Re: Re: fr1-2.16 patch for 2.6.10 kernel

"Also sprach Peter T. Breuer:"
> In article <[email protected]> you wrote:
> > So why p->rdev->bdev is NULL ?
> 
>   rdev->bdev is null?
> 
> Oh. OK. I'll take a look.

It's fairly clear that unlock_rdev() has been run.

static void unlock_rdev(mdk_rdev_t *rdev)
{
        struct block_device *bdev = rdev->bdev;
        rdev->bdev = NULL;
        ^^^^^^^^^^^^^^^^^^^
        if (!bdev)
                MD_BUG();
        bd_release(bdev);
        blkdev_put(bdev, BDEV_RAW);
}

It probably has been run from export_rdev()

static void export_rdev(mdk_rdev_t * rdev)
{
        char b[BDEVNAME_SIZE];
        printk(KERN_INFO "md: export_rdev(%s)\n",
                bdevname(rdev->bdev,b));
        if (rdev->mddev)
                MD_BUG();
        free_disk_sb(rdev);
        list_del_init(&rdev->same_set);
#ifndef MODULE
        md_autodetect_dev(rdev->bdev->bd_dev);
#endif
        unlock_rdev(rdev);
        ^^^^^^^^^^^^^^^^^^
        kfree(rdev);
}

And that could come from many places. In particular
kick_rdev_from_array().


static void kick_rdev_from_array(mdk_rdev_t * rdev)
{
        unbind_rdev_from_array(rdev);
        export_rdev(rdev);
        ^^^^^^^^^^^^^^^^^^^
}


And are you seeing this ONLY after the disk has failed to be recognized
on reinsertion, because of a different UUID?

Humour me .... can you perform your experiment NOT with scsi disks, but
with a pair of loopback devices as mirror components?

   dd if=/dev/zero bs=1k count=4k of=/tmp/core0
   dd if=/dev/zero bs=1k count=4k of=/tmp/core1

   losetup /dev/loop0 /tmp/core0
   losetup /dev/loop1 /tmp/core1

   mdadm -C -l 1 -n 2 -x 0 /dev/md0 /dev/loop{0,1}

??