Re: [PATCH] correct readlinkat attribute access (BZ #27024)

[Siddhesh Poyarekar <siddhesh@gotplt.org>]

On 1/8/21 6:19 AM, Martin Sebor via Libc-alpha wrote:
> The mode in the new attribute access added in g:06febd8c670
> to readlinkat() for the third argument is wrong: it's read_only
> but the function writes into the object.  The mode should be
> write_only.
> 
> The trivial patch below corrects this mistake.  Unless there
> are objections I'll go ahead and commit it in the next few days.
> 
> Martin
> 
> diff --git a/posix/unistd.h b/posix/unistd.h
> index 1306aded46..3f22763379 100644
> --- a/posix/unistd.h
> +++ b/posix/unistd.h
> @@ -831,7 +831,7 @@ extern int symlinkat (const char *__from, int __tofd,
>   /* Like readlink but a relative PATH is interpreted relative to FD.  */
>   extern ssize_t readlinkat (int __fd, const char *__restrict __path,
>                             char *__restrict __buf, size_t __len)
> -     __THROW __nonnull ((2, 3)) __wur __attr_access ((__read_only__, 3, 
> 4));
> +     __THROW __nonnull ((2, 3)) __wur __attr_access ((__write_only__, 
> 3, 4));
>   #endif

Why do we need an access attribute at all for __len, let alone a 
__write_only__ one?

Siddhesh