+ /* lookup for existing socket */
+ sock = sockfd_lookup(transport_fd, &err);
+ if (!sock) {
+ printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
+ return -EEXIST;
+ }
+
+ /* lookup for existing connection */
+ spin_lock_bh(&session->lock);
+ list_for_each_entry(tmp, &session->connections, item) {
+ if (tmp == conn) {
+ if (conn->c_stage != ISCSI_CONN_STOPPED ||
+ conn->stop_stage == STOP_CONN_TERM) {
+ printk(KERN_ERR "iscsi_tcp: can't bind "
+ "non-stopped connection (%d:%d)\n",
+ conn->c_stage, conn->stop_stage);
+ spin_unlock_bh(&session->lock);
+ return -EIO;
+ }
+ break;
+ }
+ }
+ if (tmp != conn) {
+ /* bind new iSCSI connection to session */
+ conn->session = session;
+
+ list_add(&conn->item, &session->connections);
+ }
+ spin_unlock_bh(&session->lock);
+
+ if (conn->stop_stage != STOP_CONN_SUSPEND) {
+ /* bind iSCSI connection and socket */
+ conn->sock = sock;
+
+ /* setup Socket parameters */
+ sk = sock->sk;
+ sk->sk_reuse = 1;
+ sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
+ sk->sk_allocation = GFP_ATOMIC;
+
+ /* FIXME: disable Nagle's algorithm */
+
+ /*
+ * Intercept TCP callbacks for sendfile like receive
+ * processing.
+ */
+ iscsi_conn_set_callbacks(conn);
+
+ conn->sendpage = conn->sock->ops->sendpage;
+
+ /*
+ * set receive state machine into initial state
+ */
+ conn->in_progress = IN_PROGRESS_WAIT_HEADER;
+ }
+
+ if (is_leading)
+ session->leadconn = conn;
+
+ /*
+ * Unblock xmitworker(), Login Phase will pass through.
+ */
+ clear_bit(SUSPEND_BIT, &conn->suspend_rx);
+ clear_bit(SUSPEND_BIT, &conn->suspend_tx);
+
+ return 0;
+}
+
+static int
+iscsi_conn_start(struct iscsi_cls_conn *cls_conn)
+{
+ struct iscsi_conn *conn = cls_conn->dd_data;
+ struct iscsi_session *session = conn->session;
+ struct sock *sk;
+
+ /* FF phase warming up... */
+
+ if (session == NULL) {
+ printk(KERN_ERR "iscsi_tcp: can't start unbound connection\n");
+ return -EPERM;
+ }
+
+ sk = conn->sock->sk;
+
+ write_lock_bh(&sk->sk_callback_lock);
+ spin_lock_bh(&session->lock);
+ conn->c_stage = ISCSI_CONN_STARTED;
+ session->state = ISCSI_STATE_LOGGED_IN;
+
+ switch(conn->stop_stage) {
+ case STOP_CONN_RECOVER:
+ /*
+ * unblock eh_abort() if it is blocked. re-try all
+ * commands after successful recovery
+ */
+ session->conn_cnt++;
+ conn->stop_stage = 0;
+ conn->tmabort_state = TMABORT_INITIAL;
+ session->age++;
+ wake_up(&conn->ehwait);
+ break;
+ case STOP_CONN_TERM:
+ session->conn_cnt++;
+ conn->stop_stage = 0;
+ break;
+ case STOP_CONN_SUSPEND:
+ conn->stop_stage = 0;
+ clear_bit(SUSPEND_BIT, &conn->suspend_rx);
+ clear_bit(SUSPEND_BIT, &conn->suspend_tx);
+ break;
+ default:
+ break;
+ }
+ spin_unlock_bh(&session->lock);
+ write_unlock_bh(&sk->sk_callback_lock);
+
+ return 0;
+}
+
+static void
+iscsi_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
+{
+ struct iscsi_conn *conn = cls_conn->dd_data;
+ struct iscsi_session *session = conn->session;
+ struct sock *sk;
+ unsigned long flags;
+
+ BUG_ON(!conn->sock);
+ sk = conn->sock->sk;
+ write_lock_bh(&sk->sk_callback_lock);
+ set_bit(SUSPEND_BIT, &conn->suspend_rx);
+ write_unlock_bh(&sk->sk_callback_lock);
+
+ mutex_lock(&conn->xmitmutex);
+
+ spin_lock_irqsave(session->host->host_lock, flags);
+ spin_lock(&session->lock);
+ conn->stop_stage = flag;
+ conn->c_stage = ISCSI_CONN_STOPPED;
+ set_bit(SUSPEND_BIT, &conn->suspend_tx);
+
+ if (flag != STOP_CONN_SUSPEND)
+ session->conn_cnt--;
+
+ if (session->conn_cnt == 0 || session->leadconn == conn)
+ session->state = ISCSI_STATE_FAILED;
+
+ spin_unlock(&session->lock);
+ spin_unlock_irqrestore(session->host->host_lock, flags);
+
+ if (flag == STOP_CONN_TERM || flag == STOP_CONN_RECOVER) {
+ struct iscsi_cmd_task *ctask;
+ struct iscsi_mgmt_task *mtask;
+
+ /*
+ * Socket must go now.
