Re: [PATCH 1/3] C implementation of the 'git' program, take two.

From: Andreas Ericsson <ae@op5.se>
Date: Nov 16, 2005, 00:42 UTC
Message-ID: <437A8067.9050308@op5.se>
In-Reply-To: <Pine.LNX.4.64.0511151603510.11232@g5.osdl.org>

Linus Torvalds wrote:

Show 15 quoted lines

> 
> On Wed, 16 Nov 2005, Andreas Ericsson wrote:
> 
>>+
>>+	/* allow relative paths, but run with exact */
>>+	if (chdir(exec_path)) {
>>+		printf("git: '%s': %s\n", exec_path, strerror(errno));
>>+		exit (1);
>>+	}
>>+
>>+	getcwd(git_command, sizeof(git_command));
>>+	chdir(wd);
> 
> 
> Argh. This is pretty horrible way to turn a path into an absolute one.

It was your idea to begin with, actually, in the thread on how to do proper path validation in the git-daemon. :)

Show 6 quoted lines

> 
> Why don't you just do
> 
> 	if (exec_path[0] != '/') {
> 		.. prepend "cwd/" to exec_path ..
>

You mean
	setenv("PATH", concat3(cwd, exec_path, old_path), 1);

Because that can fail too. Every solution is bad if you twist and turn it enough.

Show 7 quoted lines

> 
> The reason to avoid "chdir(relative) + chdir(back)" is that it totally 
> unnecessarily breaks under some extreme cases. For example, if the 
> exec_path is already absolute, and we just happen to be in a really deep 
> subdirectory, then the getcwd() could have failed due to the PATH_MAX 
> limitations.
>

True. So, would something like
char *try_goddamn_hard_to_get_working_dir(char *rel_path)
{
	size_t plen = PATH_MAX;
	char *p = malloc(PATH_MAX);
	while(p && plen < 1 << 20 && !(p = getcwd(p, plen))) {
		plen += PATH_MAX;
		p = realloc(p, plen);
	}

	return p;
}

be considered safe from this particular point of view?

Show 6 quoted lines

> Also, depending on getcwd() will not work if any parent directory is 
> unreadable or non-executable (well, under Linux it will, as long as it's 
> executable, since getcwd() is actually a system call. Not in UNIX in 
> general, though). Again, that means that unless you _have_ to know what 
> the cwd is, you should try to avoid relying on it.
>

True. I'll redo that part (tomorrow).

> 
> and the nicer thign to do is to just not try to be clever.
>

I never try. It just comes to me naturally. ;)

-- 
Andreas Ericsson                   andreas.ericsson@op5.se
OP5 AB                             www.op5.se
Tel: +46 8-230225                  Fax: +46 8-230231