tcm_vhost: Initial merge for vhost level target fabric driver
authorNicholas Bellinger <nab@linux-iscsi.org>
Wed, 18 Jul 2012 21:31:32 +0000 (14:31 -0700)
committerNicholas Bellinger <nab@linux-iscsi.org>
Sun, 29 Jul 2012 20:49:10 +0000 (13:49 -0700)
This patch adds the initial code for tcm_vhost, a Vhost level TCM
fabric driver for virtio SCSI initiators into KVM guest.

This code is currently up and running on v3.5-rc2 host+guest
from target-pending/for-next-merge.

Using tcm_vhost requires Zhi's -> Stefan -> nab's qemu vhost-scsi tree here:

  http://git.kernel.org/?p=virt/kvm/nab/qemu-kvm.git;a=shortlog;h=refs/heads/vhost-scsi

--

Changelog v4 -> v5:

  Expose ABI version via VHOST_SCSI_GET_ABI_VERSION + use Rev 0 as
  starting point for v3.6-rc code (Stefan + ALiguori + nab)
  Convert vhost_scsi_handle_vq() to vq_err() (nab + MST)
  Minor style fixes from checkpatch (nab)

Changelog v3 -> v4:

  Rename vhost_vring_target -> vhost_scsi_target (mst + nab)
  Use TRANSPORT_IQN_LEN in vhost_scsi_target->vhost_wwpn[] def (nab)
  Move back to drivers/vhost/, and just use drivers/vhost/Kconfig.tcm (mst)
  Move TCM_VHOST related ioctl defines from include/linux/vhost.h ->
  drivers/vhost/tcm_vhost.h as requested by MST (nab)
  Move Kbuild.tcm include from drivers/staging -> drivers/vhost/, and
  just use 'if STAGING' around 'source drivers/vhost/Kbuild.tcm'

Changelog v2 -> v3:

  Unlock on error in tcm_vhost_drop_nexus() (DanC)
  Fix strlen() doesn't count the terminator (DanC)
  Call kfree() on an error path (DanC)
  Convert tcm_vhost_write_pending to use target_execute_cmd (hch + nab)
  Fix another strlen() off by one in tcm_vhost_make_tport (DanC)
  Add option under drivers/staging/Kconfig, and move to drivers/vhost/tcm/
  as requested by MST (nab)

Changelog v1 -> v2:

  Fix tv_cmd completion -> release SGL memory leak (nab)
  Fix sparse warnings for static variable usage ((Fengguang Wu)
  Fix sparse warnings for min() typing + printk format specs (Fengguang Wu)
  Convert to cmwq submission for I/O dispatch (nab + hch)

Changelog v0 -> v1:

  Merge into single source + header file, and move to drivers/vhost/

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Cc: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Zhi Yong Wu <wuzhy@cn.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
drivers/vhost/Kconfig
drivers/vhost/Kconfig.tcm [new file with mode: 0644]
drivers/vhost/Makefile
drivers/vhost/tcm_vhost.c [new file with mode: 0644]
drivers/vhost/tcm_vhost.h [new file with mode: 0644]

index e4e2fd1b51076dfbc8ffc73920ac148d830c4d4f..202bba6c997cce61eb116fc74db8e842454ce1f4 100644 (file)
@@ -9,3 +9,6 @@ config VHOST_NET
          To compile this driver as a module, choose M here: the module will
          be called vhost_net.
 
