]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
IB: Add SCSI RDMA Protocol (SRP) initiator
authorRoland Dreier <rolandd@cisco.com>
Wed, 2 Nov 2005 22:07:13 +0000 (14:07 -0800)
committerRoland Dreier <rolandd@cisco.com>
Wed, 2 Nov 2005 22:07:13 +0000 (14:07 -0800)
Add an InfiniBand SCSI RDMA Protocol (SRP) initiator.  This driver is
used to talk talk to InfiniBand SRP targets (storage devices).

Signed-off-by: Roland Dreier <rolandd@cisco.com>
drivers/infiniband/Kconfig
drivers/infiniband/Makefile
drivers/infiniband/ulp/srp/Kbuild [new file with mode: 0644]
drivers/infiniband/ulp/srp/Kconfig [new file with mode: 0644]
drivers/infiniband/ulp/srp/ib_srp.c [new file with mode: 0644]
drivers/infiniband/ulp/srp/ib_srp.h [new file with mode: 0644]
include/scsi/srp.h [new file with mode: 0644]

index 325d502e25cdb30df585ecb68a642ac9aeb5818b..bdf0891a92dd5f27e2708dc961da5f2e9a7a04e9 100644 (file)
@@ -33,4 +33,6 @@ source "drivers/infiniband/hw/mthca/Kconfig"
 
 source "drivers/infiniband/ulp/ipoib/Kconfig"
 
+source "drivers/infiniband/ulp/srp/Kconfig"
+
 endmenu
index d256cf7982187415caca2e7dc9aa291d871af7ff..a43fb34cca94d01e47d3670bfefb4f35a4b45933 100644 (file)
@@ -1,3 +1,4 @@
 obj-$(CONFIG_INFINIBAND)               += core/
 obj-$(CONFIG_INFINIBAND_MTHCA)         += hw/mthca/
 obj-$(CONFIG_INFINIBAND_IPOIB)         += ulp/ipoib/
+obj-$(CONFIG_INFINIBAND_SRP)           += ulp/srp/
diff --git a/drivers/infiniband/ulp/srp/Kbuild b/drivers/infiniband/ulp/srp/Kbuild
new file mode 100644 (file)
index 0000000..a16c73c
--- /dev/null
@@ -0,0 +1 @@
+obj-$(CONFIG_INFINIBAND_SRP)                   += ib_srp.o
diff --git a/drivers/infiniband/ulp/srp/Kconfig b/drivers/infiniband/ulp/srp/Kconfig
new file mode 100644 (file)
index 0000000..8fe3be4
--- /dev/null
@@ -0,0 +1,11 @@
+config INFINIBAND_SRP
+       tristate "InfiniBand SCSI RDMA Protocol"
+       depends on INFINIBAND && SCSI
+       ---help---
+         Support for the SCSI RDMA Protocol over InfiniBand.  This
+         allows you to access storage devices that speak SRP over
+         InfiniBand.
+
+         The SRP protocol is defined by the INCITS T10 technical
+         committee.  See <http://www.t10.org/>.
+
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
new file mode 100644 (file)
index 0000000..2687e34
--- /dev/null
@@ -0,0 +1,1700 @@
+/*
+ * Copyright (c) 2005 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ib_srp.c 3932 2005-11-01 17:19:29Z roland $
+ */
+
+#include <linux/version.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/err.h>
+#include <linux/string.h>
+#include <linux/parser.h>
+#include <linux/random.h>
+
+#include <asm/atomic.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_device.h>
+#include <scsi/scsi_dbg.h>
+#include <scsi/srp.h>
+
+#include <rdma/ib_cache.h>
+
+#include "ib_srp.h"
+
+#define DRV_NAME       "ib_srp"
+#define PFX            DRV_NAME ": "
+#define DRV_VERSION    "0.2"
+#define DRV_RELDATE    "November 1, 2005"
+
+MODULE_AUTHOR("Roland Dreier");
+MODULE_DESCRIPTION("InfiniBand SCSI RDMA Protocol initiator "
+                  "v" DRV_VERSION " (" DRV_RELDATE ")");
+MODULE_LICENSE("Dual BSD/GPL");
+
+static int topspin_workarounds = 1;
+
+module_param(topspin_workarounds, int, 0444);
+MODULE_PARM_DESC(topspin_workarounds,
+                "Enable workarounds for Topspin/Cisco SRP target bugs if != 0");
+
+static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
+
+static void srp_add_one(struct ib_device *device);
+static void srp_remove_one(struct ib_device *device);
+static void srp_completion(struct ib_cq *cq, void *target_ptr);
+static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);
+
+static struct ib_client srp_client = {
+       .name   = "srp",
+       .add    = srp_add_one,
+       .remove = srp_remove_one
+};
+
+static inline struct srp_target_port *host_to_target(struct Scsi_Host *host)
+{
+       return (struct srp_target_port *) host->hostdata;
+}
+
+static const char *srp_target_info(struct Scsi_Host *host)
+{
+       return host_to_target(host)->target_name;
+}
+
+static struct srp_iu *srp_alloc_iu(struct srp_host *host, size_t size,
+                                  gfp_t gfp_mask,
+                                  enum dma_data_direction direction)
+{
+       struct srp_iu *iu;
+
+       iu = kmalloc(sizeof *iu, gfp_mask);
+       if (!iu)
+               goto out;
+
+       iu->buf = kzalloc(size, gfp_mask);
+       if (!iu->buf)
+               goto out_free_iu;
+
+       iu->dma = dma_map_single(host->dev->dma_device, iu->buf, size, direction);
+       if (dma_mapping_error(iu->dma))
+               goto out_free_buf;
+
+       iu->size      = size;
+       iu->direction = direction;
+
+       return iu;
+
+out_free_buf:
+       kfree(iu->buf);
+out_free_iu:
+       kfree(iu);
+out:
+       return NULL;
+}
+
+static void srp_free_iu(struct srp_host *host, struct srp_iu *iu)
+{
+       if (!iu)
+               return;
+
+       dma_unmap_single(host->dev->dma_device, iu->dma, iu->size, iu->direction);
+       kfree(iu->buf);
+       kfree(iu);
+}
+
+static void srp_qp_event(struct ib_event *event, void *context)
+{
+       printk(KERN_ERR PFX "QP event %d\n", event->event);
+}
+
+static int srp_init_qp(struct srp_target_port *target,
+                      struct ib_qp *qp)
+{
+       struct ib_qp_attr *attr;
+       int ret;
+
+       attr = kmalloc(sizeof *attr, GFP_KERNEL);
+       if (!attr)
+               return -ENOMEM;
+
+       ret = ib_find_cached_pkey(target->srp_host->dev,
+                                 target->srp_host->port,
+                                 be16_to_cpu(target->path.pkey),
+                                 &attr->pkey_index);
+       if (ret)
+               goto out;
+
+       attr->qp_state        = IB_QPS_INIT;
+       attr->qp_access_flags = (IB_ACCESS_REMOTE_READ |
+                                   IB_ACCESS_REMOTE_WRITE);
+       attr->port_num        = target->srp_host->port;
+
+       ret = ib_modify_qp(qp, attr,
+                          IB_QP_STATE          |
+                          IB_QP_PKEY_INDEX     |
+                          IB_QP_ACCESS_FLAGS   |
+                          IB_QP_PORT);
+
+out:
+       kfree(attr);
+       return ret;
+}
+
+static int srp_create_target_ib(struct srp_target_port *target)
+{
+       struct ib_qp_init_attr *init_attr;
+       int ret;
+
+       init_attr = kzalloc(sizeof *init_attr, GFP_KERNEL);
+       if (!init_attr)
+               return -ENOMEM;
+
+       target->cq = ib_create_cq(target->srp_host->dev, srp_completion,
+                                 NULL, target, SRP_CQ_SIZE);
+       if (IS_ERR(target->cq)) {
+               ret = PTR_ERR(target->cq);
+               goto out;
+       }
+
+       ib_req_notify_cq(target->cq, IB_CQ_NEXT_COMP);
+
+       init_attr->event_handler       = srp_qp_event;
+       init_attr->cap.max_send_wr     = SRP_SQ_SIZE;
+       init_attr->cap.max_recv_wr     = SRP_RQ_SIZE;
+       init_attr->cap.max_recv_sge    = 1;
+       init_attr->cap.max_send_sge    = 1;
+       init_attr->sq_sig_type         = IB_SIGNAL_ALL_WR;
+       init_attr->qp_type             = IB_QPT_RC;
+       init_attr->send_cq             = target->cq;
+       init_attr->recv_cq             = target->cq;
+
+       target->qp = ib_create_qp(target->srp_host->pd, init_attr);
+       if (IS_ERR(target->qp)) {
+               ret = PTR_ERR(target->qp);
+               ib_destroy_cq(target->cq);
+               goto out;
+       }
+
+       ret = srp_init_qp(target, target->qp);
+       if (ret) {
+               ib_destroy_qp(target->qp);
+               ib_destroy_cq(target->cq);
+               goto out;
+       }
+
+out:
+       kfree(init_attr);
+       return ret;
+}
+
+static void srp_free_target_ib(struct srp_target_port *target)
+{
+       int i;
+
+       ib_destroy_qp(target->qp);
+       ib_destroy_cq(target->cq);
+
+       for (i = 0; i < SRP_RQ_SIZE; ++i)
+               srp_free_iu(target->srp_host, target->rx_ring[i]);
+       for (i = 0; i < SRP_SQ_SIZE + 1; ++i)
+               srp_free_iu(target->srp_host, target->tx_ring[i]);
+}
+
+static void srp_path_rec_completion(int status,
+                                   struct ib_sa_path_rec *pathrec,
+                                   void *target_ptr)
+{
+       struct srp_target_port *target = target_ptr;
+
+       target->status = status;
+       if (status)
+               printk(KERN_ERR PFX "Got failed path rec status %d\n", status);
+       else
+               target->path = *pathrec;
+       complete(&target->done);
+}
+
