WEEKLYThis week's high-risk CVEs and how to respond | RAIMEI weekly report
ProductsRAIMONAI-WAFcmdAI PentestHONMARUAI Ops PanelSHIELDMobile App ProtectionKABUTOPC Software ProtectionSENRIGANAttack Surface Management (ASM)TAIKOUptime MonitoringINAZUMAPenetration TestingRAIMEIVulnerability IntelMOBILEMobile Security App
SolutionsGamingFinance & paymentsE-commerce & retailSaaS & startups
DevelopersDocsAPI referenceCLI・SDKCI/CD integration
PricingVulnerability IntelFree pentestSupport & contact

HTTP/2 Rapid Reset: what a WAF can and cannot stop

RAIJIN Security Research · Analysis Draft, to be checked again before launch

In October 2023, several cloud providers disclosed a denial-of-service attack against HTTP/2. It was assigned CVE-2023-44487. With very few resources, an attacker can keep a server too busy to answer normal requests.

How the attack works

HTTP/2 allows many streams to be open on a single connection at the same time. A client can send a request and then cancel it immediately with RST_STREAM.

Once cancelled, the stream no longer counts toward the concurrent-stream limit. The attacker keeps repeating "send, cancel" while the server has already started work on every request.

What a WAF can stop

A WAF or reverse proxy placed in front can count stream resets per connection. When the count is abnormal it closes the connection, so the requests never reach the backend.

What a WAF cannot stop

If attack traffic reaches an origin server directly without passing through the WAF, the WAF cannot help. The origin's IP address must not be publicly exposed.

The server software itself also needs updating. HTTP servers and libraries have released patches that limit how often streams can be reset.

How to tell whether you are affected

First check whether your public services have HTTP/2 enabled, then check the server software versions. Teams with many exposed assets can inventory them in one pass with an attack surface management tool.

Back to research

The RAIJIN products for this

RAIMON雷門Early access

AI-WAF

A cloud WAF where an AI model judges whether each request is an attack. Every block comes with a stated reason.

View details
SENRIGAN千里眼Early access

Attack Surface Management (ASM)

Continuously inventories exposed assets and finds unmanaged domains, ports and services.

View details
TAIKO太鼓Early access

Uptime Monitoring

Continuously checks the availability and response time of sites and APIs from multiple locations.

View details