On Saturday 25 October 2003 22:27, Bill Studenmund wrote:

> On Sun, Oct 26, 2003 at 02:35:43AM +0900, YAMAMOTO Takashi wrote:
> > > This implementation copies the AIX system call of the same name.
> > > Including the requirement that the file descriptor be open for write
> > > for this call. The call will cause the data in the range to be
> > > written, and also cause some file metadata to be written.
> >
> > is the AIX version a cancellation point?
> >
> > at least, not listed here:
> > http://publib16.boulder.ibm.com/pseries/en_US/aixprggd/genprogc/term_th
> >reads.htm
>
> No, but to be honest I don't see why it isn't. fsync() is, so this one
> really should be too.
>
> The one thing that comes to mind is that that list looks very
> standards-driven, so since fsync_range is an extension, it might have
> gotten overlooked.

I agree.  In particular because in POSIX fdatasync(), to which it is more 
similar in a number of ways than to fsync(), is a thread cancellation point 
as well[*].


- Klaus

[*] The current standard doesn't reflect this, but its TC2 will.
    (XSH/TC2/D5/5 [XSH ERN 109] for the technically interested.)