Solving for true-source IP with global load balancers in Google Cloud
Background Recently a customer approached us with requirements that may seem contradictory: true source IP persistence, global distribution of load balancers (LB's), TCP-only proxying, and alias IP ranges in Google Cloud. With the help of PROXY protocol support, we offered a straightforward solution that is worth documenting for others. Customer requirements We have NGINX WAF running on VM instances in Google Cloud Platform (GCP) We want to expose these to the Internet with aglobal load balancer We must know the true source IP of clients when traffic reaches our WAF We donot want so use an application (HTTP/S) load balancer in Google. i.e., we do not want to perform TLS decryption with GCP or use HTTP/HTTPS load balancing therefore, we cannot use X-Forwarded-For headers to preserve true source IP Additionally, we'd like to use Cloud Armor. How can we add on a CDN/DDoS/etc provider if needed? Let's solve for these requirements by finding the load balancer to use, and then how to preserve and use true source IP. Which load balancer type fits best? This guideoutlines our options for Google LB’s. Because our requirements includeglobal, TCP-only load balancing, we will choose the highlighted LB type of “Global external proxy Network Load Balancer”. Proxy vs Passthrough Notice that global LB’sproxytraffic. They do not preserve source IP address as apassthrough LB does. Global IP addresses are advertised from multiple, globally-distributed front end locations, using Anycast IP routing. Proxying from these locations allows traffic symmetry, but Source NAT causes loss of the original client IP address. I've added some comments into a Google diagram below to show our core problem here: PROXY protocol support with Google load balancers Google’s TCP LBdocumentationoutlines our challenge and solution: "By default, the target proxy does not preserve the original client IP address and port information. You can preserve this information by enabling the PROXY protocol on the target proxy." Without PROXY protocol support, we could only meet 2 out of 3 core requirements with any given load balancer type. PROXY protocol allows us to meet all 3 simultaneously. Setting up our environment in Google The script below configures a global TCP proxy network load balancer and associated objects. It is assumed that a VPC network, subnet, and VM instances exist already. This script assumes the VM’s are F5 BIG-IP devices, although our demo will use Ubuntu VM’s with NGINX installed. Both BIG-IP and NGINX can easily receive and parse PROXY protocol. # GCP Environment Setup Guide for Global TCP Proxy LB with Proxy Protocol. Credit to Tony Marfil, @tmarfil # Step 1: Prerequisites # Before creating the network endpoint group, ensure the following GCP resources are already configured: # # -A VPC network named my-vpc. # -A subnet within this network named outside. # -Instances ubuntu1 and ubuntu2 should have alias IP addresses configured: 10.1.2.16 and 10.1.2.17, respectively, both using port 80 and 443. # # Now, create a network endpoint group f5-neg1 in the us-east4-c zone with the default port 443. gcloud compute network-endpoint-groups create f5-neg1 \ --zone=us-east4-c \ --network=my-vpc \ --subnet=outside \ --default-port=443 # Step 2: Update the Network Endpoint Group # # Add two instances with specified IPs to the f5-neg1 group. gcloud compute network-endpoint-groups update f5-neg1 \ --zone=us-east4-c \ --add-endpoint 'instance=ubuntu1,ip=10.1.2.16,port=443' \ --add-endpoint 'instance=ubuntu2,ip=10.1.2.17,port=443' # Step 3: Create a Health Check # # Set up an HTTP health check f5-healthcheck1 that uses the serving port. gcloud compute health-checks create http f5-healthcheck1 \ --use-serving-port # Step 4: Create a Backend Service # # Configure a global backend service f5-backendservice1 with TCP protocol and attach the earlier health check. gcloud compute backend-services create f5-backendservice1 \ --global \ --health-checks=f5-healthcheck1 \ --protocol=TCP # Step 5: Add Backend to the Backend Service # # Link the network endpoint group f5-neg1 to the backend service. gcloud compute backend-services add-backend f5-backendservice1 \ --global \ --network-endpoint-group=f5-neg1 \ --network-endpoint-group-zone=us-east4-c \ --balancing-mode=CONNECTION \ --max-connections=1000 # Step 6: Create a Target TCP Proxy # # Create a global target TCP proxy f5-tcpproxy1 to handle routing to f5-backendservice1. gcloud compute target-tcp-proxies create f5-tcpproxy1 \ --backend-service=f5-backendservice1 \ --proxy-header=PROXY_V1 \ --global # Step 7: Create a Forwarding Rule # # Establish global forwarding rules for TCP traffic on port 80 & 443. gcloud compute forwarding-rules create f5-tcp-forwardingrule1 \ --ip-protocol TCP \ --ports=80 \ --global \ --target-tcp-proxy=f5-tcpproxy1 gcloud compute forwarding-rules create f5-tcp-forwardingrule2 \ --ip-protocol TCP \ --ports=443 \ --global \ --target-tcp-proxy=f5-tcpproxy1 # Step 8: Create a Firewall Rule # # Allow ingress traffic on specific ports for health checks with the rule allow-lb-health-checks. gcloud compute firewall-rules create allow-lb-health-checks \ --direction=INGRESS \ --priority=1000 \ --network=my-vpc \ --action=ALLOW \ --rules=tcp:80,tcp:443,tcp:8080,icmp \ --source-ranges=35.191.0.0/16,130.211.0.0/22 \ --target-tags=allow-health-checks # Step 9: Add Tags to Instances # # Tag instances ubuntu1 and ubuntu2 to include them in health checks. gcloud compute instances add-tags ubuntu1 --tags=allow-health-checks --zone=us-east4-c gcloud compute instances add-tags ubuntu2 --tags=allow-health-checks --zone=us-east4-c ## TO PULL THIS ALL DOWN: (uncomment the lines below) # gcloud compute firewall-rules delete allow-lb-health-checks --quiet # gcloud compute forwarding-rules delete f5-tcp-forwardingrule1 --global --quiet # gcloud compute forwarding-rules delete f5-tcp-forwardingrule2 --global --quiet # gcloud compute target-tcp-proxies delete f5-tcpproxy1 --global --quiet # gcloud compute backend-services delete f5-backendservice1 --global --quiet # gcloud compute health-checks delete f5-healthcheck1 --quiet # gcloud compute network-endpoint-groups delete f5-neg1 --zone=us-east4-c --quiet # # Then delete your VM's and VPC network if desired. Receiving PROXY protocol using NGINX We now have 2x Ubuntu VM's running in GCP that will receive traffic when we target our global TCP proxy LB's IP address. Let’s use NGINX to receive and parse the PROXY protocol traffic. When proxying and "stripping" the PROXY protocol headers from traffic, NGINX can append an additional header containing the value of the source IP obtained from PROXY protocol: server { listen 80 proxy_protocol; # tell NGINX to expect traffic with PROXY protocol server_name customer1.my-f5.com; location / { proxy_pass http://localhost:3000; proxy_http_version 1.1; proxy_set_header x-nginx-ip $server_addr; # append a header to pass the IP address of the NGINX server proxy_set_header x-proxy-protocol-source-ip $proxy_protocol_addr; # append a header to pass the src IP address obtained from PROXY protocol proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; # append a header to pass the src IP of the connection between Google's front end LB and NGINX proxy_cache_bypass $http_upgrade; } } Displaying true source IP in our web app You might notice above that NGINX is proxying to http://localhost:3000. I have a simple NodeJS app to display a page with HTTP headers: const express = require('express'); const app = express(); const port = 3000; // set the view engine to ejs app.set('view engine', 'ejs'); app.get('/', (req, res) => { const proxy_protocol_addr = req.headers['x-proxy-protocol-source-ip']; const source_ip_addr = req.headers['x-real-ip']; const array_headers = JSON.stringify(req.headers, null, 2); const nginx_ip_addr = req.headers['x-nginx-ip']; res.render('index', { proxy_protocol_addr: proxy_protocol_addr, source_ip_addr: source_ip_addr, array_headers: array_headers, nginx_ip_addr: nginx_ip_addr }); }) app.listen(port, () => { console.log('Server is listenting on port 3000'); }) For completeness, NodeJS is using the EJS template engine to build our page. The file views/index.ejs is here: <!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale-1"> <title>Demo App</title> </head> <body class="container"> <main> <h2>Hello World!</h2> <p>True source IP (the value of <code>$proxy_protocol_addr</code>) is <b><%= typeof proxy_protocol_addr != 'undefined' ? proxy_protocol_addr : '' %></b></p> <p>IP address that NGINX recieved the connection from (the value of <code>$remote_addr</code>) is <b><%= typeof source_ip_addr != 'undefined' ? source_ip_addr : '' %> </b></p> <p>IP address that NGINX is running on (the value of <code>$server_addr</code>) is <b><%= typeof nginx_ip_addr != 'undefined' ? nginx_ip_addr : '' %></b><p> <h3>Request Headers at the app:</h3> <pre><%= typeof array_headers != 'undefined' ? array_headers : '' %></pre> </main> </body> </html> Cloud Armor Cloud Armor is aneasy add-onwhen using Google load balancers. If required, an admin can: Create a Cloud Armor security policy Add rules (for example, rate limiting) to this policy Attach the policy to a TCP load balancer In this way “edge protection” is applied to your Google workloads with little effort. Our end result This small demo app shows that true source IP can be known to an application running on Google VM’s when using the Global TCP Network Load Balancer. We’ve achieved this using PROXY protocol and NGINX. We’ve used NodeJS to display a web page with proxied header values. Thanks for reading. Please reach out with any questions!