+static int srp_lookup_path(struct srp_target_port *target)
+{
+       target->path.numb_path = 1;
+
+       init_completion(&target->done);
+
+       target->path_query_id = ib_sa_path_rec_get(target->srp_host->dev,
+                                                  target->srp_host->port,
+                                                  &target->path,
+                                                  IB_SA_PATH_REC_DGID          |
+                                                  IB_SA_PATH_REC_SGID          |
+                                                  IB_SA_PATH_REC_NUMB_PATH     |
+                                                  IB_SA_PATH_REC_PKEY,
+                                                  SRP_PATH_REC_TIMEOUT_MS,
+                                                  GFP_KERNEL,
+                                                  srp_path_rec_completion,
+                                                  target, &target->path_query);
+       if (target->path_query_id < 0)
+               return target->path_query_id;
+
+       wait_for_completion(&target->done);
+
+       if (target->status < 0)
+               printk(KERN_WARNING PFX "Path record query failed\n");
+
+       return target->status;
+}
+
+static int srp_send_req(struct srp_target_port *target)
+{
+       struct {
+               struct ib_cm_req_param param;
+               struct srp_login_req   priv;
+       } *req = NULL;
+       int status;
+
+       req = kzalloc(sizeof *req, GFP_KERNEL);
+       if (!req)
+               return -ENOMEM;
+
+       req->param.primary_path               = &target->path;
+       req->param.alternate_path             = NULL;
+       req->param.service_id                 = target->service_id;
+       req->param.qp_num                     = target->qp->qp_num;
+       req->param.qp_type                    = target->qp->qp_type;
+       req->param.private_data               = &req->priv;
+       req->param.private_data_len           = sizeof req->priv;
+       req->param.flow_control               = 1;
+
+       get_random_bytes(&req->param.starting_psn, 4);
+       req->param.starting_psn              &= 0xffffff;
+
+       /*
+        * Pick some arbitrary defaults here; we could make these
+        * module parameters if anyone cared about setting them.
+        */
+       req->param.responder_resources        = 4;
+       req->param.remote_cm_response_timeout = 20;
+       req->param.local_cm_response_timeout  = 20;
+       req->param.retry_count                = 7;
+       req->param.rnr_retry_count            = 7;
+       req->param.max_cm_retries             = 15;
+
+       req->priv.opcode        = SRP_LOGIN_REQ;
+       req->priv.tag           = 0;
+       req->priv.req_it_iu_len = cpu_to_be32(SRP_MAX_IU_LEN);
+       req->priv.req_buf_fmt   = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
+                                             SRP_BUF_FORMAT_INDIRECT);
+       memcpy(req->priv.initiator_port_id, target->srp_host->initiator_port_id, 16);
+       /*
+        * Topspin/Cisco SRP targets will reject our login unless we
+        * zero out the first 8 bytes of our initiator port ID.  The
+        * second 8 bytes must be our local node GUID, but we always
+        * use that anyway.
+        */
+       if (topspin_workarounds && !memcmp(&target->ioc_guid, topspin_oui, 3)) {
+               printk(KERN_DEBUG PFX "Topspin/Cisco initiator port ID workaround "
+                      "activated for target GUID %016llx\n",
+                      (unsigned long long) be64_to_cpu(target->ioc_guid));
+               memset(req->priv.initiator_port_id, 0, 8);
+       }
+       memcpy(req->priv.target_port_id,     &target->id_ext, 8);
+       memcpy(req->priv.target_port_id + 8, &target->ioc_guid, 8);
+
+       status = ib_send_cm_req(target->cm_id, &req->param);
+
+       kfree(req);
+
+       return status;
+}
+
+static void srp_disconnect_target(struct srp_target_port *target)
+{
+       /* XXX should send SRP_I_LOGOUT request */
+
+       init_completion(&target->done);
+       ib_send_cm_dreq(target->cm_id, NULL, 0);
+       wait_for_completion(&target->done);
+}
+
+static void srp_remove_work(void *target_ptr)
+{
+       struct srp_target_port *target = target_ptr;
+
+       spin_lock_irq(target->scsi_host->host_lock);
+       if (target->state != SRP_TARGET_DEAD) {
+               spin_unlock_irq(target->scsi_host->host_lock);
+               scsi_host_put(target->scsi_host);
+               return;
+       }
+       target->state = SRP_TARGET_REMOVED;
+       spin_unlock_irq(target->scsi_host->host_lock);
+
+       down(&target->srp_host->target_mutex);
+       list_del(&target->list);
+       up(&target->srp_host->target_mutex);
+
+       scsi_remove_host(target->scsi_host);
+       ib_destroy_cm_id(target->cm_id);
+       srp_free_target_ib(target);
+       scsi_host_put(target->scsi_host);
+       /* And another put to really free the target port... */
+       scsi_host_put(target->scsi_host);
+}
+
+static int srp_connect_target(struct srp_target_port *target)
+{
+       int ret;
+
+       ret = srp_lookup_path(target);
+       if (ret)
+               return ret;
+
+       while (1) {
+               init_completion(&target->done);
+               ret = srp_send_req(target);
+               if (ret)
+                       return ret;
+               wait_for_completion(&target->done);
+
+               /*
+                * The CM event handling code will set status to
+                * SRP_PORT_REDIRECT if we get a port redirect REJ
+                * back, or SRP_DLID_REDIRECT if we get a lid/qp
+                * redirect REJ back.
+                */
+               switch (target->status) {
+               case 0:
+                       return 0;
+
+               case SRP_PORT_REDIRECT:
+                       ret = srp_lookup_path(target);
+                       if (ret)
+                               return ret;
+                       break;
+
+               case SRP_DLID_REDIRECT:
+                       break;
+
+               default:
+                       return target->status;
+               }
+       }
+}
+
+static int srp_reconnect_target(struct srp_target_port *target)
+{
+       struct ib_cm_id *new_cm_id;
+       struct ib_qp_attr qp_attr;
+       struct srp_request *req;
+       struct ib_wc wc;
+       int ret;
+       int i;
+
+       spin_lock_irq(target->scsi_host->host_lock);
+       if (target->state != SRP_TARGET_LIVE) {
+               spin_unlock_irq(target->scsi_host->host_lock);
+               return -EAGAIN;
+       }
+       target->state = SRP_TARGET_CONNECTING;
+       spin_unlock_irq(target->scsi_host->host_lock);
+
+       srp_disconnect_target(target);
+       /*
+        * Now get a new local CM ID so that we avoid confusing the
+        * target in case things are really fouled up.
+        */
+       new_cm_id = ib_create_cm_id(target->srp_host->dev,
+                                   srp_cm_handler, target);
+       if (IS_ERR(new_cm_id)) {
+               ret = PTR_ERR(new_cm_id);
+               goto err;
+       }
+       ib_destroy_cm_id(target->cm_id);
+       target->cm_id = new_cm_id;
+
+       qp_attr.qp_state = IB_QPS_RESET;
+       ret = ib_modify_qp(target->qp, &qp_attr, IB_QP_STATE);
+       if (ret)
+               goto err;
+
+       ret = srp_init_qp(target, target->qp);
+       if (ret)
+               goto err;
+
+       while (ib_poll_cq(target->cq, 1, &wc) > 0)
+               ; /* nothing */
+
+       list_for_each_entry(req, &target->req_queue, list) {
+               req->scmnd->result = DID_RESET << 16;
+               req->scmnd->scsi_done(req->scmnd);
+       }
+
+       target->rx_head  = 0;
+       target->tx_head  = 0;
+       target->tx_tail  = 0;
+       target->req_head = 0;
+       for (i = 0; i < SRP_SQ_SIZE - 1; ++i)
+               target->req_ring[i].next = i + 1;
+       target->req_ring[SRP_SQ_SIZE - 1].next = -1;
+       INIT_LIST_HEAD(&target->req_queue);
+
+       ret = srp_connect_target(target);
+       if (ret)
+               goto err;
+
+       spin_lock_irq(target->scsi_host->host_lock);
+       if (target->state == SRP_TARGET_CONNECTING) {
+               ret = 0;
+               target->state = SRP_TARGET_LIVE;
+       } else
+               ret = -EAGAIN;
+       spin_unlock_irq(target->scsi_host->host_lock);
+
+       return ret;
+
+err:
+       printk(KERN_ERR PFX "reconnect failed (%d), removing target port.\n", ret);
+
+       /*
+        * We couldn't reconnect, so kill our target port off.
