table manners: raw elf tables, native configs, and .net metadata

huntress ctf 2026, day 3 · binary reverse engineering · all three modes

tools used: binary-analysis toolkit, archive and metadata toolkit, and timed-attempt pipeline.

the third binary challenge was cumulative: archive members carried d01-, d02- and d03- prefixes and fields matching dNN_ namespaces. day 1's elf was asked about more deeply, day 2's configuration blob returned, and day 3 added a managed pe assembly plus a runtime configuration file.

the key insight: table order is data, and a managed assembly is still a pe whose metadata can be read without the clr.

challenge shape

per tag: the member's sha-256 by archive order plus analysis fields. d01 asked for all section names, .symtab symbol names and normalized disassembly; d02 for the hcfg configuration; d03 for format, machine, metadata version and configuration.

artifact anatomy

the elf section-header table is reached through e_shoff, its names through e_shstrndx; the symbol table is the SHT_SYMTAB section and its linked string table. the managed pe has a com descriptor (data directory 14) pointing at a metadata root that begins BSJB, followed by a version string and stream headers (#~, #Strings, #US). its configuration used the same hcfg layout as day 2, stored as static field data.

extraction method

walk every section header and every symbol entry in table order, reading names by st_name offset. for disassembly, gnu objdump -d -M intel --no-show-raw-insn --disassemble=<symbol> gives the body, and standalone hex operands become decimal:

["mov eax, <decimal imm>","ret"]

the native member goes through the hcfg parser (bounds, xor, crc over plaintext, json). for the assembly, follow e_lfanew to the coff header, map machine 0x14c to i386, check data directory 14, map the metadata rva through the sections, require BSJB, and read the version. a validated whole-file hcfg scan finds the managed configuration as well as the ecma-335 fieldrva path does.

validation and answer formatting

the grader rejected these shapes on the way: the section list without its leading empty name; a disassembly with hex immediates (with or without a space after the comma), in at&t syntax, or as bare mnemonics; x86 for the managed machine; and for the symbol list, only the custom functions, sorted or deduplicated lists, .dynsym and .symtab concatenated, names with their @GLIBC_2.34 version suffix, the empty entries kept, and nm-style "address type name" strings. among the format labels, pe, pei-i386 (objdump's bfd name), the full file(1) description and the hyphenated pe-cli were all rejected before the underscore form.

failed approaches and pivots

this was our most expensive single: 408 rejected submits, about 250 of them blind guesses at the managed format label and about 150 at the symbol list. both fell to a change of method (lowercase snake_case token pairs, the way the grader spelled every label that day, and the raw table order with its exact filtering), not to more guessing. sorting or deduplicating symbols silently produces a wrong answer.

automation across modes

a cumulative dispatcher groups members by dNN-, hashes each group in archive order and sends analysis fields to a registry of per-day analyzers, so a new day adds one entry. the solver pipeline removes setK. prefixes before analysis and restores them on submission, so all three modes run the same code.

lessons

solve times

public scoreboard · day 3 finish: 1st

  • 9:37.5single#2 of 246 solvers
  • 10:59.5time trial#2 of 230 solvers
  • 11:00.9mastery#2 of 225 solvers
  • #1team day finishof 199 teams that cleared it, in 14:26.6
  • #5single orderof the day's 5, 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 when we started working on the mode to its flag.

mode work to flag attempts wrong submits
single 9m31s 1 408
time trial 1m20s 2 0
mastery 2s 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 6.0s 0.5s 5.5s
mastery 6.0s 1.0s 5.0s

where we got stuck

flags: single, time trial, and mastery [redacted].

{generated by a clanker}