Re: nfs mmap adventure (was: 2.6.23-mm1)

[Trond Myklebust]

On Mon, 2007-10-15 at 15:06 +0100, David Howells wrote:
> Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> 
> > I get funny SIGBUS' like so:
> > 
> > fault
> >   if (->page_mkwrite() < 0)
> >     nfs_vm_page_mkwrite()
> >       nfs_write_begin()
> >         nfs_flush_incompatible()
> >           nfs_wb_page()
> >             nfs_wb_page_priority()
> >               nfs_sync_mapping_wait()
> >                 nfs_wait_on_request_locked()
> >                   nfs_wait_on_request()
> >                     nfs_wait_bit_interruptible()
> >                       return -ERESTARTSYS
> >     SIGBUS
> > 
> > trying to figure out what to do about this...
> > 
> 
> Hmmm...  It sounds like the fault handler should deliver the appropriate
> signal, should ->page_mkwrite() return ERESTARTSYS, and then retry the access
> instruction that caused the fault when the signal handler has finished
> running.

If you signal the process before msync() has completed, or before you
have completed unmapping the region then your writes can potentially be
lost. Why should we be providing any guarantees beyond that?

Trond

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/