Re: [PATCHv3 1/1] SCSI: update hosts module to use idr index management

[Lee Duncan <lduncan@suse.com>]

On 10/06/2015 12:19 PM, James Bottomley wrote:
> On Tue, 2015-10-06 at 12:08 -0700, Lee Duncan wrote:
>> Update the SCSI hosts module to use idr to manage
>> its host_no index instead of an ATOMIC integer. This
>> also allows using idr_find() to look up the SCSI
>> host structure given the host number.
>>
>> This means that the SCSI host number will now
>> be reclaimable.
>>
>> Signed-off-by: Lee Duncan <lduncan@suse.com>
>> Reviewed-by: Hannes Reinecke <hare@suse.com>
>> ---
>>  drivers/scsi/hosts.c | 61 ++++++++++++++++++++++++++--------------------------
>>  1 file changed, 31 insertions(+), 30 deletions(-
>>
>> diff --git a/drivers/scsi/hosts.c b/drivers/scsi/hosts.c
>> index 8bb173e01084..afe7bd962ddb 100644
>> --- a/drivers/scsi/hosts.c
>> +++ b/drivers/scsi/hosts.c
> [...]
>> +	spin_lock(&host_index_lock);
>> +	shost = idr_find(&host_index_idr, hostnum);
>> +	spin_unlock(&host_index_lock);
>> +
>> +	return shost ? scsi_host_get(shost) : NULL;
> 
> So the thing I don't like here is that there's a race between
> scsi_host_get() and the final put.  What could happen is that idr_find()
> returns the host just before but scsi_host_dev_release() is executed
> before the return.  In that instance, we'll reference freed memory in
> scsi_host_get() ... probably completely harmlessly, but it will show up
> occasionally on some of the traces ... particularly the ones doing a
> fuzz/stress test around host create/destroy.
> 
> James

Good point. The scenario you mention may be a corner case, but I also
don't like it. I cannot see another good way to synchronize lookup and
removal other than adding a new lock, which would be dumb.

I will submit my "host number" patch again, without the change to
scsi_host_lookup().

-- 
Lee Duncan