Skip to content

Day55 - SCM_RIGHTS / File Descriptor Passing

Summary

Today I learned Linux file descriptor passing using:

  • AF_UNIX socket
  • sendmsg()
  • recvmsg()
  • ancillary data
  • SCM_RIGHTS

I first created a minimal UNIX socketpair demo and verified communication between parent and child processes.

Then I implemented:

  • fdpass_send_fd()
  • fdpass_recv_fd()

using:

  • struct msghdr
  • struct iovec
  • struct cmsghdr

I also learned:

  • CMSG_SPACE()
  • CMSG_LEN()
  • CMSG_DATA()

and how ancillary data is attached to a socket message.

Important Concepts

SCM_RIGHTS does not send the fd integer itself.

Instead, Linux creates a new fd entry in the receiver process that references the same kernel struct file object.

This means:

  • file offset is shared
  • file status flags are shared
  • fd numbers are process-local

I verified this using:

  • shared lseek offset
  • O_RDONLY capability limitation

eventfd + epoll

I implemented an eventfd demo where:

  • parent created eventfd
  • parent passed eventfd to child
  • child monitored received fd using epoll

This confirmed that:

  • received fd behaves like a normal fd
  • epoll works correctly on passed fd

/dev/mypoll Integration

Finally, I integrated /dev/mypoll.

Architecture:

Parent:
    open("/dev/mypoll")
    pass fd to child

Child:
    recv fd
    epoll_wait()
    read events

The child successfully received:

  • timer events
  • manual trigger events

through epoll on the passed fd.

I also added socket disconnect handling using EPOLLHUP.

Key Takeaways

  • SCM_RIGHTS behaves like cross-process dup()
  • UNIX socket can be used as fd broker
  • epoll works normally on received fd
  • SOCK_SEQPACKET is very suitable for structured IPC
  • This architecture is commonly used in Linux daemon systems