Intermittent Net::ERR_CONNECTION_RESET Error and Incomplete Loading over HTTPS
I have an F5 load balancing setup configured with two servers. My MVC web application, which incorporates Kendo UI, Jquery, and bootstrapping, is hosted on an IIS server with an SSL certificate. However, when accessing the application via HTTPS from outside the server, it often or sometimes results in a 'net::ERR_CONNECTION_RESET' error, with intermittent failures to load javascript and CSS files to the client browser. Strangely, upon reloading the page, the assets load properly, and the page functions as expected. This issue did not occur when the application was accessed via HTTP, where it worked properly without any issues. What could be the reason behind this problem?Cookie Tampering Protection using F5 Distributed Cloud Platform
This article aims to cover the basics of cookies and then showed how intruders can tamper cookies to modify application behavior. Finally, we also showcased how F5 XC cookie tampering protection can be used to safeguard our sensitive cookie workloads.Active/Active load balancing examples with F5 BIG-IP and Azure load balancer
Background A couple years ago Iwrote an article about some practical considerations using Azure Load Balancer. Over time it's been used by customers, so I thought to add a further article that specifically discusses Active/Active load balancing options. I'll use Azure's standard load balancer as an example, but you can apply this to other cloud providers. In fact, the customer I helped most recently with this very question was running in Google Cloud. This article focuses on using standard TCP load balancers in the cloud. Why Active/Active? Most customers run 2x BIG-IP's in an Active/Standby cluster on-premises, and it's extremely common to do the same in public cloud. Since simplicity and supportability are key to successful migration projects, often it's best to stick with architectures you know and can support. However, if you are confident in your cloud engineering skills or if you want more than 2x BIG-IP's processing traffic, you may consider running them all Active. Of course, if your totalthroughput for N number of BIG-IP's exceeds the throughput thatN-1 can support, the loss of a single VM will leave you with more traffic than the remaining device(s) can handle. I recommend choosing Active/Active only if you're confident in your purpose and skillset. Let's define Active/Active Sometimes this term is used with ambiguity. I'll cover three approaches using Azure load balancer, each slightly different: multiple standalone devices Sync-Only group using Traffic Group None Sync-Failover group using Traffic Group None Each of these will use a standard TCP cloud load balancer. This article does not cover other ways to run multiple Active devices, which I've outlined at the end for completeness. Multiple standalone appliances This is a straightforward approach and an ideal target for cloud architectures. When multiple devices each receive and process traffic independently, the overhead work of disaggregating traffic to spread between the devices can be done by other solutions, like a cloud load balancer. (Other out-of-scope solutions could be ECMP, BGP, DNS load balancing, or gateway load balancers). Scaling out horizontally can be a matter of simple automation and there is no cluster configuration to maintain. The only limit to the number of BIG-IP's will be any limits of the cloud load balancer. The main disadvantage to this approach is the fear of misconfiguration by human operators. Often a customer is not confident that they can configure two separate devices consistently over time. This is why automation for configuration management is ideal. In the real world, it's also a reason customers consider our next approach. Clustering with a sync-only group A Sync-Only device group allows us to sync some configuration data between devices, but not fail over configuration objects in floating traffic groups between devices, as we would in a Sync-Failover group. With this approach, we can sync traffic objects between devices, assign them to Traffic Group None, and both devices will be considered Active. Both devices will process traffic, but changes only need to be made to a single device in the group. In the example pictured above: The 2x BIG-IP devices are in a Sync-Only group called syncGroup /Common partition isnotsynced between devices /app1 partition issynced between devices the /app1 partition has Traffic Group