+if STAGING
+source "drivers/vhost/Kconfig.tcm"
+endif
diff --git a/drivers/vhost/Kconfig.tcm b/drivers/vhost/Kconfig.tcm
new file mode 100644 (file)
index 0000000..a9c6f76
--- /dev/null
@@ -0,0 +1,6 @@
+config TCM_VHOST
+       tristate "TCM_VHOST fabric module (EXPERIMENTAL)"
+       depends on TARGET_CORE && EVENTFD && EXPERIMENTAL && m
+       default n
+       ---help---
+       Say M here to enable the TCM_VHOST fabric module for use with virtio-scsi guests
index 72dd02050bb986cf6d426828f64ea521aaff3284..a27b053bc9ab12d546c9f17bee106a916eb5e236 100644 (file)
@@ -1,2 +1,4 @@
 obj-$(CONFIG_VHOST_NET) += vhost_net.o
 vhost_net-y := vhost.o net.o
+
+obj-$(CONFIG_TCM_VHOST) += tcm_vhost.o
diff --git a/drivers/vhost/tcm_vhost.c b/drivers/vhost/tcm_vhost.c
new file mode 100644 (file)
index 0000000..fb36654
--- /dev/null
@@ -0,0 +1,1628 @@
+/*******************************************************************************
+ * Vhost kernel TCM fabric driver for virtio SCSI initiators
+ *
+ * (C) Copyright 2010-2012 RisingTide Systems LLC.
+ * (C) Copyright 2010-2012 IBM Corp.
+ *
+ * Licensed to the Linux Foundation under the General Public License (GPL) version 2.
+ *
+ * Authors: Nicholas A. Bellinger <nab@risingtidesystems.com>
+ *          Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ ****************************************************************************/
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <generated/utsrelease.h>
+#include <linux/utsname.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/kthread.h>
+#include <linux/types.h>
+#include <linux/string.h>
+#include <linux/configfs.h>
+#include <linux/ctype.h>
+#include <linux/compat.h>
+#include <linux/eventfd.h>
+#include <linux/vhost.h>
+#include <linux/fs.h>
+#include <linux/miscdevice.h>
+#include <asm/unaligned.h>
+#include <scsi/scsi.h>
+#include <scsi/scsi_tcq.h>
+#include <target/target_core_base.h>
+#include <target/target_core_fabric.h>
+#include <target/target_core_fabric_configfs.h>
+#include <target/target_core_configfs.h>
+#include <target/configfs_macros.h>
+#include <linux/vhost.h>
+#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */
+#include <linux/virtio_scsi.h>
+
+#include "vhost.c"
+#include "vhost.h"
+#include "tcm_vhost.h"
+
+struct vhost_scsi {
+       atomic_t vhost_ref_cnt;
+       struct tcm_vhost_tpg *vs_tpg;
+       struct vhost_dev dev;
+       struct vhost_virtqueue vqs[3];
+
+       struct vhost_work vs_completion_work; /* cmd completion work item */
+       struct list_head vs_completion_list;  /* cmd completion queue */
+       spinlock_t vs_completion_lock;        /* protects s_completion_list */
+};
+
+/* Local pointer to allocated TCM configfs fabric module */
+static struct target_fabric_configfs *tcm_vhost_fabric_configfs;
+
+static struct workqueue_struct *tcm_vhost_workqueue;
+
+/* Global spinlock to protect tcm_vhost TPG list for vhost IOCTL access */
+static DEFINE_MUTEX(tcm_vhost_mutex);
+static LIST_HEAD(tcm_vhost_list);
+
+static int tcm_vhost_check_true(struct se_portal_group *se_tpg)
+{
+       return 1;
+}
+
+static int tcm_vhost_check_false(struct se_portal_group *se_tpg)
+{
+       return 0;
+}
+
+static char *tcm_vhost_get_fabric_name(void)
+{
+       return "vhost";
+}
+
+static u8 tcm_vhost_get_fabric_proto_ident(struct se_portal_group *se_tpg)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_tport *tport = tpg->tport;
+
+       switch (tport->tport_proto_id) {
+       case SCSI_PROTOCOL_SAS:
+               return sas_get_fabric_proto_ident(se_tpg);
+       case SCSI_PROTOCOL_FCP:
+               return fc_get_fabric_proto_ident(se_tpg);
+       case SCSI_PROTOCOL_ISCSI:
+               return iscsi_get_fabric_proto_ident(se_tpg);
+       default:
+               pr_err("Unknown tport_proto_id: 0x%02x, using"
+                       " SAS emulation\n", tport->tport_proto_id);
+               break;
+       }
+
+       return sas_get_fabric_proto_ident(se_tpg);
+}
+
+static char *tcm_vhost_get_fabric_wwn(struct se_portal_group *se_tpg)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_tport *tport = tpg->tport;
+
+       return &tport->tport_name[0];
+}
+
+static u16 tcm_vhost_get_tag(struct se_portal_group *se_tpg)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       return tpg->tport_tpgt;
+}
+
+static u32 tcm_vhost_get_default_depth(struct se_portal_group *se_tpg)
+{
+       return 1;
+}
+
+static u32 tcm_vhost_get_pr_transport_id(
+       struct se_portal_group *se_tpg,
+       struct se_node_acl *se_nacl,
+       struct t10_pr_registration *pr_reg,
+       int *format_code,
+       unsigned char *buf)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_tport *tport = tpg->tport;
+
+       switch (tport->tport_proto_id) {
+       case SCSI_PROTOCOL_SAS:
+               return sas_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
+                                       format_code, buf);
+       case SCSI_PROTOCOL_FCP:
+               return fc_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
+                                       format_code, buf);
+       case SCSI_PROTOCOL_ISCSI:
+               return iscsi_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
+                                       format_code, buf);
+       default:
+               pr_err("Unknown tport_proto_id: 0x%02x, using"
+                       " SAS emulation\n", tport->tport_proto_id);
+               break;
+       }
+
+       return sas_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
+                       format_code, buf);
+}
+
+static u32 tcm_vhost_get_pr_transport_id_len(
+       struct se_portal_group *se_tpg,
+       struct se_node_acl *se_nacl,
+       struct t10_pr_registration *pr_reg,
+       int *format_code)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_tport *tport = tpg->tport;
+
+       switch (tport->tport_proto_id) {
+       case SCSI_PROTOCOL_SAS:
+               return sas_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
+                                       format_code);
+       case SCSI_PROTOCOL_FCP:
+               return fc_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
+                                       format_code);
+       case SCSI_PROTOCOL_ISCSI:
+               return iscsi_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
+                                       format_code);
+       default:
+               pr_err("Unknown tport_proto_id: 0x%02x, using"
+                       " SAS emulation\n", tport->tport_proto_id);
+               break;
+       }
+
+       return sas_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
+                       format_code);
+}
+
+static char *tcm_vhost_parse_pr_out_transport_id(
+       struct se_portal_group *se_tpg,
+       const char *buf,
+       u32 *out_tid_len,
+       char **port_nexus_ptr)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_tport *tport = tpg->tport;
+
+       switch (tport->tport_proto_id) {
+       case SCSI_PROTOCOL_SAS:
+               return sas_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
+                                       port_nexus_ptr);
+       case SCSI_PROTOCOL_FCP:
+               return fc_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
+                                       port_nexus_ptr);
+       case SCSI_PROTOCOL_ISCSI:
+               return iscsi_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
+                                       port_nexus_ptr);
+       default:
+               pr_err("Unknown tport_proto_id: 0x%02x, using"
+                       " SAS emulation\n", tport->tport_proto_id);
+               break;
+       }
+
+       return sas_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
+                       port_nexus_ptr);
+}
+
+static struct se_node_acl *tcm_vhost_alloc_fabric_acl(
+       struct se_portal_group *se_tpg)
+{
+       struct tcm_vhost_nacl *nacl;
+
+       nacl = kzalloc(sizeof(struct tcm_vhost_nacl), GFP_KERNEL);
+       if (!nacl) {
+               pr_err("Unable to alocate struct tcm_vhost_nacl\n");
+               return NULL;
+       }
+
+       return &nacl->se_node_acl;
+}
+
+static void tcm_vhost_release_fabric_acl(
+       struct se_portal_group *se_tpg,
+       struct se_node_acl *se_nacl)
+{
+       struct tcm_vhost_nacl *nacl = container_of(se_nacl,
+                       struct tcm_vhost_nacl, se_node_acl);
+       kfree(nacl);
+}
+
+static u32 tcm_vhost_tpg_get_inst_index(struct se_portal_group *se_tpg)
+{
+       return 1;
+}
+
+static void tcm_vhost_release_cmd(struct se_cmd *se_cmd)
+{
+       return;
+}
+
+static int tcm_vhost_shutdown_session(struct se_session *se_sess)
+{
