Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

eventfd, abstract socket and FUSE cleanup still match by fd number

Open
  • Is something wrong?
#411 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 2 days

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
35/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
c
Domain
backend

Research direction

Start by reading #408 and the fd-generation handling in src/syscall/fd.c; then trace eventfd_close and sys_eventfd2, absock_fd_cleanup and sys_close in src/syscall/net.c, and the FUSE cleanup paths in src/syscall/fuse.c. Run tests/test-special-fd-reuse.c and the eventfd reproduction described here. Done means cleanup cannot affect a replacement fd or lose its reference, and the reproduced reuse failures are fixed.

Written by the indexing model from the issue text.

Description

bug

fd_cleanup_entry runs a type's cleanup callback after fd_snapshot_and_close has freed the guest fd number, so a sibling thread's open can hold the number before the callback runs. #408 hands the callback the generation of the slot being closed, and pidfd, usbdev, timerfd, signalfd, inotify and netlink tear down by it. Three callbacks receive the generation and still match on the fd number alone:

  • eventfd_close (src/syscall/fd.c) reads eventfd_owner[guest_fd], clears it and drops a reference on that slot. When a sibling's eventfd2 has taken the number and set its own mapping, the close clears that mapping and drops that eventfd's reference. Aliases share one slot through eventfd_owner[], so the generation belongs on each owner mapping, not on the slot. sys_eventfd2 has a second window on the create side: fd_alloc publishes the fd before eventfd_owner[gfd] is written under sfd_lock. A close in that window finds no owner and drops no reference, and the write that follows leaves the slot holding a reference no fd owns. When the number has been reused by then, the write also replaces the other eventfd's mapping. eventfd_dup_fd closes this window for its destination by reserving the reference before the fd is published and committing under fd_lock and sfd_lock; sys_eventfd2 does neither.
  • absock_fd_cleanup (src/syscall/net.c) calls absock_unregister_fd(guest_fd), which unlinks the socket file of the absock_table entry holding that number and deactivates it. When a sibling has bound an abstract name on the reused number, its name is removed while its socket is open. sys_close also calls absock_unregister_fd directly, by number, ahead of the close, so keying the table by generation means that caller has to read the fd's generation first.
  • fuse_fd_cleanup (src/syscall/fuse.c) calls fuse_unbind_file_fd_locked(guest_fd) and fuse_unbind_dev_fd_locked(guest_fd), and both match guest_fd alone. A close that reaches a replacement's binding removes it and drops its alias reference. The bindings are shared between aliases through a refcount, so the per-fd match the other tables use does not carry over unchanged.

eventfd is reproduced; abstract sockets and FUSE are not. Four threads that each run eventfd, dup, close of the dup, an 8-byte write and read, and close for 1000 rounds: 3206 to 3252 of the 4000 eventfd() calls return ENOMEM over three runs, 18 to 23 of the write-read pairs fail, and with every fd closed afterwards 1 of the 32 slots can be allocated. tests/test-special-fd-reuse.c from #408 is the same shape: four threads that open, dup and close one kind of fd, each checking the fd it holds with a call that fails once the state behind it is gone. That test failed 5 of 5 runs for each of timerfd, signalfd, inotify and netlink before their tables matched by generation.

Dominant language
C
Stars
271
Forks
28
Avg merge
2d 18h
Merged PRs (30d)
19

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from sysprog21/elfuse

All issues in sysprog21/elfuse

Similar issues

More C issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.