obfuskator/crates/noxis-passes/src/strings/mod.rs
loki5512344 139aea9a8f
frames,jar,passes: verifier-driven hardening from real jars
Proven by paranoid runs over lorep (17 classes, fully green),
LoVisual (4081 classes) and oc2r (1278 classes) with per-class
java -Xverify:all, each fix first reproduced as a red test:

- merge: Top absorbs instead of Top-wins (javac declares Top for
  slots unwritten on some inflow; precise types reject those edges).
  Regression: loop_head_keeps_top_for_loop_local.
- emit: collapse [Long|Double, Top] pairs on operand stacks
  (javac writes one entry; the pair is rejected). Regression:
  wide_stack_join_emits_single_entries.
- xfer ArrayLoad: push the wide pair, fixing lcmp underflow.
  Regression: array_load_wide_pushes_pair.
- xfer <init>: convert only the receiver's own allocation site
  instead of every Uninitialized (nested construction clobbered
  the outer object). Regression: nested_construction_keeps_outer.
- xfer ANewArray: array operands gain one dimension ([B] -> [[B)
  instead of an illegal L-wrap ([L[B;). Covered by the same test.
- jar load: duplicate class names (multi-release module-info
  twins) keep the first occurrence instead of crashing save.
  Test: duplicate_class_names_keep_first.
- strings/constants: skip methods past MAX_METHOD_INSNS (8000)
  so giants cannot overflow the 64KB writer ceiling. Tests:
  oversized_methods_are_skipped in both passes.
- merge errors now name both types being merged.

Also: em-dash removed repo-wide (724 spots, rule added to
todo.md), allow-lints only in tests (scratch example deleted),
code comments migrated to English in every file this change
touches.
2026-10-09 13:47:27 +02:00

