Ingress-nginx: HTTP->HTTPS redirect does not work with use-proxy-protocol: "true"

[jpnauta]

I am currently using gcr.io/google_containers/nginx-ingress-controller:0.9.0-beta.7. I was having issues as #277, but that issue is marked as resolved. My ingress would work properly with https://, but would return an empty response with http://. This is what happened when I tried to cURL my domain:

$ curl https://mydomain.com
[html response]
$ curl http://mydomain.com
curl: (52) Empty reply from server

When I changed the use-proxy-protocol configuration from true to false, the curl worked correctly.

$ curl https://mydomain.com
[html response]
$ curl http://mydomain.com
[301 response]

Here is my original config map to reproduce the situation:

apiVersion: v1
kind: ConfigMap
metadata:
  name: my-config-map
data:
  force-ssl-redirect: "true"
  ssl-redirect: "true"
  use-proxy-protocol: "true"

[aledbf]

@jpnauta please check if the latest beta solves the issue (0.9-beta.10)

[acoshift]

@aledbf I have this problem in beta.10 too.

[acoshift]

I think It may not the problem here.
I use GCP TCP LB but it won't send PROXY protocol header for http, that why nginx return empty response https://trac.nginx.org/nginx/ticket/1048
My workaround is use custom template with disable proxy_protocolon port 80.
Is it possible to add config to disable proxy_protocol on port 80 ?

[aledbf]

@acoshift in the configmap use use-proxy-protocol: "false"

[acoshift]

@aledbf ok, right now I remove custom template, and set use-proxy-protocol: "false"
with service.beta.kubernetes.io/external-traffic: OnlyLocal
but I got 10.0.x.x ip in nginx logs.
The only way I can get real ip is to set real_ip_header proxy_protocol;

[aledbf]

@acoshift that's strange because gcp does not supports proxy protocol for http, only https.

[acoshift]

Here's all configs.

$ kubectl get -o yaml svc nginx-ingress-2
apiVersion: v1
kind: Service
metadata:
  annotations:
    service.beta.kubernetes.io/external-traffic: OnlyLocal
    service.beta.kubernetes.io/healthcheck-nodeport: "31976"
  creationTimestamp: 2017-05-28T08:40:25Z
  name: nginx-ingress-2
  namespace: default
  resourceVersion: "9970477"
  selfLink: /api/v1/namespaces/default/services/nginx-ingress-2
  uid: 4b7a0442-4381-11e7-833e-42010a94000a
spec:
  clusterIP: 10.3.255.234
  loadBalancerIP: x.x.x.x
  ports:
  - name: http
    nodePort: 30340
    port: 80
    protocol: TCP
    targetPort: 80
  - name: https
    nodePort: 31552
    port: 443
    protocol: TCP
    targetPort: 443
  selector:
    k8s-app: nginx-ingress-lb
  sessionAffinity: None
  type: LoadBalancer
status:
  loadBalancer:
    ingress:
    - ip: x.x.x.x

$ kubectl get -o yaml ing nginx-ingress
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  annotations:
    kubernetes.io/ingress.class: nginx
    kubernetes.io/tls-acme: "true"
  creationTimestamp: 2017-04-17T16:12:47Z
  generation: 29
  name: nginx-ingress
  namespace: default
  resourceVersion: "10652294"
  selfLink: /apis/extensions/v1beta1/namespaces/default/ingresses/nginx-ingress
  uid: b24bd062-2388-11e7-b9a0-42010a94000b
spec:
  rules:
  - host: x.x
    http:
      paths:
      - backend:
          serviceName: xxx
          servicePort: 8080
        path: /
  tls:
  - hosts:
    - x.x
    secretName: x.x-tls
status:
  loadBalancer:
    ingress:
    - ip: x.x.x.x

