Re: [PATCH] Fail on bad verify.

From: Shawn Lewis <shawnlewis_at_google.com>
Date: Thu, 2 Aug 2007 13:00:02 -0700

On 8/2/07, Jens Axboe <jens.axboe_at_oracle.com> wrote:
>
> On Wed, Aug 01 2007, Shawn Lewis wrote:
> > diff --git a/verify.c b/verify.c
> > index 5bdfb76..fadf1c0 100644
> > --- a/verify.c
> > +++ b/verify.c
> > @@ -349,7 +349,7 @@ int verify_io_u(struct thread_data *td,
> > struct verify_header *hdr;
> > unsigned int hdr_inc, hdr_num = 0;
> > void *p;
> > - int ret;
> > + int ret = 0;
> >
> > if (td->o.verify == VERIFY_NULL || io_u->ddir != DDIR_READ)
> > return 0;
> > @@ -401,7 +401,7 @@ int verify_io_u(struct thread_data *td,
> > hdr_num++;
> > }
> >
> > - return 0;
> > + return ret;
> > }
> >
> > static void fill_meta(struct verify_header *hdr, struct thread_data
> *td,
>
> Hmm, I think it's useful, but perhaps we also need an option to keep
> going? It may be interesting to continue verifying and see if the rest
> of the blocks are ok, or if there are other failures in that data set.
> Something like a
>
> verify_failure_fatal
>
> or whatever, I'm not very good with names. If you have a better option
> name, please tell me :-)

Actually I'd rather continue if there is a verify failure but it looks like
the original intention was to die, and it got lost somewhere. ret is now not
used. Maybe we should just always continue?

--
> Jens Axboe
>
>
Received on Thu Aug 02 2007 - 22:00:02 CEST

This archive was generated by hypermail 2.2.0 : Thu Aug 02 2007 - 22:30:01 CEST