+        * However, we have to defer the real removal because we might
+        * be in the context of the SCSI error handler now, which
+        * would deadlock if we call scsi_remove_host().
+        */
+       spin_lock_irq(target->scsi_host->host_lock);
+       if (target->state == SRP_TARGET_CONNECTING) {
+               target->state = SRP_TARGET_DEAD;
+               INIT_WORK(&target->work, srp_remove_work, target);
+               schedule_work(&target->work);
+       }
+       spin_unlock_irq(target->scsi_host->host_lock);
+
+       return ret;
+}
+
+static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target,
+                       struct srp_request *req)
+{
+       struct srp_cmd *cmd = req->cmd->buf;
+       int len;
+       u8 fmt;
+
+       if (!scmnd->request_buffer || scmnd->sc_data_direction == DMA_NONE)
+               return sizeof (struct srp_cmd);
+
+       if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
+           scmnd->sc_data_direction != DMA_TO_DEVICE) {
+               printk(KERN_WARNING PFX "Unhandled data direction %d\n",
+                      scmnd->sc_data_direction);
+               return -EINVAL;
+       }
+
+       if (scmnd->use_sg) {
+               struct scatterlist *scat = scmnd->request_buffer;
+               int n;
+               int i;
+
+               n = dma_map_sg(target->srp_host->dev->dma_device,
+                              scat, scmnd->use_sg, scmnd->sc_data_direction);
+
+               if (n == 1) {
+                       struct srp_direct_buf *buf = (void *) cmd->add_data;
+
+                       fmt = SRP_DATA_DESC_DIRECT;
+
+                       buf->va  = cpu_to_be64(sg_dma_address(scat));
+                       buf->key = cpu_to_be32(target->srp_host->mr->rkey);
+                       buf->len = cpu_to_be32(sg_dma_len(scat));
+
+                       len = sizeof (struct srp_cmd) +
+                               sizeof (struct srp_direct_buf);
+               } else {
+                       struct srp_indirect_buf *buf = (void *) cmd->add_data;
+                       u32 datalen = 0;
+
+                       fmt = SRP_DATA_DESC_INDIRECT;
+
+                       if (scmnd->sc_data_direction == DMA_TO_DEVICE)
+                               cmd->data_out_desc_cnt = n;
+                       else
+                               cmd->data_in_desc_cnt = n;
+
+                       buf->table_desc.va  = cpu_to_be64(req->cmd->dma +
+                                                         sizeof *cmd +
+                                                         sizeof *buf);
+                       buf->table_desc.key =
+                               cpu_to_be32(target->srp_host->mr->rkey);
+                       buf->table_desc.len =
+                               cpu_to_be32(n * sizeof (struct srp_direct_buf));
+
+                       for (i = 0; i < n; ++i) {
+                               buf->desc_list[i].va  = cpu_to_be64(sg_dma_address(&scat[i]));
+                               buf->desc_list[i].key =
+                                       cpu_to_be32(target->srp_host->mr->rkey);
+                               buf->desc_list[i].len = cpu_to_be32(sg_dma_len(&scat[i]));
+
+                               datalen += sg_dma_len(&scat[i]);
+                       }
+
+                       buf->len = cpu_to_be32(datalen);
+
+                       len = sizeof (struct srp_cmd) +
+                               sizeof (struct srp_indirect_buf) +
+                               n * sizeof (struct srp_direct_buf);
+               }
+       } else {
+               struct srp_direct_buf *buf = (void *) cmd->add_data;
+               dma_addr_t dma;
+
+               dma = dma_map_single(target->srp_host->dev->dma_device,
+                                    scmnd->request_buffer, scmnd->request_bufflen,
+                                    scmnd->sc_data_direction);
+               if (dma_mapping_error(dma)) {
+                       printk(KERN_WARNING PFX "unable to map %p/%d (dir %d)\n",
+                              scmnd->request_buffer, (int) scmnd->request_bufflen,
+                              scmnd->sc_data_direction);
+                       return -EINVAL;
+               }
+
+               pci_unmap_addr_set(req, direct_mapping, dma);
+
+               buf->va  = cpu_to_be64(dma);
+               buf->key = cpu_to_be32(target->srp_host->mr->rkey);
+               buf->len = cpu_to_be32(scmnd->request_bufflen);
+
+               fmt = SRP_DATA_DESC_DIRECT;
+
+               len = sizeof (struct srp_cmd) + sizeof (struct srp_direct_buf);
+       }
+
+       if (scmnd->sc_data_direction == DMA_TO_DEVICE)
+               cmd->buf_fmt = fmt << 4;
+       else
+               cmd->buf_fmt = fmt;
+
+
+       return len;
+}
+
+static void srp_unmap_data(struct scsi_cmnd *scmnd,
+                          struct srp_target_port *target,
+                          struct srp_request *req)
+{
+       if (!scmnd->request_buffer ||
+           (scmnd->sc_data_direction != DMA_TO_DEVICE &&
+            scmnd->sc_data_direction != DMA_FROM_DEVICE))
+           return;
+
+       if (scmnd->use_sg)
+               dma_unmap_sg(target->srp_host->dev->dma_device,
+                            (struct scatterlist *) scmnd->request_buffer,
+                            scmnd->use_sg, scmnd->sc_data_direction);
+       else
+               dma_unmap_single(target->srp_host->dev->dma_device,
+                                pci_unmap_addr(req, direct_mapping),
+                                scmnd->request_bufflen,
+                                scmnd->sc_data_direction);
+}
+
+static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
+{
+       struct srp_request *req;
+       struct scsi_cmnd *scmnd;
+       unsigned long flags;
+       s32 delta;
+
+       delta = (s32) be32_to_cpu(rsp->req_lim_delta);
+
+       spin_lock_irqsave(target->scsi_host->host_lock, flags);
+
+       target->req_lim += delta;
+
+       req = &target->req_ring[rsp->tag & ~SRP_TAG_TSK_MGMT];
+
+       if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
+               if (be32_to_cpu(rsp->resp_data_len) < 4)
+                       req->tsk_status = -1;
+               else
+                       req->tsk_status = rsp->data[3];
+               complete(&req->done);
+       } else {
+               scmnd         = req->scmnd;
+               if (!scmnd)
+                       printk(KERN_ERR "Null scmnd for RSP w/tag %016llx\n",
+                              (unsigned long long) rsp->tag);
+               scmnd->result = rsp->status;
+
+               if (rsp->flags & SRP_RSP_FLAG_SNSVALID) {
+                       memcpy(scmnd->sense_buffer, rsp->data +
+                              be32_to_cpu(rsp->resp_data_len),
+                              min_t(int, be32_to_cpu(rsp->sense_data_len),
+                                    SCSI_SENSE_BUFFERSIZE));
+               }
+
+               if (rsp->flags & (SRP_RSP_FLAG_DOOVER | SRP_RSP_FLAG_DOUNDER))
+                       scmnd->resid = be32_to_cpu(rsp->data_out_res_cnt);
+               else if (rsp->flags & (SRP_RSP_FLAG_DIOVER | SRP_RSP_FLAG_DIUNDER))
+                       scmnd->resid = be32_to_cpu(rsp->data_in_res_cnt);
+
+               srp_unmap_data(scmnd, target, req);
+
+               if (!req->tsk_mgmt) {
+                       req->scmnd = NULL;
+                       scmnd->host_scribble = (void *) -1L;
+                       scmnd->scsi_done(scmnd);
+
+                       list_del(&req->list);
+                       req->next = target->req_head;
+                       target->req_head = rsp->tag & ~SRP_TAG_TSK_MGMT;
+               } else
+                       req->cmd_done = 1;
+       }
+
+       spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
+}
+
+static void srp_reconnect_work(void *target_ptr)
+{
+       struct srp_target_port *target = target_ptr;
+
+       srp_reconnect_target(target);
+}
+
+static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc)
+{
+       struct srp_iu *iu;
+       u8 opcode;
+
+       iu = target->rx_ring[wc->wr_id & ~SRP_OP_RECV];
+
+       dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma,
+                               target->max_ti_iu_len, DMA_FROM_DEVICE);
+
+       opcode = *(u8 *) iu->buf;
+
+       if (0) {
+               int i;
+
+               printk(KERN_ERR PFX "recv completion, opcode 0x%02x\n", opcode);
