Reject DATAGRAMs larger than the send buffer

[zerox80]

Follow-up to #2625.

A DATAGRAM larger than datagram_send_buffer_size cannot fit in the outgoing send buffer by itself. This returns TooLarge before trying either send mode, so the behavior is the same for drop = true and drop = false.

Tested with:

cargo fmt --check
cargo test -p quinn-proto --lib
cargo test -p quinn-proto datagram

[zerox80]

Updated this too. The size check now happens before the drop branch, so drop = false returns TooLarge for a datagram that cannot fit in the send buffer. I added a test for both send modes.

[zerox80]

What is the current state?

[djc]

This looks okay to me, thanks.

Perhaps you'll want to backport it to the 0.11.x branch, too?

[zerox80]

This looks okay to me, thanks.

Perhaps you'll want to backport it to the 0.11.x branch, too?

Good point, yes, this should also be backported to 0.11.x. I checked that the same send buffer logic is still present there. I’ll address the small nits here first, and once this lands on main I’ll open a separate backport PR against 0.11.x

[djc]

Please squash the changes from review into their originating commits.

[zerox80]

Done, thanks. I rewrote the stack so the review changes are squashed into their originating commits, added the send_buffer_size local binding as the first commit, and kept the has_send_buffer_space extraction as its own separate commit.

Tested with:
cargo fmt --check
cargo test -p quinn-proto --lib
cargo test -p quinn-proto datagram

[djc]

LGTM, thanks for the iterations!