Re: [RFC 2/3] xfs_io: Add ext4 support to show FS_IOC_FSGETXATTR details

["Darrick J. Wong" <djwong@kernel.org>]

On Fri, Dec 13, 2024 at 07:44:01AM +1100, Dave Chinner wrote:
> On Thu, Dec 12, 2024 at 08:19:19AM -0800, Darrick J. Wong wrote:
> > On Thu, Dec 12, 2024 at 09:33:29AM +1100, Dave Chinner wrote:
> > > On Wed, Dec 11, 2024 at 10:17:06AM -0800, Darrick J. Wong wrote:
> > > > On Wed, Dec 11, 2024 at 01:24:03PM +0530, Ojaswin Mujoo wrote:
> > > > > Currently with stat we only show FS_IOC_FSGETXATTR details
> > > > > if the filesystem is XFS. With extsize support also coming
> > > > > to ext4 make sure to show these details when -c "stat" or "statx"
> > > > > is used.
> > > > > 
> > > > > No functional changes for filesystems other than ext4.
> > > > > 
> > > > > Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
> > > > > ---
> > > > >  io/stat.c | 38 +++++++++++++++++++++-----------------
> > > > >  1 file changed, 21 insertions(+), 17 deletions(-)
> > > > > 
> > > > > diff --git a/io/stat.c b/io/stat.c
> > > > > index 326f2822e276..d06c2186cde4 100644
> > > > > --- a/io/stat.c
> > > > > +++ b/io/stat.c
> > > > > @@ -97,14 +97,14 @@ print_file_info(void)
> > > > >  		file->flags & IO_TMPFILE ? _(",tmpfile") : "");
> > > > >  }
> > > > >  
> > > > > -static void
> > > > > -print_xfs_info(int verbose)
> > > > > +static void print_extended_info(int verbose)
> > > > >  {
> > > > > -	struct dioattr	dio;
> > > > > -	struct fsxattr	fsx, fsxa;
> > > > > +	struct dioattr dio;
> > > > > +	struct fsxattr fsx, fsxa;
> > > > > +	bool is_xfs_fd = platform_test_xfs_fd(file->fd);
> > > > >  
> > > > > -	if ((xfsctl(file->name, file->fd, FS_IOC_FSGETXATTR, &fsx)) < 0 ||
> > > > > -	    (xfsctl(file->name, file->fd, XFS_IOC_FSGETXATTRA, &fsxa)) < 0) {
> > > > > +	if ((ioctl(file->fd, FS_IOC_FSGETXATTR, &fsx)) < 0 ||
> > > > > +		(is_xfs_fd && (xfsctl(file->name, file->fd, XFS_IOC_FSGETXATTRA, &fsxa) < 0))) {
> > > > 
> > > > Urgh... perhaps we should call FS_IOC_FSGETXATTR and if it returns zero
> > > > print whatever is returned, no matter what filesystem we think is
> > > > feeding us information?
> > > 
> > > Yes, please. FS_IOC_FSGETXATTR has been generic functionality for
> > > some time, we should treat it the same way for all filesystems.
> > > 
> > > > e.g.
> > > > 
> > > > 	if (ioctl(file->fd, FS_IOC_FSGETXATTR, &fsx)) < 0) {
> > > > 		if (is_xfs_fd || (errno != EOPNOTSUPP &&
> > > > 				  errno != ENOTTY))
> > > > 			perror("FS_IOC_GETXATTR");
> > > 
> > > Why do we even need "is_xfs_fd" there? XFS will never give a
> > > EOPNOTSUPP or ENOTTY error to this or the FS_IOC_GETXATTRA ioctl...
> > 
> > Yeah, in hindsight I don't think it's needed for FS_IOC_FSGETXATTR, but
> 
> *nod*
> 
> > it's definitely nice for XFS_IOC_FSGETXATTRA (which is not implemented
> > outside xfs) so that you don't get unnecessary error messages on ext4.
> 
> I don't think we even need it for FS_IOC_GETXATTRA - if the
> filesystem does not support that ioctl, we don't print the fields,
> nor do we output an error.
> 
> After all, this "extended info" and it's only ever been printed
> for XFS, so we can define whatever semantics we want for foreign
> filesystem output right now. As long as XFS always prints the same
> info as it always has (i.e. all of it), we can do whatever we want
> with the foreign filesystem stuff.
> 
> Keep in mind that we don't need platform tests for XFS files - that
> has already been done when the file was opened and the state stored
> in file->flags via the IO_FOREIGN flag. We already use that in the
> stat_f() to determine whether we print the "xfs info" or not.
> 
> IOWs, I think all we need to do is  move where we check the
> IO_FOREIGN flag. i.e.:
> 
> print_extented_info(file)
> {
> 	struct dioattr  dio = {};
>         struct fsxattr  fsx = {}, fsxa = {};
> 
> 	if (ioctl(file->fd, FS_IOC_FSGETXATTR, &fsx)) < 0) {
> 		perror("FS_IOC_GETXATTR");
> 		exitcode = 1;
> 		return;
> 	}
> 
> 	printf(_("fsxattr.xflags = 0x%x "), fsx.fsx_xflags);
> 	printxattr(fsx.fsx_xflags, verbose, 0, file->name, 1, 1);
> 	printf(_("fsxattr.projid = %u\n"), fsx.fsx_projid);
> 	printf(_("fsxattr.extsize = %u\n"), fsx.fsx_extsize);
> 	printf(_("fsxattr.cowextsize = %u\n"), fsx.fsx_cowextsize);
> 	printf(_("fsxattr.nextents = %u\n"), fsx.fsx_nextents);
> 
> 	/* Only XFS supports FS_IOC_FSGETXATTRA and XFS_IOC_DIOINFO */
> 	if (file->flags & IO_FOREIGN)
> 		return;
> 
> 	if (ioctl(file->fd, FS_IOC_FSGETXATTRA, &fsxa)) < 0) {
> 		perror("FS_IOC_GETXATTRA");
> 		exitcode = 1;
> 		return;
> 	}
> 	if ((xfsctl(file->name, file->fd, XFS_IOC_DIOINFO, &dio)) < 0) {
> 		perror("XFS_IOC_DIOINFO");
> 		exitcode = 1;
> 		return;
> 	}
> 
> 	printf(_("fsxattr.naextents = %u\n"), fsxa.fsx_nextents);
> 	printf(_("dioattr.mem = 0x%x\n"), dio.d_mem);
> 	printf(_("dioattr.miniosz = %u\n"), dio.d_miniosz);
> 	printf(_("dioattr.maxiosz = %u\n"), dio.d_maxiosz);
> }
> 
> Thoughts?

Seems fine to me, though I'd print the fsxa before trying to call
DIOINFO.

--D

> -Dave.
> -- 
> Dave Chinner
> david@fromorbit.com
>