md: don't reset curr_resync_completed after an interrupted resync
[safe/jmp/linux-2.6] / drivers / scsi / libiscsi.c
index dafa054..b7689f3 100644 (file)
 #include <scsi/scsi_transport_iscsi.h>
 #include <scsi/libiscsi.h>
 
-static int iscsi_dbg_lib;
-module_param_named(debug_libiscsi, iscsi_dbg_lib, int, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(debug_libiscsi, "Turn on debugging for libiscsi module. "
-                "Set to 1 to turn on, and zero to turn off. Default "
-                "is off.");
+static int iscsi_dbg_lib_conn;
+module_param_named(debug_libiscsi_conn, iscsi_dbg_lib_conn, int,
+                  S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(debug_libiscsi_conn,
+                "Turn on debugging for connections in libiscsi module. "
+                "Set to 1 to turn on, and zero to turn off. Default is off.");
+
+static int iscsi_dbg_lib_session;
+module_param_named(debug_libiscsi_session, iscsi_dbg_lib_session, int,
+                  S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(debug_libiscsi_session,
+                "Turn on debugging for sessions in libiscsi module. "
+                "Set to 1 to turn on, and zero to turn off. Default is off.");
+
+static int iscsi_dbg_lib_eh;
+module_param_named(debug_libiscsi_eh, iscsi_dbg_lib_eh, int,
+                  S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(debug_libiscsi_eh,
+                "Turn on debugging for error handling in libiscsi module. "
+                "Set to 1 to turn on, and zero to turn off. Default is off.");
 
 #define ISCSI_DBG_CONN(_conn, dbg_fmt, arg...)                 \
        do {                                                    \
-               if (iscsi_dbg_lib)                              \
+               if (iscsi_dbg_lib_conn)                         \
                        iscsi_conn_printk(KERN_INFO, _conn,     \
                                             "%s " dbg_fmt,     \
                                             __func__, ##arg);  \
@@ -54,7 +69,15 @@ MODULE_PARM_DESC(debug_libiscsi, "Turn on debugging for libiscsi module. "
 
 #define ISCSI_DBG_SESSION(_session, dbg_fmt, arg...)                   \
        do {                                                            \
-               if (iscsi_dbg_lib)                                      \
+               if (iscsi_dbg_lib_session)                              \
+                       iscsi_session_printk(KERN_INFO, _session,       \
+                                            "%s " dbg_fmt,             \
+                                            __func__, ##arg);          \
+       } while (0);
+
+#define ISCSI_DBG_EH(_session, dbg_fmt, arg...)                                \
+       do {                                                            \
+               if (iscsi_dbg_lib_eh)                                   \
                        iscsi_session_printk(KERN_INFO, _session,       \
                                             "%s " dbg_fmt,             \
                                             __func__, ##arg);          \
@@ -86,12 +109,9 @@ inline void iscsi_conn_queue_work(struct iscsi_conn *conn)
 }
 EXPORT_SYMBOL_GPL(iscsi_conn_queue_work);
 
-void
-iscsi_update_cmdsn(struct iscsi_session *session, struct iscsi_nopin *hdr)
+static void __iscsi_update_cmdsn(struct iscsi_session *session,
+                                uint32_t exp_cmdsn, uint32_t max_cmdsn)
 {
-       uint32_t max_cmdsn = be32_to_cpu(hdr->max_cmdsn);
-       uint32_t exp_cmdsn = be32_to_cpu(hdr->exp_cmdsn);
-
        /*
         * standard specifies this check for when to update expected and
         * max sequence numbers
@@ -115,6 +135,12 @@ iscsi_update_cmdsn(struct iscsi_session *session, struct iscsi_nopin *hdr)
                        iscsi_conn_queue_work(session->leadconn);
        }
 }
+
+void iscsi_update_cmdsn(struct iscsi_session *session, struct iscsi_nopin *hdr)
+{
+       __iscsi_update_cmdsn(session, be32_to_cpu(hdr->exp_cmdsn),
+                            be32_to_cpu(hdr->max_cmdsn));
+}
 EXPORT_SYMBOL_GPL(iscsi_update_cmdsn);
 
 /**
@@ -240,6 +266,88 @@ static int iscsi_prep_bidi_ahs(struct iscsi_task *task)
 }
 
 /**
+ * iscsi_check_tmf_restrictions - check if a task is affected by TMF
+ * @task: iscsi task
+ * @opcode: opcode to check for
+ *
+ * During TMF a task has to be checked if it's affected.
+ * All unrelated I/O can be passed through, but I/O to the
+ * affected LUN should be restricted.
+ * If 'fast_abort' is set we won't be sending any I/O to the
+ * affected LUN.
+ * Otherwise the target is waiting for all TTTs to be completed,
+ * so we have to send all outstanding Data-Out PDUs to the target.
+ */
+static int iscsi_check_tmf_restrictions(struct iscsi_task *task, int opcode)
+{
+       struct iscsi_conn *conn = task->conn;
+       struct iscsi_tm *tmf = &conn->tmhdr;
+       unsigned int hdr_lun;
+
+       if (conn->tmf_state == TMF_INITIAL)
+               return 0;
+
+       if ((tmf->opcode & ISCSI_OPCODE_MASK) != ISCSI_OP_SCSI_TMFUNC)
+               return 0;
+
+       switch (ISCSI_TM_FUNC_VALUE(tmf)) {
+       case ISCSI_TM_FUNC_LOGICAL_UNIT_RESET:
+               /*
+                * Allow PDUs for unrelated LUNs
+                */
+               hdr_lun = scsilun_to_int((struct scsi_lun *)tmf->lun);
+               if (hdr_lun != task->sc->device->lun)
+                       return 0;
+               /* fall through */
+       case ISCSI_TM_FUNC_TARGET_WARM_RESET:
+               /*
+                * Fail all SCSI cmd PDUs
+                */
+               if (opcode != ISCSI_OP_SCSI_DATA_OUT) {
+                       iscsi_conn_printk(KERN_INFO, conn,
+                                         "task [op %x/%x itt "
+                                         "0x%x/0x%x] "
+                                         "rejected.\n",
+                                         task->hdr->opcode, opcode,
+                                         task->itt, task->hdr_itt);
+                       return -EACCES;
+               }
+               /*
+                * And also all data-out PDUs in response to R2T
+                * if fast_abort is set.
+                */
+               if (conn->session->fast_abort) {
+                       iscsi_conn_printk(KERN_INFO, conn,
+                                         "task [op %x/%x itt "
+                                         "0x%x/0x%x] fast abort.\n",
+                                         task->hdr->opcode, opcode,
+                                         task->itt, task->hdr_itt);
+                       return -EACCES;
+               }
+               break;
+       case ISCSI_TM_FUNC_ABORT_TASK:
+               /*
+                * the caller has already checked if the task
+                * they want to abort was in the pending queue so if
+                * we are here the cmd pdu has gone out already, and
+                * we will only hit this for data-outs
+                */
+               if (opcode == ISCSI_OP_SCSI_DATA_OUT &&
+                   task->hdr_itt == tmf->rtt) {
+                       ISCSI_DBG_SESSION(conn->session,
+                                         "Preventing task %x/%x from sending "
+                                         "data-out due to abort task in "
+                                         "progress\n", task->itt,
+                                         task->hdr_itt);
+                       return -EACCES;
+               }
+               break;
+       }
+
+       return 0;
+}
+
+/**
  * iscsi_prep_scsi_cmd_pdu - prep iscsi scsi cmd pdu
  * @task: iscsi task
  *
@@ -256,6 +364,10 @@ static int iscsi_prep_scsi_cmd_pdu(struct iscsi_task *task)
        itt_t itt;
        int rc;
 
