Lucene search

K
githubGitHub Advisory DatabaseGHSA-8FMJ-33GW-G7PW
HistoryMay 28, 2024 - 4:55 p.m.

Denial of service of Minder Server from maliciously crafted GitHub attestations

2024-05-2816:55:02
CWE-770
GitHub Advisory Database
github.com
6
github attestations
denial of service
minder server vulnerability
sigstore verifier
attack vector
response parsing

5.3 Medium

CVSS3

Attack Vector

NETWORK

Attack Complexity

HIGH

Privileges Required

LOW

User Interaction

NONE

Scope

UNCHANGED

Confidentiality Impact

NONE

Integrity Impact

NONE

Availability Impact

HIGH

CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:N/A:H

6.7 Medium

AI Score

Confidence

High

0.0004 Low

EPSS

Percentile

15.7%

Minder is vulnerable to a denial-of-service (DoS) attack which could allow an attacker to crash the Minder server and deny other users access to it.

The root cause of the vulnerability is that Minders sigstore verifier reads an untrusted response entirely into memory without enforcing a limit on the response body. An attacker can exploit this by making Minder make a request to an attacker-controlled endpoint which returns a response with a large body which will crash the Minder server.

Specifically, the point of failure is where Minder parses the response from the GitHub attestations endpoint in getAttestationReply. Here, Minder makes a request to the orgs/$owner/attestations/$checksumref GitHub endpoint (line 285) and then parses the response into the AttestationReply (line 295):

https://github.com/stacklok/minder/blob/daccbc12e364e2d407d56b87a13f7bb24cbdb074/internal/verifier/sigstore/container/container.go#L271-L300

The way Minder parses the response on line 295 makes it prone to DoS if the response is large enough. Essentially, the response needs to be larger than the machine has available memory.

To demonstrate this in an isolated way, consider the following example:

package main

import (
        "encoding/json"
        "fmt"
        "io"
        "strings"
)

type Attestation struct {
        Bundle json.RawMessage `json:"bundle"`
}

type AttestationReply struct {
        Attestations []Attestation `json:"attestations"`
}

func main() {
        m1 := strings.NewReader("{\"attestations\":[")
        maliciousBody := strings.NewReader(strings.Repeat("{\"bundle\":{\"k\": \"v\"{{,", 100000000))
        m2 := strings.NewReader("{\"bundle\":{\"k\": \"v\"}}]}")
        maliciousBodyReader := io.MultiReader(m1, maliciousBody, maliciousBody, maliciousBody, m2)
        fmt.Println("Created malicious body")

        var attestationReply AttestationReply
        _ = json.NewDecoder(maliciousBodyReader).Decode(&attestationReply)
}

This example mimics the behavior of Minders getAttestationReply and how a malicious response body passed to getAttestationReply’s parsing of the response will cause DoS.

When running this script locally on my system, Go incrementally increases memory consumption up to above 90%, freezes the machine and then performs a sigkill.

Attack vector

The content that is hosted at the orgs/$owner/attestations/$checksumref GitHub attestation endpoint is controlled by users including unauthenticated users to Minders threat model. However, a user will need to configure their own Minder settings to cause Minder to make Minder send a request to fetch the attestations. The user would need to know of a package whose attestations were configured in such a way that they would return a large response when fetching them. As such, the steps needed to carry out this attack would look as such:

  1. The attacker adds a package to ghcr.io with attestations that can be fetched via the orgs/$owner/attestations/$checksumref GitHub endpoint.
  2. The attacker registers on Minder and makes Minder fetch the attestations.
  3. Minder fetches attestations and crashes thereby being denied of service.

Affected configurations

Vulners
Node
stacklokminderRange<0.0.51
CPENameOperatorVersion
github.com/stacklok/minderlt0.0.51

5.3 Medium

CVSS3

Attack Vector

NETWORK

Attack Complexity

HIGH

Privileges Required

LOW

User Interaction

NONE

Scope

UNCHANGED

Confidentiality Impact

NONE

Integrity Impact

NONE

Availability Impact

HIGH

CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:N/A:H

6.7 Medium

AI Score

Confidence

High

0.0004 Low

EPSS

Percentile

15.7%

Related for GHSA-8FMJ-33GW-G7PW