Add Filesystem::syncfs() for FUSE_SYNCFS

[cberner]

Fixes #359.

FUSE_SYNCFS is opcode 50, ABI 7.34, carrying struct fuse_syncfs_in { uint64_t padding; } and answered with an empty reply. fuser reports 7.44, so nothing needs version gating on the receive side. Adds the opcode, the argument struct, an Operation::SyncFs, the dispatch, and Filesystem::syncfs().

Leaving syncfs() unimplemented reports ENOSYS, which is what the unknown-opcode path already produced, and the kernel then clears fc->sync_fs and stops asking for the lifetime of the connection. So this changes nothing for a filesystem that does not want it.

One thing worth knowing before merging

The kernel only propagates syncfs(2) on fuseblk and virtiofs connections. fc->sync_fs is set in exactly two places -- fuse_fill_super_common() under if (ctx->is_bdev), and virtio_fs.c -- and fuse_sync_fs() returns 0 immediately when it is unset. fuser mounts filesystem type fuse, never fuseblk, so on a filesystem fuser mounts the kernel never sends this and syncfs(2) succeeds without a round trip.

That leaves one reachable path: a session built with Session::from_fd on a fuseblk or virtiofs connection. The trait method and the changelog both say so, so nobody implements it expecting it to fire on an ordinary mount.

It is still worth having -- it is protocol coverage, it is what the issue asked for, and from_fd is public API -- but it is not something I can demonstrate end to end, and I would rather you knew that than found it later.

Testing

ll::request::tests::syncfs and syncfs_without_its_argument, in the byte-level style of the existing request tests, for both endiannesses. They cover the parse and that a request whose header declares no argument is rejected rather than parsed. Confirmed the second fails if the argument fetch is dropped.

No e2e test, for the reason above: the kernel will not send this to anything fuser can mount, so a fuser-tests case would assert nothing.

cargo fmt, all three clippy invocations from make pre, cargo doc, cargo test --all with and without libfuse, the musl build, cargo check --target x86_64-apple-darwin --features=macos-no-mount, and make test_passthrough are green. Docker was unavailable here, so mount_tests/pjdfs_tests/xfstests did not run locally; this adds an opcode none of them exercise.

🤖 Generated with Claude Code

https://claude.ai/code/session_01H5akmePV1VnD9YRQLZH4gv

---

Generated by Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 12cedba171

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".