205 lines
7.8 KiB
Rust

//! The strings pass: `Const::String` becomes runtime-decoder calls.
//!
//! The scheme rotates per string (`salt % n`), the salt is deterministic
//! (KDF or RNG stream - see `derive_key`). The decoder is generated once per
//! program and only if at least one string was encrypted.
mod crypt;
#[cfg(test)]
mod tests;
use std::collections::BTreeMap;
use noxis_analysis::ClassHierarchy;
use noxis_classfile::{Const, Insn, MAX_METHOD_INSNS, Program, compute_frames};
use noxis_core::{Context, CoreError, NameStyle, Pass, PassId, PassReport, derive_seed};
use noxis_runtime::{Scheme, decoder_class, decoder_owner, encrypt};
use crate::namegen::NameGenerator;
use crypt::{decode_sequence, salt_for_scheme, split_chunks};
/// The string encryption pass.
#[derive(Debug, Default, Clone, Copy)]
pub struct StringsPass;
impl Pass for StringsPass {
fn id(&self) -> PassId {
PassId::Strings
}
fn apply(&self, ctx: &mut Context, program: &mut Program) -> Result<PassReport, CoreError> {
let cfg = ctx.config.strings.clone();
let mut schemes = Vec::with_capacity(cfg.schemes.len());
for name in &cfg.schemes {
schemes.push(Scheme::parse(name).map_err(|e| pass_error(e.to_string()))?);
}
if schemes.is_empty() {
return Err(pass_error("no encryption schemes"));
}
// Phase 1: collect every encryptable string (position + text).
let mut targets: Vec<(usize, usize, usize, String)> = Vec::new();
let mut skipped = 0usize;
for (ci, class) in program.classes.iter().enumerate() {
for (mi, method) in class.methods.iter().enumerate() {
let Some(code) = method.code.as_ref() else {
continue;
};
for (ii, insn) in code.insns.iter().enumerate() {
if let Insn::Const(Const::String(text)) = insn {
if text.chars().count() >= cfg.min_len && !is_skipped(text, &cfg.skip) {
targets.push((ci, mi, ii, text.clone()));
} else {
skipped += 1;
}
}
}
}
}
if targets.is_empty() {
return Ok(report(
0,
skipped,
vec!["no strings to encrypt".to_string()],
));
}
// Phase 2: encrypt and substitute call sequences.
let master = ctx.master_seed;
let mut rng = ctx.rng("strings");
let owner = decoder_name(ctx, program);
let mut chunk_counter = 0usize;
let mut salt_counter = 0u64;
let mut replacements: BTreeMap<(usize, usize, usize), Vec<Insn>> = BTreeMap::new();
for (ci, mi, ii, text) in &targets {
let mut parts = Vec::new();
for chunk in split_chunks(text, cfg.split_above) {
let scheme = chunk_counter % schemes.len();
let raw = next_salt(&mut rng, master, &mut salt_counter, cfg.derive_key);
let salt = salt_for_scheme(raw, scheme, schemes.len());
let blob = encrypt(schemes[scheme], master, salt, &chunk);
parts.push((blob.cipher, salt));
chunk_counter += 1;
}
replacements.insert((*ci, *mi, *ii), decode_sequence(&owner, &parts));
}
let mut mutated: Vec<(usize, usize)> = Vec::new();
let mut applied = 0usize;
let mut per_method: BTreeMap<(usize, usize), usize> = BTreeMap::new();
for (ci, mi, _, _) in &targets {
*per_method.entry((*ci, *mi)).or_default() += 1;
}
for (ci, class) in program.classes.iter_mut().enumerate() {
for (mi, method) in class.methods.iter_mut().enumerate() {
let Some(code) = method.code.as_mut() else {
continue;
};
let original = std::mem::take(&mut code.insns);
let method_targets = per_method.get(&(ci, mi)).copied().unwrap_or(0);
let mut fresh = Vec::with_capacity(original.len());
let mut touched = false;
for (ii, insn) in original.iter().enumerate() {
if let Some(sequence) = replacements.remove(&(ci, mi, ii)) {
fresh.extend(sequence);
touched = true;
} else {
fresh.push(insn.clone());
}
}
if fresh.len() > MAX_METHOD_INSNS {
// Would risk the 64KB writer ceiling: keep the original
// code (and its strings) untouched.
code.insns = original;
skipped += 1;
continue;
}
code.insns = fresh;
if touched {
applied += method_targets;
code.max_stack = code.max_stack.saturating_add(2);
mutated.push((ci, mi));
}
}
}
// Phase 3: one decoder class per program.
let scheme_names: Vec<&str> = cfg.schemes.iter().map(String::as_str).collect();
let decoder =
decoder_class(&owner, master, &scheme_names).map_err(|e| pass_error(e.to_string()))?;
program.classes.push(decoder);
ctx.note(format!("string decoder: {owner}"));
// Phase 4: frames of the touched methods.
let hierarchy = ClassHierarchy::build(program);
for (ci, mi) in &mutated {
let (name, snapshot) = {
let class = program
.classes
.get(*ci)
.ok_or_else(|| pass_error("class vanished"))?;
(
class.this_class.clone(),
class
.methods
.get(*mi)
.cloned()
.ok_or_else(|| pass_error("method vanished"))?,
)
};
let frames = compute_frames(&name, &snapshot, &hierarchy)
.map_err(|e| pass_error(e.to_string()))?;
program
.classes
.get_mut(*ci)
.and_then(|c| c.methods.get_mut(*mi))
.and_then(|m| m.code.as_mut())
.ok_or_else(|| pass_error("method vanished"))?
.frames = frames;
}
Ok(report(
applied,
skipped,
vec![format!("strings encrypted: {applied}, decoder: {owner}")],
))
}
}
fn report(transformed: usize, skipped: usize, notes: Vec<String>) -> PassReport {
PassReport {
pass: PassId::Strings,
transformed,
skipped,
notes,
}
}
fn pass_error(msg: impl Into<String>) -> CoreError {
CoreError::pass(PassId::Strings, msg)
}
/// A string matches the `skip` patterns of the config.
fn is_skipped(text: &str, skip: &[String]) -> bool {
skip.iter().any(|pattern| text.contains(pattern))
}
/// A chunk salt: KDF from the position or an RNG stream (see `derive_key`).
fn next_salt(rng: &mut noxis_core::Rng, master: u64, counter: &mut u64, derive: bool) -> u64 {
let value = if derive {
derive_seed(master, "strings/salt", *counter)
} else {
rng.next_u64()
};
*counter += 1;
value
}
/// A collision-free deterministic decoder name.
fn decoder_name(ctx: &mut Context, program: &Program) -> String {
let mut names = NameGenerator::new(ctx.rng("strings/decoder"), NameStyle::ShortLatin);
for class in &program.classes {
names.reserve(class.this_class.clone());
}
loop {
let owner = decoder_owner(&format!("{}/{}", names.next_name(), names.next_name()));
if program.class(&owner).is_none() {
return owner;
}
}
}