+       return 0;
+}
+
+static void tcm_vhost_close_session(struct se_session *se_sess)
+{
+       return;
+}
+
+static u32 tcm_vhost_sess_get_index(struct se_session *se_sess)
+{
+       return 0;
+}
+
+static int tcm_vhost_write_pending(struct se_cmd *se_cmd)
+{
+       /* Go ahead and process the write immediately */
+       target_execute_cmd(se_cmd);
+       return 0;
+}
+
+static int tcm_vhost_write_pending_status(struct se_cmd *se_cmd)
+{
+       return 0;
+}
+
+static void tcm_vhost_set_default_node_attrs(struct se_node_acl *nacl)
+{
+       return;
+}
+
+static u32 tcm_vhost_get_task_tag(struct se_cmd *se_cmd)
+{
+       return 0;
+}
+
+static int tcm_vhost_get_cmd_state(struct se_cmd *se_cmd)
+{
+       return 0;
+}
+
+static void vhost_scsi_complete_cmd(struct tcm_vhost_cmd *);
+
+static int tcm_vhost_queue_data_in(struct se_cmd *se_cmd)
+{
+       struct tcm_vhost_cmd *tv_cmd = container_of(se_cmd,
+                               struct tcm_vhost_cmd, tvc_se_cmd);
+       vhost_scsi_complete_cmd(tv_cmd);
+       return 0;
+}
+
+static int tcm_vhost_queue_status(struct se_cmd *se_cmd)
+{
+       struct tcm_vhost_cmd *tv_cmd = container_of(se_cmd,
+                               struct tcm_vhost_cmd, tvc_se_cmd);
+       vhost_scsi_complete_cmd(tv_cmd);
+       return 0;
+}
+
+static int tcm_vhost_queue_tm_rsp(struct se_cmd *se_cmd)
+{
+       return 0;
+}
+
+static u16 tcm_vhost_set_fabric_sense_len(struct se_cmd *se_cmd,
+       u32 sense_length)
+{
+       return 0;
+}
+
+static u16 tcm_vhost_get_fabric_sense_len(void)
+{
+       return 0;
+}
+
+static void vhost_scsi_free_cmd(struct tcm_vhost_cmd *tv_cmd)
+{
+       struct se_cmd *se_cmd = &tv_cmd->tvc_se_cmd;
+
+       /* TODO locking against target/backend threads? */
+       transport_generic_free_cmd(se_cmd, 1);
+
+       if (tv_cmd->tvc_sgl_count) {
+               u32 i;
+               for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
+                       put_page(sg_page(&tv_cmd->tvc_sgl[i]));
+
+               kfree(tv_cmd->tvc_sgl);
+       }
+
+       kfree(tv_cmd);
+}
+
+/* Dequeue a command from the completion list */
+static struct tcm_vhost_cmd *vhost_scsi_get_cmd_from_completion(
+       struct vhost_scsi *vs)
+{
+       struct tcm_vhost_cmd *tv_cmd = NULL;
+
+       spin_lock_bh(&vs->vs_completion_lock);
+       if (list_empty(&vs->vs_completion_list)) {
+               spin_unlock_bh(&vs->vs_completion_lock);
+               return NULL;
+       }
+
+       list_for_each_entry(tv_cmd, &vs->vs_completion_list,
+                           tvc_completion_list) {
+               list_del(&tv_cmd->tvc_completion_list);
+               break;
+       }
+       spin_unlock_bh(&vs->vs_completion_lock);
+       return tv_cmd;
+}
+
+/* Fill in status and signal that we are done processing this command
+ *
+ * This is scheduled in the vhost work queue so we are called with the owner
+ * process mm and can access the vring.
+ */
+static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
+{
+       struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
+                                       vs_completion_work);
+       struct tcm_vhost_cmd *tv_cmd;
+
+       while ((tv_cmd = vhost_scsi_get_cmd_from_completion(vs)) != NULL) {
+               struct virtio_scsi_cmd_resp v_rsp;
+               struct se_cmd *se_cmd = &tv_cmd->tvc_se_cmd;
+               int ret;
+
+               pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
+                       tv_cmd, se_cmd->residual_count, se_cmd->scsi_status);
+
+               memset(&v_rsp, 0, sizeof(v_rsp));
+               v_rsp.resid = se_cmd->residual_count;
+               /* TODO is status_qualifier field needed? */
+               v_rsp.status = se_cmd->scsi_status;
+               v_rsp.sense_len = se_cmd->scsi_sense_length;
+               memcpy(v_rsp.sense, tv_cmd->tvc_sense_buf,
+                      v_rsp.sense_len);
+               ret = copy_to_user(tv_cmd->tvc_resp, &v_rsp, sizeof(v_rsp));
+               if (likely(ret == 0))
+                       vhost_add_used(&vs->vqs[2], tv_cmd->tvc_vq_desc, 0);
+               else
+                       pr_err("Faulted on virtio_scsi_cmd_resp\n");
+
+               vhost_scsi_free_cmd(tv_cmd);
+       }
+
+       vhost_signal(&vs->dev, &vs->vqs[2]);
+}
+
+static void vhost_scsi_complete_cmd(struct tcm_vhost_cmd *tv_cmd)
+{
+       struct vhost_scsi *vs = tv_cmd->tvc_vhost;
+
+       pr_debug("%s tv_cmd %p\n", __func__, tv_cmd);
+
+       spin_lock_bh(&vs->vs_completion_lock);
+       list_add_tail(&tv_cmd->tvc_completion_list, &vs->vs_completion_list);
+       spin_unlock_bh(&vs->vs_completion_lock);
+
+       vhost_work_queue(&vs->dev, &vs->vs_completion_work);
+}
+
+static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(
+       struct tcm_vhost_tpg *tv_tpg,
+       struct virtio_scsi_cmd_req *v_req,
+       u32 exp_data_len,
+       int data_direction)
+{
+       struct tcm_vhost_cmd *tv_cmd;
+       struct tcm_vhost_nexus *tv_nexus;
+       struct se_portal_group *se_tpg = &tv_tpg->se_tpg;
+       struct se_session *se_sess;
+       struct se_cmd *se_cmd;
+       int sam_task_attr;
+
+       tv_nexus = tv_tpg->tpg_nexus;
+       if (!tv_nexus) {
+               pr_err("Unable to locate active struct tcm_vhost_nexus\n");
+               return ERR_PTR(-EIO);
+       }
+       se_sess = tv_nexus->tvn_se_sess;
+
+       tv_cmd = kzalloc(sizeof(struct tcm_vhost_cmd), GFP_ATOMIC);
+       if (!tv_cmd) {
+               pr_err("Unable to allocate struct tcm_vhost_cmd\n");
+               return ERR_PTR(-ENOMEM);
+       }
+       INIT_LIST_HEAD(&tv_cmd->tvc_completion_list);
+       tv_cmd->tvc_tag = v_req->tag;
+
+       se_cmd = &tv_cmd->tvc_se_cmd;
+       /*
+        * Locate the SAM Task Attr from virtio_scsi_cmd_req
+        */
+       sam_task_attr = v_req->task_attr;
+       /*
+        * Initialize struct se_cmd descriptor from TCM infrastructure
+        */
+       transport_init_se_cmd(se_cmd, se_tpg->se_tpg_tfo, se_sess, exp_data_len,
+                               data_direction, sam_task_attr,
+                               &tv_cmd->tvc_sense_buf[0]);
+
+#if 0  /* FIXME: vhost_scsi_allocate_cmd() BIDI operation */
+       if (bidi)
+               se_cmd->se_cmd_flags |= SCF_BIDI;
+#endif
+       return tv_cmd;
+}
+
+/*
+ * Map a user memory range into a scatterlist
+ *
+ * Returns the number of scatterlist entries used or -errno on error.
+ */
+static int vhost_scsi_map_to_sgl(struct scatterlist *sgl,
+       unsigned int sgl_count, void __user *ptr, size_t len, int write)
+{
+       struct scatterlist *sg = sgl;
+       unsigned int npages = 0;
+       int ret;
+
+       while (len > 0) {
+               struct page *page;
+               unsigned int offset = (uintptr_t)ptr & ~PAGE_MASK;
+               unsigned int nbytes = min_t(unsigned int,
+                               PAGE_SIZE - offset, len);
+
+               if (npages == sgl_count) {
+                       ret = -ENOBUFS;
+                       goto err;
+               }
+
+               ret = get_user_pages_fast((unsigned long)ptr, 1, write, &page);
+               BUG_ON(ret == 0); /* we should either get our page or fail */
+               if (ret < 0)
+                       goto err;
+
+               sg_set_page(sg, page, nbytes, offset);
+               ptr += nbytes;
+               len -= nbytes;
+               sg++;
+               npages++;
+       }
+       return npages;
+
+err:
+       /* Put pages that we hold */
+       for (sg = sgl; sg != &sgl[npages]; sg++)
+               put_page(sg_page(sg));
+       return ret;
+}
+
+static int vhost_scsi_map_iov_to_sgl(struct tcm_vhost_cmd *tv_cmd,
+       struct iovec *iov, unsigned int niov, int write)
+{
+       int ret;
+       unsigned int i;
+       u32 sgl_count;
+       struct scatterlist *sg;
+
+       /*
+        * Find out how long sglist needs to be
+        */
+       sgl_count = 0;
+       for (i = 0; i < niov; i++) {
+               sgl_count += (((uintptr_t)iov[i].iov_base + iov[i].iov_len +
+                               PAGE_SIZE - 1) >> PAGE_SHIFT) -
+                               ((uintptr_t)iov[i].iov_base >> PAGE_SHIFT);
+       }
+       /* TODO overflow checking */
+
+       sg = kmalloc(sizeof(tv_cmd->tvc_sgl[0]) * sgl_count, GFP_ATOMIC);
+       if (!sg)
+               return -ENOMEM;
+       pr_debug("%s sg %p sgl_count %u is_err %ld\n", __func__,
+              sg, sgl_count, IS_ERR(sg));
+       sg_init_table(sg, sgl_count);
+
+       tv_cmd->tvc_sgl = sg;
+       tv_cmd->tvc_sgl_count = sgl_count;
+
+       pr_debug("Mapping %u iovecs for %u pages\n", niov, sgl_count);
+       for (i = 0; i < niov; i++) {
