Received-SPF: pass (google.com: domain of linux-kernel+bounces-189180-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Fri, 24 May 2024 23:56:40 +0100
From: Al Viro <viro@zeniv.linux.org.uk>
To: Alice Ryhl <aliceryhl@google.com>
Cc: brauner@kernel.org, a.hindborg@samsung.com, alex.gaynor@gmail.com,
	arve@android.com, benno.lossin@proton.me, bjorn3_gh@protonmail.com,
	boqun.feng@gmail.com, cmllamas@google.com, dan.j.williams@intel.com,
	dxu@dxuuu.xyz, gary@garyguo.net, gregkh@linuxfoundation.org,
	joel@joelfernandes.org, keescook@chromium.org,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	maco@android.com, ojeda@kernel.org, peterz@infradead.org,
	rust-for-linux@vger.kernel.org, surenb@google.com,
	tglx@linutronix.de, tkjos@android.com, tmgross@umich.edu,
	wedsonaf@gmail.com, willy@infradead.org, yakoyoku@gmail.com
Subject: Re: [PATCH v6 3/8] rust: file: add Rust abstraction for `struct file`
Message-ID: <20240524225640.GU2118490@ZenIV>
References: <20240524-anhieb-bundesweit-e1b0227fd3ed@brauner>
 <20240524191714.2950286-1-aliceryhl@google.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240524191714.2950286-1-aliceryhl@google.com>
Sender: Al Viro <viro@ftp.linux.org.uk>

On Fri, May 24, 2024 at 07:17:13PM +0000, Alice Ryhl wrote:
> > And then those both implement a get_file() method so the caller can take
> > an explicit long-term reference to the file.
> 
> Even if you call `get_file` to get a long-term reference from something
> you have an fdget_pos reference to, that doesn't necessarily mean that
> you can share that long-term reference with other threads. You would
> need to release the fdget_pos reference first. For that reason, the
> long-term reference returned by `get_file` would still need to have the
> `File<MaybeFdgetPos>` type.

Why would you want such a bizarre requirement?

> Note that since it forgets which fd and fd table it came from, calls to
> `fdget` are actually not a problem for sending our long-term references
> across threads. The `fdget` requirements only care about things that
> touch the entry in the file descriptor table, such as closing the fd.
> The `ARef<File>` type does not provide any methods that could lead to
> that happening, so sharing it across threads is okay *even if* there is
> an light reference. That's why I have an `MaybeFdgetPos` but no
> `MaybeFdget`.

Huh?

What is "the entry in the file descriptor table"?  Which one and in which one?

> 	let file = File::fget(my_fd)?;
> 	// SAFETY: We know that there are no active `fdget_pos` calls on
> 	// the current thread, since this is an ioctl and we have not
> 	// called `fdget_pos` inside the Binder driver.
> 	let thread_safe_file = unsafe { file.assume_no_fdget_pos() };
> 
> (search for File::from_fd in the RFC to find where this would go)
> 
> The `assume_no_fdget_pos` call has no effect at runtime - it is purely a
> compile-time thing to force the user to use unsafe to "promise" that
> there aren't any `fdget_pos` calls on the same fd.

Why does fdget_pos() even matter?  The above makes no sense...

Again, cloning a reference and sending it to another thread is perfectly
fine.  And what's so special about fdget_pos()/fdput_pos() compared to
fdget()/fdput()?

_What_ memory safety issues are you talking about?