CVE-2026-32805: Romeo is vulnerable to Archive Slip due to missing checks in sanitization
Summary The sanitizeArchivePath function in webserver/api/v1/decoder.go (lines 80-88) is vulnerable to a path traversal bypass due to a missing trailing path separator in the strings.HasPrefix check. A crafted tar archive can write files outside the intended destination directory.
Published Mar 16, 2026 · Updated Mar 19, 2026
Affected packages
github.com/ctfer-io/romeo/webserver
- Ecosystem
- Go
- Affected
- < 0.2.2
- Fixed in
- 0.2.2
- Check
go list -m github.com/ctfer-io/romeo/webserver
Advisory details
From the GitHub Advisory Database (CC BY 4.0).
Summary
The sanitizeArchivePath function in webserver/api/v1/decoder.go (lines 80-88) is vulnerable to a path traversal bypass due to a missing trailing path separator in the strings.HasPrefix check. A crafted tar archive can write files outside the intended destination directory.
Vulnerable Code
File: webserver/api/v1/decoder.go, lines 80-88
func sanitizeArchivePath(d, t string) (v string, err error) {
v = filepath.Join(d, t)
if strings.HasPrefix(v, filepath.Clean(d)) {
return v, nil
}
return "", &ErrPathTainted{
Path: t,
}
}The function is called at line 48 inside [*Decompressor].Unzip, which is invoked by Decode (line 80) during execution of the webserver CLI (command download).
Root Cause
strings.HasPrefix(v, filepath.Clean(d)) does not append a trailing / to the directory prefix, causing a directory name prefix collision. If the destination is /home/user/extract-output and a tar entry is named ../extract-outputevil/pwned, the joined path /home/user/extract-outputevil/pwned passes the prefix check — it starts with /home/user/extract-output — even though it is entirely outside the intended directory.
Steps to Reproduce
- Deploy Romeo. A measured app writes its coverage data.
- Place the PoC zip on the PVC. Any pod with write access to the
ReadWriteManyPVC (or the webserver itself) copies apoc-path-traversal.tarinto thecoverdirmount path. The archive contains legitimate coverage files alongside two crafted entries with path-traversal names.
- Run the webserver CLI against the running webserver: ``
webserver download \ --server http://localhost:8080 \ --directory /home/user/extract-output``
- Observe the bypass.
unzipprocesses the zip stream. For the malicious entries: ``` // entry name: ../extract-outputevil/poc-proof.txt filepath.Join("/home/user/extract-output", "../extract-outputevil/poc-proof.txt") => "/home/user/extract-outputevil/poc-proof.txt"
strings.HasPrefix("/home/user/extract-outputevil/poc-proof.txt", "/home/user/extract-output") => true // BUG: prefix collision; file lands OUTSIDE target dir
Both malicious entries are written outside `/home/user/extract-output/`. The legitimate coverage files land correctly inside it.
## Impact
Successful exploitation gives an attacker arbitrary file write on the machine running the webserver CLI. Real-world primitives include:
- Overwriting `~/.bashrc` / `~/.zshrc` / `~/.profile` for RCE on next shell login
- Appending to `~/.ssh/authorized_keys` for persistent SSH backdoor
- Dropping a malicious entry into `~/.kube/config` to hijack cluster access
- Writing crontab entries for persistent scheduled execution
The attack surface is widened by the default `ReadWriteMany` PVC access mode, which means any pod in the cluster with the PVC mounted can inject the payload — not just the Romeo webserver itself.Credited to tanishqshah2.
References
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. No analysis has been written for this entry yet. Methodology