+       rc = iscsi_check_tmf_restrictions(task, ISCSI_OP_SCSI_CMD);
+       if (rc)
+               return rc;
+
        if (conn->session->tt->alloc_pdu) {
                rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_CMD);
                if (rc)
@@ -278,8 +390,6 @@ static int iscsi_prep_scsi_cmd_pdu(struct iscsi_task *task)
        hdr->flags = ISCSI_ATTR_SIMPLE;
        int_to_scsilun(sc->device->lun, (struct scsi_lun *)hdr->lun);
        memcpy(task->lun, hdr->lun, sizeof(task->lun));
-       hdr->cmdsn = task->cmdsn = cpu_to_be32(session->cmdsn);
-       session->cmdsn++;
        hdr->exp_statsn = cpu_to_be32(conn->exp_statsn);
        cmd_len = sc->cmd_len;
        if (cmd_len < ISCSI_CDB_SIZE)
@@ -365,6 +475,8 @@ static int iscsi_prep_scsi_cmd_pdu(struct iscsi_task *task)
                return -EIO;
 
        task->state = ISCSI_TASK_RUNNING;
+       hdr->cmdsn = task->cmdsn = cpu_to_be32(session->cmdsn);
+       session->cmdsn++;
 
        conn->scsicmd_pdus_cnt++;
        ISCSI_DBG_SESSION(session, "iscsi prep [%s cid %d sc %p cdb 0x%x "
@@ -393,10 +505,12 @@ static void iscsi_free_task(struct iscsi_task *task)
        struct iscsi_session *session = conn->session;
        struct scsi_cmnd *sc = task->sc;
 
+       ISCSI_DBG_SESSION(session, "freeing task itt 0x%x state %d sc %p\n",
+                         task->itt, task->state, task->sc);
+
        session->tt->cleanup_task(task);
        task->state = ISCSI_TASK_FREE;
        task->sc = NULL;
-
        /*
         * login task is preallocated so do not free
         */
@@ -443,18 +557,23 @@ EXPORT_SYMBOL_GPL(iscsi_put_task);
 /**
  * iscsi_complete_task - finish a task
  * @task: iscsi cmd task
+ * @state: state to complete task with
  *
  * Must be called with session lock.
  */
-static void iscsi_complete_task(struct iscsi_task *task)
+static void iscsi_complete_task(struct iscsi_task *task, int state)
 {
        struct iscsi_conn *conn = task->conn;
 
-       if (task->state == ISCSI_TASK_COMPLETED)
+       ISCSI_DBG_SESSION(conn->session,
+                         "complete task itt 0x%x state %d sc %p\n",
+                         task->itt, task->state, task->sc);
+       if (task->state == ISCSI_TASK_COMPLETED ||
+           task->state == ISCSI_TASK_ABRT_TMF ||
+           task->state == ISCSI_TASK_ABRT_SESS_RECOV)
                return;
        WARN_ON_ONCE(task->state == ISCSI_TASK_FREE);
-
-       task->state = ISCSI_TASK_COMPLETED;
+       task->state = state;
 
        if (!list_empty(&task->running))
                list_del_init(&task->running);
@@ -469,6 +588,31 @@ static void iscsi_complete_task(struct iscsi_task *task)
        __iscsi_put_task(task);
 }
 
+/**
+ * iscsi_complete_scsi_task - finish scsi task normally
+ * @task: iscsi task for scsi cmd
+ * @exp_cmdsn: expected cmd sn in cpu format
+ * @max_cmdsn: max cmd sn in cpu format
+ *
+ * This is used when drivers do not need or cannot perform
+ * lower level pdu processing.
+ *
+ * Called with session lock
+ */
+void iscsi_complete_scsi_task(struct iscsi_task *task,
+                             uint32_t exp_cmdsn, uint32_t max_cmdsn)
+{
+       struct iscsi_conn *conn = task->conn;
+
+       ISCSI_DBG_SESSION(conn->session, "[itt 0x%x]\n", task->itt);
+
+       conn->last_recv = jiffies;
+       __iscsi_update_cmdsn(conn->session, exp_cmdsn, max_cmdsn);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
+}
+EXPORT_SYMBOL_GPL(iscsi_complete_scsi_task);
+
+
 /*
  * session lock must be held and if not called for a task that is
  * still pending or from the xmit thread, then xmit thread must
@@ -478,6 +622,7 @@ static void fail_scsi_task(struct iscsi_task *task, int err)
 {
        struct iscsi_conn *conn = task->conn;
        struct scsi_cmnd *sc;
+       int state;
 
        /*
         * if a command completes and we get a successful tmf response
@@ -488,14 +633,20 @@ static void fail_scsi_task(struct iscsi_task *task, int err)
        if (!sc)
                return;
 
-       if (task->state == ISCSI_TASK_PENDING)
+       if (task->state == ISCSI_TASK_PENDING) {
                /*
                 * cmd never made it to the xmit thread, so we should not count
                 * the cmd in the sequencing
                 */
                conn->session->queued_cmdsn--;
+               /* it was never sent so just complete like normal */
+               state = ISCSI_TASK_COMPLETED;
+       } else if (err == DID_TRANSPORT_DISRUPTED)
+               state = ISCSI_TASK_ABRT_SESS_RECOV;
+       else
+               state = ISCSI_TASK_ABRT_TMF;
 
-       sc->result = err;
+       sc->result = err << 16;
        if (!scsi_bidi_cmnd(sc))
                scsi_set_resid(sc, scsi_bufflen(sc));
        else {
@@ -503,7 +654,7 @@ static void fail_scsi_task(struct iscsi_task *task, int err)
                scsi_in(sc)->resid = scsi_in(sc)->length;
        }
 
-       iscsi_complete_task(task);
+       iscsi_complete_task(task, state);
 }
 
 static int iscsi_prep_mgmt_task(struct iscsi_conn *conn,
@@ -512,12 +663,12 @@ static int iscsi_prep_mgmt_task(struct iscsi_conn *conn,
        struct iscsi_session *session = conn->session;
        struct iscsi_hdr *hdr = task->hdr;
        struct iscsi_nopout *nop = (struct iscsi_nopout *)hdr;
+       uint8_t opcode = hdr->opcode & ISCSI_OPCODE_MASK;
 
        if (conn->session->state == ISCSI_STATE_LOGGING_OUT)
                return -ENOTCONN;
 