+               ret = vhost_scsi_map_to_sgl(sg, sgl_count, iov[i].iov_base,
+                                       iov[i].iov_len, write);
+               if (ret < 0) {
+                       for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
+                               put_page(sg_page(&tv_cmd->tvc_sgl[i]));
+                       kfree(tv_cmd->tvc_sgl);
+                       tv_cmd->tvc_sgl = NULL;
+                       tv_cmd->tvc_sgl_count = 0;
+                       return ret;
+               }
+
+               sg += ret;
+               sgl_count -= ret;
+       }
+       return 0;
+}
+
+static void tcm_vhost_submission_work(struct work_struct *work)
+{
+       struct tcm_vhost_cmd *tv_cmd =
+               container_of(work, struct tcm_vhost_cmd, work);
+       struct se_cmd *se_cmd = &tv_cmd->tvc_se_cmd;
+       struct scatterlist *sg_ptr, *sg_bidi_ptr = NULL;
+       int rc, sg_no_bidi = 0;
+       /*
+        * Locate the struct se_lun pointer based on v_req->lun, and
+        * attach it to struct se_cmd
+        */
+       rc = transport_lookup_cmd_lun(&tv_cmd->tvc_se_cmd, tv_cmd->tvc_lun);
+       if (rc < 0) {
+               pr_err("Failed to look up lun: %d\n", tv_cmd->tvc_lun);
+               transport_send_check_condition_and_sense(&tv_cmd->tvc_se_cmd,
+                       tv_cmd->tvc_se_cmd.scsi_sense_reason, 0);
+               transport_generic_free_cmd(se_cmd, 0);
+               return;
+       }
+
+       rc = target_setup_cmd_from_cdb(se_cmd, tv_cmd->tvc_cdb);
+       if (rc == -ENOMEM) {
+               transport_send_check_condition_and_sense(se_cmd,
+                               TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
+               transport_generic_free_cmd(se_cmd, 0);
+               return;
+       } else if (rc < 0) {
+               if (se_cmd->se_cmd_flags & SCF_SCSI_RESERVATION_CONFLICT)
+                       tcm_vhost_queue_status(se_cmd);
+               else
+                       transport_send_check_condition_and_sense(se_cmd,
+                                       se_cmd->scsi_sense_reason, 0);
+               transport_generic_free_cmd(se_cmd, 0);
+               return;
+       }
+
+       if (tv_cmd->tvc_sgl_count) {
+               sg_ptr = tv_cmd->tvc_sgl;
+               /*
+                * For BIDI commands, pass in the extra READ buffer
+                * to transport_generic_map_mem_to_cmd() below..
+                */
+/* FIXME: Fix BIDI operation in tcm_vhost_submission_work() */
+#if 0
+               if (se_cmd->se_cmd_flags & SCF_BIDI) {
+                       sg_bidi_ptr = NULL;
+                       sg_no_bidi = 0;
+               }
+#endif
+       } else {
+               sg_ptr = NULL;
+       }
+
+       rc = transport_generic_map_mem_to_cmd(se_cmd, sg_ptr,
+                               tv_cmd->tvc_sgl_count, sg_bidi_ptr,
+                               sg_no_bidi);
+       if (rc < 0) {
+               transport_send_check_condition_and_sense(se_cmd,
+                               se_cmd->scsi_sense_reason, 0);
+               transport_generic_free_cmd(se_cmd, 0);
+               return;
+       }
+       transport_handle_cdb_direct(se_cmd);
+}
+
+static void vhost_scsi_handle_vq(struct vhost_scsi *vs)
+{
+       struct vhost_virtqueue *vq = &vs->vqs[2];
+       struct virtio_scsi_cmd_req v_req;
+       struct tcm_vhost_tpg *tv_tpg;
+       struct tcm_vhost_cmd *tv_cmd;
+       u32 exp_data_len, data_first, data_num, data_direction;
+       unsigned out, in, i;
+       int head, ret;
+
+       /* Must use ioctl VHOST_SCSI_SET_ENDPOINT */
+       tv_tpg = vs->vs_tpg;
+       if (unlikely(!tv_tpg)) {
+               pr_err("%s endpoint not set\n", __func__);
+               return;
+       }
+
+       mutex_lock(&vq->mutex);
+       vhost_disable_notify(&vs->dev, vq);
+
+       for (;;) {
+               head = vhost_get_vq_desc(&vs->dev, vq, vq->iov,
+                                       ARRAY_SIZE(vq->iov), &out, &in,
+                                       NULL, NULL);
+               pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
+                                       head, out, in);
+               /* On error, stop handling until the next kick. */
+               if (unlikely(head < 0))
+                       break;
+               /* Nothing new?  Wait for eventfd to tell us they refilled. */
+               if (head == vq->num) {
+                       if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
+                               vhost_disable_notify(&vs->dev, vq);
+                               continue;
+                       }
+                       break;
+               }
+
+/* FIXME: BIDI operation */
+               if (out == 1 && in == 1) {
+                       data_direction = DMA_NONE;
+                       data_first = 0;
+                       data_num = 0;
+               } else if (out == 1 && in > 1) {
+                       data_direction = DMA_FROM_DEVICE;
+                       data_first = out + 1;
+                       data_num = in - 1;
+               } else if (out > 1 && in == 1) {
+                       data_direction = DMA_TO_DEVICE;
+                       data_first = 1;
+                       data_num = out - 1;
+               } else {
+                       vq_err(vq, "Invalid buffer layout out: %u in: %u\n",
+                                       out, in);
+                       break;
+               }
+
+               /*
+                * Check for a sane resp buffer so we can report errors to
+                * the guest.
+                */
+               if (unlikely(vq->iov[out].iov_len !=
+                                       sizeof(struct virtio_scsi_cmd_resp))) {
+                       vq_err(vq, "Expecting virtio_scsi_cmd_resp, got %zu"
+                               " bytes\n", vq->iov[out].iov_len);
+                       break;
+               }
+
+               if (unlikely(vq->iov[0].iov_len != sizeof(v_req))) {
+                       vq_err(vq, "Expecting virtio_scsi_cmd_req, got %zu"
+                               " bytes\n", vq->iov[0].iov_len);
+                       break;
+               }
+               pr_debug("Calling __copy_from_user: vq->iov[0].iov_base: %p,"
+                       " len: %zu\n", vq->iov[0].iov_base, sizeof(v_req));
+               ret = __copy_from_user(&v_req, vq->iov[0].iov_base,
+                               sizeof(v_req));
+               if (unlikely(ret)) {
+                       vq_err(vq, "Faulted on virtio_scsi_cmd_req\n");
+                       break;
+               }
+
+               exp_data_len = 0;
+               for (i = 0; i < data_num; i++)
+                       exp_data_len += vq->iov[data_first + i].iov_len;
+
+               tv_cmd = vhost_scsi_allocate_cmd(tv_tpg, &v_req,
+                                       exp_data_len, data_direction);
+               if (IS_ERR(tv_cmd)) {
+                       vq_err(vq, "vhost_scsi_allocate_cmd failed %ld\n",
+                                       PTR_ERR(tv_cmd));
+                       break;
+               }
+               pr_debug("Allocated tv_cmd: %p exp_data_len: %d, data_direction"
+                       ": %d\n", tv_cmd, exp_data_len, data_direction);
+
+               tv_cmd->tvc_vhost = vs;
+
+               if (unlikely(vq->iov[out].iov_len !=
+                               sizeof(struct virtio_scsi_cmd_resp))) {
+                       vq_err(vq, "Expecting virtio_scsi_cmd_resp, got %zu"
+                               " bytes, out: %d, in: %d\n",
+                               vq->iov[out].iov_len, out, in);
+                       break;
+               }
+
+               tv_cmd->tvc_resp = vq->iov[out].iov_base;
+
+               /*
+                * Copy in the recieved CDB descriptor into tv_cmd->tvc_cdb
+                * that will be used by tcm_vhost_new_cmd_map() and down into
+                * target_setup_cmd_from_cdb()
+                */
+               memcpy(tv_cmd->tvc_cdb, v_req.cdb, TCM_VHOST_MAX_CDB_SIZE);
+               /*
+                * Check that the recieved CDB size does not exceeded our
+                * hardcoded max for tcm_vhost
+                */
+               /* TODO what if cdb was too small for varlen cdb header? */
+               if (unlikely(scsi_command_size(tv_cmd->tvc_cdb) >
+                                       TCM_VHOST_MAX_CDB_SIZE)) {
+                       vq_err(vq, "Received SCSI CDB with command_size: %d that"
+                               " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
+                               scsi_command_size(tv_cmd->tvc_cdb),