None selected the /app1 partition has the Sync-Only group syncGroup selected Both devices are Active and will process traffic received on Traffic Group None The disadvantage to this approach is that you can create an invalid configuration by referring to objects that are not synced. For example, if Nodes are created in/Common, they will exist on the device on which they were created, but not on other devices. If a Pool in /app1 then references Nodes from /Common, the resulting configuration will be invalid for devices that do not have these Nodes configured. Another consideration is that an operator must use and understand partitions. These are simple and should be embraced. However, not all customers understand the use of partitions and many prefer to use /Common only, if possible. The big advantage here is that changes only need to be made on a single device, and they will be replicated to other devices (up to 32 devices in a Sync-Only group). The risk of inconsistent configuration due to human error is reduced. Each device has a small green "Active" icon in the top left hand of the console, reminding operators that each device is Active and will process incoming traffic onTraffic Group None. Failover clustering using Traffic Group None Our third approach is very similar to our second approach. However, instead of a Sync-Only group, we will use a Sync-Failover group. A Sync-Failover group will sync all traffic objects in the default /Common partition, allowing us to keep all traffic objects in the default partition and avoid the use of additional partitions. This creates a traditional Active/Standby pair for a failover traffic group, and a Standby device will not respond to data plane traffic. So how do we make this Active/Active? When we create our VIPs in Traffic Group None, all devices will process traffic received on these Virtual Servers. One device will show "Active" and the other "Standby" in their console, but this is only the status for the floating traffic group. We don't need to use the floating traffic group, and by using Traffic Group None we have an Active/Active configuration in terms of traffic flow. The advantage here is similar to the previous example: human operators only need to configure objects in a single device, and all changes are synced between device group members (up to 8 in a Sync-Failover group). Another advantage is that you can use the/Common partition, which was not possible with the previous example. The main disadvantage here is that the console will show the word "Active" and "Standby" on devices, and this can confuse an operator that is familiar only with Active/Standby clusters using traffic groups for failover. While this third approach is a very legitimate approach and technically sound, it's worth considering if your daily operations and support teams have the knowledge to support this. Other considerations Source NAT (SNAT) It is almost always a requirement that you SNAT traffic when using Active/Active architecture, and this especially applies to the public cloud, where our options for other networking tricks are limited. If you have a requirement to see true source IPandneed to use multiple devices in Active/Active fashion, consider using Azure or AWS Gateway Load Balancer options. Alternative solutions like NGINX and F5 Distributed Cloud may also be worth considering in high-value, hard-requirement situations. Alternatives to a cloud load balancer This article is not referring to F5 with Azure Gateway Load Balancer, or to F5 with AWS Gateway Load Balancer. Those gateway load balancer solutions are another way for customers to run appliances as multiple standalone devices in the cloud. However, they typically requirerouting, not proxying the traffic (ie, they don't allow destination NAT, which many customers intend with BIG-IP). This article is also not referring to other ways you might achieve Active/Active architectures, such as DNS-based high availability, or using routing protocols, like BGP or ECMP. Note that using multiple traffic groups to achieve Active/Active BIG-IP's - the traditional approach on-prem or in private cloud - is not practical in public cloud, as briefly outlined below. Failover of traffic groups with Cloud Failover Extension (CFE) One option for Active/Standby high availability of BIG-IP is to use the CFE , which can programmatically update IP addresses and routes in Azure at time of device failure. Since CFE does not support Active/Active scenarios, it is appropriate only for failover of a single traffic group (ie., Active/Standby). Conclusion Thanks for reading! In general I see that Active/Standby solutions work for many customers, but if you are confident in your skills and have a need for Active/Active F5 BIG-IP devices in the cloud, please reach out if you'd like me to walk you through these options and explore any other possibilities. Related articles Practical Considerations using F5 BIG-IP and Azure Load Balancer Deploying F5 BIG-IP with Azure Cross-Region Load Balancer