+ */
+ sock_hold(conn->sock->sk);
+ iscsi_conn_restore_callbacks(conn);
+ sock_put(conn->sock->sk);
+
+ /*
+ * flush xmit queues.
+ */
+ spin_lock_bh(&session->lock);
+ while (__kfifo_get(conn->writequeue, (void*)&ctask,
+ sizeof(void*)) ||
+ __kfifo_get(conn->xmitqueue, (void*)&ctask,
+ sizeof(void*))) {
+ struct iscsi_r2t_info *r2t;
+
+ /*
+ * flush ctask's r2t queues
+ */
+ while (__kfifo_get(ctask->r2tqueue, (void*)&r2t,
+ sizeof(void*)))
+ __kfifo_put(ctask->r2tpool.queue, (void*)&r2t,
+ sizeof(void*));
+
+ spin_unlock_bh(&session->lock);
+ local_bh_disable();
+ iscsi_ctask_cleanup(conn, ctask);
+ local_bh_enable();
+ spin_lock_bh(&session->lock);
+ }
+ conn->ctask = NULL;
+ while (__kfifo_get(conn->immqueue, (void*)&mtask,
+ sizeof(void*)) ||
+ __kfifo_get(conn->mgmtqueue, (void*)&mtask,
+ sizeof(void*))) {
+ __kfifo_put(session->mgmtpool.queue,
+ (void*)&mtask, sizeof(void*));
+ }
+ conn->mtask = NULL;
+ spin_unlock_bh(&session->lock);
+
+ /*
+ * release socket only after we stopped data_xmit()
+ * activity and flushed all outstandings
+ */
+ sock_release(conn->sock);
+ conn->sock = NULL;
+
+ /*
+ * for connection level recovery we should not calculate
+ * header digest. conn->hdr_size used for optimization
+ * in hdr_extract() and will be re-negotiated at
+ * set_param() time.
+ */
+ if (flag == STOP_CONN_RECOVER) {
+ conn->hdr_size = sizeof(struct iscsi_hdr);
+ conn->hdrdgst_en = 0;
+ conn->datadgst_en = 0;
+ }
+ }
+ mutex_unlock(&conn->xmitmutex);
+}
+
+static int
+iscsi_conn_send_generic(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
+ char *data, uint32_t data_size)
+{
+ struct iscsi_session *session = conn->session;
+ struct iscsi_nopout *nop = (struct iscsi_nopout *)hdr;
+ struct iscsi_mgmt_task *mtask;
+
+ spin_lock_bh(&session->lock);
+ if (session->state == ISCSI_STATE_TERMINATE) {
+ spin_unlock_bh(&session->lock);
+ return -EPERM;
+ }
+ if (hdr->opcode == (ISCSI_OP_LOGIN | ISCSI_OP_IMMEDIATE) ||
+ hdr->opcode == (ISCSI_OP_TEXT | ISCSI_OP_IMMEDIATE))
+ /*
+ * Login and Text are sent serially, in
+ * request-followed-by-response sequence.
+ * Same mtask can be used. Same ITT must be used.
+ * Note that login_mtask is preallocated at conn_create().
+ */
+ mtask = conn->login_mtask;
+ else {
+ BUG_ON(conn->c_stage == ISCSI_CONN_INITIAL_STAGE);
+ BUG_ON(conn->c_stage == ISCSI_CONN_STOPPED);
+
+ if (!__kfifo_get(session->mgmtpool.queue,
+ (void*)&mtask, sizeof(void*))) {
+ spin_unlock_bh(&session->lock);
+ return -ENOSPC;
+ }
+ }
+
+ /*
+ * pre-format CmdSN and ExpStatSN for outgoing PDU.