-       if (hdr->opcode != (ISCSI_OP_LOGIN | ISCSI_OP_IMMEDIATE) &&
-           hdr->opcode != (ISCSI_OP_TEXT | ISCSI_OP_IMMEDIATE))
+       if (opcode != ISCSI_OP_LOGIN && opcode != ISCSI_OP_TEXT)
                nop->exp_statsn = cpu_to_be32(conn->exp_statsn);
        /*
         * pre-format CmdSN for outgoing PDU.
@@ -525,9 +676,12 @@ static int iscsi_prep_mgmt_task(struct iscsi_conn *conn,
        nop->cmdsn = cpu_to_be32(session->cmdsn);
        if (hdr->itt != RESERVED_ITT) {
                /*
-                * TODO: We always use immediate, so we never hit this.
+                * TODO: We always use immediate for normal session pdus.
                 * If we start to send tmfs or nops as non-immediate then
                 * we should start checking the cmdsn numbers for mgmt tasks.
+                *
+                * During discovery sessions iscsid sends TEXT as non immediate,
+                * but we always only send one PDU at a time.
                 */
                if (conn->c_stage == ISCSI_CONN_STARTED &&
                    !(hdr->opcode & ISCSI_OP_IMMEDIATE)) {
@@ -555,22 +709,28 @@ __iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
 {
        struct iscsi_session *session = conn->session;
        struct iscsi_host *ihost = shost_priv(session->host);
+       uint8_t opcode = hdr->opcode & ISCSI_OPCODE_MASK;
        struct iscsi_task *task;
        itt_t itt;
 
        if (session->state == ISCSI_STATE_TERMINATE)
                return NULL;
 
-       if (hdr->opcode == (ISCSI_OP_LOGIN | ISCSI_OP_IMMEDIATE) ||
-           hdr->opcode == (ISCSI_OP_TEXT | ISCSI_OP_IMMEDIATE))
+       if (opcode == ISCSI_OP_LOGIN || opcode == ISCSI_OP_TEXT) {
                /*
                 * Login and Text are sent serially, in
                 * request-followed-by-response sequence.
                 * Same task can be used. Same ITT must be used.
                 * Note that login_task is preallocated at conn_create().
                 */
+               if (conn->login_task->state != ISCSI_TASK_FREE) {
+                       iscsi_conn_printk(KERN_ERR, conn, "Login/Text in "
+                                         "progress. Cannot start new task.\n");
+                       return NULL;
+               }
+
                task = conn->login_task;
-       else {
+       else {
                if (session->state != ISCSI_STATE_LOGGED_IN)
                        return NULL;
 
@@ -731,7 +891,7 @@ out:
        ISCSI_DBG_SESSION(session, "cmd rsp done [sc %p res %d itt 0x%x]\n",
                          sc, sc->result, task->itt);
        conn->scsirsp_pdus_cnt++;
-       iscsi_complete_task(task);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
 }
 
 /**
@@ -769,7 +929,7 @@ iscsi_data_in_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
                          "[sc %p res %d itt 0x%x]\n",
                          sc, sc->result, task->itt);
        conn->scsirsp_pdus_cnt++;
-       iscsi_complete_task(task);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
 }
 
 static void iscsi_tmf_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
@@ -820,27 +980,102 @@ static void iscsi_send_nopout(struct iscsi_conn *conn, struct iscsi_nopin *rhdr)
        }
 }
 
+static int iscsi_nop_out_rsp(struct iscsi_task *task,
+                            struct iscsi_nopin *nop, char *data, int datalen)
+{
+       struct iscsi_conn *conn = task->conn;
+       int rc = 0;
+
+       if (conn->ping_task != task) {
+               /*
+                * If this is not in response to one of our
+                * nops then it must be from userspace.
+                */
+               if (iscsi_recv_pdu(conn->cls_conn, (struct iscsi_hdr *)nop,
+                                  data, datalen))
+                       rc = ISCSI_ERR_CONN_FAILED;
+       } else
+               mod_timer(&conn->transport_timer, jiffies + conn->recv_timeout);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
+       return rc;
+}
+
 static int iscsi_handle_reject(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
                               char *data, int datalen)
 {
        struct iscsi_reject *reject = (struct iscsi_reject *)hdr;
        struct iscsi_hdr rejected_pdu;
+       int opcode, rc = 0;
 
        conn->exp_statsn = be32_to_cpu(reject->statsn) + 1;
 
-       if (reject->reason == ISCSI_REASON_DATA_DIGEST_ERROR) {
-               if (ntoh24(reject->dlength) > datalen)
-                       return ISCSI_ERR_PROTO;
+       if (ntoh24(reject->dlength) > datalen ||
+           ntoh24(reject->dlength) < sizeof(struct iscsi_hdr)) {
+               iscsi_conn_printk(KERN_ERR, conn, "Cannot handle rejected "
+                                 "pdu. Invalid data length (pdu dlength "
+                                 "%u, datalen %d\n", ntoh24(reject->dlength),
+                                 datalen);
+               return ISCSI_ERR_PROTO;
+       }
+       memcpy(&rejected_pdu, data, sizeof(struct iscsi_hdr));
+       opcode = rejected_pdu.opcode & ISCSI_OPCODE_MASK;
+
+       switch (reject->reason) {
+       case ISCSI_REASON_DATA_DIGEST_ERROR:
+               iscsi_conn_printk(KERN_ERR, conn,
+                                 "pdu (op 0x%x itt 0x%x) rejected "
+                                 "due to DataDigest error.\n",
+                                 rejected_pdu.itt, opcode);
+               break;
+       case ISCSI_REASON_IMM_CMD_REJECT:
+               iscsi_conn_printk(KERN_ERR, conn,
+                                 "pdu (op 0x%x itt 0x%x) rejected. Too many "
+                                 "immediate commands.\n",
+                                 rejected_pdu.itt, opcode);
+               /*
+                * We only send one TMF at a time so if the target could not
+                * handle it, then it should get fixed (RFC mandates that
+                * a target can handle one immediate TMF per conn).
+                *
+                * For nops-outs, we could have sent more than one if
+                * the target is sending us lots of nop-ins
+                */
+               if (opcode != ISCSI_OP_NOOP_OUT)
+                       return 0;
 
-               if (ntoh24(reject->dlength) >= sizeof(struct iscsi_hdr)) {
-                       memcpy(&rejected_pdu, data, sizeof(struct iscsi_hdr));
-                       iscsi_conn_printk(KERN_ERR, conn,
-                                         "pdu (op 0x%x) rejected "
-                                         "due to DataDigest error.\n",
-                                         rejected_pdu.opcode);
+                if (rejected_pdu.itt == cpu_to_be32(ISCSI_RESERVED_TAG))
+                       /*
+                        * nop-out in response to target's nop-out rejected.
+                        * Just resend.
+                        */
+                       iscsi_send_nopout(conn,
+                                         (struct iscsi_nopin*)&rejected_pdu);
+               else {
+                       struct iscsi_task *task;
+                       /*
+                        * Our nop as ping got dropped. We know the target
+                        * and transport are ok so just clean up
+                        */
+                       task = iscsi_itt_to_task(conn, rejected_pdu.itt);
+                       if (!task) {
+                               iscsi_conn_printk(KERN_ERR, conn,
+                                                "Invalid pdu reject. Could "
+                                                "not lookup rejected task.\n");
+                               rc = ISCSI_ERR_BAD_ITT;
+                       } else
+                               rc = iscsi_nop_out_rsp(task,
+                                       (struct iscsi_nopin*)&rejected_pdu,
+                                       NULL, 0);
                }
+               break;
+       default:
+               iscsi_conn_printk(KERN_ERR, conn,
+                                 "pdu (op 0x%x itt 0x%x) rejected. Reason "
+                                 "code 0x%x\n", rejected_pdu.itt,
+                                 rejected_pdu.opcode, reject->reason);
+               break;
        }
-       return 0;
+       return rc;
 }
 
