Gitea: Unbounded Arch package file metadata can cause resource amplification in Gitea package uploads
Summary
Hello Gitea Security Team,
Thank you for your continued work on Gitea. I would like to responsibly report a potential availability-impact issue that I observed in Gitea’s Arch package registry implementation.
During local testing, I noticed that Gitea records non-dot regular file entries from an uploaded Arch package archive into package file metadata. I could not identify an explicit limit on the number of recorded file entries or on the cumulative size of recorded file names before this metadata is serialized, stored, and later used during repository index generation.
As a result, a relatively small compressed .pkg.tar.gz archive may lead to significantly larger server-side metadata processing and storage. I tested this only against a local self-hosted Gitea instance and have not tested this against any third-party or production service.
Suggested Severity
Suggested severity: Medium
Suggested CVSS 3.1 vector:
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:L
Suggested CVSS score: 4.3
This assessment is only a suggestion. The issue appears to require an authenticated user with package publishing permission. However, once that condition is met, the behavior is reachable over the network, does not require user interaction, and may affect availability through amplified metadata parsing, serialization, database storage, and repository index generation.
Affected Component
- Gitea package registry
- Arch package upload endpoint
- Arch package metadata parsing
- Arch repository index generation
Technical Details
The upload flow appears to accept an Arch package archive, parse its contents, record file entries into package metadata, and later reuse that metadata when generating the Arch repository index.
The relevant flow appears to include:
routers/api/packages/arch/arch.go:46 accepts the upload stream.routers/api/packages/arch/arch.go:55 copies the upload into a HashedBuffer.routers/api/packages/arch/arch.go:62 parses the archive with arch_module.ParsePackage.modules/packages/arch/metadata.go:149 appends each non-dot regular tar entry name to files.modules/packages/arch/metadata.go:158 stores the full list as p.FileMetadata.Files.routers/api/packages/arch/arch.go:77 JSON-marshals the file metadata.routers/api/packages/arch/arch.go:143 persists the metadata as arch_module.PropertyMetadata.services/packages/arch/repository.go:302 deserializes the metadata during index generation.services/packages/arch/repository.go:365 joins the full file list into the generated files entry.
From my review, the package upload size limit can reduce the maximum compressed archive size that is accepted, but it does not appear to directly limit the number of file entries or the expanded metadata size for archives that remain below the compressed upload limit.
Impact
An authenticated user with permission to publish Arch packages may be able to upload an archive containing a valid .PKGINFO file and a large number of empty regular file entries.
In my local test environment, Gitea accepted such packages and stored the full file list as package metadata. This caused the server-side metadata size and generated repository files index content to become much larger than the compressed upload size.
The practical impact appears to be resource amplification affecting:
- CPU usage during parsing and index generation
- memory usage during metadata handling
- database storage due to large serialized metadata
- repository index generation size and processing time
This seems most relevant for instances where untrusted or semi-trusted users are allowed to publish packages.
Local Validation Results
I tested this only on a local self-hosted Gitea instance.
A 470,403 byte archive containing 100,000 empty file entries was accepted by the Arch package upload endpoint. It produced a 4,500,112 byte arch.metadata database property and a generated repository index whose files member contained 100,001 lines.
A larger 2,349,767 byte archive containing 500,000 empty file entries was also accepted in the default configuration. It produced a 22,500,112 byte arch.metadata database property and a generated repository files member with 500,001 lines.
Proof of Concept
The following proof of concept is intended only for a local self-hosted test instance.
Save the following script as generate_arch_metadata_test_package.py:
```python
#!/usr/bin/env python3
from __future__ import annotations
import argparse
import gzip
import io
import tarfile
from pathlib import Path
PKGINFO = """pkgname = gitea-metadata-test
pkgbase = gitea-metadata-test
pkgver = 1.0.0-1
pkgdesc = Local metadata scaling test package
url = https://example.invalid/
packager = local test
arch = x86_64
license = MIT
builddate = 1714521600
size = 0
"""
def add_bytes(tar: tarfile.
