Description of problem:
scsi hotplug add/remove panics in sysfs_hash_and_remove() due to a NULL
dentry pointer dereference when tearing down the sysfs entry for the sg
device node. We have a reliable reproducer script and a verified patch
from upstream, both (will be) attached to this BZ.

Version-Release number of selected component (if applicable):
All RHEL4 kernels. RHEL5 is not affected.

How reproducible:
The attached reproducer script can reliably reproduce this in seconds.

Steps to Reproduce:
1. boot up any RHEL4 kernel on a system with at least one unused
   scsi device. Note: cannot use qemu virtual scsi drives due to
   a different bug, but a scsi_debug device is OK and so is a real
   scsi device.
2. run the reproducer script.
3. splat
  
Actual results:
RIP: 0010:[<ffffffff801b5203>]
<ffffffff801b5203>{sysfs_hash_and_remove+14}
...
Call Trace:<ffffffff8024e2ba>{class_device_del+156}
<ffffffff8024e33e>{class_device_unregister+9}
<ffffffffa0009f3e>{:scsi_mod:scsi_remove_device+78}
<ffffffffa0009fd3>{:scsi_mod:sdev_store_delete+16}
<ffffffff8024c6a7>{dev_attr_store+29}
<ffffffff801b554f>{sysfs_write_file+194}
<ffffffff8017af0e>{vfs_write+207}
<ffffffff8017aff6>{sys_write+69}
<ffffffff8011026a>{system_call+126}

Expected results:
reproducer.sh runs forever