 /**
@@ -940,6 +1175,7 @@ int __iscsi_complete_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
                task = iscsi_itt_to_ctask(conn, hdr->itt);
                if (!task)
                        return ISCSI_ERR_BAD_ITT;
+               task->last_xfer = jiffies;
                break;
        case ISCSI_OP_R2T:
                /*
@@ -990,7 +1226,7 @@ int __iscsi_complete_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
                }
 
                iscsi_tmf_rsp(conn, hdr);
-               iscsi_complete_task(task);
+               iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
                break;
        case ISCSI_OP_NOOP_IN:
                iscsi_update_cmdsn(session, (struct iscsi_nopin*)hdr);
@@ -1000,15 +1236,8 @@ int __iscsi_complete_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
                }
                conn->exp_statsn = be32_to_cpu(hdr->statsn) + 1;
 
-               if (conn->ping_task != task)
-                       /*
-                        * If this is not in response to one of our
-                        * nops then it must be from userspace.
-                        */
-                       goto recv_pdu;
-
-               mod_timer(&conn->transport_timer, jiffies + conn->recv_timeout);
-               iscsi_complete_task(task);
+               rc = iscsi_nop_out_rsp(task, (struct iscsi_nopin*)hdr,
+                                      data, datalen);
                break;
        default:
                rc = ISCSI_ERR_BAD_OPCODE;
@@ -1020,7 +1249,7 @@ out:
 recv_pdu:
        if (iscsi_recv_pdu(conn->cls_conn, hdr, data, datalen))
                rc = ISCSI_ERR_CONN_FAILED;
-       iscsi_complete_task(task);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
        return rc;
 }
 EXPORT_SYMBOL_GPL(__iscsi_complete_pdu);
@@ -1174,14 +1403,19 @@ static int iscsi_xmit_task(struct iscsi_conn *conn)
        struct iscsi_task *task = conn->task;
        int rc;
 
+       if (test_bit(ISCSI_SUSPEND_BIT, &conn->suspend_tx))
+               return -ENODATA;
+
        __iscsi_get_task(task);
        spin_unlock_bh(&conn->session->lock);
        rc = conn->session->tt->xmit_task(task);
        spin_lock_bh(&conn->session->lock);
-       __iscsi_put_task(task);
-       if (!rc)
+       if (!rc) {
                /* done with this task */
+               task->last_xfer = jiffies;
                conn->task = NULL;
+       }
+       __iscsi_put_task(task);
        return rc;
 }
 
@@ -1218,10 +1452,11 @@ EXPORT_SYMBOL_GPL(iscsi_requeue_task);
  **/
 static int iscsi_data_xmit(struct iscsi_conn *conn)
 {
+       struct iscsi_task *task;
        int rc = 0;
 
        spin_lock_bh(&conn->session->lock);
-       if (unlikely(conn->suspend_tx)) {
+       if (test_bit(ISCSI_SUSPEND_BIT, &conn->suspend_tx)) {
                ISCSI_DBG_SESSION(conn->session, "Tx suspended!\n");
                spin_unlock_bh(&conn->session->lock);
                return -ENODATA;
@@ -1230,7 +1465,7 @@ static int iscsi_data_xmit(struct iscsi_conn *conn)
        if (conn->task) {
                rc = iscsi_xmit_task(conn);
                if (rc)
-                       goto again;
+                       goto done;
        }
 
        /*
@@ -1250,35 +1485,32 @@ check_mgmt:
                }
                rc = iscsi_xmit_task(conn);
                if (rc)
-                       goto again;
+                       goto done;
        }
 
        /* process pending command queue */
        while (!list_empty(&conn->cmdqueue)) {
-               if (conn->tmf_state == TMF_QUEUED)
-                       break;
-
-               conn->task = list_entry(conn->cmdqueue.next,
-                                        struct iscsi_task, running);
+               conn->task = list_entry(conn->cmdqueue.next, struct iscsi_task,
+                                       running);
                list_del_init(&conn->task->running);
                if (conn->session->state == ISCSI_STATE_LOGGING_OUT) {
-                       fail_scsi_task(conn->task, DID_IMM_RETRY << 16);
+                       fail_scsi_task(conn->task, DID_IMM_RETRY);
                        continue;
                }
                rc = iscsi_prep_scsi_cmd_pdu(conn->task);
                if (rc) {
-                       if (rc == -ENOMEM) {
+                       if (rc == -ENOMEM || rc == -EACCES) {
                                list_add_tail(&conn->task->running,
                                              &conn->cmdqueue);
                                conn->task = NULL;
-                               goto again;
+                               goto done;
                        } else
-                               fail_scsi_task(conn->task, DID_ABORT << 16);
+                               fail_scsi_task(conn->task, DID_ABORT);
                        continue;
                }
                rc = iscsi_xmit_task(conn);
                if (rc)
-                       goto again;
+                       goto done;
                /*
                 * we could continuously get new task requests so
                 * we need to check the mgmt queue for nops that need to
@@ -1289,31 +1521,30 @@ check_mgmt:
        }
 
        while (!list_empty(&conn->requeue)) {
-               if (conn->session->fast_abort && conn->tmf_state != TMF_INITIAL)
-                       break;
-
                /*
                 * we always do fastlogout - conn stop code will clean up.
                 */
                if (conn->session->state == ISCSI_STATE_LOGGING_OUT)
                        break;
 
-               conn->task = list_entry(conn->requeue.next,
-                                        struct iscsi_task, running);
+               task = list_entry(conn->requeue.next, struct iscsi_task,
+                                 running);
+               if (iscsi_check_tmf_restrictions(task, ISCSI_OP_SCSI_DATA_OUT))
+                       break;
+
+               conn->task = task;
                list_del_init(&conn->task->running);
                conn->task->state = ISCSI_TASK_RUNNING;
                rc = iscsi_xmit_task(conn);
                if (rc)
-                       goto again;
+                       goto done;
                if (!list_empty(&conn->mgmtqueue))
                        goto check_mgmt;
        }
        spin_unlock_bh(&conn->session->lock);
        return -ENODATA;
 
-again:
-       if (unlikely(conn->suspend_tx))
-               rc = -ENODATA;
+done:
        spin_unlock_bh(&conn->session->lock);
        return rc;
 }
@@ -1347,6 +1578,9 @@ static inline struct iscsi_task *iscsi_alloc_task(struct iscsi_conn *conn,
        task->state = ISCSI_TASK_PENDING;
        task->conn = conn;
        task->sc = sc;
+       task->have_checked_conn = false;
+       task->last_timeout = jiffies;
+       task->last_xfer = jiffies;
        INIT_LIST_HEAD(&task->running);
        return task;
 }
@@ -1431,6 +1665,12 @@ int iscsi_queuecommand(struct scsi_cmnd *sc, void (*done)(struct scsi_cmnd *))
                goto fault;
        }
 