+
+               for (i = 0; i < wc->byte_len; ++i) {
+                       if (i % 8 == 0)
+                               printk(KERN_ERR "  [%02x] ", i);
+                       printk(" %02x", ((u8 *) iu->buf)[i]);
+                       if ((i + 1) % 8 == 0)
+                               printk("\n");
+               }
+
+               if (wc->byte_len % 8)
+                       printk("\n");
+       }
+
+       switch (opcode) {
+       case SRP_RSP:
+               srp_process_rsp(target, iu->buf);
+               break;
+
+       case SRP_T_LOGOUT:
+               /* XXX Handle target logout */
+               printk(KERN_WARNING PFX "Got target logout request\n");
+               break;
+
+       default:
+               printk(KERN_WARNING PFX "Unhandled SRP opcode 0x%02x\n", opcode);
+               break;
+       }
+
+       dma_sync_single_for_device(target->srp_host->dev->dma_device, iu->dma,
+                                  target->max_ti_iu_len, DMA_FROM_DEVICE);
+}
+
+static void srp_completion(struct ib_cq *cq, void *target_ptr)
+{
+       struct srp_target_port *target = target_ptr;
+       struct ib_wc wc;
+       unsigned long flags;
+
+       ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+       while (ib_poll_cq(cq, 1, &wc) > 0) {
+               if (wc.status) {
+                       printk(KERN_ERR PFX "failed %s status %d\n",
+                              wc.wr_id & SRP_OP_RECV ? "receive" : "send",
+                              wc.status);
+                       spin_lock_irqsave(target->scsi_host->host_lock, flags);
+                       if (target->state == SRP_TARGET_LIVE)
+                               schedule_work(&target->work);
+                       spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
+                       break;
+               }
+
+               if (wc.wr_id & SRP_OP_RECV)
+                       srp_handle_recv(target, &wc);
+               else
+                       ++target->tx_tail;
+       }
+}
+
+static int __srp_post_recv(struct srp_target_port *target)
+{
+       struct srp_iu *iu;
+       struct ib_sge list;
+       struct ib_recv_wr wr, *bad_wr;
+       unsigned int next;
+       int ret;
+
+       next     = target->rx_head & (SRP_RQ_SIZE - 1);
+       wr.wr_id = next | SRP_OP_RECV;
+       iu       = target->rx_ring[next];
+
+       list.addr   = iu->dma;
+       list.length = iu->size;
+       list.lkey   = target->srp_host->mr->lkey;
+
+       wr.next     = NULL;
+       wr.sg_list  = &list;
+       wr.num_sge  = 1;
+
+       ret = ib_post_recv(target->qp, &wr, &bad_wr);
+       if (!ret)
+               ++target->rx_head;
+
+       return ret;
+}
+
+static int srp_post_recv(struct srp_target_port *target)
+{
+       unsigned long flags;
+       int ret;
+
+       spin_lock_irqsave(target->scsi_host->host_lock, flags);
+       ret = __srp_post_recv(target);
+       spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
+
+       return ret;
+}
+
+/*
+ * Must be called with target->scsi_host->host_lock held to protect
+ * req_lim and tx_head.
+ */
+static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target)
+{
+       if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE)
+               return NULL;
+
+       return target->tx_ring[target->tx_head & SRP_SQ_SIZE];
+}
+
+/*
+ * Must be called with target->scsi_host->host_lock held to protect
+ * req_lim and tx_head.
+ */
+static int __srp_post_send(struct srp_target_port *target,
+                          struct srp_iu *iu, int len)
+{
+       struct ib_sge list;
+       struct ib_send_wr wr, *bad_wr;
+       int ret = 0;
+
+       if (target->req_lim < 1) {
+               printk(KERN_ERR PFX "Target has req_lim %d\n", target->req_lim);
+               return -EAGAIN;
+       }
+
+       list.addr   = iu->dma;
+       list.length = len;
+       list.lkey   = target->srp_host->mr->lkey;
+
+       wr.next       = NULL;
+       wr.wr_id      = target->tx_head & SRP_SQ_SIZE;
+       wr.sg_list    = &list;
+       wr.num_sge    = 1;
+       wr.opcode     = IB_WR_SEND;
+       wr.send_flags = IB_SEND_SIGNALED;
+
+       ret = ib_post_send(target->qp, &wr, &bad_wr);
+
+       if (!ret) {
+               ++target->tx_head;
+               --target->req_lim;
+       }
+
+       return ret;
+}
+
+static int srp_queuecommand(struct scsi_cmnd *scmnd,
+                           void (*done)(struct scsi_cmnd *))
+{
+       struct srp_target_port *target = host_to_target(scmnd->device->host);
+       struct srp_request *req;
+       struct srp_iu *iu;
+       struct srp_cmd *cmd;
+       long req_index;
+       int len;
+
+       if (target->state == SRP_TARGET_CONNECTING)
+               goto err;
+
+       if (target->state == SRP_TARGET_DEAD ||
+           target->state == SRP_TARGET_REMOVED) {
+               scmnd->result = DID_BAD_TARGET << 16;
+               done(scmnd);
+               return 0;
+       }
+
+       iu = __srp_get_tx_iu(target);
+       if (!iu)
+               goto err;
+
+       dma_sync_single_for_cpu(target->srp_host->dev->dma_device, iu->dma,
+                               SRP_MAX_IU_LEN, DMA_TO_DEVICE);
+
+       req_index = target->req_head;
+
+       scmnd->scsi_done     = done;
+       scmnd->result        = 0;
+       scmnd->host_scribble = (void *) req_index;
+
+       cmd = iu->buf;
+       memset(cmd, 0, sizeof *cmd);
+
+       cmd->opcode = SRP_CMD;
+       cmd->lun    = cpu_to_be64((u64) scmnd->device->lun << 48);
+       cmd->tag    = req_index;
+       memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);
+
+       req = &target->req_ring[req_index];
+
+       req->scmnd    = scmnd;
+       req->cmd      = iu;
+       req->cmd_done = 0;
+       req->tsk_mgmt = NULL;
+
+       len = srp_map_data(scmnd, target, req);
+       if (len < 0) {
+               printk(KERN_ERR PFX "Failed to map data\n");
+               goto err;
+       }
+
+       if (__srp_post_recv(target)) {
+               printk(KERN_ERR PFX "Recv failed\n");
+               goto err_unmap;
+       }
+
+       dma_sync_single_for_device(target->srp_host->dev->dma_device, iu->dma,
+                                  SRP_MAX_IU_LEN, DMA_TO_DEVICE);
+
+       if (__srp_post_send(target, iu, len)) {
+               printk(KERN_ERR PFX "Send failed\n");
+               goto err_unmap;
+       }
+
+       target->req_head = req->next;
+       list_add_tail(&req->list, &target->req_queue);
+
+       return 0;
+
+err_unmap:
+       srp_unmap_data(scmnd, target, req);
+
+err:
+       return SCSI_MLQUEUE_HOST_BUSY;
+}
+
+static int srp_alloc_iu_bufs(struct srp_target_port *target)
+{
+       int i;
+
+       for (i = 0; i < SRP_RQ_SIZE; ++i) {
+               target->rx_ring[i] = srp_alloc_iu(target->srp_host,
+                                                 target->max_ti_iu_len,
+                                                 GFP_KERNEL, DMA_FROM_DEVICE);
+               if (!target->rx_ring[i])
+                       goto err;
+       }
+
+       for (i = 0; i < SRP_SQ_SIZE + 1; ++i) {
+               target->tx_ring[i] = srp_alloc_iu(target->srp_host,
+                                                 SRP_MAX_IU_LEN,
+                                                 GFP_KERNEL, DMA_TO_DEVICE);
+               if (!target->tx_ring[i])
+                       goto err;
+       }
+
+       return 0;
+
+err:
+       for (i = 0; i < SRP_RQ_SIZE; ++i) {
+               srp_free_iu(target->srp_host, target->rx_ring[i]);
+               target->rx_ring[i] = NULL;
+       }
+
+       for (i = 0; i < SRP_SQ_SIZE + 1; ++i) {
+               srp_free_iu(target->srp_host, target->tx_ring[i]);
+               target->tx_ring[i] = NULL;
+       }
+
+       return -ENOMEM;
+}
+
+static void srp_cm_rej_handler(struct ib_cm_id *cm_id,
+                              struct ib_cm_event *event,
+                              struct srp_target_port *target)
+{
+       struct ib_class_port_info *cpi;
+       int opcode;
+
+       switch (event->param.rej_rcvd.reason) {
+       case IB_CM_REJ_PORT_CM_REDIRECT:
+               cpi = event->param.rej_rcvd.ari;
+               target->path.dlid = cpi->redirect_lid;
+               target->path.pkey = cpi->redirect_pkey;
+               cm_id->remote_cm_qpn = be32_to_cpu(cpi->redirect_qp) & 0x00ffffff;
+               memcpy(target->path.dgid.raw, cpi->redirect_gid, 16);
+
+               target->status = target->path.dlid ?