+ */
+ if (hdr->itt != cpu_to_be32(ISCSI_RESERVED_TAG)) {
+ hdr->itt = mtask->itt | (conn->id << CID_SHIFT) |
+ (session->age << AGE_SHIFT);
+ nop->cmdsn = cpu_to_be32(session->cmdsn);
+ if (conn->c_stage == ISCSI_CONN_STARTED &&
+ !(hdr->opcode & ISCSI_OP_IMMEDIATE))
+ session->cmdsn++;
+ } else
+ /* do not advance CmdSN */
+ nop->cmdsn = cpu_to_be32(session->cmdsn);
+
+ nop->exp_statsn = cpu_to_be32(conn->exp_statsn);
+
+ memcpy(&mtask->hdr, hdr, sizeof(struct iscsi_hdr));
+
+ iscsi_buf_init_virt(&mtask->headbuf, (char*)&mtask->hdr,
+ sizeof(struct iscsi_hdr));
+
+ spin_unlock_bh(&session->lock);
+
+ if (data_size) {
+ memcpy(mtask->data, data, data_size);
+ mtask->data_count = data_size;
+ } else
+ mtask->data_count = 0;
+
+ mtask->xmstate = XMSTATE_IMM_HDR;
+
+ if (mtask->data_count) {
+ iscsi_buf_init_iov(&mtask->sendbuf, (char*)mtask->data,
+ mtask->data_count);
+ }
+
+ debug_scsi("mgmtpdu [op 0x%x hdr->itt 0x%x datalen %d]\n",
+ hdr->opcode, hdr->itt, data_size);
+
+ /*
+ * since send_pdu() could be called at least from two contexts,
+ * we need to serialize __kfifo_put, so we don't have to take
+ * additional lock on fast data-path
+ */
+ if (hdr->opcode & ISCSI_OP_IMMEDIATE)
+ __kfifo_put(conn->immqueue, (void*)&mtask, sizeof(void*));
+ else
+ __kfifo_put(conn->mgmtqueue, (void*)&mtask, sizeof(void*));
+
+ scsi_queue_work(session->host, &conn->xmitwork);
+ return 0;
+}
+
+static int
+iscsi_eh_host_reset(struct scsi_cmnd *sc)
+{
+ struct iscsi_cmd_task *ctask = (struct iscsi_cmd_task *)sc->SCp.ptr;
+ struct iscsi_conn *conn = ctask->conn;
+ struct iscsi_session *session = conn->session;
+
+ spin_lock_bh(&session->lock);
+ if (session->state == ISCSI_STATE_TERMINATE) {
+ debug_scsi("failing host reset: session terminated "
+ "[CID %d age %d]", conn->id, session->age);
+ spin_unlock_bh(&session->lock);
+ return FAILED;
+ }
+ spin_unlock_bh(&session->lock);
+
+ debug_scsi("failing connection CID %d due to SCSI host reset "
+ "[itt 0x%x age %d]", conn->id, ctask->itt,
+ session->age);
+ iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
+
+ return SUCCESS;
+}
+
+static void
+iscsi_tmabort_timedout(unsigned long data)
+{
+ struct iscsi_cmd_task *ctask = (struct iscsi_cmd_task *)data;
+ struct iscsi_conn *conn = ctask->conn;
+ struct iscsi_session *session = conn->session;
+
+ spin_lock(&session->lock);
+ if (conn->tmabort_state == TMABORT_INITIAL) {
+ __kfifo_put(session->mgmtpool.queue,
+ (void*)&ctask->mtask, sizeof(void*));
+ conn->tmabort_state = TMABORT_TIMEDOUT;
+ debug_scsi("tmabort timedout [sc %lx itt 0x%x]\n",
+ (long)ctask->sc, ctask->itt);
+ /* unblock eh_abort() */
+ wake_up(&conn->ehwait);
+ }
+ spin_unlock(&session->lock);
+}
+
+static int
+iscsi_eh_abort(struct scsi_cmnd *sc)
+{
+ int rc;
+ struct iscsi_cmd_task *ctask = (struct iscsi_cmd_task *)sc->SCp.ptr;
+ struct iscsi_conn *conn = ctask->conn;
+ struct iscsi_session *session = conn->session;
+
+ conn->eh_abort_cnt++;
+ debug_scsi("aborting [sc %lx itt 0x%x]\n", (long)sc, ctask->itt);
+
+ /*
+ * two cases for ERL=0 here:
+ *
+ * 1) connection-level failure;
+ * 2) recovery due protocol error;
+ */
+ mutex_lock(&conn->xmitmutex);
+ spin_lock_bh(&session->lock);
+ if (session->state != ISCSI_STATE_LOGGED_IN) {
+ if (session->state == ISCSI_STATE_TERMINATE) {
+ spin_unlock_bh(&session->lock);
+ mutex_unlock(&conn->xmitmutex);
+ goto failed;
+ }
+ spin_unlock_bh(&session->lock);
+ } else {
+ struct iscsi_tm *hdr = &conn->tmhdr;
+
+ /*
+ * Still LOGGED_IN...
