Am 26.09.2013 um 18:46 hat Benoît Canet geschrieben:
> Le Friday 08 Feb 2013 à 13:07:03 (+0100), Kevin Wolf a écrit :
> > Am 28.01.2013 18:07, schrieb Benoît Canet:
> > > Use gnutls's SHA-256 to compare versions.
> > >
> > > Signed-off-by: Benoit Canet <ben...@irqsave.net>
> > > ---
> > > block/quorum.c | 303
> > > +++++++++++++++++++++++++++++++++++++++++++++++++++++++-
> > > configure | 22 ++++
> > > 2 files changed, 324 insertions(+), 1 deletion(-)
> > > +static bool quorum_iovec_compare(QEMUIOVector *a, QEMUIOVector *b)
> > > +{
> > > + int i;
> > > + int result;
> > > +
> > > + assert(a->niov == b->niov);
> > > + for (i = 0; i < a->niov; i++) {
> > > + assert(a->iov[i].iov_len == b->iov[i].iov_len);
> > > + result = memcmp(a->iov[i].iov_base,
> > > + b->iov[i].iov_base,
> > > + a->iov[i].iov_len);
> > > + if (result) {
> > > + return false;
> > > + }
> > > + }
> > > +
> > > + return true;
> > > +}
> >
> > qemu_iovec_compare() seems to do exactly the same, except that this
> > doesn't return the offset of the first difference.
> >
> > Why is it a good idea to duplicate the code?
>
> Quorum need speed: qemu_iovec_compare will do byte level comparison whereas
> memcmp will use SSE on large block.
Fair enough.
> > > + /* we have a winner: copy it */
> > > + quorum_copy_qiov(acb->qiov, &acb->aios[winner->index].qiov);
> > > +
> > > + /* some versions are bad print them */
> > > + quorum_print_bad_versions(acb, &winner->value);
> >
> > Same here. Is this driver meant to be used in production or only for
> > debugging? Maybe it should have a debug mode that must explicitly be
> > enabled and messages are only printed in this mode?
>
> Quorum needs to communicate some event to the user for maintainance.
> Would thowing QMP events be better ?
Yes, absolutely. Error messages on stderr end up in a log file at best
and can only be read by human administrators. Using QMP enables
management tools to take action.
Kevin