+       if (test_bit(ISCSI_SUSPEND_BIT, &conn->suspend_tx)) {
+               reason = FAILURE_SESSION_IN_RECOVERY;
+               sc->result = DID_REQUEUE;
+               goto fault;
+       }
+
        if (iscsi_check_cmdsn_window_closed(conn)) {
                reason = FAILURE_WINDOW_CLOSED;
                goto reject;
@@ -1445,7 +1685,7 @@ int iscsi_queuecommand(struct scsi_cmnd *sc, void (*done)(struct scsi_cmnd *))
        if (!ihost->workq) {
                reason = iscsi_prep_scsi_cmd_pdu(task);
                if (reason) {
-                       if (reason == -ENOMEM) {
+                       if (reason == -ENOMEM ||  reason == -EACCES) {
                                reason = FAILURE_OOM;
                                goto prepd_reject;
                        } else {
@@ -1454,6 +1694,7 @@ int iscsi_queuecommand(struct scsi_cmnd *sc, void (*done)(struct scsi_cmnd *))
                        }
                }
                if (session->tt->xmit_task(task)) {
+                       session->cmdsn--;
                        reason = FAILURE_SESSION_NOT_READY;
                        goto prepd_reject;
                }
@@ -1469,7 +1710,7 @@ int iscsi_queuecommand(struct scsi_cmnd *sc, void (*done)(struct scsi_cmnd *))
 
 prepd_reject:
        sc->scsi_done = NULL;
-       iscsi_complete_task(task);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
 reject:
        spin_unlock(&session->lock);
        ISCSI_DBG_SESSION(session, "cmd 0x%x rejected (%d)\n",
@@ -1479,7 +1720,7 @@ reject:
 
 prepd_fault:
        sc->scsi_done = NULL;
-       iscsi_complete_task(task);
+       iscsi_complete_task(task, ISCSI_TASK_COMPLETED);
 fault:
        spin_unlock(&session->lock);
        ISCSI_DBG_SESSION(session, "iscsi: cmd 0x%x is not queued (%d)\n",
@@ -1496,9 +1737,21 @@ fault:
 }
 EXPORT_SYMBOL_GPL(iscsi_queuecommand);
 
-int iscsi_change_queue_depth(struct scsi_device *sdev, int depth)
+int iscsi_change_queue_depth(struct scsi_device *sdev, int depth, int reason)
 {
-       scsi_adjust_queue_depth(sdev, scsi_get_tag_type(sdev), depth);
+       switch (reason) {
+       case SCSI_QDEPTH_DEFAULT:
+               scsi_adjust_queue_depth(sdev, scsi_get_tag_type(sdev), depth);
+               break;
+       case SCSI_QDEPTH_QFULL:
+               scsi_track_queue_full(sdev, depth);
+               break;
+       case SCSI_QDEPTH_RAMP_UP:
+               scsi_adjust_queue_depth(sdev, scsi_get_tag_type(sdev), depth);
+               break;
+       default:
+               return -EOPNOTSUPP;
+       }
        return sdev->queue_depth;
 }
 EXPORT_SYMBOL_GPL(iscsi_change_queue_depth);
@@ -1513,72 +1766,6 @@ int iscsi_target_alloc(struct scsi_target *starget)
 }
 EXPORT_SYMBOL_GPL(iscsi_target_alloc);
 
-void iscsi_session_recovery_timedout(struct iscsi_cls_session *cls_session)
-{
-       struct iscsi_session *session = cls_session->dd_data;
-
-       spin_lock_bh(&session->lock);
-       if (session->state != ISCSI_STATE_LOGGED_IN) {
-               session->state = ISCSI_STATE_RECOVERY_FAILED;
-               if (session->leadconn)
-                       wake_up(&session->leadconn->ehwait);
-       }
-       spin_unlock_bh(&session->lock);
-}
-EXPORT_SYMBOL_GPL(iscsi_session_recovery_timedout);
-
-int iscsi_eh_target_reset(struct scsi_cmnd *sc)
-{
-       struct iscsi_cls_session *cls_session;
-       struct iscsi_session *session;
-       struct iscsi_conn *conn;
-
-       cls_session = starget_to_session(scsi_target(sc->device));
-       session = cls_session->dd_data;
-       conn = session->leadconn;
-
-       mutex_lock(&session->eh_mutex);
-       spin_lock_bh(&session->lock);
-       if (session->state == ISCSI_STATE_TERMINATE) {
-failed:
-               iscsi_session_printk(KERN_INFO, session,
-                                    "failing target reset: Could not log "
-                                    "back into target [age %d]\n",
-                                    session->age);
-               spin_unlock_bh(&session->lock);
-               mutex_unlock(&session->eh_mutex);
-               return FAILED;
-       }
-
-       spin_unlock_bh(&session->lock);
-       mutex_unlock(&session->eh_mutex);
-       /*
-        * we drop the lock here but the leadconn cannot be destoyed while
-        * we are in the scsi eh
-        */
-       iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
-
-       ISCSI_DBG_SESSION(session, "wait for relogin\n");
-       wait_event_interruptible(conn->ehwait,
-                                session->state == ISCSI_STATE_TERMINATE ||
-                                session->state == ISCSI_STATE_LOGGED_IN ||
-                                session->state == ISCSI_STATE_RECOVERY_FAILED);
-       if (signal_pending(current))
-               flush_signals(current);
-
-       mutex_lock(&session->eh_mutex);
-       spin_lock_bh(&session->lock);
-       if (session->state == ISCSI_STATE_LOGGED_IN)
-               iscsi_session_printk(KERN_INFO, session,
-                                    "target reset succeeded\n");
-       else
-               goto failed;
-       spin_unlock_bh(&session->lock);
-       mutex_unlock(&session->eh_mutex);
-       return SUCCESS;
-}
-EXPORT_SYMBOL_GPL(iscsi_eh_target_reset);
-
 static void iscsi_tmf_timedout(unsigned long data)
 {
        struct iscsi_conn *conn = (struct iscsi_conn *)data;
@@ -1587,7 +1774,7 @@ static void iscsi_tmf_timedout(unsigned long data)
        spin_lock(&session->lock);
        if (conn->tmf_state == TMF_QUEUED) {
                conn->tmf_state = TMF_TIMEDOUT;
-               ISCSI_DBG_SESSION(session, "tmf timedout\n");
+               ISCSI_DBG_EH(session, "tmf timedout\n");
                /* unblock eh_abort() */
                wake_up(&conn->ehwait);
        }
@@ -1607,7 +1794,7 @@ static int iscsi_exec_task_mgmt_fn(struct iscsi_conn *conn,
                spin_unlock_bh(&session->lock);
                iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
                spin_lock_bh(&session->lock);
-               ISCSI_DBG_SESSION(session, "tmf exec failure\n");
+               ISCSI_DBG_EH(session, "tmf exec failure\n");
                return -EPERM;
        }
        conn->tmfcmd_pdus_cnt++;
@@ -1615,7 +1802,7 @@ static int iscsi_exec_task_mgmt_fn(struct iscsi_conn *conn,
        conn->tmf_timer.function = iscsi_tmf_timedout;
        conn->tmf_timer.data = (unsigned long)conn;
        add_timer(&conn->tmf_timer);
-       ISCSI_DBG_SESSION(session, "tmf set timeout\n");
+       ISCSI_DBG_EH(session, "tmf set timeout\n");
 
        spin_unlock_bh(&session->lock);
        mutex_unlock(&session->eh_mutex);
@@ -1665,10 +1852,37 @@ static void fail_scsi_tasks(struct iscsi_conn *conn, unsigned lun,
                ISCSI_DBG_SESSION(conn->session,
                                  "failing sc %p itt 0x%x state %d\n",
                                  task->sc, task->itt, task->state);
-               fail_scsi_task(task, error << 16);
+               fail_scsi_task(task, error);
        }
 }
 