+                               TCM_VHOST_MAX_CDB_SIZE);
+                       break; /* TODO */
+               }
+               tv_cmd->tvc_lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
+
+               pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
+                       tv_cmd->tvc_cdb[0], tv_cmd->tvc_lun);
+
+               if (data_direction != DMA_NONE) {
+                       ret = vhost_scsi_map_iov_to_sgl(tv_cmd,
+                                       &vq->iov[data_first], data_num,
+                                       data_direction == DMA_TO_DEVICE);
+                       if (unlikely(ret)) {
+                               vq_err(vq, "Failed to map iov to sgl\n");
+                               break; /* TODO */
+                       }
+               }
+
+               /*
+                * Save the descriptor from vhost_get_vq_desc() to be used to
+                * complete the virtio-scsi request in TCM callback context via
+                * tcm_vhost_queue_data_in() and tcm_vhost_queue_status()
+                */
+               tv_cmd->tvc_vq_desc = head;
+               /*
+                * Dispatch tv_cmd descriptor for cmwq execution in process
+                * context provided by tcm_vhost_workqueue.  This also ensures
+                * tv_cmd is executed on the same kworker CPU as this vhost
+                * thread to gain positive L2 cache locality effects..
+                */
+               INIT_WORK(&tv_cmd->work, tcm_vhost_submission_work);
+               queue_work(tcm_vhost_workqueue, &tv_cmd->work);
+       }
+
+       mutex_unlock(&vq->mutex);
+}
+
+static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
+{
+       pr_err("%s: The handling func for control queue.\n", __func__);
+}
+
+static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
+{
+       pr_err("%s: The handling func for event queue.\n", __func__);
+}
+
+static void vhost_scsi_handle_kick(struct vhost_work *work)
+{
+       struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
+                                               poll.work);
+       struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
+
+       vhost_scsi_handle_vq(vs);
+}
+
+/*
+ * Called from vhost_scsi_ioctl() context to walk the list of available
+ * tcm_vhost_tpg with an active struct tcm_vhost_nexus
+ */
+static int vhost_scsi_set_endpoint(
+       struct vhost_scsi *vs,
+       struct vhost_scsi_target *t)
+{
+       struct tcm_vhost_tport *tv_tport;
+       struct tcm_vhost_tpg *tv_tpg;
+       int index;
+
+       mutex_lock(&vs->dev.mutex);
+       /* Verify that ring has been setup correctly. */
+       for (index = 0; index < vs->dev.nvqs; ++index) {
+               /* Verify that ring has been setup correctly. */
+               if (!vhost_vq_access_ok(&vs->vqs[index])) {
+                       mutex_unlock(&vs->dev.mutex);
+                       return -EFAULT;
+               }
+       }
+
+       if (vs->vs_tpg) {
+               mutex_unlock(&vs->dev.mutex);
+               return -EEXIST;
+       }
+       mutex_unlock(&vs->dev.mutex);
+
+       mutex_lock(&tcm_vhost_mutex);
+       list_for_each_entry(tv_tpg, &tcm_vhost_list, tv_tpg_list) {
+               mutex_lock(&tv_tpg->tv_tpg_mutex);
+               if (!tv_tpg->tpg_nexus) {
+                       mutex_unlock(&tv_tpg->tv_tpg_mutex);
+                       continue;
+               }
+               if (atomic_read(&tv_tpg->tv_tpg_vhost_count)) {
+                       mutex_unlock(&tv_tpg->tv_tpg_mutex);
+                       continue;
+               }
+               tv_tport = tv_tpg->tport;
+
+               if (!strcmp(tv_tport->tport_name, t->vhost_wwpn) &&
+                   (tv_tpg->tport_tpgt == t->vhost_tpgt)) {
+                       atomic_inc(&tv_tpg->tv_tpg_vhost_count);
+                       smp_mb__after_atomic_inc();
+                       mutex_unlock(&tv_tpg->tv_tpg_mutex);
+                       mutex_unlock(&tcm_vhost_mutex);
+
+                       mutex_lock(&vs->dev.mutex);
+                       vs->vs_tpg = tv_tpg;
+                       atomic_inc(&vs->vhost_ref_cnt);
+                       smp_mb__after_atomic_inc();
+                       mutex_unlock(&vs->dev.mutex);
+                       return 0;
+               }
+               mutex_unlock(&tv_tpg->tv_tpg_mutex);
+       }
+       mutex_unlock(&tcm_vhost_mutex);
+       return -EINVAL;
+}
+
+static int vhost_scsi_clear_endpoint(
+       struct vhost_scsi *vs,
+       struct vhost_scsi_target *t)
+{
+       struct tcm_vhost_tport *tv_tport;
+       struct tcm_vhost_tpg *tv_tpg;
+       int index;
+
+       mutex_lock(&vs->dev.mutex);
+       /* Verify that ring has been setup correctly. */
+       for (index = 0; index < vs->dev.nvqs; ++index) {
+               if (!vhost_vq_access_ok(&vs->vqs[index])) {
+                       mutex_unlock(&vs->dev.mutex);
+                       return -EFAULT;
+               }
+       }
+
+       if (!vs->vs_tpg) {
+               mutex_unlock(&vs->dev.mutex);
+               return -ENODEV;
+       }
+       tv_tpg = vs->vs_tpg;
+       tv_tport = tv_tpg->tport;
+
+       if (strcmp(tv_tport->tport_name, t->vhost_wwpn) ||
+           (tv_tpg->tport_tpgt != t->vhost_tpgt)) {
+               mutex_unlock(&vs->dev.mutex);
+               pr_warn("tv_tport->tport_name: %s, tv_tpg->tport_tpgt: %hu"
+                       " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
+                       tv_tport->tport_name, tv_tpg->tport_tpgt,
+                       t->vhost_wwpn, t->vhost_tpgt);
+               return -EINVAL;
+       }
+       atomic_dec(&tv_tpg->tv_tpg_vhost_count);
+       vs->vs_tpg = NULL;
+       mutex_unlock(&vs->dev.mutex);
+
+       return 0;
+}
+
+static int vhost_scsi_open(struct inode *inode, struct file *f)
+{
+       struct vhost_scsi *s;
+       int r;
+
+       s = kzalloc(sizeof(*s), GFP_KERNEL);
+       if (!s)
+               return -ENOMEM;
+
+       vhost_work_init(&s->vs_completion_work, vhost_scsi_complete_cmd_work);
+       INIT_LIST_HEAD(&s->vs_completion_list);
+       spin_lock_init(&s->vs_completion_lock);
+
+       s->vqs[0].handle_kick = vhost_scsi_ctl_handle_kick;
+       s->vqs[1].handle_kick = vhost_scsi_evt_handle_kick;
+       s->vqs[2].handle_kick = vhost_scsi_handle_kick;
+       r = vhost_dev_init(&s->dev, s->vqs, 3);
+       if (r < 0) {
+               kfree(s);
+               return r;
+       }
+
+       f->private_data = s;
+       return 0;
+}
+
+static int vhost_scsi_release(struct inode *inode, struct file *f)
+{
+       struct vhost_scsi *s = f->private_data;
+
+       if (s->vs_tpg && s->vs_tpg->tport) {
+               struct vhost_scsi_target backend;
+
+               memcpy(backend.vhost_wwpn, s->vs_tpg->tport->tport_name,
+                               sizeof(backend.vhost_wwpn));
+               backend.vhost_tpgt = s->vs_tpg->tport_tpgt;
+               vhost_scsi_clear_endpoint(s, &backend);
+       }
+
+       vhost_dev_cleanup(&s->dev, false);
+       kfree(s);
+       return 0;
+}
+
+static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
+{
+       if (features & ~VHOST_FEATURES)
+               return -EOPNOTSUPP;
+
+       mutex_lock(&vs->dev.mutex);
+       if ((features & (1 << VHOST_F_LOG_ALL)) &&
+           !vhost_log_access_ok(&vs->dev)) {
+               mutex_unlock(&vs->dev.mutex);
+               return -EFAULT;
+       }
+       vs->dev.acked_features = features;
+       /* TODO possibly smp_wmb() and flush vqs */
+       mutex_unlock(&vs->dev.mutex);
+       return 0;
+}
+
+static long vhost_scsi_ioctl(struct file *f, unsigned int ioctl,
+                               unsigned long arg)
+{
+       struct vhost_scsi *vs = f->private_data;
+       struct vhost_scsi_target backend;
+       void __user *argp = (void __user *)arg;
+       u64 __user *featurep = argp;
+       u64 features;
+       int r;
+
+       switch (ioctl) {
+       case VHOST_SCSI_SET_ENDPOINT:
+               if (copy_from_user(&backend, argp, sizeof backend))
+                       return -EFAULT;
+
+               return vhost_scsi_set_endpoint(vs, &backend);
+       case VHOST_SCSI_CLEAR_ENDPOINT:
+               if (copy_from_user(&backend, argp, sizeof backend))
+                       return -EFAULT;
+
+               return vhost_scsi_clear_endpoint(vs, &backend);
+       case VHOST_SCSI_GET_ABI_VERSION:
+               if (copy_from_user(&backend, argp, sizeof backend))
+                       return -EFAULT;
+
+               backend.abi_version = VHOST_SCSI_ABI_VERSION;
+
+               if (copy_to_user(argp, &backend, sizeof backend))