+                       SRP_DLID_REDIRECT : SRP_PORT_REDIRECT;
+               break;
+
+       case IB_CM_REJ_PORT_REDIRECT:
+               if (topspin_workarounds &&
+                   !memcmp(&target->ioc_guid, topspin_oui, 3)) {
+                       /*
+                        * Topspin/Cisco SRP gateways incorrectly send
+                        * reject reason code 25 when they mean 24
+                        * (port redirect).
+                        */
+                       memcpy(target->path.dgid.raw,
+                              event->param.rej_rcvd.ari, 16);
+
+                       printk(KERN_DEBUG PFX "Topspin/Cisco redirect to target port GID %016llx%016llx\n",
+                              (unsigned long long) be64_to_cpu(target->path.dgid.global.subnet_prefix),
+                              (unsigned long long) be64_to_cpu(target->path.dgid.global.interface_id));
+
+                       target->status = SRP_PORT_REDIRECT;
+               } else {
+                       printk(KERN_WARNING "  REJ reason: IB_CM_REJ_PORT_REDIRECT\n");
+                       target->status = -ECONNRESET;
+               }
+               break;
+
+       case IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID:
+               printk(KERN_WARNING "  REJ reason: IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID\n");
+               target->status = -ECONNRESET;
+               break;
+
+       case IB_CM_REJ_CONSUMER_DEFINED:
+               opcode = *(u8 *) event->private_data;
+               if (opcode == SRP_LOGIN_REJ) {
+                       struct srp_login_rej *rej = event->private_data;
+                       u32 reason = be32_to_cpu(rej->reason);
+
+                       if (reason == SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE)
+                               printk(KERN_WARNING PFX
+                                      "SRP_LOGIN_REJ: requested max_it_iu_len too large\n");
+                       else
+                               printk(KERN_WARNING PFX
+                                      "SRP LOGIN REJECTED, reason 0x%08x\n", reason);
+               } else
+                       printk(KERN_WARNING "  REJ reason: IB_CM_REJ_CONSUMER_DEFINED,"
+                              " opcode 0x%02x\n", opcode);
+               target->status = -ECONNRESET;
+               break;
+
+       default:
+               printk(KERN_WARNING "  REJ reason 0x%x\n",
+                      event->param.rej_rcvd.reason);
+               target->status = -ECONNRESET;
+       }
+}
+
+static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
+{
+       struct srp_target_port *target = cm_id->context;
+       struct ib_qp_attr *qp_attr = NULL;
+       int attr_mask = 0;
+       int comp = 0;
+       int opcode = 0;
+
+       switch (event->event) {
+       case IB_CM_REQ_ERROR:
+               printk(KERN_DEBUG PFX "Sending CM REQ failed\n");
+               comp = 1;
+               target->status = -ECONNRESET;
+               break;
+
+       case IB_CM_REP_RECEIVED:
+               comp = 1;
+               opcode = *(u8 *) event->private_data;
+
+               if (opcode == SRP_LOGIN_RSP) {
+                       struct srp_login_rsp *rsp = event->private_data;
+
+                       target->max_ti_iu_len = be32_to_cpu(rsp->max_ti_iu_len);
+                       target->req_lim       = be32_to_cpu(rsp->req_lim_delta);
+
+                       target->scsi_host->can_queue = min(target->req_lim,
+                                                          target->scsi_host->can_queue);
+               } else {
+                       printk(KERN_WARNING PFX "Unhandled RSP opcode %#x\n", opcode);
+                       target->status = -ECONNRESET;
+                       break;
+               }
+
+               target->status = srp_alloc_iu_bufs(target);
+               if (target->status)
+                       break;
+
+               qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL);
+               if (!qp_attr) {
+                       target->status = -ENOMEM;
+                       break;
+               }
+
+               qp_attr->qp_state = IB_QPS_RTR;
+               target->status = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
+               if (target->status)
+                       break;
+
+               target->status = ib_modify_qp(target->qp, qp_attr, attr_mask);
+               if (target->status)
+                       break;
+
+               target->status = srp_post_recv(target);
+               if (target->status)
+                       break;
+
+               qp_attr->qp_state = IB_QPS_RTS;
+               target->status = ib_cm_init_qp_attr(cm_id, qp_attr, &attr_mask);
+               if (target->status)
+                       break;
+
+               target->status = ib_modify_qp(target->qp, qp_attr, attr_mask);
+               if (target->status)
+                       break;
+
+               target->status = ib_send_cm_rtu(cm_id, NULL, 0);
+               if (target->status)
+                       break;
+
+               break;
+
+       case IB_CM_REJ_RECEIVED:
+               printk(KERN_DEBUG PFX "REJ received\n");
+               comp = 1;
+
+               srp_cm_rej_handler(cm_id, event, target);
+               break;
+
+       case IB_CM_MRA_RECEIVED:
+               printk(KERN_ERR PFX "MRA received\n");
+               break;
+
+       case IB_CM_DREP_RECEIVED:
+               break;
+
+       case IB_CM_TIMEWAIT_EXIT:
+               printk(KERN_ERR PFX "connection closed\n");
+
+               comp = 1;
+               target->status = 0;
+               break;
+
+       default:
+               printk(KERN_WARNING PFX "Unhandled CM event %d\n", event->event);
+               break;
+       }
+
+       if (comp)
+               complete(&target->done);
+
+       kfree(qp_attr);
+
+       return 0;
+}
+
+static int srp_send_tsk_mgmt(struct scsi_cmnd *scmnd, u8 func)
+{
+       struct srp_target_port *target = host_to_target(scmnd->device->host);
+       struct srp_request *req;
+       struct srp_iu *iu;
+       struct srp_tsk_mgmt *tsk_mgmt;
+       int req_index;
+       int ret = FAILED;
+
+       spin_lock_irq(target->scsi_host->host_lock);
+
+       if (scmnd->host_scribble == (void *) -1L)
+               goto out;
+
+       req_index = (long) scmnd->host_scribble;
+       printk(KERN_ERR "Abort for req_index %d\n", req_index);
+
+       req = &target->req_ring[req_index];
+       init_completion(&req->done);
+
+       iu = __srp_get_tx_iu(target);
+       if (!iu)
+               goto out;
+
+       tsk_mgmt = iu->buf;
+       memset(tsk_mgmt, 0, sizeof *tsk_mgmt);
+
+       tsk_mgmt->opcode        = SRP_TSK_MGMT;
+       tsk_mgmt->lun           = cpu_to_be64((u64) scmnd->device->lun << 48);
+       tsk_mgmt->tag           = req_index | SRP_TAG_TSK_MGMT;
+       tsk_mgmt->tsk_mgmt_func = func;
+       tsk_mgmt->task_tag      = req_index;
+
+       if (__srp_post_send(target, iu, sizeof *tsk_mgmt))
+               goto out;
+
+       req->tsk_mgmt = iu;
+
+       spin_unlock_irq(target->scsi_host->host_lock);
+       if (!wait_for_completion_timeout(&req->done,
+                                        msecs_to_jiffies(SRP_ABORT_TIMEOUT_MS)))
+               return FAILED;
+       spin_lock_irq(target->scsi_host->host_lock);
+
+       if (req->cmd_done) {
+               list_del(&req->list);
+               req->next = target->req_head;
+               target->req_head = req_index;
+
+               scmnd->scsi_done(scmnd);
+       } else if (!req->tsk_status) {
+               scmnd->result = DID_ABORT << 16;
+               ret = SUCCESS;
+       }
+
+out:
+       spin_unlock_irq(target->scsi_host->host_lock);
+       return ret;
+}
+
+static int srp_abort(struct scsi_cmnd *scmnd)
+{
+       printk(KERN_ERR "SRP abort called\n");
+
+       return srp_send_tsk_mgmt(scmnd, SRP_TSK_ABORT_TASK);
+}
+
+static int srp_reset_device(struct scsi_cmnd *scmnd)
+{
+       printk(KERN_ERR "SRP reset_device called\n");
+
+       return srp_send_tsk_mgmt(scmnd, SRP_TSK_LUN_RESET);
+}
+
+static int srp_reset_host(struct scsi_cmnd *scmnd)
+{
+       struct srp_target_port *target = host_to_target(scmnd->device->host);
+       int ret = FAILED;
+
+       printk(KERN_ERR PFX "SRP reset_host called\n");
+
+       if (!srp_reconnect_target(target))
+               ret = SUCCESS;
+
+       return ret;
+}
+
+static struct scsi_host_template srp_template = {
+       .module                         = THIS_MODULE,
+       .name                           = DRV_NAME,
+       .info                           = srp_target_info,
+       .queuecommand                   = srp_queuecommand,
+       .eh_abort_handler               = srp_abort,
+       .eh_device_reset_handler        = srp_reset_device,
+       .eh_host_reset_handler          = srp_reset_host,
+       .can_queue                      = SRP_SQ_SIZE,
+       .this_id                        = -1,
+       .sg_tablesize                   = SRP_MAX_INDIRECT,
+       .cmd_per_lun                    = SRP_SQ_SIZE,
+       .use_clustering                 = ENABLE_CLUSTERING
+};
+
+static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
+{
+       sprintf(target->target_name, "SRP.T10:%016llX",
+                (unsigned long long) be64_to_cpu(target->id_ext));
+
+       if (scsi_add_host(target->scsi_host, host->dev->dma_device))
+               return -ENODEV;
+
+       down(&host->target_mutex);
+       list_add_tail(&target->list, &host->target_list);
+       up(&host->target_mutex);
+
+       target->state = SRP_TARGET_LIVE;
+
+       /* XXX: are we supposed to have a definition of SCAN_WILD_CARD ?? */
+       scsi_scan_target(&target->scsi_host->shost_gendev,
+                        0, target->scsi_id, ~0, 0);
+
+       return 0;
+}
+
+static void srp_release_class_dev(struct class_device *class_dev)
+{
+       struct srp_host *host =
+               container_of(class_dev, struct srp_host, class_dev);
+
+       complete(&host->released);
+}
+
+static struct class srp_class = {
+       .name    = "infiniband_srp",
+       .release = srp_release_class_dev
+};
+
+/*
+ * Target ports are added by writing
+ *
+ *     id_ext=<SRP ID ext>,ioc_guid=<SRP IOC GUID>,dgid=<dest GID>,
+ *     pkey=<P_Key>,service_id=<service ID>
+ *
+ * to the add_target sysfs attribute.
