GHSA-mm7c-rhg6-qr4r

Gitea: Authorization Bypass via "Allow edits from maintainers" allows unauthorized commits to any readable repo

## Summary Any authenticated low-privilege user with read access to a repository can push arbitrary commits directly to that repository, bypassing all write-access checks. ## Vulnerability Gitea's "Allow edits from maintainers" PR option can be abused via reverse-fork PRs: 1. The web UI PR-create endpoint binds `allow_maintainer_edit=true` **without** verifying that the submitter has write access to the HEAD repository. 2. Gitea allows creating a PR where **BASE = attacker's fork** and **HEAD = upstream target**. The attacker is "maintainer" of the BASE (their own fork), so the flag is set against the upstream HEAD. 3. On `git push` over HTTP/SSH, Gitea relaxes the required access mode to `Read` when `SupportProcReceive` is enabled ([`routers/web/repo/githttp.go`](https://github.com/go-gitea/gitea/blob/v1.25.5/routers/web/repo/githttp.go#L189), [`routers/private/serv.go`](https://github.com/go-gitea/gitea/blob/v1.25.5/routers/private/serv.go#L337)) and defers enforcement to the pre-receive hook. 4. The pre-receive hook calls [`CanMaintainerWriteToBranch`](https://github.com/go-gitea/gitea/blob/v1.25.5/models/issues/pull_list.go#L72) (`models/issues/pull_list.go`), which finds the malicious PR, sees `AllowMaintainerEdit=true`, and checks whether the pusher has write access to the **BASE** repo. Since BASE is the attacker's own fork, the check passes and the push is authorized against the upstream. ## Exploitation 1. Attacker forks the target repository. 2. Attacker visits the web compare endpoint and creates a PR with `BASE = their_fork`, `HEAD = upstream`, and "Allow edits from maintainers" checked. 3. Attacker clones their fork, makes a commit, and runs `git push <upstream_url> <branch>` — the push is accepted. ## Reproduction ```bash python3 poc.py --repo http://gitea:3000/victim/repo --user attacker --password attacker_pass ``` [poc.py](https://github.com/user-attachments/files/26641541/poc.py) Expected output: ``` [+] target: victim/my_repo default branch: main [*] forking -> attacker/my_repo_pocfork (202) [+] fork ready [+] malicious PR created (BASE=attacker fork, HEAD=upstream) remote: . Processing 1 references remote: Processed 1 references in total To http://192.168.101.20:3000/victim/my_repo.git e5c07b3..9a0b884 main -> main [+] latest commit on victim/my_repo@main: 'PoC: unauthorized commit via maintainer-edit bypass' [+] CONFIRMED: unauthorized push to upstream succeeded. ``` A `PWNED.txt` file will appear on the target repo's default branch, committed by the attacker who has no write access. ## Impact Full repository compromise. Any logged-in user can backdoor any repository they can read, including all public repositories on the instance. ## Suggested Fix Two independent checks are missing; both should be added for defense in depth: 1. **At PR creation:** before setting `AllowMaintainerEdit = true`, verify the submitter has write access to the **HEAD** repository. 2. **In `CanMaintainerWriteToBranch`:** verify that the PR's HEAD repo matches the repository being pushed to, and that the PR was opened by a legitimate owner/writer of the HEAD repository. Do not trust `AllowMaintainerEdit` solely based on BASE write access.