noq-udp panics on musl targets: assertion failed: align_of::<T>() <= align_of::<C>()

[ypopivniak]

Describe the bug
noq-udp panics on the first received packet on any musl target.

cmsg::decode() asserts align_of::<T>() <= align_of::<C>(). musl declares cmsghdr as { socklen_t cmsg_len, c_int __pad1, c_int cmsg_level, c_int cmsg_type } → align 4, whereas glibc uses size_t cmsg_len → align 8. The SCM_TIMESTAMPNS arm decodes libc::timespec (align 8), so the assert fails. Both operands are compile-time constants, so this is deterministic rather than data-dependent, and SO_TIMESTAMPNS is set unconditionally on Linux, so the path is always live.

Introduced by b454d1799 feat(udp): implement kernel receive timestamps on Linux/Android.

This is not memory unsafety: the control buffer is #[repr(align(8))], so the payload really is 8-aligned in practice. The assert compares the declared alignment of cmsghdr rather than the buffer alignment the code deliberately enforces.

Reproduces in noq's own test suite:

docker run --rm -v "$PWD:/w" -w /w rust:alpine \
  sh -c 'apk add --no-cache musl-dev >/dev/null; cargo test -p noq-udp'

Relevant Logs

thread 'gso' panicked at noq-udp/src/cmsg/mod.rs:81:5:
assertion failed: align_of::<T>() <= align_of::<C>()
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

failures:
    basic
    basic_src_ip
    ecn_v4
    ecn_v4_mapped_v6
    ecn_v6
    ecn_v6_dualstack
    gso
    recv_survives_icmp_unreachable_from_prior_send
    socket_buffers

test result: FAILED. 0 passed; 9 failed; 0 ignored; 0 measured; 0 filtered out

Expected behavior

• Packets are received on musl targets
• cargo test -p noq-udp passes

noq

Version: 1.1.0 (also reproduces on main @ 5cf07f4)

noq-udp v1.1.0
├── noq-udp feature "default" (command-line)
├── noq-udp feature "tracing"
│   ├── noq-udp feature "default" (command-line)
│   └── noq-udp feature "tracing-log"
│       └── noq-udp feature "default" (command-line)
└── noq-udp feature "tracing-log" (*)

Platform(s)
Desktop:

• OS: Alpine Linux (Docker rust:alpine, rustc 1.97.1) reproduced on both aarch64-unknown-linux-musl and x86_64-unknown-linux-musl
• Version 3.24.1

Desktop (also affected):

• OS: Ubuntu x86_64 running the statically-linked musl iroh-relay build
• Version 24.04.3 LTS

Additional Context / Screenshots / GIFs
Real-world impact: the official iroh-relay v1.0.3 static binary (from the n0computer/iroh-relay:v1.0.3 image) is musl-linked and crash-loops on first traffic. Because the cmsg decode happens before any protocol parsing, a single unauthenticated UDP datagram to the QUIC port is enough to abort the process so no handshake required. I confirmed that with nc against two otherwise-identical hosts, one patched and one not.

lolwut

• This issue was created by a human that thought critically about
  the issue being reported and wrote this as concisely and clearly
  as they could. Taking full responsibility for the issue being
  accurate.
• This issue isn't slop, and is backed up by evidence.

[flub]

This is pretty severe breakage, and the correct fix will take quite a bit of work I think from having a quick look (see my comment on the PR). So I wonder if we should do a smaller fix that just disables the recv timestamps for now unconditionally, and include the musl CI to that. That way we might be able to do a quick bugfix release while we take our time figuring the best way to do this properly.