+/**
+ * iscsi_suspend_queue - suspend iscsi_queuecommand
+ * @conn: iscsi conn to stop queueing IO on
+ *
+ * This grabs the session lock to make sure no one is in
+ * xmit_task/queuecommand, and then sets suspend to prevent
+ * new commands from being queued. This only needs to be called
+ * by offload drivers that need to sync a path like ep disconnect
+ * with the iscsi_queuecommand/xmit_task. To start IO again libiscsi
+ * will call iscsi_start_tx and iscsi_unblock_session when in FFP.
+ */
+void iscsi_suspend_queue(struct iscsi_conn *conn)
+{
+       spin_lock_bh(&conn->session->lock);
+       set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_tx);
+       spin_unlock_bh(&conn->session->lock);
+}
+EXPORT_SYMBOL_GPL(iscsi_suspend_queue);
+
+/**
+ * iscsi_suspend_tx - suspend iscsi_data_xmit
+ * @conn: iscsi conn tp stop processing IO on.
+ *
+ * This function sets the suspend bit to prevent iscsi_data_xmit
+ * from sending new IO, and if work is queued on the xmit thread
+ * it will wait for it to be completed.
+ */
 void iscsi_suspend_tx(struct iscsi_conn *conn)
 {
        struct Scsi_Host *shost = conn->session->host;
@@ -1702,17 +1916,18 @@ static int iscsi_has_ping_timed_out(struct iscsi_conn *conn)
                return 0;
 }
 
-static enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd *scmd)
+static enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd *sc)
 {
+       enum blk_eh_timer_return rc = BLK_EH_NOT_HANDLED;
+       struct iscsi_task *task = NULL;
        struct iscsi_cls_session *cls_session;
        struct iscsi_session *session;
        struct iscsi_conn *conn;
-       enum blk_eh_timer_return rc = BLK_EH_NOT_HANDLED;
 
-       cls_session = starget_to_session(scsi_target(scmd->device));
+       cls_session = starget_to_session(scsi_target(sc->device));
        session = cls_session->dd_data;
 
-       ISCSI_DBG_SESSION(session, "scsi cmd %p timedout\n", scmd);
+       ISCSI_DBG_EH(session, "scsi cmd %p timedout\n", sc);
 
        spin_lock(&session->lock);
        if (session->state != ISCSI_STATE_LOGGED_IN) {
@@ -1731,6 +1946,26 @@ static enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd *scmd)
                goto done;
        }
 
+       task = (struct iscsi_task *)sc->SCp.ptr;
+       if (!task)
+               goto done;
+       /*
+        * If we have sent (at least queued to the network layer) a pdu or
+        * recvd one for the task since the last timeout ask for
+        * more time. If on the next timeout we have not made progress
+        * we can check if it is the task or connection when we send the
+        * nop as a ping.
+        */
+       if (time_after_eq(task->last_xfer, task->last_timeout)) {
+               ISCSI_DBG_EH(session, "Command making progress. Asking "
+                            "scsi-ml for more time to complete. "
+                            "Last data recv at %lu. Last timeout was at "
+                            "%lu\n.", task->last_xfer, task->last_timeout);
+               task->have_checked_conn = false;
+               rc = BLK_EH_RESET_TIMER;
+               goto done;
+       }
+
        if (!conn->recv_timeout && !conn->ping_timeout)
                goto done;
        /*
@@ -1741,23 +1976,32 @@ static enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd *scmd)
                rc = BLK_EH_RESET_TIMER;
                goto done;
        }
+
+       /* Assumes nop timeout is shorter than scsi cmd timeout */
+       if (task->have_checked_conn)
+               goto done;
+
        /*
-        * if we are about to check the transport then give the command
-        * more time
+        * Checking the transport already or nop from a cmd timeout still
+        * running
         */
-       if (time_before_eq(conn->last_recv + (conn->recv_timeout * HZ),
-                          jiffies)) {
+       if (conn->ping_task) {
+               task->have_checked_conn = true;
                rc = BLK_EH_RESET_TIMER;
                goto done;
        }
 
-       /* if in the middle of checking the transport then give us more time */
-       if (conn->ping_task)
-               rc = BLK_EH_RESET_TIMER;
+       /* Make sure there is a transport check done */
+       iscsi_send_nopout(conn, NULL);
+       task->have_checked_conn = true;
+       rc = BLK_EH_RESET_TIMER;
+
 done:
+       if (task)
+               task->last_timeout = jiffies;
        spin_unlock(&session->lock);
-       ISCSI_DBG_SESSION(session, "return %s\n", rc == BLK_EH_RESET_TIMER ?
-                         "timer reset" : "nh");
+       ISCSI_DBG_EH(session, "return %s\n", rc == BLK_EH_RESET_TIMER ?
+                    "timer reset" : "nh");
        return rc;
 }
 
@@ -1827,6 +2071,8 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
        cls_session = starget_to_session(scsi_target(sc->device));
        session = cls_session->dd_data;
 
+       ISCSI_DBG_EH(session, "aborting sc %p\n", sc);
+
        mutex_lock(&session->eh_mutex);
        spin_lock_bh(&session->lock);
        /*
@@ -1834,8 +2080,8 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
         * got the command.
         */
        if (!sc->SCp.ptr) {
-               ISCSI_DBG_SESSION(session, "sc never reached iscsi layer or "
-                                 "it completed.\n");
+               ISCSI_DBG_EH(session, "sc never reached iscsi layer or "
+                                     "it completed.\n");
                spin_unlock_bh(&session->lock);
                mutex_unlock(&session->eh_mutex);
                return SUCCESS;
@@ -1849,6 +2095,8 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
            sc->SCp.phase != session->age) {
                spin_unlock_bh(&session->lock);
                mutex_unlock(&session->eh_mutex);
+               ISCSI_DBG_EH(session, "failing abort due to dropped "
+                                 "session.\n");
                return FAILED;
        }
 
@@ -1857,18 +2105,17 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
        age = session->age;
 
        task = (struct iscsi_task *)sc->SCp.ptr;
-       ISCSI_DBG_SESSION(session, "aborting [sc %p itt 0x%x]\n",
-                         sc, task->itt);
+       ISCSI_DBG_EH(session, "aborting [sc %p itt 0x%x]\n",
+                    sc, task->itt);
 
        /* task completed before time out */
        if (!task->sc) {
-               ISCSI_DBG_SESSION(session, "sc completed while abort in "
-                                 "progress\n");
+               ISCSI_DBG_EH(session, "sc completed while abort in progress\n");
                goto success;
        }
 
        if (task->state == ISCSI_TASK_PENDING) {
-               fail_scsi_task(task, DID_ABORT << 16);
+               fail_scsi_task(task, DID_ABORT);
                goto success;
        }
 
