sahvx655-wq opened a new pull request, #3758:
URL: https://github.com/apache/brpc/pull/3758
### What problem does this PR solve?
Issue Number: resolve
Problem Summary:
`http_parser_execute` (src/brpc/details/http_parser.cpp) skips everything
after `;` or SP on a chunk-size line until it sees CR (`s_chunk_parameters`),
so a bare LF inside a chunk extension is swallowed as extension bytes. Every
other line-scanning state in the parser treats LF as a line end, and front-ends
that do the same on the chunk-size line read the bytes after the LF as chunk
data. The two sides then disagree on chunk boundaries and on where the request
ends, which is the chunk-extension variant of HTTP request smuggling ("funky
chunks").
Concrete trigger, request body under `Transfer-Encoding: chunked`:
```
2;\nxx\r\n
31\r\n
0\r\n\r\nGET /smuggled HTTP/1.1\r\nHost: a.com\r\nX: \r\n\r\n
0\r\n\r\n
```
A proxy that ends the line at LF forwards one request with chunks `xx` and
0x31 bytes. brpc reads `31` as the 2-byte chunk, ends the request at
`0\r\n\r\n` and parses `GET /smuggled` as the next request on the connection,
bypassing whatever routing or access rules the proxy applied to it. RFC 9112
7.1.1 allows no CTL other than HTAB anywhere in chunk-ext (token,
quoted-string, BWS).
### What is changed and the side effects?
Changed:
- `s_chunk_parameters` rejects LF and other CTLs (everything
`IS_HEADER_CHAR` refuses) with `HPE_INVALID_CHUNK_SIZE`. CR still ends the line
as before. The check sits in the parser state itself because that is the only
place that knows the bytes belong to a chunk extension.
- `HttpMessageTest.ctl_in_chunk_extension` covers the smuggling payload
above, a bare LF and DEL in a normal chunk and a bare LF on the last-chunk
line, plus extensions that must keep parsing (`name=val`, quoted-string
containing SP and `;`, HTAB as BWS).
Verified by compiling http_parser.cpp on its own with a harness that pauses
on message_complete the same way `HttpMessage` does: on master the smuggling
payload completes the first request with body `31` and the other three rejected
cases are accepted; with the fix all four fail with `HPE_INVALID_CHUNK_SIZE`
and the accepted cases still parse. The new gtest was not run in a full brpc
build.
Side effects:
- Performance effects: one extra byte-class check per chunk-extension byte;
none for chunks without extensions.
- Breaking backward compatibility: peers that put raw control characters
other than HTAB in chunk extensions are now rejected. Those messages are
malformed under RFC 9112.
---
### Check List:
- Please make sure your changes are compilable.
- When providing us with a new feature, it is best to add related tests.
- Please follow [Contributor Covenant Code of
Conduct](https://github.com/apache/brpc/blob/master/CODE_OF_CONDUCT.md).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]