On Tue, Feb 20, 2001 at 04:34:33PM +0100, Steffen Neumann wrote:
> Jan Harkes <jaharkes_at_cs.cmu.edu> writes:
> 
> > On Fri, Feb 02, 2001 at 04:00:13PM +0100, Steffen Neumann wrote:
> [...]
> > Coda should support all basic read/write/mmap related system calls. It
> > looks like a subtle pagecache problem in the kernel. The data might be
> > written into the Coda inode page mappings, and read from the underlying
> > container inode pages. Adding an fsync right before the close might
> > actually show that the problem disappears.
> 
> Hi,
> 
> We now got mysql to work. I don't really know why and how,
> but telling config.h not to use pwrite (outcomment HAVE_PWRITE)
> solved the problem.
> 
> Does it make sense to look deeper into that ?
> I'd be willing to do so, if told what and how. 

I just looked at the kernel code (fs/read_write.c) and I can't see how
it could lead to any different behaviour in Coda.

    sys_write  calls f->f_op->write(f, buf, count, &f->f_pos)
    sys_pwrite calls f->f_op->write(f, buf, count, &pos)

Jan