[LTP] [PATCH v2 1/2] syscalls/sync_file_range: add partial file sync test-case

[Sumit Garg]

On Wed, 6 Mar 2019 at 14:24, Cyril Hrubis <chrubis@suse.cz> wrote:
>
> Hi!
> > > Looking at this the function is nearly the same as the other one, I
> > > guess that we may as well define the function as:
> > >
> > > static void verify_sync_file_range(off64_t off, off64_t size, char byte)
> > > {
> > >         ...
> > > }
> > >
> >
> > Yeah it could be made common. But we may not be able to reuse it for
> > third case you mentioned below.
>
> We may as well add offset and size for the write as well.
>

Ok.

> > > Also I'm not sure I was clear enough, but I was suggesting to check for
> > > upper bound for the synced size as well, which is why I suggested to do
> > > full write, sync only part of it, then check that the size was within
> > > bounds, i.e. >= size and  <= size + epsilon.
> > >
> >
> > Do you see any value add of upper bound check? AFAIK, device writes
> > continue in back-end and we might not be sure about appropriate value
> > for "epsilon".
>
> AFAIK this makes sense as far the file written fits into RAM buffers.
> We do write followed by a sync that asks particular range to be written
> to the device, so unless we manage to write significant portion of the
> file while we are calling the write syscall in a loop we will get quite
> close to the expectations. I guess that we can also check how much data
> was written right after we finished writing to the file and if that
> number is neglectible it would be fairly safe to check for upper bound.
> Does that sound reasonable to you?
>

Yes, I see number is negligible after write to file. So will add this
upper bound check with epsilon = 10%.

-Sumit

> > > I guess that we can even extend this to call the sync over a range that
> > > has been only partially written, but for that we would have to be
> > > careful and make sure all the data has been either synced at the end of
> > > the test function or use a different file for each test.
> > >
> >
> > I think using different file for each case looks more appropriate.
>
> Agreed.
>
> --
> Cyril Hrubis
> chrubis@suse.cz