$ kubectl get -o yaml ds nginx-ingress-controller-ds
apiVersion: extensions/v1beta1
kind: DaemonSet
metadata:
  creationTimestamp: 2017-07-02T09:42:56Z
  generation: 2
  labels:
    k8s-app: nginx-ingress-lb
  name: nginx-ingress-controller-ds
  namespace: default
  resourceVersion: "10652196"
  selfLink: /apis/extensions/v1beta1/namespaces/default/daemonsets/nginx-ingress-controller-ds
  uid: d3bbeb6b-5f0a-11e7-ad52-42010a94000a
spec:
  selector:
    matchLabels:
      k8s-app: nginx-ingress-lb
  template:
    metadata:
      creationTimestamp: null
      labels:
        k8s-app: nginx-ingress-lb
    spec:
      containers:
      - args:
        - /nginx-ingress-controller
        - --default-backend-service=$(POD_NAMESPACE)/default-http-backend
        - --configmap=$(POD_NAMESPACE)/nginx-config
        env:
        - name: POD_NAME
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: metadata.name
        - name: POD_NAMESPACE
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: metadata.namespace
        image: gcr.io/google-containers/nginx-ingress-controller:0.9.0-beta.10
        imagePullPolicy: Always
        livenessProbe:
          failureThreshold: 3
          httpGet:
            path: /healthz
            port: 10254
            scheme: HTTP
          initialDelaySeconds: 10
          periodSeconds: 10
          successThreshold: 1
          timeoutSeconds: 5
        name: nginx-ingress-controller
        ports:
        - containerPort: 80
          protocol: TCP
        - containerPort: 443
          protocol: TCP
        readinessProbe:
          failureThreshold: 3
          httpGet:
            path: /healthz
            port: 10254
            scheme: HTTP
          periodSeconds: 10
          successThreshold: 1
          timeoutSeconds: 1
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
      dnsPolicy: ClusterFirst
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext: {}
      terminationGracePeriodSeconds: 60
  templateGeneration: 2
  updateStrategy:
    rollingUpdate:
      maxUnavailable: 1
    type: RollingUpdate
status:
  currentNumberScheduled: 3
  desiredNumberScheduled: 3
  numberAvailable: 3
  numberMisscheduled: 0
  numberReady: 3
  observedGeneration: 2
  updatedNumberScheduled: 3

Some logs in nginx pod

// this is https
2017-07-03T02:48:24.670527712Z 127.0.0.1 - [127.0.0.1] - - [03/Jul/2017:02:48:24 +0000] "GET / HTTP/2.0" 200 4764 "-" "Mozilla/5.0 (Windows NT 6.3; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/59.0.3071.115 Safari/537.36" 249 0.010 [default-xxx-8080] 10.0.3.7:8080 19918 0.010 200

// this is http got real ip
2017-07-03T02:50:58.695187914Z 14.207.109.160 - [14.207.x.x] - - [03/Jul/2017:02:50:58 +0000] "GET / HTTP/1.1" 301 178 "-" "curl/7.51.0" 74 0.000 [default-xxx-8080] - - - -

[aledbf]

I use GCP TCP LB but it won't send PROXY protocol header for http, that why nginx return empty response h

Please change the gcp lb to http. In that mode the load balancer sends the X-Forwarded-For

[acoshift]

tyvm for helping me, but for my use-case, I can not use gcp lb http because I want ingress controller to handle TLS (from kube-lego). Right now I have to use custom template for workaround.

My knowledge is very limited but I doubt this line

{{/* Listen on 442 because port 443 is used in the TLS sni server */}}
{{/* This listener must always have proxy_protocol enabled, because the SNI listener forwards on source IP info in it. */}}

"TLS sni server send source IP on port 442", maybe set real_ip_header proxy_protocol; for only port 442 should solve the problem but idk how.

[jpnauta]

@aledbf Unfortunately upgrading to 0.9-beta.10 did not work. However, instead of an empty reply from the server, now I get a 502 error as follows:

<HEAD><TITLE>Server Hangup</TITLE></HEAD>
<BODY BGCOLOR="white" FGCOLOR="black">
<FONT FACE="Helvetica,Arial"><B>

[aledbf]

@jpnauta I cannot reproduce this error. Not sure where are you running but this is the full script to provision a cluster in aws.

Create a cluster using kops in us-west

export MASTER_ZONES=us-west-2a
export WORKER_ZONES=us-west-2a,us-west-2b
export KOPS_STATE_STORE=s3://k8s-xxxxxx-01
export AWS_DEFAULT_REGION=us-west-2

kops create cluster \
 --name uswest2-01.xxxxxxx.io \
 --cloud aws \
 --master-zones $MASTER_ZONES \
 --node-count 2 \
 --zones $WORKER_ZONES \
 --master-size m3.medium \
 --node-size m4.large \
 --ssh-public-key ~/.ssh/id_rsa.pub \
 --image coreos.com/CoreOS-stable-1409.5.0-hvm \
 --yes

Create the echoheaders deployment

echo "
apiVersion: v1
kind: Service
metadata:
  name: echoheaders
  labels:
    app: echoheaders
spec:
  type: NodePort
  ports:
  - port: 80
    targetPort: 8080
    protocol: TCP
    name: http
  selector:
    app: echoheaders

---

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: echoheaders
spec:
  replicas: 1
  template:
    metadata:
      labels:
        app: echoheaders
    spec:
      containers:
      - name: echoheaders
        image: gcr.io/google_containers/echoserver:1.4
        ports:
        - containerPort: 8080

---

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: echoheaders-nginx
  annotations:
    kubernetes.io/ingress.class: nginx
spec:
  tls:
  - hosts:
    - echoheaders.uswest2-01.rocket-science.io
    secretName: echoserver-tls
  rules:
  - host: echoheaders.uswest2-01.xxxxx-xxxx.io
    http:
      paths:
      - backend:
          serviceName: echoheaders
          servicePort: 80
" | kubectl create -f -

