[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [address@hidden: Re: serious bug. Evolution and Microsoft mentality.

From: Marcus Brinkmann
Subject: Re: [address@hidden: Re: serious bug. Evolution and Microsoft mentality.]
Date: Fri, 11 Jan 2002 22:57:34 +0100
User-agent: Mutt/1.3.25i

On Fri, Jan 11, 2002 at 10:21:17PM +0100, Jeroen Dekkers wrote:
> On Fri, Jan 11, 2002 at 03:20:20AM +0000, Adam Olsen wrote:
> > The basic question is whether realpath() uses _POSIX_PATH_MAX as the
> > limit, which is a useful behavior (atleast as far as realpath goes),
> > or whether it uses no limit at all, which is as useless as gets().
> > 
> > As for portability, I think the best is to use canonicalize_file_name
> > when it's available, and fallback to realpath if it's not.  Otherwise,
> > although your program would compile, it'd be broken if you ever got a
> > path that was too long (either by crashing or getting the file wrong).
> > And don't say it won't happen.  The hurd is *alot* more flexible than
> > traditional systems, so it's more than possible somebody could dream
> > up a use for it in the future.
> I read the austin draft 7. The with realpath() generated pathname is
> stored as a nul-terminated string, up to PATH_MAX bytes. If PATH_MAX
> doesn't exist, there is just no limit. You never know how big the
> returned string and realpath() would just cause a buffer
> overflow. _POSIX_PATH_MAX has nothing to do with it, it just specifies
> a minimum for PATH_MAX, a system may not define a lower value.

Note that our current glibc implementation (stdlib/canonicalize.c) will
hardcode a fixed max size of 1024 for the string returned by realpath if
PATH_MAX and pathconf don't indicate a limit.

Yes, if you define your own PATH_MAX in your application, and use 512 (or
anything smaller than 1024), you have a potential buffer overflow for each
realpath() invocation if your program runs on the Hurd.

Obviously, realpath() behaviour on non-PATH_MAX systems is not specified.
Looks like a bug in POSIX to me.

On the GNU system, using canonicalize_file_name is the way to go.  All other
systems I know are safe because they define PATH_MAX.


`Rhubarb is no Egyptian god.' Debian
Marcus Brinkmann              GNU

reply via email to

[Prev in Thread] Current Thread [Next in Thread]