Re: [PATCH] Fix is_gitfile() for files larger than PATH_MAX

[Junio C Hamano <gitster@pobox.com>]

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

> The logic to check whether a file is a gitfile used the heuristics that
> the file cannot be larger than PATH_MAX. But in that case it returned the
> wrong value. Our test cases do not cover this, as the bundle files
> produced are smaller than PATH_MAX. Except on Windows.
>
> While at it, fix the faulty logic that the path stored in a gitfile cannot
> be larger than PATH_MAX-sizeof("gitfile: ").
>
> Problem identified by running the test suite in msysGit, offending commit
> identified by Jörg Rosenkranz.
>
> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> ---
> 	This patch should apply cleanly to 'next', which we track in
> 	msysgit/git.
>
> 	The task of adding a test case is something I leave to someone who
> 	wants to get involved with Git development and needs an easy way
> 	in.
>
>  transport.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/transport.c b/transport.c
> index f3195c0..57138d9 100644
> --- a/transport.c
> +++ b/transport.c
> @@ -868,8 +868,8 @@ static int is_gitfile(const char *url)
>  		return 0;
>  	if (!S_ISREG(st.st_mode))
>  		return 0;
> -	if (st.st_size < 10 || st.st_size > PATH_MAX)
> -		return 1;
> +	if (st.st_size < 10 || st.st_size > 9 + PATH_MAX)
> +		return 0;

We are asked if the file is likely to be a single-liner "gitfile: <path>",
and were answering yes when it is a very short file (less than 10 bytes)
that cannot possibly even contain "gitfile: " prefix.

I suspect that we can and should get rid of the "cannot be very long"
check altogether---we do open and check the file, and after all it is not
like we are throwing different strings as "url" argument to this function
at random and this function needs heuristics to reject bogus input
early. The argument is an input from the user.

Quite an embarrasing bug. Thanks for spotting.