* optimize queue->lock, use atomic operations if possible
* do we need ctask->sg_count? can we use ctask->sg instead?
* optimize HEAD + SENDBUF logic, no need to have two "if" statements:


			}
			ctask->r2t = r2t;
			ctask->in_progress = IN_PROGRESS_SOLICIT_WRITE;
			/* can  goto to fixme */
		}

		if (ctask->in_progress == IN_PROGRESS_SOLICIT_WRITE) {
			r2t = ctask->r2t;
_solicit_again:
			/*
			 * send Data-Out's payload whitnin this R2T sequence.
			 */
			if (r2t->data_count) {
fixme:
				len = r2t->sendbuf.size - r2t->sendbuf.sent;
				if (len > r2t->data_count) {
					len = r2t->data_count;

* data_xmit(): recover sendpage() error.
* make sense to split data_recv() and exctract data-in processing into
  separated function. Call it from header processing context directly if
  there is an iSCSI data and from data_recv() later on conn state machine
  processing.
* restructure header processing code into set of small functions.
* prefetch in tcp_copy_bits().
* support residual_count in case of scatter-gather Data-In. As of today, only
  non-sg Scsi_Cmnd's are working. But I can amazing some target which will
  send residual_count in the middle of scatter-gather Data-In.

* daemon should manage its own ITT-space
* send_pdu() should use new "anon" poll of mtasks. the anon-mtask will be freed
  right after its xmited
* make InitiatorAlias configurable: see get_iscsi_initiatorname()
* Ming> ./usr/iscsiadm -m discovery -t st -p littleboy.iplink.net
        iscsiadm: cannot resolve host name littleboy.iplink