Move connections to another node via the API
I am very new to F5 and the API. I am looking for a way to move our Application's active sessions/connections from one node to another via the native API. I need to do this so I can shut down the node after all connections are moved over, so I can upgrade it. Is this possible? Thank you!
F5 equivalent of Alteon's buddy helthchecks
Hi all, I'm in the process of migrating old Radware Alteon load balancers onto F5 kit. On the Alteons they have what's known as a "buddy health check". Basically what it does is monitor a second node, and if that goes down then the first node is marked as down too. For example, Server A is configured to have Server B as its buddy. If the health check for Server B fails, the load balancer will mark Server A as down. I'm aware this could be done using Priority Group Activation, however in this case Server A is being health checked on port 3000 whilst Server B is health checked on port 3001. Data for both is on port 7021. Is this setup achievable on an F5? Thanks!
HTTP Pipelining: A security risk without real performance benefits
Everyone wants web sites and applications to load faster, and there’s no shortage of folks out there looking for ways to do just that. But all that glitters is not gold, and not all acceleration techniques actually do all that much to accelerate the delivery of web sites and applications. Worse, some actual incur risk in the form of leaving servers open to exploitation. A BRIEF HISTORY Back in the day when HTTP was still evolving, someone came up with the concept of persistent connections. See, in ancient times – when administrators still wore togas in the data center – HTTP 1.0 required one TCP connection for every object on a page. That was okay, until pages started comprising ten, twenty, and more objects. So someone added an HTTP header, Keep-Alive, which basically told the server not to close the TCP connection until (a) the browser told it to or (b) it didn’t hear from the browser for X number of seconds (a time out). This eventually became the default behavior when HTTP 1.1 was written and became a standard. I told you it was a brief history. This capability is known as a persistent connection, because the connection persists across multiple requests. This is not the same as pipelining, though the two are closely related. Pipelining takes the concept of persistent connections and then ignores the traditional request – reply relationship inherent in HTTP and throws it out the window. The general line of thought goes like this: “Whoa. What if we just shoved all the requests from a page at the server and then waited for them all to come back rather than doing it one at a time? We could make things even faster!” Tada! HTTP pipelining. In technical terms, HTTP pipelining is initiated by the browser by opening a connection to the server and then sending multiple requests to the server without waiting for a response. Once the requests are all sent then the browser starts listening for responses. The reason this is considered an acceleration technique is that by shoving all the requests at the server at once you essentially save the RTT (Round Trip Time) on the connection waiting for a response after each request is sent. WHY IT JUST DOESN’T MATTER ANYMORE (AND MAYBE NEVER DID) Unfortunately, pipelining was conceived of and implemented before broadband connections were widely utilized as a method of accessing the Internet. Back then, the RTT was significant enough to have a negative impact on application and web site performance and the overall user-experience was improved by the use of pipelining. Today, however, most folks have a comfortable speed at which they access the Internet and the RTT impact on most web application’s performance, despite the increasing number of objects per page, is relatively low. There is no arguing, however, that some reduction in time to load is better than none. Too, anyone who’s had to access the Internet via high latency links can tell you anything that makes that experience faster has got to be a Good Thing. So what’s the problem? The problem is that pipelining isn’t actually treated any differently on the server than regular old persistent connections. In fact, the HTTP 1.1 specification requires that a “server MUST send its responses to those requests in the same order that the requests were received.” In other words, the requests are return in