+ */
+enum {
+       SRP_OPT_ERR             = 0,
+       SRP_OPT_ID_EXT          = 1 << 0,
+       SRP_OPT_IOC_GUID        = 1 << 1,
+       SRP_OPT_DGID            = 1 << 2,
+       SRP_OPT_PKEY            = 1 << 3,
+       SRP_OPT_SERVICE_ID      = 1 << 4,
+       SRP_OPT_MAX_SECT        = 1 << 5,
+       SRP_OPT_ALL             = (SRP_OPT_ID_EXT       |
+                                  SRP_OPT_IOC_GUID     |
+                                  SRP_OPT_DGID         |
+                                  SRP_OPT_PKEY         |
+                                  SRP_OPT_SERVICE_ID),
+};
+
+static match_table_t srp_opt_tokens = {
+       { SRP_OPT_ID_EXT,       "id_ext=%s"     },
+       { SRP_OPT_IOC_GUID,     "ioc_guid=%s"   },
+       { SRP_OPT_DGID,         "dgid=%s"       },
+       { SRP_OPT_PKEY,         "pkey=%x"       },
+       { SRP_OPT_SERVICE_ID,   "service_id=%s" },
+       { SRP_OPT_MAX_SECT,     "max_sect=%d"   },
+       { SRP_OPT_ERR,          NULL            }
+};
+
+static int srp_parse_options(const char *buf, struct srp_target_port *target)
+{
+       char *options, *sep_opt;
+       char *p;
+       char dgid[3];
+       substring_t args[MAX_OPT_ARGS];
+       int opt_mask = 0;
+       int token;
+       int ret = -EINVAL;
+       int i;
+
+       options = kstrdup(buf, GFP_KERNEL);
+       if (!options)
+               return -ENOMEM;
+
+       sep_opt = options;
+       while ((p = strsep(&sep_opt, ",")) != NULL) {
+               if (!*p)
+                       continue;
+
+               token = match_token(p, srp_opt_tokens, args);
+               opt_mask |= token;
+
+               switch (token) {
+               case SRP_OPT_ID_EXT:
+                       p = match_strdup(args);
+                       target->id_ext = cpu_to_be64(simple_strtoull(p, NULL, 16));
+                       kfree(p);
+                       break;
+
+               case SRP_OPT_IOC_GUID:
+                       p = match_strdup(args);
+                       target->ioc_guid = cpu_to_be64(simple_strtoull(p, NULL, 16));
+                       kfree(p);
+                       break;
+
+               case SRP_OPT_DGID:
+                       p = match_strdup(args);
+                       if (strlen(p) != 32) {
+                               printk(KERN_WARNING PFX "bad dest GID parameter '%s'\n", p);
+                               goto out;
+                       }
+
+                       for (i = 0; i < 16; ++i) {
+                               strlcpy(dgid, p + i * 2, 3);
+                               target->path.dgid.raw[i] = simple_strtoul(dgid, NULL, 16);
+                       }
+                       break;
+
+               case SRP_OPT_PKEY:
+                       if (match_hex(args, &token)) {
+                               printk(KERN_WARNING PFX "bad P_Key parameter '%s'\n", p);
+                               goto out;
+                       }
+                       target->path.pkey = cpu_to_be16(token);
+                       break;
+
+               case SRP_OPT_SERVICE_ID:
+                       p = match_strdup(args);
+                       target->service_id = cpu_to_be64(simple_strtoull(p, NULL, 16));
+                       kfree(p);
+                       break;
+
+               case SRP_OPT_MAX_SECT:
+                       if (match_int(args, &token)) {
+                               printk(KERN_WARNING PFX "bad max sect parameter '%s'\n", p);
+                               goto out;
+                       }
+                       target->scsi_host->max_sectors = token;
+                       break;
+
+               default:
+                       printk(KERN_WARNING PFX "unknown parameter or missing value "
+                              "'%s' in target creation request\n", p);
+                       goto out;
+               }
+       }
+
+       if ((opt_mask & SRP_OPT_ALL) == SRP_OPT_ALL)
+               ret = 0;
+       else
+               for (i = 0; i < ARRAY_SIZE(srp_opt_tokens); ++i)
+                       if ((srp_opt_tokens[i].token & SRP_OPT_ALL) &&
+                           !(srp_opt_tokens[i].token & opt_mask))
+                               printk(KERN_WARNING PFX "target creation request is "
+                                      "missing parameter '%s'\n",
+                                      srp_opt_tokens[i].pattern);
+
+out:
+       kfree(options);
+       return ret;
+}
+
+static ssize_t srp_create_target(struct class_device *class_dev,
+                                const char *buf, size_t count)
+{
+       struct srp_host *host =
+               container_of(class_dev, struct srp_host, class_dev);
+       struct Scsi_Host *target_host;
+       struct srp_target_port *target;
+       int ret;
+       int i;
+
+       target_host = scsi_host_alloc(&srp_template,
+                                     sizeof (struct srp_target_port));
+       if (!target_host)
+               return -ENOMEM;
+
+       target = host_to_target(target_host);
+       memset(target, 0, sizeof *target);
+
+       target->scsi_host  = target_host;
+       target->srp_host   = host;
+
+       INIT_WORK(&target->work, srp_reconnect_work, target);
+
+       for (i = 0; i < SRP_SQ_SIZE - 1; ++i)
+               target->req_ring[i].next = i + 1;
+       target->req_ring[SRP_SQ_SIZE - 1].next = -1;
+       INIT_LIST_HEAD(&target->req_queue);
+
+       ret = srp_parse_options(buf, target);
+       if (ret)
+               goto err;
+
+       ib_get_cached_gid(host->dev, host->port, 0, &target->path.sgid);
+
+       printk(KERN_DEBUG PFX "new target: id_ext %016llx ioc_guid %016llx pkey %04x "
+              "service_id %016llx dgid %04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n",
+              (unsigned long long) be64_to_cpu(target->id_ext),
+              (unsigned long long) be64_to_cpu(target->ioc_guid),
+              be16_to_cpu(target->path.pkey),
+              (unsigned long long) be64_to_cpu(target->service_id),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[0]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[2]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[4]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[6]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[8]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[10]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[12]),
+              (int) be16_to_cpu(*(__be16 *) &target->path.dgid.raw[14]));
+
+       ret = srp_create_target_ib(target);
+       if (ret)
+               goto err;
+
+       target->cm_id = ib_create_cm_id(host->dev, srp_cm_handler, target);
+       if (IS_ERR(target->cm_id)) {
+               ret = PTR_ERR(target->cm_id);
+               goto err_free;
+       }
+
+       ret = srp_connect_target(target);
+       if (ret) {
+               printk(KERN_ERR PFX "Connection failed\n");
+               goto err_cm_id;
+       }
+
+       ret = srp_add_target(host, target);
+       if (ret)
+               goto err_disconnect;
+
+       return count;
+
+err_disconnect:
+       srp_disconnect_target(target);
+
+err_cm_id:
+       ib_destroy_cm_id(target->cm_id);
+
+err_free:
+       srp_free_target_ib(target);
+
+err:
+       scsi_host_put(target_host);
+
+       return ret;
+}
+
+static CLASS_DEVICE_ATTR(add_target, S_IWUSR, NULL, srp_create_target);
+
+static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
+{
+       struct srp_host *host =
+               container_of(class_dev, struct srp_host, class_dev);
+
+       return sprintf(buf, "%s\n", host->dev->name);
+}
+
+static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
+
+static ssize_t show_port(struct class_device *class_dev, char *buf)
+{
+       struct srp_host *host =
+               container_of(class_dev, struct srp_host, class_dev);
+
+       return sprintf(buf, "%d\n", host->port);
+}
+
+static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
+
+static struct srp_host *srp_add_port(struct ib_device *device,
+                                    __be64 node_guid, u8 port)
+{
+       struct srp_host *host;
+
+       host = kzalloc(sizeof *host, GFP_KERNEL);
+       if (!host)
+               return NULL;
+
+       INIT_LIST_HEAD(&host->target_list);
+       init_MUTEX(&host->target_mutex);
+       init_completion(&host->released);
+       host->dev  = device;
+       host->port = port;
+
+       host->initiator_port_id[7] = port;
+       memcpy(host->initiator_port_id + 8, &node_guid, 8);
+
+       host->pd   = ib_alloc_pd(device);
+       if (IS_ERR(host->pd))
+               goto err_free;
+
+       host->mr   = ib_get_dma_mr(host->pd,
+                                  IB_ACCESS_LOCAL_WRITE |
+                                  IB_ACCESS_REMOTE_READ |
+                                  IB_ACCESS_REMOTE_WRITE);
+       if (IS_ERR(host->mr))
+               goto err_pd;
+
+       host->class_dev.class = &srp_class;
+       host->class_dev.dev   = device->dma_device;
