On Wed, Sep 30, 2009 at 11:24:38AM -0400, Christopher Faylor wrote:
>On Wed, Sep 30, 2009 at 06:07:29AM -0600, Eric Blake wrote:
>>-----BEGIN PGP SIGNED MESSAGE-----
>>Hash: SHA1
>>
>>My testing on rename found another corner case: we rejected
>>rename("dir","a/./") but accepted rename("dir","a/.//"). OK to commit?
>>
>>For reference, the test I am writing for hammering rename() and renameat()
>>corner cases is currently visible here; it will be part of the next
>>coreutils release, among other places. It currently stands at 400+ lines,
>>and exposes bugs in NetBSD, Solaris 10, mingw, and cygwin 1.5, but passes
>>on cygwin 1.7 (after this patch) and on Linux:
>>http://repo.or.cz/w/gnulib/ericb.git?a=blob;f=tests/test-rename.h
>>
>>2009-09-30 Eric Blake <e...@byu.net>
>>
>> * path.cc (has_dot_last_component): Detect "a/.//".
>
>No, I don't think so. I don't think this function is right. It
>shouldn't be doing a strrchr(dir, '//). And the formatting is off
>slightly.
>
>Is this function supposed to detect just "." or "*/."?
Assuming the answer is yes, then how about the below? I added a bunch of
comments but the function is still fairly small.
I've attached the function as-is since I basically rewrote it.
cgf
bool
has_dot_last_component (const char *dir, bool test_dot_dot)
{
/* SUSv3: . and .. are not allowed as last components in various system
calls. Don't test for backslash path separator since that's a Win32
path following Win32 rules. */
const char *last_comp = strrchr (dir, '\0');
if (last_comp == dir)
return false; /* Empty string. Probably shouldn't happen here? */
/* Detect run of trailing slashes */
while (last_comp > dir && *--last_comp == '/')
continue;
/* Detect just a run of slashes or a path that does not end with a slash. */
if (*last_comp != '.')
return false;
/* We know we have a trailing dot here. Check that it really is a standalone
"."
path component by checking that it is at the beginning of the string or is
preceded by a "/" */
if (last_comp == dir || *--last_comp == '/')
return true;
/* If we're not checking for '..' we're done. Ditto if we're now pointing to
a non-dot. */
if (!test_dot_dot || *last_comp != '.')
return false; /* either not testing for .. or this was not
'..' */
/* Repeat previous test for standalone or path component. */
return last_comp == dir || last_comp[-1] == '/';
}
