How to gracefully return 413 with RequestBodyLimitLayer for chunked transfer encoding

[banool]

• I have looked for existing issues (including closed) about this

I read this old thread but it didn't really address this problem: tokio-rs/axum#853.

Bug Report

Version

$ cargo tree | grep tower-http
└── tower-http v0.6.6

Platform

$ uname -a
Darwin dportMacbookM4.local 25.0.0 Darwin Kernel Version 25.0.0: Wed Sep 17 21:41:26 PDT 2025; root:xnu-12377.1.9~141/RELEASE_ARM64_T6041 arm64

Crates

axum = "0.7"
tokio = { version = "1", features = ["full"] }
tower = "0.5.2"
tower-http = { version = "0.6.6", features = ["limit"] }
hyper = { version = "1.0", features = ["client", "http1"] }
hyper-util = { version = "0.1", features = ["client-legacy", "http1", "tokio"] }
http-body-util = "0.1"
bytes = "1"
futures-util = "0.3"

Description

I have example code here:

https://github.com/banool/rust-examples/blob/e0e103404ffd1678613fdbe2fe02c12f71ae750a/chunked-limit-demo/src/main.rs

Right now when using the request limit layer you get a client error, rather than a clean 413 response. I was hoping RequestBodyLimitLayer would handle this on my behalf, is there a any chance the layer could be modified to handle this more gracefully? Like making the limited body return an error that results in a 413 response?

[banool]

I suppose the issue is the middleware isn't fully integrated into Axum, as opposed to something like DefaultBodyLimit. So you get a hyper error, which also isn't converted into an appropriate axum error nicely.

[jplatte]

I suppose the issue is the middleware isn't fully integrated into Axum, as opposed to something like DefaultBodyLimit.

No, that has nothing to do with it. DefaultBodyLimit behaves exactly the same. The issue is more that tower (and maybe hyper too) doesn't really support sending a response and closing the connection, and if you don't close the connection you're not stopping the client from sending more data in the request, whether you have already sent a response or not. Plus I think many HTTP clients don't start reading the response if they haven't finished sending the request yet.

If you're only concerned with well-behaving clients, you can make your own variation of the Limited body type that continues to read the body when reaching the limit, but throws away all the body data in that case and stores the information that this has occurred. Then you can use this body type, handle it if within the limit, or return an error response if not.