On Tuesday 20 October 2015 15:22:22 Richard W.M. Jones wrote:
On Tue, Oct 20, 2015 at 03:50:31PM +0200, Pino Toscano wrote:
> On Tuesday 20 October 2015 14:43:53 Richard W.M. Jones wrote:
> > On Tue, Oct 20, 2015 at 01:59:10PM +0200, Pino Toscano wrote:
> > > Use RelativePathnameList as type for lists of relative paths, as used in
> > > some listing-alike APIs. This way we can ensure absolute paths in those
> > > lists are rejects outright.
> > >
> > > As a consequence, test-big-dirs.pl does not need to prepend the
> > > directory name anymore before calling listing-alike APIs: previously
> > > they didn't fail, but the returned lists contained only invalid
> > > elements (and only their size was checked).
> >
> > Are these all relative pathnames, or are they in fact just filenames
> > without any path at all. That is to say: is "foo/bar" permitted, or
> > just "bar"?
>
> At least with *lstat*list and *readlinklist functions, the file names
> are considered as relative wrt the path specified, as they are resolved
> against the file descriptor of the directory.
>
> In case of *lxattrlist, the absolute path+name for each is built and
> used as path within the guest.
>
> So yes, "bar", "foo/bar", and "../bar" too, should
work.
I think I really meant -- is it a caller bug if the parameter contains
a slash in it?
I don't know that :) I mean, considering even our code (see my change
in test-big-dirs.pl) was passing more than just a filename in the
specified path, then I thought relative filenames were somehow intended,
and thus I just rejected absolute ones.
I'm open to both positions though, either what I proposed in the patches
or just reject any file name not in the current directory.
All the *list functions were really intended as optimizations for
fuse/guestmount, and IIRC it was intended that only filenames (not
even relative paths) be used there.
They are used also in diff & ls nowadays (still with filenames only
though).
--
Pino Toscano