@@ -1898,10 +2145,11 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
                 * good and have never sent us a successful tmf response
                 * then sent more data for the cmd.
                 */
-               spin_lock(&session->lock);
-               fail_scsi_task(task, DID_ABORT << 16);
+               spin_lock_bh(&session->lock);
+               fail_scsi_task(task, DID_ABORT);
                conn->tmf_state = TMF_INITIAL;
-               spin_unlock(&session->lock);
+               memset(hdr, 0, sizeof(*hdr));
+               spin_unlock_bh(&session->lock);
                iscsi_start_tx(conn);
                goto success_unlocked;
        case TMF_TIMEDOUT:
@@ -1911,9 +2159,10 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
        case TMF_NOT_FOUND:
                if (!sc->SCp.ptr) {
                        conn->tmf_state = TMF_INITIAL;
+                       memset(hdr, 0, sizeof(*hdr));
                        /* task completed before tmf abort response */
-                       ISCSI_DBG_SESSION(session, "sc completed while abort "
-                                         "in progress\n");
+                       ISCSI_DBG_EH(session, "sc completed while abort in "
+                                             "progress\n");
                        goto success;
                }
                /* fall through */
@@ -1925,16 +2174,16 @@ int iscsi_eh_abort(struct scsi_cmnd *sc)
 success:
        spin_unlock_bh(&session->lock);
 success_unlocked:
-       ISCSI_DBG_SESSION(session, "abort success [sc %p itt 0x%x]\n",
-                         sc, task->itt);
+       ISCSI_DBG_EH(session, "abort success [sc %p itt 0x%x]\n",
+                    sc, task->itt);
        mutex_unlock(&session->eh_mutex);
        return SUCCESS;
 
 failed:
        spin_unlock_bh(&session->lock);
 failed_unlocked:
-       ISCSI_DBG_SESSION(session, "abort failed [sc %p itt 0x%x]\n", sc,
-                         task ? task->itt : 0);
+       ISCSI_DBG_EH(session, "abort failed [sc %p itt 0x%x]\n", sc,
+                    task ? task->itt : 0);
        mutex_unlock(&session->eh_mutex);
        return FAILED;
 }
@@ -1961,8 +2210,7 @@ int iscsi_eh_device_reset(struct scsi_cmnd *sc)
        cls_session = starget_to_session(scsi_target(sc->device));
        session = cls_session->dd_data;
 
-       ISCSI_DBG_SESSION(session, "LU Reset [sc %p lun %u]\n",
-                         sc, sc->device->lun);
+       ISCSI_DBG_EH(session, "LU Reset [sc %p lun %u]\n", sc, sc->device->lun);
 
        mutex_lock(&session->eh_mutex);
        spin_lock_bh(&session->lock);
@@ -2006,6 +2254,7 @@ int iscsi_eh_device_reset(struct scsi_cmnd *sc)
        iscsi_suspend_tx(conn);
 
        spin_lock_bh(&session->lock);
+       memset(hdr, 0, sizeof(*hdr));
        fail_scsi_tasks(conn, sc->device->lun, DID_ERROR);
        conn->tmf_state = TMF_INITIAL;
        spin_unlock_bh(&session->lock);
@@ -2016,13 +2265,179 @@ int iscsi_eh_device_reset(struct scsi_cmnd *sc)
 unlock:
        spin_unlock_bh(&session->lock);
 done:
-       ISCSI_DBG_SESSION(session, "dev reset result = %s\n",
-                        rc == SUCCESS ? "SUCCESS" : "FAILED");
+       ISCSI_DBG_EH(session, "dev reset result = %s\n",
+                    rc == SUCCESS ? "SUCCESS" : "FAILED");
        mutex_unlock(&session->eh_mutex);
        return rc;
 }
 EXPORT_SYMBOL_GPL(iscsi_eh_device_reset);
 