serial, despite the fact that some web servers may actually process those requests in parallel. Because the server MUST return responses to requests in order that the server has to do some extra processing to ensure compliance with this part of the HTTP 1.1 specification. It has to queue up the responses and make certain responses are returned properly, which essentially negates the performance gained by reducing the number of round trips using pipelining. Depending on the order in which requests are sent, if a request requiring particularly lengthy processing – say a database query – were sent relatively early in the pipeline, this could actually cause a degradation in performance because all the other responses have to wait for the lengthy one to finish before the others can be sent back. Application intermediaries such as proxies, application delivery controllers, and general load-balancers can and do support pipelining, but they, too, will adhere to the protocol specification and return responses in the proper order according to how the requests were received. This limitation on the server side actually inhibits a potentially significant boost in performance because we know that processing dynamic requests takes longer than processing a request for static content. If this limitation were removed it is possible that the server would become more efficient and the user would experience non-trivial improvements in performance. Or, if intermediaries were smart enough to rearrange requests such that they their execution were optimized (I seem to recall I was required to design and implement a solution to a similar example in graduate school) then we’d maintain the performance benefits gained by pipelining. But that would require an understanding of the application that goes far beyond what even today’s most intelligent application delivery controllers are capable of providing. THE SILVER LINING At this point it may be fairly disappointing to learn that HTTP pipelining today does not result in as significant a performance gain as it might at first seem to offer (except over high latency links like satellite or dial-up, which are rapidly dwindling in usage). But that may very well be a good thing. As miscreants have become smarter and more intelligent about exploiting protocols and not just application code, they’ve learned to take advantage of the protocol to “trick” servers into believing their requests are legitimate, even though the desired result is usually malicious. In the case of pipelining, it would be a simple thing to exploit the capability to enact a layer 7 DoS attack on the server in question. Because pipelining assumes that requests will be sent one after the other and that the client is not waiting for the response until the end, it would have a difficult time distinguishing between someone attempting to consume resources and a legitimate request. Consider that the server has no understanding of a “page”. It understands individual requests. It has no way of knowing that a “page” consists of only 50 objects, and therefore a client pipelining requests for the maximum allowed – by default 100 for Apache – may not be seen as out of the ordinary. Several clients opening connections and pipelining hundreds or thousands of requests every second without caring if they receive any of the responses could quickly consume the server’s resources or available bandwidth and result in a denial of service to legitimate users. So perhaps the fact that pipelining is not really all that useful to most folks is a good thing, as server administrators can disable the feature without too much concern and thereby mitigate the risk of the feature being leveraged as an attack method against them. Pipelining as it is specified and implemented today is more of a security risk than it is a performance enhancement. There are, however, tweaks to the specification that could be made in the future that might make it more useful. Those tweaks do not address the potential security risk, however, so perhaps given that there are so many other optimizations and acceleration techniques that can be used to improve performance that incur no measurable security risk that we simply let sleeping dogs lie. IMAGES COURTESTY WIKIPEDIA COMMONSUnable to Connect to VIP after Load Balancing - F5 BIG IP
I have VIP 192.168.8.9 on BIG IP F5 which is Pinging and able to talk to Pool members(webservers) 8.1 and 8.2 on on port 80. Pool members able to ping VIP and telnet on port 80. yet, the VIP (loadblancer) is not Routing the HTTP requests to webservers. all the Pool members are green, so is the virtual server. Your Thoughts and advivce is appreciated. any further detail required, please do let me know.