Create the nginx ingress controller

$ kubectl create -f https://raw.githubusercontent.com/kubernetes/ingress/master/examples/aws/nginx/nginx-ingress-controller.yaml

Install kube-lego

kubectl create -f https://raw.githubusercontent.com/jetstack/kube-lego/master/examples/nginx/lego/00-namespace.yaml

Configure kube-lego

wget https://raw.githubusercontent.com/jetstack/kube-lego/master/examples/nginx/lego/configmap.yaml
nano configmap.yaml 

Install

kubectl create -f configmap.yaml 
kubectl create -f https://raw.githubusercontent.com/jetstack/kube-lego/master/examples/nginx/lego/deployment.yaml

Run the tests*

$ curl -v echoheaders.uswest2-01.rocket-science.io
* Rebuilt URL to: echoheaders.uswest2-01.rocket-science.io/
*   Trying 52.32.132.20...
* TCP_NODELAY set
* Connected to echoheaders.uswest2-01.rocket-science.io (52.32.132.20) port 80 (#0)
> GET / HTTP/1.1
> Host: echoheaders.uswest2-01.rocket-science.io
> User-Agent: curl/7.52.1
> Accept: */*
> 
< HTTP/1.1 301 Moved Permanently
< Server: nginx/1.13.2
< Date: Thu, 06 Jul 2017 01:54:57 GMT
< Content-Type: text/html
< Content-Length: 185
< Connection: keep-alive
< Location: https://echoheaders.uswest2-01.rocket-science.io/
< Strict-Transport-Security: max-age=15724800; includeSubDomains;
< 
<html>
<head><title>301 Moved Permanently</title></head>
<body bgcolor="white">
<center><h1>301 Moved Permanently</h1></center>
<hr><center>nginx/1.13.2</center>
</body>
</html>
* Curl_http_done: called premature == 0

Delete the cluster

kops delete cluster  --name uswest2-01.xxxxxxxx.io --yes

[aledbf]

@jpnauta if you are running in GCE or GKE you cannot enable proxy protocol because it only works with HTTPS

[jpnauta]

Ahhh okay good to know 👍 I'm on GKE, thanks for your help @aledbf

[icereval]

FYI if you want to configure a load balancer manually (not for the faint of heart). You can workaround this limitation by sharing an External IP between the L7 GLBC Ingress w/ a custom HTTP backend to redirect all traffic to HTTPS, and another manually created L4 LB w/ TCP Proxy Protocol for your HTTPS traffic (to the Nginx Ingress Controller).

[allanharris]

There is the same issue on Azure (AKS). Redirection doesn't work.

[anurag]

@aledbf since proxy-protocol doesn't work over HTTP in GKE, is it possible to get the client IP with GCE's TCP load balancer and ssl-passthrough with proxy-protocol disabled?

[aledbf]

@anurag not sure. If you want to test this please make sure you use externalTrafficPolicy: Local in the service spec of the ingress controller

[dm3ch]

I've got same problem.
When I'm enabling proxy-protocol I'm getting proxy protocol header error for requests on 80 port.

The problem is that I can't disable proxy protocol because if i'll disable it, I will break client IP detection for https (because I need to use ssl-passthrough for some backends).

The only way I see now is to use haproxy for proxying traffic for 80 port using proxy-protocol.

Maybe we can add two different config annotation to enable proxy protocol listening for 80 and 443 ports separately

[dm3ch]

@aledbf ^

[coolersport]

When running nginx ingress on GKE (with a TCP Load Balancer), the only way to get real client IP is to turn on proxy protocol. However, it will stop http->https redirection. In fact, http requests will end up with empty response on client side, and broken header on nginx side. Confirmed the issue still exists with the latest release 0.17.1.

My solution is:

1. Set use-proxy-protocol to true
2. Add another nginx side car with just https redirection rule and listen on a different port, e.g. 8080
3. Update nginx ingress controller service to map 80 to 8080

Voila.

[artushin]

@coolersport Is that with the regional or global TCP LB?

[coolersport]

It is regional in my case. However, this solution address it at nginx layer, nothing to do with GCP LB. In fact, LB is auto-provisioned by GKE.

[artushin]

That's weird, I get real IPs from my regional GCP TCP LB in both x-forwarded-for and x-real-ip, without use-proxy-protocol. You just have to guard it against spoofing with proxy-real-ip-cidr. The global TCP LB doesn't pass through IPs in x-forwarded-for though.

[thomascooper]

I am having this same problem with 0.19

@coolersport I have tried your approach but i believe that it relies on the GCP TCP Proxy which is only available for global, not regional static IP's and Forwarding rules.

