Skip to content

Building 70 Projects ranging from beginner to advanced so anyone can β€” learn from, build upon, use as a reference, or even copy directly. Owned by my cybersecurity company πŸ‘‡

docs(lisdex): name the stream-driven scope boundary instead of failing quietly

CarterPerez-devd0636816be52
diff --git a/PROJECTS/advanced/zero-day-vulnerability-scanner/README.md b/PROJECTS/advanced/zero-day-vulnerability-scanner/README.md
index d79b0de0..1ab4c9ae 100644
--- a/PROJECTS/advanced/zero-day-vulnerability-scanner/README.md
+++ b/PROJECTS/advanced/zero-day-vulnerability-scanner/README.md
@@ -85,6 +85,15 @@ These are deliberate, and they are the reason to trust the rest of the output.
parameter blocked it.
- **It will not report a planted bug.** Crashes from a benchmark-instrumented
build are stamped NOT DISCLOSABLE with no filing route.
+- **It will not invent a stream for a library that never takes bytes as an
+ argument.** lisdex drives entry points that receive the input as a
+ *parameter*. Some libraries only accept input the caller has already
+ assembled into a stream: openjpeg wants read, seek and skip callbacks over
+ its own user-data struct, and libtiff's own fuzz target opens a C++
+ `std::istream` through `TIFFStreamOpen`. Neither shape is a parameter, so
+ the ranker grades those entry points `NoDataInput` and says so, rather than
+ guessing a call sequence no header states. Measured on the six validation
+ libraries: four synthesize, those two do not.
## The optional booster
diff --git a/PROJECTS/advanced/zero-day-vulnerability-scanner/learn/04-CHALLENGES.md b/PROJECTS/advanced/zero-day-vulnerability-scanner/learn/04-CHALLENGES.md
index 4e8623d3..0a7e8dcd 100644
--- a/PROJECTS/advanced/zero-day-vulnerability-scanner/learn/04-CHALLENGES.md
+++ b/PROJECTS/advanced/zero-day-vulnerability-scanner/learn/04-CHALLENGES.md
@@ -209,6 +209,38 @@ lose, because the OSS-Fuzz harness was hand-written by someone who knows the
library. The useful result is *by how much*, and whether the gap is in the
harness or in the campaign.
+### 11. Assemble a stream for a library that will not take one as an argument
+
+lisdex drives entry points that receive their bytes as a parameter. openjpeg
+never offers one: `opj_read_header(opj_stream_t *, opj_codec_t *, opj_image_t **)`
+takes no buffer, and the ranker is being honest when it grades that `NoDataInput`.
+The library's own fuzz target builds the stream instead β€” a `MemFile` struct, then
+`opj_stream_set_read_function`, `opj_stream_set_seek_function` and
+`opj_stream_set_user_data_length`, and only then a decode.
+
+The interesting part is where the callback's *role* comes from. You cannot read it
+off the type. libtiff declares one typedef for two different jobs β€”
+
+```c
+typedef tmsize_t (*TIFFReadWriteProc)(thandle_t, void *, tmsize_t);
+```
+
+β€” and hands it to `TIFFClientOpen` as both `readproc` and `writeproc`, so no
+type-level rule can ever separate them. openjpeg is friendlier: the role is in the
+*setter's* name, `set_read_function` against `set_seek_function`, which is the kind
+of signal `naming.rs` already reasons about.
+
+So build it for openjpeg, not for libtiff, and let the ladder replay the author's
+call order rather than inventing one. The oracles decide whether you succeeded: a
+stream adapter that lies β€” a size callback reporting more bytes than the read
+callback will deliver β€” makes the parser read uninitialised memory and reports a
+crash your harness manufactured. `judge_execution` rejects anything that crashes on
+a valid seed, and `judge_inter_seed_divergence` rejects anything that ignores the
+corpus. Getting past both is the whole exercise.
+
+Do not bother with libtiff. Its maintainers fuzz it through a C++ `std::istream`
+via `TIFFStreamOpen`, which is a different capability again.
+
## What Not To Build
Two things that look like obvious improvements and are not.