What should listxattr() return if there are no xattrs?

What should listxattr() return if there are no xattrs?

[Theodore Ts'o]

Wang Sheng-Hui has sent some patches that change listxattr for the extN
file systems to return ENODATA when an inode has no extended attributes
(instead of zero).

I've noted that neither btrfs nor XFS do this, and would have rejected
the patches on that basis. However, I'll note the man page for the
listxattr(2) syscall is ambiguous:

    On  success,  a  positive number is returned indicating the size of the
    extended attribute name list.  On failure, -1 is returned and errno  is
    set appropriately.

It states that on success a _positive_ number is returned.  I've looked
and this wordining is in SGI's man page as well.  I suspect the man page
should be corrected to sate that "non-negative" number is returned?

							- Ted

Re: What should listxattr() return if there are no xattrs?

[Aneesh Kumar K. V]

On Mon, 20 Dec 2010 08:26:53 -0500, "Theodore Ts'o" <tytso@mit.edu> wrote:
> 
> Wang Sheng-Hui has sent some patches that change listxattr for the extN
> file systems to return ENODATA when an inode has no extended attributes
> (instead of zero).
> 
> I've noted that neither btrfs nor XFS do this, and would have rejected
> the patches on that basis. However, I'll note the man page for the
> listxattr(2) syscall is ambiguous:
> 
>     On  success,  a  positive number is returned indicating the size of the
>     extended attribute name list.  On failure, -1 is returned and errno  is
>     set appropriately.
> 
> It states that on success a _positive_ number is returned.  I've looked
> and this wordining is in SGI's man page as well.  I suspect the man page
> should be corrected to sate that "non-negative" number is returned?

I agree that man page needs update. Note that getfattr code does

	length = do_listxattr(path, NULL, 0);
	if (length < 0) {
		fprintf(stderr, "%s: %s: %s\n", progname, xquote(path, "\n\r"),
			strerror_ea(errno));
		had_errors++;
		return 1;
	} else if (length == 0)
		return 0;
		

-aneesh

Re: What should listxattr() return if there are no xattrs?

[Jeremy Allison]

On Mon, Dec 20, 2010 at 08:26:53AM -0500, Theodore Ts'o wrote:
> 
> Wang Sheng-Hui has sent some patches that change listxattr for the extN
> file systems to return ENODATA when an inode has no extended attributes
> (instead of zero).
> 
> I've noted that neither btrfs nor XFS do this, and would have rejected
> the patches on that basis. However, I'll note the man page for the
> listxattr(2) syscall is ambiguous:
> 
>     On  success,  a  positive number is returned indicating the size of the
>     extended attribute name list.  On failure, -1 is returned and errno  is
>     set appropriately.
> 
> It states that on success a _positive_ number is returned.  I've looked
> and this wordining is in SGI's man page as well.  I suspect the man page
> should be corrected to sate that "non-negative" number is returned?

Samba calls listxattr() when deciding what EA's are present on
a file. Right now examining our code I think we'd survive a
return of -1/ENODATA on an absense of EA's, but it's a risky
change to make for existing applications.

Keeping consistency with btrfs and XFS would seem to be a
better idea (IMHO).

Jeremy.

Re: What should listxattr() return if there are no xattrs?

[Michael Kerrisk (man-pages)]

(Following up long after the fact...)

Thanks for CCing me Ted.

On Mon, Dec 20, 2010 at 2:26 PM, Theodore Ts'o <tytso@mit.edu> wrote:
>
> Wang Sheng-Hui has sent some patches that change listxattr for the extN
> file systems to return ENODATA when an inode has no extended attributes
> (instead of zero).
>
> I've noted that neither btrfs nor XFS do this, and would have rejected
> the patches on that basis. However, I'll note the man page for the
> listxattr(2) syscall is ambiguous:
>
>     On  success,  a  positive number is returned indicating the size of the
>     extended attribute name list.  On failure, -1 is returned and errno  is
>     set appropriately.
>
> It states that on success a _positive_ number is returned.  I've looked
> and this wordining is in SGI's man page as well.  I suspect the man page
> should be corrected to sate that "non-negative" number is returned?

Thanks. Fixed now, to say "nonnegative".

Cheers,

Michael


-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Author of "The Linux Programming Interface"; http://man7.org/tlpi/