+                       return -EFAULT;
+               return 0;
+       case VHOST_GET_FEATURES:
+               features = VHOST_FEATURES;
+               if (copy_to_user(featurep, &features, sizeof features))
+                       return -EFAULT;
+               return 0;
+       case VHOST_SET_FEATURES:
+               if (copy_from_user(&features, featurep, sizeof features))
+                       return -EFAULT;
+               return vhost_scsi_set_features(vs, features);
+       default:
+               mutex_lock(&vs->dev.mutex);
+               r = vhost_dev_ioctl(&vs->dev, ioctl, arg);
+               mutex_unlock(&vs->dev.mutex);
+               return r;
+       }
+}
+
+static const struct file_operations vhost_scsi_fops = {
+       .owner          = THIS_MODULE,
+       .release        = vhost_scsi_release,
+       .unlocked_ioctl = vhost_scsi_ioctl,
+       /* TODO compat ioctl? */
+       .open           = vhost_scsi_open,
+       .llseek         = noop_llseek,
+};
+
+static struct miscdevice vhost_scsi_misc = {
+       MISC_DYNAMIC_MINOR,
+       "vhost-scsi",
+       &vhost_scsi_fops,
+};
+
+static int __init vhost_scsi_register(void)
+{
+       return misc_register(&vhost_scsi_misc);
+}
+
+static int vhost_scsi_deregister(void)
+{
+       return misc_deregister(&vhost_scsi_misc);
+}
+
+static char *tcm_vhost_dump_proto_id(struct tcm_vhost_tport *tport)
+{
+       switch (tport->tport_proto_id) {
+       case SCSI_PROTOCOL_SAS:
+               return "SAS";
+       case SCSI_PROTOCOL_FCP:
+               return "FCP";
+       case SCSI_PROTOCOL_ISCSI:
+               return "iSCSI";
+       default:
+               break;
+       }
+
+       return "Unknown";
+}
+
+static int tcm_vhost_port_link(
+       struct se_portal_group *se_tpg,
+       struct se_lun *lun)
+{
+       struct tcm_vhost_tpg *tv_tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+
+       atomic_inc(&tv_tpg->tv_tpg_port_count);
+       smp_mb__after_atomic_inc();
+
+       return 0;
+}
+
+static void tcm_vhost_port_unlink(
+       struct se_portal_group *se_tpg,
+       struct se_lun *se_lun)
+{
+       struct tcm_vhost_tpg *tv_tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+
+       atomic_dec(&tv_tpg->tv_tpg_port_count);
+       smp_mb__after_atomic_dec();
+}
+
+static struct se_node_acl *tcm_vhost_make_nodeacl(
+       struct se_portal_group *se_tpg,
+       struct config_group *group,
+       const char *name)
+{
+       struct se_node_acl *se_nacl, *se_nacl_new;
+       struct tcm_vhost_nacl *nacl;
+       u64 wwpn = 0;
+       u32 nexus_depth;
+
+       /* tcm_vhost_parse_wwn(name, &wwpn, 1) < 0)
+               return ERR_PTR(-EINVAL); */
+       se_nacl_new = tcm_vhost_alloc_fabric_acl(se_tpg);
+       if (!se_nacl_new)
+               return ERR_PTR(-ENOMEM);
+
+       nexus_depth = 1;
+       /*
+        * se_nacl_new may be released by core_tpg_add_initiator_node_acl()
+        * when converting a NodeACL from demo mode -> explict
+        */
+       se_nacl = core_tpg_add_initiator_node_acl(se_tpg, se_nacl_new,
+                               name, nexus_depth);
+       if (IS_ERR(se_nacl)) {
+               tcm_vhost_release_fabric_acl(se_tpg, se_nacl_new);
+               return se_nacl;
+       }
+       /*
+        * Locate our struct tcm_vhost_nacl and set the FC Nport WWPN
+        */
+       nacl = container_of(se_nacl, struct tcm_vhost_nacl, se_node_acl);
+       nacl->iport_wwpn = wwpn;
+
+       return se_nacl;
+}
+
+static void tcm_vhost_drop_nodeacl(struct se_node_acl *se_acl)
+{
+       struct tcm_vhost_nacl *nacl = container_of(se_acl,
+                               struct tcm_vhost_nacl, se_node_acl);
+       core_tpg_del_initiator_node_acl(se_acl->se_tpg, se_acl, 1);
+       kfree(nacl);
+}
+
+static int tcm_vhost_make_nexus(
+       struct tcm_vhost_tpg *tv_tpg,
+       const char *name)
+{
+       struct se_portal_group *se_tpg;
+       struct tcm_vhost_nexus *tv_nexus;
+
+       mutex_lock(&tv_tpg->tv_tpg_mutex);
+       if (tv_tpg->tpg_nexus) {
+               mutex_unlock(&tv_tpg->tv_tpg_mutex);
+               pr_debug("tv_tpg->tpg_nexus already exists\n");
+               return -EEXIST;
+       }
+       se_tpg = &tv_tpg->se_tpg;
+
+       tv_nexus = kzalloc(sizeof(struct tcm_vhost_nexus), GFP_KERNEL);
+       if (!tv_nexus) {
+               mutex_unlock(&tv_tpg->tv_tpg_mutex);
+               pr_err("Unable to allocate struct tcm_vhost_nexus\n");
+               return -ENOMEM;
+       }
+       /*
+        *  Initialize the struct se_session pointer
+        */
+       tv_nexus->tvn_se_sess = transport_init_session();
+       if (IS_ERR(tv_nexus->tvn_se_sess)) {
+               mutex_unlock(&tv_tpg->tv_tpg_mutex);
+               kfree(tv_nexus);
+               return -ENOMEM;
+       }
+       /*
+        * Since we are running in 'demo mode' this call with generate a
+        * struct se_node_acl for the tcm_vhost struct se_portal_group with
+        * the SCSI Initiator port name of the passed configfs group 'name'.
+        */
+       tv_nexus->tvn_se_sess->se_node_acl = core_tpg_check_initiator_node_acl(
+                               se_tpg, (unsigned char *)name);
+       if (!tv_nexus->tvn_se_sess->se_node_acl) {
+               mutex_unlock(&tv_tpg->tv_tpg_mutex);
+               pr_debug("core_tpg_check_initiator_node_acl() failed"
+                               " for %s\n", name);
+               transport_free_session(tv_nexus->tvn_se_sess);
+               kfree(tv_nexus);
+               return -ENOMEM;
+       }
+       /*
+        * Now register the TCM vHost virtual I_T Nexus as active with the
+        * call to __transport_register_session()
+        */
+       __transport_register_session(se_tpg, tv_nexus->tvn_se_sess->se_node_acl,
+                       tv_nexus->tvn_se_sess, tv_nexus);
+       tv_tpg->tpg_nexus = tv_nexus;
+
+       mutex_unlock(&tv_tpg->tv_tpg_mutex);
+       return 0;
+}
+
+static int tcm_vhost_drop_nexus(
+       struct tcm_vhost_tpg *tpg)
+{
+       struct se_session *se_sess;
+       struct tcm_vhost_nexus *tv_nexus;
+
+       mutex_lock(&tpg->tv_tpg_mutex);
+       tv_nexus = tpg->tpg_nexus;
+       if (!tv_nexus) {
+               mutex_unlock(&tpg->tv_tpg_mutex);
+               return -ENODEV;
+       }
+
+       se_sess = tv_nexus->tvn_se_sess;
+       if (!se_sess) {
+               mutex_unlock(&tpg->tv_tpg_mutex);
+               return -ENODEV;
+       }
+
+       if (atomic_read(&tpg->tv_tpg_port_count)) {
+               mutex_unlock(&tpg->tv_tpg_mutex);
+               pr_err("Unable to remove TCM_vHost I_T Nexus with"
+                       " active TPG port count: %d\n",
+                       atomic_read(&tpg->tv_tpg_port_count));
+               return -EPERM;
+       }
+
+       if (atomic_read(&tpg->tv_tpg_vhost_count)) {
+               mutex_unlock(&tpg->tv_tpg_mutex);
+               pr_err("Unable to remove TCM_vHost I_T Nexus with"
+                       " active TPG vhost count: %d\n",
+                       atomic_read(&tpg->tv_tpg_vhost_count));
+               return -EPERM;
+       }
+
+       pr_debug("TCM_vHost_ConfigFS: Removing I_T Nexus to emulated"
+               " %s Initiator Port: %s\n", tcm_vhost_dump_proto_id(tpg->tport),
+               tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
+       /*
+        * Release the SCSI I_T Nexus to the emulated vHost Target Port
+        */
+       transport_deregister_session(tv_nexus->tvn_se_sess);
+       tpg->tpg_nexus = NULL;
+       mutex_unlock(&tpg->tv_tpg_mutex);
+
+       kfree(tv_nexus);
+       return 0;
+}
+
+static ssize_t tcm_vhost_tpg_show_nexus(
+       struct se_portal_group *se_tpg,
+       char *page)
+{
+       struct tcm_vhost_tpg *tv_tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_nexus *tv_nexus;
+       ssize_t ret;
+
+       mutex_lock(&tv_tpg->tv_tpg_mutex);
+       tv_nexus = tv_tpg->tpg_nexus;
+       if (!tv_nexus) {
+               mutex_unlock(&tv_tpg->tv_tpg_mutex);
+               return -ENODEV;
+       }
+       ret = snprintf(page, PAGE_SIZE, "%s\n",
+                       tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
+       mutex_unlock(&tv_tpg->tv_tpg_mutex);
+
+       return ret;
+}
+
+static ssize_t tcm_vhost_tpg_store_nexus(
+       struct se_portal_group *se_tpg,
+       const char *page,
+       size_t count)
+{
+       struct tcm_vhost_tpg *tv_tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+       struct tcm_vhost_tport *tport_wwn = tv_tpg->tport;
+       unsigned char i_port[TCM_VHOST_NAMELEN], *ptr, *port_ptr;
+       int ret;
+       /*
+        * Shutdown the active I_T nexus if 'NULL' is passed..
+        */