+       snprintf(host->class_dev.class_id, BUS_ID_SIZE, "srp-%s-%d",
+                device->name, port);
+
+       if (class_device_register(&host->class_dev))
+               goto err_mr;
+       if (class_device_create_file(&host->class_dev, &class_device_attr_add_target))
+               goto err_class;
+       if (class_device_create_file(&host->class_dev, &class_device_attr_ibdev))
+               goto err_class;
+       if (class_device_create_file(&host->class_dev, &class_device_attr_port))
+               goto err_class;
+
+       return host;
+
+err_class:
+       class_device_unregister(&host->class_dev);
+
+err_mr:
+       ib_dereg_mr(host->mr);
+
+err_pd:
+       ib_dealloc_pd(host->pd);
+
+err_free:
+       kfree(host);
+
+       return NULL;
+}
+
+static void srp_add_one(struct ib_device *device)
+{
+       struct list_head *dev_list;
+       struct srp_host *host;
+       struct ib_device_attr *dev_attr;
+       int s, e, p;
+
+       dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
+       if (!dev_attr)
+               return;
+
+       if (ib_query_device(device, dev_attr)) {
+               printk(KERN_WARNING PFX "Couldn't query node GUID for %s.\n",
+                      device->name);
+               goto out;
+       }
+
+       dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
+       if (!dev_list)
+               goto out;
+
+       INIT_LIST_HEAD(dev_list);
+
+       if (device->node_type == IB_NODE_SWITCH) {
+               s = 0;
+               e = 0;
+       } else {
+               s = 1;
+               e = device->phys_port_cnt;
+       }
+
+       for (p = s; p <= e; ++p) {
+               host = srp_add_port(device, dev_attr->node_guid, p);
+               if (host)
+                       list_add_tail(&host->list, dev_list);
+       }
+
+       ib_set_client_data(device, &srp_client, dev_list);
+
+out:
+       kfree(dev_attr);
+}
+
+static void srp_remove_one(struct ib_device *device)
+{
+       struct list_head *dev_list;
+       struct srp_host *host, *tmp_host;
+       LIST_HEAD(target_list);
+       struct srp_target_port *target, *tmp_target;
+       unsigned long flags;
+
+       dev_list = ib_get_client_data(device, &srp_client);
+
+       list_for_each_entry_safe(host, tmp_host, dev_list, list) {
+               class_device_unregister(&host->class_dev);
+               /*
+                * Wait for the sysfs entry to go away, so that no new
+                * target ports can be created.
+                */
+               wait_for_completion(&host->released);
+
+               /*
+                * Mark all target ports as removed, so we stop queueing
+                * commands and don't try to reconnect.
+                */
+               down(&host->target_mutex);
+               list_for_each_entry_safe(target, tmp_target,
+                                        &host->target_list, list) {
+                       spin_lock_irqsave(target->scsi_host->host_lock, flags);
+                       if (target->state != SRP_TARGET_REMOVED)
+                               target->state = SRP_TARGET_REMOVED;
+                       spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
+               }
+               up(&host->target_mutex);
+
+               /*
+                * Wait for any reconnection tasks that may have
+                * started before we marked our target ports as
+                * removed, and any target port removal tasks.
+                */
+               flush_scheduled_work();
+
+               list_for_each_entry_safe(target, tmp_target,
+                                        &host->target_list, list) {
+                       scsi_remove_host(target->scsi_host);
+                       srp_disconnect_target(target);
+                       ib_destroy_cm_id(target->cm_id);
+                       srp_free_target_ib(target);
+                       scsi_host_put(target->scsi_host);
+               }
+
+               ib_dereg_mr(host->mr);
+               ib_dealloc_pd(host->pd);
+               kfree(host);
+       }
+
+       kfree(dev_list);
+}
+
+static int __init srp_init_module(void)
+{
+       int ret;
+
+       ret = class_register(&srp_class);
+       if (ret) {
+               printk(KERN_ERR PFX "couldn't register class infiniband_srp\n");
+               return ret;
+       }
+
+       ret = ib_register_client(&srp_client);
+       if (ret) {
+               printk(KERN_ERR PFX "couldn't register IB client\n");
+               class_unregister(&srp_class);
+               return ret;
+       }
+
+       return 0;
+}
+
+static void __exit srp_cleanup_module(void)
+{
+       ib_unregister_client(&srp_client);
+       class_unregister(&srp_class);
+}
+
+module_init(srp_init_module);
+module_exit(srp_cleanup_module);
diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h
new file mode 100644 (file)
index 0000000..4fec28a
--- /dev/null
@@ -0,0 +1,150 @@
+/*
+ * Copyright (c) 2005 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: ib_srp.h 3932 2005-11-01 17:19:29Z roland $
+ */
+
+#ifndef IB_SRP_H
+#define IB_SRP_H
+
+#include <linux/types.h>
+#include <linux/list.h>
+
+#include <asm/semaphore.h>
+
+#include <scsi/scsi_host.h>
+#include <scsi/scsi_cmnd.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_sa.h>
+#include <rdma/ib_cm.h>
+
+enum {
+       SRP_PATH_REC_TIMEOUT_MS = 1000,
+       SRP_ABORT_TIMEOUT_MS    = 5000,
+
+       SRP_PORT_REDIRECT       = 1,
+       SRP_DLID_REDIRECT       = 2,
+
+       SRP_MAX_IU_LEN          = 256,
+
+       SRP_RQ_SHIFT            = 6,
+       SRP_RQ_SIZE             = 1 << SRP_RQ_SHIFT,
+       SRP_SQ_SIZE             = SRP_RQ_SIZE - 1,
+       SRP_CQ_SIZE             = SRP_SQ_SIZE + SRP_RQ_SIZE,
+
+       SRP_TAG_TSK_MGMT        = 1 << (SRP_RQ_SHIFT + 1)
+};
+
+#define SRP_OP_RECV            (1 << 31)
+#define SRP_MAX_INDIRECT       ((SRP_MAX_IU_LEN -                      \
+                                 sizeof (struct srp_cmd) -             \
+                                 sizeof (struct srp_indirect_buf)) / 16)
+
+enum srp_target_state {
+       SRP_TARGET_LIVE,
+       SRP_TARGET_CONNECTING,
+       SRP_TARGET_DEAD,
+       SRP_TARGET_REMOVED
+};
+
+struct srp_host {
+       u8                      initiator_port_id[16];
+       struct ib_device       *dev;
+       u8                      port;
+       struct ib_pd           *pd;
+       struct ib_mr           *mr;
+       struct class_device     class_dev;
+       struct list_head        target_list;
+       struct semaphore        target_mutex;
+       struct completion       released;
+       struct list_head        list;
+};
+
+struct srp_request {
+       struct list_head        list;
+       struct scsi_cmnd       *scmnd;
+       struct srp_iu          *cmd;
+       struct srp_iu          *tsk_mgmt;
+       DECLARE_PCI_UNMAP_ADDR(direct_mapping)
+       struct completion       done;
+       short                   next;
+       u8                      cmd_done;
+       u8                      tsk_status;
+};
+
+struct srp_target_port {
+       __be64                  id_ext;
+       __be64                  ioc_guid;
+       __be64                  service_id;
+       struct srp_host        *srp_host;
+       struct Scsi_Host       *scsi_host;
+       char                    target_name[32];
+       unsigned int            scsi_id;
+
+       struct ib_sa_path_rec   path;
+       struct ib_sa_query     *path_query;
+       int                     path_query_id;
+
+       struct ib_cm_id        *cm_id;
+       struct ib_cq           *cq;
+       struct ib_qp           *qp;
+
+       int                     max_ti_iu_len;
+       s32                     req_lim;
+
+       unsigned                rx_head;
+       struct srp_iu          *rx_ring[SRP_RQ_SIZE];
+
+       unsigned                tx_head;
+       unsigned                tx_tail;
+       struct srp_iu          *tx_ring[SRP_SQ_SIZE + 1];
+
+       int                     req_head;
+       struct list_head        req_queue;
+       struct srp_request      req_ring[SRP_SQ_SIZE];
+
+       struct work_struct      work;
+
+       struct list_head        list;
+       struct completion       done;
+       int                     status;
+       enum srp_target_state   state;
+};
+
+struct srp_iu {
+       dma_addr_t              dma;
+       void                   *buf;
+       size_t                  size;
+       enum dma_data_direction direction;
+};
+
+#endif /* IB_SRP_H */
diff --git a/include/scsi/srp.h b/include/scsi/srp.h
new file mode 100644 (file)
index 0000000..6c2681d
--- /dev/null
@@ -0,0 +1,226 @@
+/*
+ * Copyright (c) 2005 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id$
+ */
+
+#ifndef SCSI_SRP_H
+#define SCSI_SRP_H
+
+/*
+ * Structures and constants for the SCSI RDMA Protocol (SRP) as
+ * defined by the INCITS T10 committee.  This file was written using
+ * draft Revision 16a of the SRP standard.