+void iscsi_session_recovery_timedout(struct iscsi_cls_session *cls_session)
+{
+       struct iscsi_session *session = cls_session->dd_data;
+
+       spin_lock_bh(&session->lock);
+       if (session->state != ISCSI_STATE_LOGGED_IN) {
+               session->state = ISCSI_STATE_RECOVERY_FAILED;
+               if (session->leadconn)
+                       wake_up(&session->leadconn->ehwait);
+       }
+       spin_unlock_bh(&session->lock);
+}
+EXPORT_SYMBOL_GPL(iscsi_session_recovery_timedout);
+
+/**
+ * iscsi_eh_session_reset - drop session and attempt relogin
+ * @sc: scsi command
+ *
+ * This function will wait for a relogin, session termination from
+ * userspace, or a recovery/replacement timeout.
+ */
+static int iscsi_eh_session_reset(struct scsi_cmnd *sc)
+{
+       struct iscsi_cls_session *cls_session;
+       struct iscsi_session *session;
+       struct iscsi_conn *conn;
+
+       cls_session = starget_to_session(scsi_target(sc->device));
+       session = cls_session->dd_data;
+       conn = session->leadconn;
+
+       mutex_lock(&session->eh_mutex);
+       spin_lock_bh(&session->lock);
+       if (session->state == ISCSI_STATE_TERMINATE) {
+failed:
+               ISCSI_DBG_EH(session,
+                            "failing session reset: Could not log back into "
+                            "%s, %s [age %d]\n", session->targetname,
+                            conn->persistent_address, session->age);
+               spin_unlock_bh(&session->lock);
+               mutex_unlock(&session->eh_mutex);
+               return FAILED;
+       }
+
+       spin_unlock_bh(&session->lock);
+       mutex_unlock(&session->eh_mutex);
+       /*
+        * we drop the lock here but the leadconn cannot be destoyed while
+        * we are in the scsi eh
+        */
+       iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
+
+       ISCSI_DBG_EH(session, "wait for relogin\n");
+       wait_event_interruptible(conn->ehwait,
+                                session->state == ISCSI_STATE_TERMINATE ||
+                                session->state == ISCSI_STATE_LOGGED_IN ||
+                                session->state == ISCSI_STATE_RECOVERY_FAILED);
+       if (signal_pending(current))
+               flush_signals(current);
+
+       mutex_lock(&session->eh_mutex);
+       spin_lock_bh(&session->lock);
+       if (session->state == ISCSI_STATE_LOGGED_IN) {
+               ISCSI_DBG_EH(session,
+                            "session reset succeeded for %s,%s\n",
+                            session->targetname, conn->persistent_address);
+       } else
+               goto failed;
+       spin_unlock_bh(&session->lock);
+       mutex_unlock(&session->eh_mutex);
+       return SUCCESS;
+}
+
+static void iscsi_prep_tgt_reset_pdu(struct scsi_cmnd *sc, struct iscsi_tm *hdr)
+{
+       memset(hdr, 0, sizeof(*hdr));
+       hdr->opcode = ISCSI_OP_SCSI_TMFUNC | ISCSI_OP_IMMEDIATE;
+       hdr->flags = ISCSI_TM_FUNC_TARGET_WARM_RESET & ISCSI_FLAG_TM_FUNC_MASK;
+       hdr->flags |= ISCSI_FLAG_CMD_FINAL;
+       hdr->rtt = RESERVED_ITT;
+}
+
+/**
+ * iscsi_eh_target_reset - reset target
+ * @sc: scsi command
+ *
+ * This will attempt to send a warm target reset. If that fails
+ * then we will drop the session and attempt ERL0 recovery.
+ */
+int iscsi_eh_target_reset(struct scsi_cmnd *sc)
+{
+       struct iscsi_cls_session *cls_session;
+       struct iscsi_session *session;
+       struct iscsi_conn *conn;
+       struct iscsi_tm *hdr;
+       int rc = FAILED;
+
+       cls_session = starget_to_session(scsi_target(sc->device));
+       session = cls_session->dd_data;
+
+       ISCSI_DBG_EH(session, "tgt Reset [sc %p tgt %s]\n", sc,
+                    session->targetname);
+
+       mutex_lock(&session->eh_mutex);
+       spin_lock_bh(&session->lock);
+       /*
+        * Just check if we are not logged in. We cannot check for
+        * the phase because the reset could come from a ioctl.
+        */
+       if (!session->leadconn || session->state != ISCSI_STATE_LOGGED_IN)
+               goto unlock;
+       conn = session->leadconn;
+
+       /* only have one tmf outstanding at a time */
+       if (conn->tmf_state != TMF_INITIAL)
+               goto unlock;
+       conn->tmf_state = TMF_QUEUED;
+
+       hdr = &conn->tmhdr;
+       iscsi_prep_tgt_reset_pdu(sc, hdr);
+
+       if (iscsi_exec_task_mgmt_fn(conn, hdr, session->age,
+                                   session->tgt_reset_timeout)) {
+               rc = FAILED;
+               goto unlock;
+       }
+
+       switch (conn->tmf_state) {
+       case TMF_SUCCESS:
+               break;
+       case TMF_TIMEDOUT:
+               spin_unlock_bh(&session->lock);
+               iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
+               goto done;
+       default:
+               conn->tmf_state = TMF_INITIAL;
+               goto unlock;
+       }
+
+       rc = SUCCESS;
+       spin_unlock_bh(&session->lock);
+
+       iscsi_suspend_tx(conn);
+
+       spin_lock_bh(&session->lock);
+       memset(hdr, 0, sizeof(*hdr));
+       fail_scsi_tasks(conn, -1, DID_ERROR);
+       conn->tmf_state = TMF_INITIAL;
+       spin_unlock_bh(&session->lock);
+
+       iscsi_start_tx(conn);
+       goto done;
+
+unlock:
+       spin_unlock_bh(&session->lock);
+done:
+       ISCSI_DBG_EH(session, "tgt %s reset result = %s\n", session->targetname,
+                    rc == SUCCESS ? "SUCCESS" : "FAILED");
+       mutex_unlock(&session->eh_mutex);
+
+       if (rc == FAILED)
+               rc = iscsi_eh_session_reset(sc);
+       return rc;
+}
+EXPORT_SYMBOL_GPL(iscsi_eh_target_reset);
+
 /*
  * Pre-allocate a pool of @max items of @item_size. By default, the pool
  * should be accessed via kfifo_{get,put} on q->queue.
@@ -2230,7 +2645,7 @@ static void iscsi_host_dec_session_cnt(struct Scsi_Host *shost)
  */
 struct iscsi_cls_session *
 iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost,
-                   uint16_t cmds_max, int cmd_task_size,
+                   uint16_t cmds_max, int dd_size, int cmd_task_size,
                    uint32_t initial_cmdsn, unsigned int id)
 {
        struct iscsi_host *ihost = shost_priv(shost);
@@ -2280,7 +2695,8 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost,
        scsi_cmds = total_cmds - ISCSI_MGMT_CMDS_MAX;
 
        cls_session = iscsi_alloc_session(shost, iscsit,
-                                         sizeof(struct iscsi_session));
+                                         sizeof(struct iscsi_session) +
+                                         dd_size);
        if (!cls_session)
                goto dec_session_count;
        session = cls_session->dd_data;
@@ -2288,6 +2704,7 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost,
        session->host = shost;
        session->state = ISCSI_STATE_FREE;
        session->fast_abort = 1;
+       session->tgt_reset_timeout = 30;
        session->lu_reset_timeout = 15;
        session->abort_timeout = 10;
        session->scsi_cmds_max = scsi_cmds;
@@ -2297,6 +2714,7 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost,
        session->max_cmdsn = initial_cmdsn + 1;
        session->max_r2t = 1;
        session->tt = iscsit;
+       session->dd_data = cls_session->dd_data + sizeof(*session);
        mutex_init(&session->eh_mutex);
        spin_lock_init(&session->lock);
 
@@ -2572,7 +2990,7 @@ static void
 fail_mgmt_tasks(struct iscsi_session *session, struct iscsi_conn *conn)
 {
        struct iscsi_task *task;
-       int i;
+       int i, state;
 
        for (i = 0; i < conn->session->cmds_max; i++) {
                task = conn->session->cmds[i];
@@ -2585,7 +3003,11 @@ fail_mgmt_tasks(struct iscsi_session *session, struct iscsi_conn *conn)
                ISCSI_DBG_SESSION(conn->session,
                                  "failing mgmt itt 0x%x state %d\n",
                                  task->itt, task->state);
-               iscsi_complete_task(task);
+               state = ISCSI_TASK_ABRT_SESS_RECOV;
+               if (task->state == ISCSI_TASK_PENDING)
+                       state = ISCSI_TASK_COMPLETED;
+               iscsi_complete_task(task, state);
+
        }
 }
 
@@ -2642,11 +3064,9 @@ static void iscsi_start_session_recovery(struct iscsi_session *session,
         * flush queues.
         */
        spin_lock_bh(&session->lock);
-       if (flag == STOP_CONN_RECOVER)
-               fail_scsi_tasks(conn, -1, DID_TRANSPORT_DISRUPTED);
-       else
-               fail_scsi_tasks(conn, -1, DID_ERROR);
+       fail_scsi_tasks(conn, -1, DID_TRANSPORT_DISRUPTED);
        fail_mgmt_tasks(session, conn);
+       memset(&conn->tmhdr, 0, sizeof(conn->tmhdr));
        spin_unlock_bh(&session->lock);
        mutex_unlock(&session->eh_mutex);
 }
@@ -2723,6 +3143,9 @@ int iscsi_set_param(struct iscsi_cls_conn *cls_conn,
        case ISCSI_PARAM_LU_RESET_TMO:
                sscanf(buf, "%d", &session->lu_reset_timeout);
                break;
+       case ISCSI_PARAM_TGT_RESET_TMO:
+               sscanf(buf, "%d", &session->tgt_reset_timeout);
+               break;
        case ISCSI_PARAM_PING_TMO:
                sscanf(buf, "%d", &conn->ping_timeout);
                break;
@@ -2822,6 +3245,9 @@ int iscsi_session_get_param(struct iscsi_cls_session *cls_session,
        case ISCSI_PARAM_LU_RESET_TMO:
                len = sprintf(buf, "%d\n", session->lu_reset_timeout);
                break;
+       case ISCSI_PARAM_TGT_RESET_TMO:
+               len = sprintf(buf, "%d\n", session->tgt_reset_timeout);
+               break;
        case ISCSI_PARAM_INITIAL_R2T_EN:
                len = sprintf(buf, "%d\n", session->initial_r2t_en);
                break;