+ */
+
+ if (!ctask->sc || sc->SCp.phase != session->age) {
+ /*
+ * 1) ctask completed before time out. But session
+ * is still ok => Happy Retry.
+ * 2) session was re-open during time out of ctask.
+ */
+ spin_unlock_bh(&session->lock);
+ mutex_unlock(&conn->xmitmutex);
+ goto success;
+ }
+ conn->tmabort_state = TMABORT_INITIAL;
+ spin_unlock_bh(&session->lock);
+
+ /*
+ * ctask timed out but session is OK
+ * ERL=0 requires task mgmt abort to be issued on each
+ * failed command. requests must be serialized.
+ */
+ memset(hdr, 0, sizeof(struct iscsi_tm));
+ hdr->opcode = ISCSI_OP_SCSI_TMFUNC | ISCSI_OP_IMMEDIATE;
+ hdr->flags = ISCSI_TM_FUNC_ABORT_TASK;
+ hdr->flags |= ISCSI_FLAG_CMD_FINAL;
+ memcpy(hdr->lun, ctask->hdr.lun, sizeof(hdr->lun));
+ hdr->rtt = ctask->hdr.itt;
+ hdr->refcmdsn = ctask->hdr.cmdsn;
+
+ rc = iscsi_conn_send_generic(conn, (struct iscsi_hdr *)hdr,
+ NULL, 0);
+ if (rc) {
+ iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
+ debug_scsi("abort sent failure [itt 0x%x]", ctask->itt);
+ } else {
+ struct iscsi_r2t_info *r2t;
+
+ /*
+ * TMF abort vs. TMF response race logic
+ */
+ spin_lock_bh(&session->lock);
+ ctask->mtask = (struct iscsi_mgmt_task *)
+ session->mgmt_cmds[(hdr->itt & ITT_MASK) -
+ ISCSI_MGMT_ITT_OFFSET];
+ /*
+ * have to flush r2tqueue to avoid r2t leaks
+ */
+ while (__kfifo_get(ctask->r2tqueue, (void*)&r2t,
+ sizeof(void*))) {
+ __kfifo_put(ctask->r2tpool.queue, (void*)&r2t,
+ sizeof(void*));
+ }
+ if (conn->tmabort_state == TMABORT_INITIAL) {
+ conn->tmfcmd_pdus_cnt++;
+ conn->tmabort_timer.expires = 3*HZ + jiffies;
+ conn->tmabort_timer.function =
+ iscsi_tmabort_timedout;
+ conn->tmabort_timer.data = (unsigned long)ctask;
+ add_timer(&conn->tmabort_timer);
+ debug_scsi("abort sent [itt 0x%x]", ctask->itt);
+ } else {
+ if (!ctask->sc ||
+ conn->tmabort_state == TMABORT_SUCCESS) {
+ conn->tmabort_state = TMABORT_INITIAL;
+ spin_unlock_bh(&session->lock);
+ mutex_unlock(&conn->xmitmutex);
+ goto success;
+ }
+ conn->tmabort_state = TMABORT_INITIAL;
+ iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
+ }
+ spin_unlock_bh(&session->lock);
+ }
+ }
+ mutex_unlock(&conn->xmitmutex);
+
+
+ /*
+ * block eh thread until:
+ *
+ * 1) abort response;
+ * 2) abort timeout;
+ * 3) session re-opened;
+ * 4) session terminated;
+ */
+ for (;;) {
+ int p_state = session->state;
+
+ rc = wait_event_interruptible(conn->ehwait,
+ (p_state == ISCSI_STATE_LOGGED_IN ?
