Re: [PATCH]qla4xxx:Add support for Async Message PDUs

[Mike Christie <michaelc@cs.wisc.edu>]

David C Somayajulu wrote:
> On Thu, 2008-04-24 at 11:40 -0500, Mike Christie wrote:
>> David C Somayajulu wrote:
>>> On Tue, 2008-04-22 at 11:20 -0500, Mike Christie wrote:
>>>> David C Somayajulu wrote:
>>>>> +int qla4xxx_conn_close_sess_logout(struct scsi_qla_host * ha,
>>>>> +	   uint16_t fw_ddb_index, uint16_t connection_id, uint16_t option)
>>>>> +{
>>>>> +	uint32_t mbox_cmd[MBOX_REG_COUNT];
>>>>> +	uint32_t mbox_sts[MBOX_REG_COUNT];
>>>>> +
>>>>> +	memset(&mbox_cmd, 0, sizeof(mbox_cmd));
>>>>> +	memset(&mbox_sts, 0, sizeof(mbox_sts));
>>>>> +
>>>>> +	mbox_cmd[0] = MBOX_CMD_CONN_CLOSE_SESS_LOGOUT;
>>>>> +	mbox_cmd[1] = fw_ddb_index;
>>>>> +	mbox_cmd[2] = connection_id;
>>>>> +	mbox_cmd[3] = LOGOUT_OPTION_RESET;
>>>>> +
>>>>> +	if (qla4xxx_mailbox_command(ha, MBOX_REG_COUNT, 2, &mbox_cmd[0],
>>>>> +		&mbox_sts[0]) != QLA_SUCCESS) {
>>>>> +		DEBUG2(printk("scsi%ld: %s: MBOX_CMD_CONN_CLOSE_SESS_LOGOUT "
>>>>> +			      "option %04x failed sts %04X %04X",
>>>>> +			      ha->host_no, __func__,
>>>>> +			      option, mbox_sts[0], mbox_sts[1]));
>>>>> +		if (mbox_sts[0] == 0x4005)
>>>>> +			DEBUG2(printk("%s reason %04X\n", __func__,
>>>>> +				      mbox_sts[1]));
>>>>> +	}
>>>>> +	return QLA_SUCCESS;
>>>> Does this logout function logout a session and stop it from being logged 
>>>> back in, or does it logout the session then the firmware tries to log 
>>>> back in?
>>> Once the session is logged out, it stays in that state till the user
>>> initiates a login. The firmware does not automatically try to login
>>> again.
>> I think we want to login in this case. Did you guys test with this 
>> Equalogic boxes, or what boxes did you try that send a logout request 
>> and always do not want you to relogin?
>>
>> EQL boxes have that magic target load balancing, where if they detect a 
>> problem the box will send the async logout request. It then expects the 
>> initiator to try a relogin, and at that time it will use the login 
>> redirect feature to send us to a new portal.
> Mike, you are correct. Async Logout is for recovery (stated in RFC5048).
> I have since confirmed with one of the storage vendors as well. I am
> trying to see if there is a Time2Wait (seconds) type thing the initiator
> needs to honor before it does the re-login. I will repost the patch.

There is. I think in the dropping connections related asyn pdus there is 
a time2wait value, and with the logout request asyn pdu, you will get a 
time2wait in the logout response.