+ */
+
+#include <linux/types.h>
+
+enum {
+       SRP_LOGIN_REQ   = 0x00,
+       SRP_TSK_MGMT    = 0x01,
+       SRP_CMD         = 0x02,
+       SRP_I_LOGOUT    = 0x03,
+       SRP_LOGIN_RSP   = 0xc0,
+       SRP_RSP         = 0xc1,
+       SRP_LOGIN_REJ   = 0xc2,
+       SRP_T_LOGOUT    = 0x80,
+       SRP_CRED_REQ    = 0x81,
+       SRP_AER_REQ     = 0x82,
+       SRP_CRED_RSP    = 0x41,
+       SRP_AER_RSP     = 0x42
+};
+
+enum {
+       SRP_BUF_FORMAT_DIRECT   = 1 << 1,
+       SRP_BUF_FORMAT_INDIRECT = 1 << 2
+};
+
+enum {
+       SRP_NO_DATA_DESC        = 0,
+       SRP_DATA_DESC_DIRECT    = 1,
+       SRP_DATA_DESC_INDIRECT  = 2
+};
+
+enum {
+       SRP_TSK_ABORT_TASK      = 0x01,
+       SRP_TSK_ABORT_TASK_SET  = 0x02,
+       SRP_TSK_CLEAR_TASK_SET  = 0x04,
+       SRP_TSK_LUN_RESET       = 0x08,
+       SRP_TSK_CLEAR_ACA       = 0x40
+};
+
+enum srp_login_rej_reason {
+       SRP_LOGIN_REJ_UNABLE_ESTABLISH_CHANNEL          = 0x00010000,
+       SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES            = 0x00010001,
+       SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE        = 0x00010002,
+       SRP_LOGIN_REJ_UNABLE_ASSOCIATE_CHANNEL          = 0x00010003,
+       SRP_LOGIN_REJ_UNSUPPORTED_DESCRIPTOR_FMT        = 0x00010004,
+       SRP_LOGIN_REJ_MULTI_CHANNEL_UNSUPPORTED         = 0x00010005,
+       SRP_LOGIN_REJ_CHANNEL_LIMIT_REACHED             = 0x00010006
+};
+
+struct srp_direct_buf {
+       __be64  va;
+       __be32  key;
+       __be32  len;
+};
+
+/*
+ * We need the packed attribute because the SRP spec puts the list of
+ * descriptors at an offset of 20, which is not aligned to the size
+ * of struct srp_direct_buf.
+ */
+struct srp_indirect_buf {
+       struct srp_direct_buf   table_desc;
+       __be32                  len;
+       struct srp_direct_buf   desc_list[0] __attribute__((packed));
+};
+
+enum {
+       SRP_MULTICHAN_SINGLE = 0,
+       SRP_MULTICHAN_MULTI  = 1
+};
+
+struct srp_login_req {
+       u8      opcode;
+       u8      reserved1[7];
+       u64     tag;
+       __be32  req_it_iu_len;
+       u8      reserved2[4];
+       __be16  req_buf_fmt;
+       u8      req_flags;
+       u8      reserved3[5];
+       u8      initiator_port_id[16];
+       u8      target_port_id[16];
+};
+
+struct srp_login_rsp {
+       u8      opcode;
+       u8      reserved1[3];
+       __be32  req_lim_delta;
+       u64     tag;
+       __be32  max_it_iu_len;
+       __be32  max_ti_iu_len;
+       __be16  buf_fmt;
+       u8      rsp_flags;
+       u8      reserved2[25];
+};
+
+struct srp_login_rej {
+       u8      opcode;
+       u8      reserved1[3];
+       __be32  reason;
+       u64     tag;
+       u8      reserved2[8];
+       __be16  buf_fmt;
+       u8      reserved3[6];
+};
+
+struct srp_i_logout {
+       u8      opcode;
+       u8      reserved[7];
+       u64     tag;
+};
+
+struct srp_t_logout {
+       u8      opcode;
+       u8      sol_not;
+       u8      reserved[2];
+       __be32  reason;
+       u64     tag;
+};
+
+/*
+ * We need the packed attribute because the SRP spec only aligns the
+ * 8-byte LUN field to 4 bytes.
+ */
+struct srp_tsk_mgmt {
+       u8      opcode;
+       u8      sol_not;
+       u8      reserved1[6];
+       u64     tag;
+       u8      reserved2[4];
+       __be64  lun __attribute__((packed));
+       u8      reserved3[2];
+       u8      tsk_mgmt_func;
+       u8      reserved4;
+       u64     task_tag;
+       u8      reserved5[8];
+};
+
+/*
+ * We need the packed attribute because the SRP spec only aligns the
+ * 8-byte LUN field to 4 bytes.
+ */
+struct srp_cmd {
+       u8      opcode;
+       u8      sol_not;
+       u8      reserved1[3];
+       u8      buf_fmt;
+       u8      data_out_desc_cnt;
+       u8      data_in_desc_cnt;
+       u64     tag;
+       u8      reserved2[4];
+       __be64  lun __attribute__((packed));
+       u8      reserved3;
+       u8      task_attr;
+       u8      reserved4;
+       u8      add_cdb_len;
+       u8      cdb[16];
+       u8      add_data[0];
+};
+
+enum {
+       SRP_RSP_FLAG_RSPVALID = 1 << 0,
+       SRP_RSP_FLAG_SNSVALID = 1 << 1,
+       SRP_RSP_FLAG_DOOVER   = 1 << 2,
+       SRP_RSP_FLAG_DOUNDER  = 1 << 3,
+       SRP_RSP_FLAG_DIOVER   = 1 << 4,
+       SRP_RSP_FLAG_DIUNDER  = 1 << 5
+};
+
+struct srp_rsp {
+       u8      opcode;
+       u8      sol_not;
+       u8      reserved1[2];
+       __be32  req_lim_delta;
+       u64     tag;
+       u8      reserved2[2];
+       u8      flags;
+       u8      status;
+       __be32  data_out_res_cnt;
+       __be32  data_in_res_cnt;
+       __be32  sense_data_len;
+       __be32  resp_data_len;
+       u8      data[0];
+};
+
+#endif /* SCSI_SRP_H */