+       if (!strncmp(page, "NULL", 4)) {
+               ret = tcm_vhost_drop_nexus(tv_tpg);
+               return (!ret) ? count : ret;
+       }
+       /*
+        * Otherwise make sure the passed virtual Initiator port WWN matches
+        * the fabric protocol_id set in tcm_vhost_make_tport(), and call
+        * tcm_vhost_make_nexus().
+        */
+       if (strlen(page) >= TCM_VHOST_NAMELEN) {
+               pr_err("Emulated NAA Sas Address: %s, exceeds"
+                               " max: %d\n", page, TCM_VHOST_NAMELEN);
+               return -EINVAL;
+       }
+       snprintf(&i_port[0], TCM_VHOST_NAMELEN, "%s", page);
+
+       ptr = strstr(i_port, "naa.");
+       if (ptr) {
+               if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
+                       pr_err("Passed SAS Initiator Port %s does not"
+                               " match target port protoid: %s\n", i_port,
+                               tcm_vhost_dump_proto_id(tport_wwn));
+                       return -EINVAL;
+               }
+               port_ptr = &i_port[0];
+               goto check_newline;
+       }
+       ptr = strstr(i_port, "fc.");
+       if (ptr) {
+               if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
+                       pr_err("Passed FCP Initiator Port %s does not"
+                               " match target port protoid: %s\n", i_port,
+                               tcm_vhost_dump_proto_id(tport_wwn));
+                       return -EINVAL;
+               }
+               port_ptr = &i_port[3]; /* Skip over "fc." */
+               goto check_newline;
+       }
+       ptr = strstr(i_port, "iqn.");
+       if (ptr) {
+               if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
+                       pr_err("Passed iSCSI Initiator Port %s does not"
+                               " match target port protoid: %s\n", i_port,
+                               tcm_vhost_dump_proto_id(tport_wwn));
+                       return -EINVAL;
+               }
+               port_ptr = &i_port[0];
+               goto check_newline;
+       }
+       pr_err("Unable to locate prefix for emulated Initiator Port:"
+                       " %s\n", i_port);
+       return -EINVAL;
+       /*
+        * Clear any trailing newline for the NAA WWN
+        */
+check_newline:
+       if (i_port[strlen(i_port)-1] == '\n')
+               i_port[strlen(i_port)-1] = '\0';
+
+       ret = tcm_vhost_make_nexus(tv_tpg, port_ptr);
+       if (ret < 0)
+               return ret;
+
+       return count;
+}
+
+TF_TPG_BASE_ATTR(tcm_vhost, nexus, S_IRUGO | S_IWUSR);
+
+static struct configfs_attribute *tcm_vhost_tpg_attrs[] = {
+       &tcm_vhost_tpg_nexus.attr,
+       NULL,
+};
+
+static struct se_portal_group *tcm_vhost_make_tpg(
+       struct se_wwn *wwn,
+       struct config_group *group,
+       const char *name)
+{
+       struct tcm_vhost_tport *tport = container_of(wwn,
+                       struct tcm_vhost_tport, tport_wwn);
+
+       struct tcm_vhost_tpg *tpg;
+       unsigned long tpgt;
+       int ret;
+
+       if (strstr(name, "tpgt_") != name)
+               return ERR_PTR(-EINVAL);
+       if (kstrtoul(name + 5, 10, &tpgt) || tpgt > UINT_MAX)
+               return ERR_PTR(-EINVAL);
+
+       tpg = kzalloc(sizeof(struct tcm_vhost_tpg), GFP_KERNEL);
+       if (!tpg) {
+               pr_err("Unable to allocate struct tcm_vhost_tpg");
+               return ERR_PTR(-ENOMEM);
+       }
+       mutex_init(&tpg->tv_tpg_mutex);
+       INIT_LIST_HEAD(&tpg->tv_tpg_list);
+       tpg->tport = tport;
+       tpg->tport_tpgt = tpgt;
+
+       ret = core_tpg_register(&tcm_vhost_fabric_configfs->tf_ops, wwn,
+                               &tpg->se_tpg, tpg, TRANSPORT_TPG_TYPE_NORMAL);
+       if (ret < 0) {
+               kfree(tpg);
+               return NULL;
+       }
+       mutex_lock(&tcm_vhost_mutex);
+       list_add_tail(&tpg->tv_tpg_list, &tcm_vhost_list);
+       mutex_unlock(&tcm_vhost_mutex);
+
+       return &tpg->se_tpg;
+}
+
+static void tcm_vhost_drop_tpg(struct se_portal_group *se_tpg)
+{
+       struct tcm_vhost_tpg *tpg = container_of(se_tpg,
+                               struct tcm_vhost_tpg, se_tpg);
+
+       mutex_lock(&tcm_vhost_mutex);
+       list_del(&tpg->tv_tpg_list);
+       mutex_unlock(&tcm_vhost_mutex);
+       /*
+        * Release the virtual I_T Nexus for this vHost TPG
+        */
+       tcm_vhost_drop_nexus(tpg);
+       /*
+        * Deregister the se_tpg from TCM..
+        */
+       core_tpg_deregister(se_tpg);
+       kfree(tpg);
+}
+
+static struct se_wwn *tcm_vhost_make_tport(
+       struct target_fabric_configfs *tf,
+       struct config_group *group,
+       const char *name)
+{
+       struct tcm_vhost_tport *tport;
+       char *ptr;
+       u64 wwpn = 0;
+       int off = 0;
+
+       /* if (tcm_vhost_parse_wwn(name, &wwpn, 1) < 0)
+               return ERR_PTR(-EINVAL); */
+
+       tport = kzalloc(sizeof(struct tcm_vhost_tport), GFP_KERNEL);
+       if (!tport) {
+               pr_err("Unable to allocate struct tcm_vhost_tport");
+               return ERR_PTR(-ENOMEM);
+       }
+       tport->tport_wwpn = wwpn;
+       /*
+        * Determine the emulated Protocol Identifier and Target Port Name
+        * based on the incoming configfs directory name.
+        */
+       ptr = strstr(name, "naa.");
+       if (ptr) {
+               tport->tport_proto_id = SCSI_PROTOCOL_SAS;
+               goto check_len;
+       }
+       ptr = strstr(name, "fc.");
+       if (ptr) {
+               tport->tport_proto_id = SCSI_PROTOCOL_FCP;
+               off = 3; /* Skip over "fc." */
+               goto check_len;
+       }
+       ptr = strstr(name, "iqn.");
+       if (ptr) {
+               tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
+               goto check_len;
+       }
+
+       pr_err("Unable to locate prefix for emulated Target Port:"
+                       " %s\n", name);
+       kfree(tport);
+       return ERR_PTR(-EINVAL);
+
+check_len:
+       if (strlen(name) >= TCM_VHOST_NAMELEN) {
+               pr_err("Emulated %s Address: %s, exceeds"
+                       " max: %d\n", name, tcm_vhost_dump_proto_id(tport),
+                       TCM_VHOST_NAMELEN);
+               kfree(tport);
+               return ERR_PTR(-EINVAL);
+       }
+       snprintf(&tport->tport_name[0], TCM_VHOST_NAMELEN, "%s", &name[off]);
+
+       pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
+               " %s Address: %s\n", tcm_vhost_dump_proto_id(tport), name);
+
+       return &tport->tport_wwn;
+}
+
+static void tcm_vhost_drop_tport(struct se_wwn *wwn)
+{
+       struct tcm_vhost_tport *tport = container_of(wwn,
+                               struct tcm_vhost_tport, tport_wwn);
+
+       pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
+               " %s Address: %s\n", tcm_vhost_dump_proto_id(tport),
+               tport->tport_name);
+
+       kfree(tport);
+}
+
+static ssize_t tcm_vhost_wwn_show_attr_version(
+       struct target_fabric_configfs *tf,
+       char *page)
+{
+       return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
+               "on "UTS_RELEASE"\n", TCM_VHOST_VERSION, utsname()->sysname,
+               utsname()->machine);
+}
+
+TF_WWN_ATTR_RO(tcm_vhost, version);
+
+static struct configfs_attribute *tcm_vhost_wwn_attrs[] = {
+       &tcm_vhost_wwn_version.attr,
+       NULL,
+};
+
+static struct target_core_fabric_ops tcm_vhost_ops = {
+       .get_fabric_name                = tcm_vhost_get_fabric_name,
+       .get_fabric_proto_ident         = tcm_vhost_get_fabric_proto_ident,
+       .tpg_get_wwn                    = tcm_vhost_get_fabric_wwn,
+       .tpg_get_tag                    = tcm_vhost_get_tag,
+       .tpg_get_default_depth          = tcm_vhost_get_default_depth,
+       .tpg_get_pr_transport_id        = tcm_vhost_get_pr_transport_id,
+       .tpg_get_pr_transport_id_len    = tcm_vhost_get_pr_transport_id_len,
+       .tpg_parse_pr_out_transport_id  = tcm_vhost_parse_pr_out_transport_id,
+       .tpg_check_demo_mode            = tcm_vhost_check_true,
+       .tpg_check_demo_mode_cache      = tcm_vhost_check_true,
+       .tpg_check_demo_mode_write_protect = tcm_vhost_check_false,
+       .tpg_check_prod_mode_write_protect = tcm_vhost_check_false,
+       .tpg_alloc_fabric_acl           = tcm_vhost_alloc_fabric_acl,
+       .tpg_release_fabric_acl         = tcm_vhost_release_fabric_acl,
+       .tpg_get_inst_index             = tcm_vhost_tpg_get_inst_index,
+       .release_cmd                    = tcm_vhost_release_cmd,
+       .shutdown_session               = tcm_vhost_shutdown_session,
