Re: [PATCH 6/8] fsstress: implement the clonerange/deduperange ioctls

[Eryu Guan <eguan@redhat.com>]

On Thu, Dec 14, 2017 at 08:39:38AM +0200, Amir Goldstein wrote:
> On Wed, Dec 13, 2017 at 8:03 AM, Darrick J. Wong
> <darrick.wong@oracle.com> wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> >
> > Mix it up a bit by reflinking and deduping data blocks when possible.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> >  ltp/fsstress.c |  440 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 440 insertions(+)
> >
> >
> > diff --git a/ltp/fsstress.c b/ltp/fsstress.c
> > index 96f48b1..e2dfa5e 100644
> > --- a/ltp/fsstress.c
> > +++ b/ltp/fsstress.c
> > @@ -68,7 +68,9 @@ typedef enum {
> >         OP_BULKSTAT,
> >         OP_BULKSTAT1,
> >         OP_CHOWN,
> > +       OP_CLONERANGE,
> >         OP_CREAT,
> > +       OP_DEDUPERANGE,
> >         OP_DREAD,
> >         OP_DWRITE,
> >         OP_FALLOCATE,
> > @@ -174,7 +176,9 @@ void        awrite_f(int, long);
> >  void   bulkstat_f(int, long);
> >  void   bulkstat1_f(int, long);
> >  void   chown_f(int, long);
> > +void   clonerange_f(int, long);
> >  void   creat_f(int, long);
> > +void   deduperange_f(int, long);
> >  void   dread_f(int, long);
> >  void   dwrite_f(int, long);
> >  void   fallocate_f(int, long);
> > @@ -221,7 +225,9 @@ opdesc_t    ops[] = {
> >         { OP_BULKSTAT, "bulkstat", bulkstat_f, 1, 0 },
> >         { OP_BULKSTAT1, "bulkstat1", bulkstat1_f, 1, 0 },
> >         { OP_CHOWN, "chown", chown_f, 3, 1 },
> > +       { OP_CLONERANGE, "clonerange", clonerange_f, 4, 1 },
> >         { OP_CREAT, "creat", creat_f, 4, 1 },
> > +       { OP_DEDUPERANGE, "deduperange", deduperange_f, 4, 1},
> >         { OP_DREAD, "dread", dread_f, 4, 0 },
> >         { OP_DWRITE, "dwrite", dwrite_f, 4, 1 },
> >         { OP_FALLOCATE, "fallocate", fallocate_f, 1, 1 },
> > @@ -1312,6 +1318,16 @@ make_freq_table(void)
> >         }
> >  }
> >
> > +void
> > +free_freq_table(void)
> > +{
> > +       if (!freq_table)
> > +               return;
> > +       free(freq_table);
> > +       freq_table = NULL;
> > +       freq_table_size = 0;
> > +}
> > +
> >  int
> >  mkdir_path(pathname_t *name, mode_t mode)
> >  {
> > @@ -2189,6 +2205,430 @@ chown_f(int opno, long r)
> >         free_pathname(&f);
> >  }
> >
> > +static void
> > +disable_op(opty_t opt)
> > +{
> > +       opdesc_t        *p;
> > +
> > +       for (p = ops; p < ops_end; p++) {
> > +               if (opt == p->op) {
> > +                       p->freq = 0;
> > +                       free_freq_table();
> > +                       make_freq_table();
> > +                       return;
> > +               }
> > +       }
> > +}
> > +
> 
> If we want to go down the path of runtime disable ops, the question is:
> Why disable_op clonerange/deduperange and not disable_op
> insert/collapse/zero/punch? there are probably other ops as well.
> This is also inconsistent by the fact that build time disable of
> neither FALLOC nor LONERANGE/DEDUPERANGE will not
> change the random op sequence, while runtime disable will.

That's a good point. I don't think we want to disable unsupported
operations dynamically, fsstress just accepts/ignores whatever result
the operation gives and just prints the errno to log in verbose mode.

Thanks,
Eryu

> 
> This is related to the conversation of the golden output of xfs/068.
> Will take my arguments there now...
> 
> Amir.