Re: [PATCH] Don't segfault if the given SHA1 name is longer than 40 characters

From: Junio C Hamano <junkio@cox.net>
Date: Oct 19, 2006, 04:13 UTC
Message-ID: <7vwt6xvt7q.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20061019013441.GB9379@localhost>

pclouds@gmail.com writes:

Show 18 quoted lines

> Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
> ---
>  sha1_name.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/sha1_name.c b/sha1_name.c
> index 9b226e3..6ffee22 100644
> --- a/sha1_name.c
> +++ b/sha1_name.c
> @@ -157,7 +157,7 @@ static int get_short_sha1(const char *na
>  	char canonical[40];
>  	unsigned char res[20];
>  
> -	if (len < MINIMUM_ABBREV)
> +	if (len < MINIMUM_ABBREV || len > 40)
>  		return -1;
>  	hashclr(res);
>  	memset(canonical, 'x', 40);

Thanks.

The patch is correct, but it needs a better explanation in the proposed commit log message than just "Don't segfault".

get_describe_name() can be fed a string foo-gXXX...XXX with a very looooooooong hexstring. It calls get_short_sha1() without checking if XXX...XXX part is longer than 40-bytes (in which case it cannot be a valid object name).

get_sha1_1() has the same problem. Given a hexstring longer than 40-bytes, it calls the function with it because earlier ones such as get_sha1_hex() would reject the input.