p/invoke and ffi in between. pay the troll: trace every crossing and map what goes over." the day's new tag (d22) is a .net module; the prompt describes its bridge into native code, and the questions ask the module's identity and the configuration it carries, beside the earlier days' tags.
the key insight: a managed pe is still a pe. its format and its runtime version come from two small structures the pe points at: the cli header (data directory 14) and the metadata root, whose version string is the runtime the module was built against.
challenge shape
single 264 fields 22 tags x 12 fields: member sha-256s, then
each tag's analysis fields
time trial 264 fields the same, fresh bytes
mastery 1584 fields the same over six sets
the d22 fields: three member hashes, format, machine, metadata_version, the configuration block's campaign, domain, mutex, url and crc32_hex, and hcfg_version.
artifact anatomy
the evidence zip's d22 members are a managed dll (d22--sample.dll, a pe with a cli header, written in f#), its runtime configuration file (d22--sample.runtimeconfig.json) and the f# core library it references (d22--FSharp.Core.dll); the earlier tags' members sit beside them, and the zip is password-protected like every binary bundle. the managed module's metadata root starts with the BSJB signature, then a length-prefixed version string. the configuration block is the same HCFG structure the lane has read since its first days.
cli_rva, _ = data_directories[14] # IMAGE_DIRECTORY_ENTRY_COM_DESCRIPTOR
meta_rva = struct.unpack_from("<I", pe, rva_to_off(cli_rva) + 8)[0]
root = rva_to_off(meta_rva)
assert pe[root:root + 4] == b"BSJB"
n = struct.unpack_from("<I", pe, root + 12)[0]
version = pe[root + 16:root + 16 + n].rstrip(b"\0").decode()
extraction method
every analysis question names its member the same way: "use the unique d22-- primary managed assembly whose basename starts d22-- and ends .dll; .runtimeconfig.json and dependency dll members are delivery support only." two members end in .dll, so the primary one is the sample, not the f# core library it depends on.
the binary analysis producers read each tag's fields from its own format: the pe's file header gives the machine (0x14c, graded as i386), the cli header makes the format the managed pe label (pe_cli, from the enum elf, pe, pe_cli, jar, wasm, neko_bytecode), the metadata root gives the version, and the configuration block is located by its magic in the sample dll. the version question spells the read: "read the clr metadata-root version string at bsjb + 16 using its declared length, remove trailing nul bytes, and submit the exact text"; it graded v4.0.30319, the clr 4 metadata version, in every set.
validation and answer formatting
hashes are taken over the exact member bytes in central-directory order and written as "exactly 64 lowercase hexadecimal characters; no prefix"; labels come from the question's enum; the configuration's fields are answered as the block stores them, the crc32 as 0x plus eight lowercase hex digits and the hcfg version as a base-10 integer.
failed approaches and pivots
none on day 22: the new tag was read by producers that already existed.
automation across modes
the same producers answer every field of every set.
lessons
- a "new" platform often reuses an old container: find the container's own pointer to the new structure before writing a parser.
solve times
public scoreboard · day 22 finish: below top 10
- 0:09.0singlefirst blood, +6.5 s
- 0:15.7time trialfirst blood, +11.3 s
- 0:25.3masteryfirst blood, +32.0 s
- #26team day finishof 145 teams that cleared it, in 10:39.5
- #5single orderof the day's 7, 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 start on the game master to its correct submit.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | 7.2s | 1 | 0 |
| time trial | 5.0s | 1 | 0 |
| mastery | 7.7s | 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 | 80.0s | 5.0s | 75.0s |
| mastery | 400.0s | 7.7s | 392.3s |
where we got stuck
- nowhere on day 22.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}