From: Junio C Hamano <junkio@cox.net>
Date: Sat, 24 Feb 2007 08:51:34 GMT
Subject: Re: [PATCH] Limit filename for format-patch
Message-ID: <7vtzxcym3t.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20070223222606.28660.46208.stgit@lathund.dewire.com>

Robin Rosenberg <robin.rosenberg@dewire.com> writes:

> -static void reopen_stdout(struct commit *commit, int nr, int keep_subject)
> +static int reopen_stdout(struct commit *commit, int nr, int keep_subject)
>  {
> -	char filename[1024];
> +	char filename[PATH_MAX];
>  	char *sol;
>  	int len = 0;
> -	int suffix_len = strlen(fmt_patch_suffix) + 10; /* ., NUL and slop */
> +	int suffix_len = strlen(fmt_patch_suffix) + 1;
>  
>  	if (output_directory) {
> -		strlcpy(filename, output_directory, 1000);
> +		if (strlen(output_directory) >= sizeof(filename) - 64 - suffix_len)
> +			return error("name of output directory is too long");
> +		strlcpy(filename, output_directory, sizeof(filename) - suffix_len);

Sorry for a late doubt, but I started wondering if we should use
NAME_MAX instead of hardcoded 64.  Purists might argue for using
pathconf() but I think it is an overkill.

NAME_MAX is 255 on Linux, POSIX says it should be 14 at least
(and further says if the platform supports only smaller max,
NAME_MAX should not be defined -- heh).  I do not know how
universal NAME_MAX is defined, and I hate dealing with header
incompatibility across different systems, so I am tempted to
just do something like:

#define FORMAT_PATCH_NAME_MAX 64

locally, and use that instead.