Re: FAILED: patch "[PATCH] io_uring: Clean up a false-positive warning from GCC 9.3.0" failed to apply to 5.10-stable tree

Re: FAILED: patch "[PATCH] io_uring: Clean up a false-positive warning from GCC 9.3.0" failed to apply to 5.10-stable tree

[Alviro Iskandar Setiawan]

On Sun, Jan 22, 2023 at 9:44 PM <gregkh@linuxfoundation.org> wrote:
> The patch below does not apply to the 5.10-stable tree.
> If someone wants it applied there, or to any other stable or longterm
> tree, then please email the backport, including the original git commit
> id to <stable@vger.kernel.org>.

That uninitialized reading is living in 5.10.y branch now
https://github.com/gregkh/linux/blob/v5.10.162/io_uring/io_uring.c#L4989-L5017

If this:

   ret = import_single_range(RE AD, buf, sr->len, &iov, &msg.msg_iter);

fails, this one (flags & MSG_WAITALL) may read an uninitialized
variable because @flags is uninitialized.

Fortunately, if import_single_range() fails, (ret < min_ret) is always
true, so this:

    ret < min_ret || ((flags & MSG_WAITALL)

will always short circuit. But no one tells the compiler if @ret is
always less than @min_ret in that case. So it can't prove that @flags
is never actually read. That still falls to undefined behavior anyway,
the compiler may emit "ud2" or similar trap for that or behave
randomly. IDK...

Re: FAILED: patch "[PATCH] io_uring: Clean up a false-positive warning from GCC 9.3.0" failed to apply to 5.10-stable tree

[Jens Axboe]

On 1/22/23 8:43 AM, Alviro Iskandar Setiawan wrote:
> On Sun, Jan 22, 2023 at 9:44 PM <gregkh@linuxfoundation.org> wrote:
>> The patch below does not apply to the 5.10-stable tree.
>> If someone wants it applied there, or to any other stable or longterm
>> tree, then please email the backport, including the original git commit
>> id to <stable@vger.kernel.org>.
> 
> That uninitialized reading is living in 5.10.y branch now
> https://github.com/gregkh/linux/blob/v5.10.162/io_uring/io_uring.c#L4989-L5017
> 
> If this:
> 
>    ret = import_single_range(RE AD, buf, sr->len, &iov, &msg.msg_iter);
> 
> fails, this one (flags & MSG_WAITALL) may read an uninitialized
> variable because @flags is uninitialized.
> 
> Fortunately, if import_single_range() fails, (ret < min_ret) is always
> true, so this:
> 
>     ret < min_ret || ((flags & MSG_WAITALL)
> 
> will always short circuit. But no one tells the compiler if @ret is
> always less than @min_ret in that case. So it can't prove that @flags
> is never actually read. That still falls to undefined behavior anyway,
> the compiler may emit "ud2" or similar trap for that or behave
> randomly. IDK...

Now handled for both trees.

-- 
Jens Axboe