/* We can't use the large write optimization if there are any failures * in the stripe. In the declustered layout, there is no way to * immediately determine what disks constitute a stripe, so we * actually have to hunt through the stripe looking for failures. The * reason we map the parity instead of just using asm->parityInfo->col * is because the latter may have been already redirected to a spare * drive, which would mess up the computation of the stripe offset. * * ASSUMES AT MOST ONE FAILURE IN THE STRIPE. */