change_sacl_perms() and ACLs from Solaris to 2.6 Linux

[Andrew Gideon]

On Mon, 02 Oct 2006 20:56:34 -0400, Matt McCutchen wrote:

> Rsync expects that stat(2) on a file whose ACL contains a mask entry will
> return the mask entry as the S_IRWXG mode bits.  Perhaps Solaris returns
> the group-owner entry no matter what; that would explain the trouble. 
> Would you please test and see if this is the case?

If I understand what you're asking, I have looked at this.  The proper
data is being sent to, and received by, the rsync running on the Linux
side.  

I displayed to stderr the ACL for the file before and after the code I
cited in change_sacl_perms().  The ACL is correct before that chunk and
wrong after that chunk.

I've found yet another problem along these lines, but it is different. 

I'm copying a directory and one file within the directory.  I've managed
to make the file copy have the correct ACL by changing the condition in
change_sacl_perms() to racl->mask != ACL_NO_ENTRY.  However, the
directory is showing a similar symptom: the mask is being set to the
group mode.

This is different, though, because it involves a second call
to set_file_attrs() for the directory.  The directory's ACL is correct up
to that point.  But within set_file_attrs() is a call to do_chmod() which
is apparently setting the mask of the directory incorrectly.

I'm still chasing that down, so I'm not sure.  I do know that the ACL of
the directory is correct on disk before the chunk of code within
HAVE_CHMOD and wrong afterward.

There's a first call to  set_file_attrs() for the directory.  There's
then a call to set_file_attrs() for the file.  And then there's a second
call to set_file_attrs() for the directory.  Before that second call for
the directory, the directory exists and has the proper ACL.  Afterward, it
has the wrong mask.

I don't know why there is a second call to set_file_attrs() for the
directory, or why it is doing something different.  I also don't know how
do_chmod() (which is ACL-ignorant) is changing the mask.  But as I wrote,
this one I'm still chasing.

BTW, these problems didn't exist in 2.6.2.

	- Andrew