Emulated Netlink sockets cause infinite busy loops under event-loop

[doanbaotrung]

When guest applications (such as avahi-daemon) run event loops that
wait for netlink messages using ppoll() or select(), the emulated
descriptor immediately returns POLLIN because the write end is
closed. The application's subsequent recvmsg or read call returns 0
bytes.

Because no sender credentials can be resolved from a 0-byte read,
the application ignores the read and immediately queries ppoll()
again. This causes the guest application to spin in an infinite
busy-loop consuming 100% CPU.

To address this, implement a non-blocking self-pipe signaling
mechanism inside src/syscall/netlink.c:

1. Store and use pipe_rd in netlink_state_t directly to avoid
   lockless global state reads in netlink_clear_readable.
2. Honor MSG_DONTWAIT and O_NONBLOCK flags and implement proper
   blocking semantics by polling the pipe outside nl_lock.
3. Signal readability only on empty-to-nonempty transition to
   prevent pipe buffer drift.
4. Handle zero-length reads/recvs before empty-buffer check.

Close #105

---

Summary by cubic

Fixes infinite busy loops in emulated Netlink sockets by signaling readiness only when data exists and enforcing correct blocking. Stops 100% CPU spin in ppoll/select loops (e.g., avahi-daemon) and honors MSG_DONTWAIT/O_NONBLOCK.

• Bug Fixes
  • Add a per-socket non-blocking self-pipe; store rd/wr in state (avoid lockless reads), set both O_NONBLOCK, keep write end open, and clean up on close.
  • Signal only on empty→non-empty; drain the pipe when buffers empty; handle 0-length read/recv/recvmsg first to avoid spurious POLLIN.
  • Enforce blocking correctly: return -EAGAIN for non-blocking; otherwise poll the pipe outside nl_lock; pass flags from sys_recvfrom to netlink_recv.

^{Written for commit 035e92a. Summary will update on new commits.}

[doanbaotrung]

Fix #105

[jserv]

Netlink busy-loop fix review. The core approach is sound: keeping the pipe write end open and signaling readiness only when a response is buffered fixes the spurious peer-closed POLLIN that spun avahi-daemon. A few correctness and robustness points below.

Append Close #105 at the end of git commit messages.

[jserv]

netlink_signal_readable() writes a token on every successful send. Since each request overwrites ns->buf (buf_pos reset to 0), repeated sends without an intervening recv keep pushing bytes into the pipe until it fills and the writes silently drop on EAGAIN. Harmless today, but it makes the pipe level drift from ns->buf state and can mask a real missed-wakeup later. Signal only on the empty-to-nonempty transition: capture bool was_empty = ns->buf_pos >= ns->buf_len before nl_process_request() and write only when it goes non-empty. Same at line 667 (netlink_send).

[jserv]

fd_set_nonblock() failure returns -LINUX_EMFILE, which is misleading -- this is an fcntl failure, not fd exhaustion. Preserve the real errno (return -linux_errno() from the failing call) after closing both pipe fds.

[doanbaotrung]

Address netlink review feedback on busy-loop fix

1. Store and use pipe_rd in netlink_state_t directly to avoid
   lockless global state reads in netlink_clear_readable.
2. Honor MSG_DONTWAIT and O_NONBLOCK flags and implement proper
   blocking semantics by polling the pipe outside nl_lock.
3. Signal readability only on empty-to-nonempty transition to
   prevent pipe buffer drift.
4. Handle zero-length reads/recvs before empty-buffer check.

[jserv]

Squash commits.

[jserv]

Add proper comments.

[doanbaotrung]

Added

[jserv]

Be aware of preferable code style.

[jserv]

Use brackets only when necessary.

[doanbaotrung]

fixed

[jserv]

Use brackets only when necessary.

[doanbaotrung]

fixed

[jserv]

Use brackets only when necessary.

[doanbaotrung]

fixed

[jserv]

Use brackets only when necessary.

[doanbaotrung]

fixed

[jserv]

Use brackets only when necessary.

[doanbaotrung]

fixed

[jserv]

Use brackets only when necessary.

[doanbaotrung]

fixed

[doanbaotrung]

Noted. I'll aware of using brackets only when necessary.