Here is an overview of our setup, we have 900+ static IPs for our network, each of these have a manually created regional Forwarding rules (80-443) targeting all required instance groups.

We have 10 nginx-ingress controllers, each with 100+ static IP's configured via ExternalIPs on the service. (This was Google designed and suggested due to a hardcoded limitation of 50 live health checks per cluster)

We use cert-manager (updated version of kube-lego) to automatically provision certs with ingress annotations.

Everything in this scenario works aside from getting the clients actual IP into our app.

If we enable "use-proxy-protocol" in our configMap, then we immediately start getting "broken header:" error messages, I've tried every combination of proxy-real-ip-cidr possible with no results. We Cannot re-provision all 900+ static IPs as global due to multiple issues iincluding quota and the fallout of propagation across all of the domains. Looking for any help we can get at all.

[Spittal]

@artushin what version of ingress-nginx are you using where you don't need to use proxy protocol?

[artushin]

@Spittal I'm on 0.13.0 but, again, that works only on regional TCP LBs. I don't know if it works with proxy protocol on a global LB but it definitely doesn't without it.

[esseti]

Hi all, even if i turn on the use-proxy-protocol

Name:         nginx-ingress-controller
Namespace:    default
Labels:       app=nginx-ingress
              chart=nginx-ingress-0.29.2
              component=controller
              heritage=Tiller
              release=nginx-ingress
Annotations:  <none>

Data
====
enable-vts-status:
----
false
force-ssl-redirect:
----
false
ssl-redirect:
----
false
use-proxy-protocol:
----
true
Events:
  Type    Reason  Age   From                      Message
  ----    ------  ----  ----                      -------
  Normal  CREATE  3m    nginx-ingress-controller  ConfigMap default/nginx-ingress-controller

in the controller i still receive errors and the server does not respond (HTTPS call)

2018/10/26 09:39:10 [error] 91#91: *353 broken header: g9��Z��%�_���9��8��y�;v�D�C��<�n�/�+�0�,��'g�(k�$��
����jih9876�2�.�*�&���=" while reading PROXY protocol, client: 79.0.54.49, server: 0.0.0.0:443

the result for HTTP call has the same error but you seethe content of the requests + headers.

any idea why I can't make it work even with the flag set?

PS: I'm using the helm version

[coolersport]

That's weird, I get real IPs from my regional GCP TCP LB in both x-forwarded-for and x-real-ip, without use-proxy-protocol. You just have to guard it against spoofing with proxy-real-ip-cidr. The global TCP LB doesn't pass through IPs in x-forwarded-for though.

I just noticed that it works in a cluster which uses secure-backend. When setting up a new cluster which has no SSL-passthru, broken header issue reappears.

[roboticsound]

@coolersport I know this is from a while ago but I am having this exact issue. I am quite new to kubernetes and I wonder if you could clarify how you set up the sidecar?

This has been driving me crazy for 2 weeks now!

[coolersport]

@roboticsound, here they are. Sorry, I can't post full YAML files. Hope this gives you the idea.

--- pod container (sidecar) ---
- name: https-redirector
  image: nginx:1.15-alpine
  imagePullPolicy: IfNotPresent
  ports:
  - containerPort: 8080
    name: redirector
  securityContext:
    allowPrivilegeEscalation: false
  volumeMounts:
  - name: nginx-redirector
    mountPath: /etc/nginx/nginx.conf
    subPath: nginx.conf
    readOnly: true
--- service ---
ports:
- name: http
  port: 80
  targetPort: redirector
--- configmap ---
nginx.conf: |
  events {
      worker_connections  128;
  }
  http {
    server {
      listen 8080;
      server_name _;
      return 301 https://$host$request_uri;
    }
  }

[roboticsound]

@coolersport Thanks! helped a lot

[dano0b]

In case somebody didn't see the better solution: https://kubernetes.io/docs/tasks/access-application-cluster/create-external-load-balancer/#preserving-the-client-source-ip

[miquelbar]

@dano0b maybe I'm missing something but I configured kuberntes-ingress in that way and it didn't work: I'm using GKE and when connecting using HTTP I got the real IP but when I'm connecting using HTTPS I'm always getting 127.0.0.1 as the remote IP.

In my opinion, the best solution right now is the one that @coolersport providedÇ

UPDATED After disabled --enable-ssl-passthrough flag I was getting the real request IP as @dano0b pointed

[spacejam101]

Do we have like a standard way of doing this?

[SoulSkare]

This is a real headache, I've followed https://kubernetes.io/docs/tasks/access-application-cluster/create-external-load-balancer/#preserving-the-client-source-ip

I've modified our configMap

apiVersion: v1
data:
real-ip-header: X-Forwared-For
real-ip-recursive: "true"
use-forwarded-headers: "true"
use-proxy-protocol: "true"
kind: ConfigMap

Still nothing only showing the local ip.