+ (session->state == ISCSI_STATE_TERMINATE ||
+ conn->tmabort_state != TMABORT_INITIAL) :
+ (session->state == ISCSI_STATE_TERMINATE ||
+ session->state == ISCSI_STATE_LOGGED_IN)));
+ if (rc) {
+ /* shutdown.. */
+ session->state = ISCSI_STATE_TERMINATE;
+ goto failed;
+ }
+
+ if (signal_pending(current))
+ flush_signals(current);
+
+ if (session->state == ISCSI_STATE_TERMINATE)
+ goto failed;
+
+ spin_lock_bh(&session->lock);
+ if (sc->SCp.phase == session->age &&
+ (conn->tmabort_state == TMABORT_TIMEDOUT ||
+ conn->tmabort_state == TMABORT_FAILED)) {
+ conn->tmabort_state = TMABORT_INITIAL;
+ if (!ctask->sc) {
+ /*
+ * ctask completed before tmf abort response or
+ * time out.
+ * But session is still ok => Happy Retry.
+ */
+ spin_unlock_bh(&session->lock);
+ break;
+ }
+ spin_unlock_bh(&session->lock);
+ iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
+ continue;
+ }
+ spin_unlock_bh(&session->lock);
+ break;
+ }
+
+success:
+ debug_scsi("abort success [sc %lx itt 0x%x]\n", (long)sc, ctask->itt);
+ rc = SUCCESS;
+ goto exit;
+
+failed:
+ debug_scsi("abort failed [sc %lx itt 0x%x]\n", (long)sc, ctask->itt);
+ rc = FAILED;
+
+exit:
+ del_timer_sync(&conn->tmabort_timer);
+
+ mutex_lock(&conn->xmitmutex);
+ if (conn->sock) {
+ struct sock *sk = conn->sock->sk;
+
+ write_lock_bh(&sk->sk_callback_lock);
+ iscsi_ctask_cleanup(conn, ctask);
+ write_unlock_bh(&sk->sk_callback_lock);
+ }
+ mutex_unlock(&conn->xmitmutex);
+ return rc;
+}
+
+static int
+iscsi_r2tpool_alloc(struct iscsi_session *session)
+{
+ int i;
+ int cmd_i;
+
+ /*
+ * initialize per-task: R2T pool and xmit queue
+ */
+ for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
+ struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
+
+ /*
+ * pre-allocated x4 as much r2ts to handle race when
+ * target acks DataOut faster than we data_xmit() queues
+ * could replenish r2tqueue.
+ */
+
+ /* R2T pool */
+ if (iscsi_pool_init(&ctask->r2tpool, session->max_r2t * 4,
+ (void***)&ctask->r2ts, sizeof(struct iscsi_r2t_info))) {
+ goto r2t_alloc_fail;
+ }
+
+ /* R2T xmit queue */
+ ctask->r2tqueue = kfifo_alloc(
+ session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
+ if (ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
+ iscsi_pool_free(&ctask->r2tpool, (void**)ctask->r2ts);
+ goto r2t_alloc_fail;
+ }
+
+ /*
+ * number of
+ * Data-Out PDU's within R2T-sequence can be quite big;
+ * using mempool
+ */
+ ctask->datapool = mempool_create_slab_pool(ISCSI_DTASK_DEFAULT_MAX,
+ taskcache);
+ if (ctask->datapool == NULL) {
+ kfifo_free(ctask->r2tqueue);
+ iscsi_pool_free(&ctask->r2tpool, (void**)ctask->r2ts);
+ goto r2t_alloc_fail;
+ }
+ INIT_LIST_HEAD(&ctask->dataqueue);
+ }
+
+ return 0;
+
+r2t_alloc_fail:
+ for (i = 0; i < cmd_i; i++) {
+ mempool_destroy(session->cmds[i]->datapool);
+ kfifo_free(session->cmds[i]->r2tqueue);
+ iscsi_pool_free(&session->cmds[i]->r2tpool,
+ (void**)session->cmds[i]->r2ts);
+ }
+ return -ENOMEM;
+}
+
+static void
+iscsi_r2tpool_free(struct iscsi_session *session)
+{
+ int i;
+
+ for (i = 0; i < session->cmds_max; i++) {
+ mempool_destroy(session->cmds[i]->datapool);
+ kfifo_free(session->cmds[i]->r2tqueue);
+ iscsi_pool_free(&session->cmds[i]->r2tpool,
+ (void**)session->cmds[i]->r2ts);
+ }
+}
+
+static struct scsi_host_template iscsi_sht = {
+ .name = "iSCSI Initiator over TCP/IP, v."
+ ISCSI_VERSION_STR,
+ .queuecommand = iscsi_queuecommand,