[libvirt] [PATCH v2] test_driver: implement virDomainGetDiskErrors

[Ilias Stamatis]

On Wed, May 15, 2019 at 10:14 AM Michal Privoznik <mprivozn at redhat.com> wrote:
>
> On 5/14/19 5:24 PM, Ilias Stamatis wrote:
> > On Tue, May 14, 2019 at 5:04 PM Michal Privoznik <mprivozn at redhat.com> wrote:
> >>
> >> On 5/14/19 12:50 PM, Ilias Stamatis wrote:
> >>> On Tue, May 14, 2019 at 12:40 PM John Ferlan <jferlan at redhat.com> wrote:
> >>>>
> >>>>
> >>>>
> >>>> On 5/13/19 9:04 AM, Ilias Stamatis wrote:
> >>>>> On Mon, May 13, 2019 at 2:38 PM Michal Privoznik <mprivozn at redhat.com> wrote:
> >>>>>>
> >>>>>> On 5/13/19 1:26 AM, Ilias Stamatis wrote:
> >>>>>>> Return the number of disks present in the configuration of the test
> >>>>>>> domain when called with @errors as NULL and @maxerrors as 0.
> >>>>>>>
> >>>>>>> Otherwise report an error for every second disk, assigning available
> >>>>>>> error codes in a cyclic order.
> >>>>>>>
> >>>>>>> Signed-off-by: Ilias Stamatis <stamatis.iliass at gmail.com>
> >>>>>>> ---
> >>>>>>>     src/test/test_driver.c | 42 ++++++++++++++++++++++++++++++++++++++++++
> >>>>>>>     1 file changed, 42 insertions(+)
> >>>>>>>
> >>>>>>> diff --git a/src/test/test_driver.c b/src/test/test_driver.c
> >>>>>>> index a06d1fc402..527c2f5d3b 100644
> >>>>>>> --- a/src/test/test_driver.c
> >>>>>>> +++ b/src/test/test_driver.c
> >>>>>>> @@ -3046,6 +3046,47 @@ static int testDomainSetAutostart(virDomainPtr domain,
> >>>>>>>         return 0;
> >>>>>>>     }
> >>>>>>>
> >>>>>>> +static int testDomainGetDiskErrors(virDomainPtr dom,
> >>>>>>> +                                   virDomainDiskErrorPtr errors,
> >>>>>>> +                                   unsigned int maxerrors,
> >>>>>>> +                                   unsigned int flags)
> >>>>>>> +{
> >>>>
> >>>> [...]
> >>>>
> >>>>>>> +            n++;
> >>>>>>> +        }
> >>>>>>> +        ret = n;
> >>>>>>> +    }
> >>>>>>> +
> >>>>>>> + cleanup:
> >>>>>>> +    virDomainObjEndAPI(&vm);
> >>>>>>> +    if (ret < 0) {
> >>>>>>> +        for (i = 0; i < n; i++)
> >>>>>>> +            VIR_FREE(errors[i].disk);
> >>>>>>> +    }
> >>>>
> >>>> The above got changed to :
> >>>>
> >>>> + cleanup:
> >>>> +    virDomainObjEndAPI(&vm);
> >>>> +    if (ret < 0) {
> >>>> +        for (i = 0; i < MIN(vm->def->ndisks, maxerrors); i++)
> >>>> +            VIR_FREE(errors[i].disk);
> >>>> +    }
> >>>
> >>> I think this change is incorrect and a bug lies in here.
> >>>
> >>> If VIR_STRDUP fails above, memory for less than MIN(vm->def->ndisks,
> >>> maxerrors) will have been allocated, and then in the cleanup code
> >>> we'll call VIR_FREE with pointers that haven't been previously
> >>> allocated.
> >>
> >> That isn't a problem. User has to passed an array that we can touch. If
> >> they store some data in it, well, their fault - how are we supposed to
> >> return anything if we can't touch the array?
> >
> > I'm not sure I understand exactly what you mean.
> >
> > We can touch the array of course.
> >
> > What I'm saying is that we allocate memory with VIR_STRDUP for each
> > errors[i].disk, but if the call fails we free this memory on our own.
> >
> > However how it is implemented now we might call VIR_FREE on pointers
> > for which we have *not* allocated any memory.
> >
> > Because in the first loop, VIR_STRDUP might fail and send us to
> > "cleanup". But then on cleanup we iterate over the whole errors array.
> >
> > Isn't this incorrect? Do I understand something wrong?
>
>
> Ah, now I get it. If user passes an array that is not zeroed out then we
> might end up passing a random pointer to free(). How about this then?
>
>      if (ret < 0) {
>          while (i > 0)
>              VIR_FREE(errors[i--].disk);
>      }
>

Yes, this would work I think. And then the other changes in the
cleanup etc are not needed.

Ie it can be again:

    if (!(vm = testDomObjFromDomain(dom)))
        goto cleanup;

instead of "return -1" which is more consistent with the rest of the code.

However the code now returns errors for all disks. I thought we wanted
to report errors only for some of them?

In that case we would need to use the @n variable, as it was
initially. It is used in the same way in the qemu driver.