fix: handle invalid HTTP/2 headers gracefully (#4356)#5077
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #5077 +/- ##
==========================================
- Coverage 93.10% 93.08% -0.03%
==========================================
Files 110 110
Lines 35770 35853 +83
==========================================
+ Hits 33303 33372 +69
- Misses 2467 2481 +14 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
| // An error here means the server sent invalid HTTP/2 headers | ||
| // (e.g., HTTP/1 headers like "http2-settings"). This is a server error. | ||
| // We destroy the socket/session and error the request so the client | ||
| // can retry with a new connection. | ||
| util.errorRequest(client, request, err) | ||
| const socket = session[kSocket] | ||
| session.destroy(err) | ||
| util.destroy(socket, err) | ||
| return false |
There was a problem hiding this comment.
I don’t know how h2 works, but did this cause other "valid", ongoing requests that use the same socket/session to be interrupted?
When session.request() throws synchronously due to invalid HTTP/2 headers (e.g., duplicate headers like 'content-type' and 'Content-Type'), the error was not being caught, causing an uncaught exception. The fix wraps all session.request() calls with try-catch blocks. When an error is caught, only the request is errored (via util.errorRequest) and the function returns false. The session is not destroyed because the error is caused by the client sending invalid headers, not the server. This allows the client to continue using the same session for subsequent requests, and the error is properly caught by user code. Closes #4356
mcollina
force-pushed
the
fix/http2-invalid-headers
branch
from
dcf9d03 to
500eb60
Compare
metcoder95
left a comment
metcoder95
left a comment
There was a problem hiding this comment.
Tests seems, failing; rest lgtm
|
@metcoder95 PTAL |
cesarvspr
left a comment
cesarvspr
left a comment
There was a problem hiding this comment.
Thanks for this. I read through it and ran it locally.
The fix looks right to me. Wrapping session.request and only erroring the request, while leaving the session alive, matches the existing pattern just above at the upgrade check, and not destroying the session on a client side header mistake seems like the correct call. The second test is a nice regression guard: when I reverted the patch locally it failed exactly as expected, with an uncaughtException and code ERR_HTTP2_HEADER_SINGLE_VALUE, then passed again once restored.
One small thing, please take it or leave it. The first test, "invalid headers should be recoverable", sends two valid GETs and never actually sends an invalid header, so it stays green even with the fix reverted. It looks like the real coverage lives entirely in the second test. If it is useful, that first test could send a bad request, catch the throw, then reuse the same client for a good request, which would prove the session really did recover. Completely optional and non blocking.
Thanks again for sorting this out.
|
just conflicts to solve and ready to merge |
|
It was fixed in:
This one should be closed. |
5 participants
When session.request() throws synchronously due to invalid HTTP/2 headers (e.g., duplicate headers like "content-type" and "Content-Type"), the error was not being caught, causing an uncaught exception.
The fix wraps all session.request() calls with try-catch blocks. When an error is caught, only the request is errored (via util.errorRequest) and the function returns false. The session is not destroyed because the error is caused by the client sending invalid headers, not the server.
This allows the client to continue using the same session for subsequent requests, and the error is properly caught by user code.
Closes #4356