+       .close_session                  = tcm_vhost_close_session,
+       .sess_get_index                 = tcm_vhost_sess_get_index,
+       .sess_get_initiator_sid         = NULL,
+       .write_pending                  = tcm_vhost_write_pending,
+       .write_pending_status           = tcm_vhost_write_pending_status,
+       .set_default_node_attributes    = tcm_vhost_set_default_node_attrs,
+       .get_task_tag                   = tcm_vhost_get_task_tag,
+       .get_cmd_state                  = tcm_vhost_get_cmd_state,
+       .queue_data_in                  = tcm_vhost_queue_data_in,
+       .queue_status                   = tcm_vhost_queue_status,
+       .queue_tm_rsp                   = tcm_vhost_queue_tm_rsp,
+       .get_fabric_sense_len           = tcm_vhost_get_fabric_sense_len,
+       .set_fabric_sense_len           = tcm_vhost_set_fabric_sense_len,
+       /*
+        * Setup callers for generic logic in target_core_fabric_configfs.c
+        */
+       .fabric_make_wwn                = tcm_vhost_make_tport,
+       .fabric_drop_wwn                = tcm_vhost_drop_tport,
+       .fabric_make_tpg                = tcm_vhost_make_tpg,
+       .fabric_drop_tpg                = tcm_vhost_drop_tpg,
+       .fabric_post_link               = tcm_vhost_port_link,
+       .fabric_pre_unlink              = tcm_vhost_port_unlink,
+       .fabric_make_np                 = NULL,
+       .fabric_drop_np                 = NULL,
+       .fabric_make_nodeacl            = tcm_vhost_make_nodeacl,
+       .fabric_drop_nodeacl            = tcm_vhost_drop_nodeacl,
+};
+
+static int tcm_vhost_register_configfs(void)
+{
+       struct target_fabric_configfs *fabric;
+       int ret;
+
+       pr_debug("TCM_VHOST fabric module %s on %s/%s"
+               " on "UTS_RELEASE"\n", TCM_VHOST_VERSION, utsname()->sysname,
+               utsname()->machine);
+       /*
+        * Register the top level struct config_item_type with TCM core
+        */
+       fabric = target_fabric_configfs_init(THIS_MODULE, "vhost");
+       if (IS_ERR(fabric)) {
+               pr_err("target_fabric_configfs_init() failed\n");
+               return PTR_ERR(fabric);
+       }
+       /*
+        * Setup fabric->tf_ops from our local tcm_vhost_ops
+        */
+       fabric->tf_ops = tcm_vhost_ops;
+       /*
+        * Setup default attribute lists for various fabric->tf_cit_tmpl
+        */
+       TF_CIT_TMPL(fabric)->tfc_wwn_cit.ct_attrs = tcm_vhost_wwn_attrs;
+       TF_CIT_TMPL(fabric)->tfc_tpg_base_cit.ct_attrs = tcm_vhost_tpg_attrs;
+       TF_CIT_TMPL(fabric)->tfc_tpg_attrib_cit.ct_attrs = NULL;
+       TF_CIT_TMPL(fabric)->tfc_tpg_param_cit.ct_attrs = NULL;
+       TF_CIT_TMPL(fabric)->tfc_tpg_np_base_cit.ct_attrs = NULL;
+       TF_CIT_TMPL(fabric)->tfc_tpg_nacl_base_cit.ct_attrs = NULL;
+       TF_CIT_TMPL(fabric)->tfc_tpg_nacl_attrib_cit.ct_attrs = NULL;
+       TF_CIT_TMPL(fabric)->tfc_tpg_nacl_auth_cit.ct_attrs = NULL;
+       TF_CIT_TMPL(fabric)->tfc_tpg_nacl_param_cit.ct_attrs = NULL;
+       /*
+        * Register the fabric for use within TCM
+        */
+       ret = target_fabric_configfs_register(fabric);
+       if (ret < 0) {
+               pr_err("target_fabric_configfs_register() failed"
+                               " for TCM_VHOST\n");
+               return ret;
+       }
+       /*
+        * Setup our local pointer to *fabric
+        */
+       tcm_vhost_fabric_configfs = fabric;
+       pr_debug("TCM_VHOST[0] - Set fabric -> tcm_vhost_fabric_configfs\n");
+       return 0;
+};
+
+static void tcm_vhost_deregister_configfs(void)
+{
+       if (!tcm_vhost_fabric_configfs)
+               return;
+
+       target_fabric_configfs_deregister(tcm_vhost_fabric_configfs);
+       tcm_vhost_fabric_configfs = NULL;
+       pr_debug("TCM_VHOST[0] - Cleared tcm_vhost_fabric_configfs\n");
+};
+
+static int __init tcm_vhost_init(void)
+{
+       int ret = -ENOMEM;
+
+       tcm_vhost_workqueue = alloc_workqueue("tcm_vhost", 0, 0);
+       if (!tcm_vhost_workqueue)
+               goto out;
+
+       ret = vhost_scsi_register();
+       if (ret < 0)
+               goto out_destroy_workqueue;
+
+       ret = tcm_vhost_register_configfs();
+       if (ret < 0)
+               goto out_vhost_scsi_deregister;
+
+       return 0;
+
+out_vhost_scsi_deregister:
+       vhost_scsi_deregister();
+out_destroy_workqueue:
+       destroy_workqueue(tcm_vhost_workqueue);
+out:
+       return ret;
+};
+
+static void tcm_vhost_exit(void)
+{
+       tcm_vhost_deregister_configfs();
+       vhost_scsi_deregister();
+       destroy_workqueue(tcm_vhost_workqueue);
+};
+
+MODULE_DESCRIPTION("TCM_VHOST series fabric driver");
+MODULE_LICENSE("GPL");
+module_init(tcm_vhost_init);
+module_exit(tcm_vhost_exit);
diff --git a/drivers/vhost/tcm_vhost.h b/drivers/vhost/tcm_vhost.h
new file mode 100644 (file)
index 0000000..c983ed2
--- /dev/null
@@ -0,0 +1,101 @@
+#define TCM_VHOST_VERSION  "v0.1"
+#define TCM_VHOST_NAMELEN 256
+#define TCM_VHOST_MAX_CDB_SIZE 32
+
+struct tcm_vhost_cmd {
+       /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
+       int tvc_vq_desc;
+       /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
+       u64 tvc_tag;
+       /* The number of scatterlists associated with this cmd */
+       u32 tvc_sgl_count;
+       /* Saved unpacked SCSI LUN for tcm_vhost_submission_work() */
+       u32 tvc_lun;
+       /* Pointer to the SGL formatted memory from virtio-scsi */
+       struct scatterlist *tvc_sgl;
+       /* Pointer to response */
+       struct virtio_scsi_cmd_resp __user *tvc_resp;
+       /* Pointer to vhost_scsi for our device */
+       struct vhost_scsi *tvc_vhost;
+       /* The TCM I/O descriptor that is accessed via container_of() */
+       struct se_cmd tvc_se_cmd;
+       /* work item used for cmwq dispatch to tcm_vhost_submission_work() */
+       struct work_struct work;
+       /* Copy of the incoming SCSI command descriptor block (CDB) */
+       unsigned char tvc_cdb[TCM_VHOST_MAX_CDB_SIZE];
+       /* Sense buffer that will be mapped into outgoing status */
+       unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
+       /* Completed commands list, serviced from vhost worker thread */
+       struct list_head tvc_completion_list;
+};
+
+struct tcm_vhost_nexus {
+       /* Pointer to TCM session for I_T Nexus */
+       struct se_session *tvn_se_sess;
+};
+
+struct tcm_vhost_nacl {
+       /* Binary World Wide unique Port Name for Vhost Initiator port */
+       u64 iport_wwpn;
+       /* ASCII formatted WWPN for Sas Initiator port */
+       char iport_name[TCM_VHOST_NAMELEN];
+       /* Returned by tcm_vhost_make_nodeacl() */
+       struct se_node_acl se_node_acl;
+};
+
+struct tcm_vhost_tpg {
+       /* Vhost port target portal group tag for TCM */
+       u16 tport_tpgt;
+       /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
+       atomic_t tv_tpg_port_count;
+       /* Used for vhost_scsi device reference to tpg_nexus */
+       atomic_t tv_tpg_vhost_count;
+       /* list for tcm_vhost_list */
+       struct list_head tv_tpg_list;
+       /* Used to protect access for tpg_nexus */
+       struct mutex tv_tpg_mutex;
+       /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
+       struct tcm_vhost_nexus *tpg_nexus;
+       /* Pointer back to tcm_vhost_tport */
+       struct tcm_vhost_tport *tport;
+       /* Returned by tcm_vhost_make_tpg() */
+       struct se_portal_group se_tpg;
+};
+
+struct tcm_vhost_tport {
+       /* SCSI protocol the tport is providing */
+       u8 tport_proto_id;
+       /* Binary World Wide unique Port Name for Vhost Target port */
+       u64 tport_wwpn;
+       /* ASCII formatted WWPN for Vhost Target port */
+       char tport_name[TCM_VHOST_NAMELEN];
+       /* Returned by tcm_vhost_make_tport() */
+       struct se_wwn tport_wwn;
+};
+
+/*
+ * As per request from MST, keep TCM_VHOST related ioctl defines out of
+ * linux/vhost.h (user-space) for now..
+ */
+
+#include <linux/vhost.h>
+
+/*
+ * Used by QEMU userspace to ensure a consistent vhost-scsi ABI.
+ *
+ * ABI Rev 0: July 2012 version starting point for v3.6-rc merge candidate +
+ *            RFC-v2 vhost-scsi userspace.  Add GET_ABI_VERSION ioctl usage
+ */
+
+#define VHOST_SCSI_ABI_VERSION 0
+
+struct vhost_scsi_target {
+       int abi_version;
+       unsigned char vhost_wwpn[TRANSPORT_IQN_LEN];
+       unsigned short vhost_tpgt;
+};
+
+/* VHOST_SCSI specific defines */
+#define VHOST_SCSI_SET_ENDPOINT _IOW(VHOST_VIRTIO, 0x40, struct vhost_scsi_target)
+#define VHOST_SCSI_CLEAR_ENDPOINT _IOW(VHOST_VIRTIO, 0x41, struct vhost_scsi_target)
+#define VHOST_SCSI_GET_ABI_VERSION _IOW(VHOST_VIRTIO, 0x42, struct vhost_scsi_target)