nvme-fabrics: Add host support for FC transport
authorJames Smart <jsmart2021@gmail.com>
Fri, 2 Dec 2016 08:28:42 +0000 (00:28 -0800)
committerSagi Grimberg <sagi@grimberg.me>
Tue, 6 Dec 2016 08:17:56 +0000 (10:17 +0200)
Implements the FC-NVME T11 definition of how nvme fabric capsules are
performed on an FC fabric. Utilizes a lower-layer API to FC host adapters
to send/receive FC-4 LS operations and FCP operations that comprise NVME
over FC operation.

The T11 definitions for FC-4 Link Services are implemented which create
NVMeOF connections.  Implements the hooks with blk-mq to then submit admin
and io requests to the different connections.

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Jay Freyensee <james_p_freyensee@linux.intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
MAINTAINERS
drivers/nvme/host/Kconfig
drivers/nvme/host/Makefile
drivers/nvme/host/fc.c [new file with mode: 0644]

index 539502e7089de4ca4ed5f0855bdcd975a46da17f..00505700b30a31889370e480c785fe0c49e7ae28 100644 (file)
@@ -8665,6 +8665,7 @@ L:        linux-nvme@lists.infradead.org
 S:     Supported
 F:     include/linux/nvme-fc.h
 F:     include/linux/nvme-fc-driver.h
+F:     drivers/nvme/host/fc.c
 
 NVMEM FRAMEWORK
 M:     Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
index f7d37a62f874b780c94a0bbbf3d2607845c0c5ab..90745a616df7c984c74317d72cd0e8832a8b0476 100644 (file)
@@ -43,3 +43,20 @@ config NVME_RDMA
          from https://github.com/linux-nvme/nvme-cli.
 
          If unsure, say N.
+
+config NVME_FC
+       tristate "NVM Express over Fabrics FC host driver"
+       depends on BLOCK
+       depends on HAS_DMA
+       select NVME_CORE
+       select NVME_FABRICS
+       select SG_POOL
+       help
+         This provides support for the NVMe over Fabrics protocol using
+         the FC transport.  This allows you to use remote block devices
+         exported using the NVMe protocol set.
+
+         To configure a NVMe over Fabrics controller use the nvme-cli tool
+         from https://github.com/linux-nvme/nvme-cli.
+
+         If unsure, say N.
index 47abcec23514baedf3d48ce01a42e3301297cac3..f1a7d945fbb6624212c2c357a1b7f5a66eaa4b96 100644 (file)
@@ -2,6 +2,7 @@ obj-$(CONFIG_NVME_CORE)                 += nvme-core.o
 obj-$(CONFIG_BLK_DEV_NVME)             += nvme.o
 obj-$(CONFIG_NVME_FABRICS)             += nvme-fabrics.o
 obj-$(CONFIG_NVME_RDMA)                        += nvme-rdma.o
+obj-$(CONFIG_NVME_FC)                  += nvme-fc.o
 
 nvme-core-y                            := core.o
 nvme-core-$(CONFIG_BLK_DEV_NVME_SCSI)  += scsi.o
@@ -12,3 +13,5 @@ nvme-y                                        += pci.o
 nvme-fabrics-y                         += fabrics.o
 
 nvme-rdma-y                            += rdma.o
+
+nvme-fc-y                              += fc.o
diff --git a/drivers/nvme/host/fc.c b/drivers/nvme/host/fc.c
new file mode 100644 (file)
index 0000000..771e2e7
--- /dev/null
@@ -0,0 +1,2586 @@
+/*
+ * Copyright (c) 2016 Avago Technologies.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful.
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND WARRANTIES,
+ * INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS FOR A
+ * PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE DISCLAIMED, EXCEPT TO
+ * THE EXTENT THAT SUCH DISCLAIMERS ARE HELD TO BE LEGALLY INVALID.
+ * See the GNU General Public License for more details, a copy of which
+ * can be found in the file COPYING included with this package
+ *
+ */
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+#include <linux/module.h>
+#include <linux/parser.h>
+#include <uapi/scsi/fc/fc_fs.h>
+#include <uapi/scsi/fc/fc_els.h>
+
+#include "nvme.h"
+#include "fabrics.h"
+#include <linux/nvme-fc-driver.h>
+#include <linux/nvme-fc.h>
+
+
+/* *************************** Data Structures/Defines ****************** */
+
+
+/*
+ * We handle AEN commands ourselves and don't even let the
+ * block layer know about them.
+ */
+#define NVME_FC_NR_AEN_COMMANDS        1
+#define NVME_FC_AQ_BLKMQ_DEPTH \
+       (NVMF_AQ_DEPTH - NVME_FC_NR_AEN_COMMANDS)
+#define AEN_CMDID_BASE         (NVME_FC_AQ_BLKMQ_DEPTH + 1)
+
+enum nvme_fc_queue_flags {
+       NVME_FC_Q_CONNECTED = (1 << 0),
+};
+
+#define NVMEFC_QUEUE_DELAY     3               /* ms units */
+
+struct nvme_fc_queue {
+       struct nvme_fc_ctrl     *ctrl;
+       struct device           *dev;
+       struct blk_mq_hw_ctx    *hctx;
+       void                    *lldd_handle;
+       int                     queue_size;
+       size_t                  cmnd_capsule_len;
+       u32                     qnum;
+       u32                     rqcnt;
+       u32                     seqno;
+
+       u64                     connection_id;
+       atomic_t                csn;
+
+       unsigned long           flags;
+} __aligned(sizeof(u64));      /* alignment for other things alloc'd with */
+
+struct nvmefc_ls_req_op {
+       struct nvmefc_ls_req    ls_req;
+
+       struct nvme_fc_ctrl     *ctrl;
+       struct nvme_fc_queue    *queue;
+       struct request          *rq;
+
+       int                     ls_error;
+       struct completion       ls_done;
+       struct list_head        lsreq_list;     /* ctrl->ls_req_list */
+       bool                    req_queued;
+};
+
+enum nvme_fcpop_state {
+       FCPOP_STATE_UNINIT      = 0,
+       FCPOP_STATE_IDLE        = 1,
+       FCPOP_STATE_ACTIVE      = 2,
+       FCPOP_STATE_ABORTED     = 3,
+};
+
+struct nvme_fc_fcp_op {
+       struct nvme_request     nreq;           /*
+                                                * nvme/host/core.c
+                                                * requires this to be
+                                                * the 1st element in the
+                                                * private structure
+                                                * associated with the
+                                                * request.
+                                                */
+       struct nvmefc_fcp_req   fcp_req;
+
+       struct nvme_fc_ctrl     *ctrl;
+       struct nvme_fc_queue    *queue;
+       struct request          *rq;
+
+       atomic_t                state;
+       u32                     rqno;
+       u32                     nents;
+
+       struct nvme_fc_cmd_iu   cmd_iu;
+       struct nvme_fc_ersp_iu  rsp_iu;
+};
+
+struct nvme_fc_lport {
+       struct nvme_fc_local_port       localport;
+
+       struct ida                      endp_cnt;
+       struct list_head                port_list;      /* nvme_fc_port_list */
+       struct list_head                endp_list;
+       struct device                   *dev;   /* physical device for dma */
+       struct nvme_fc_port_template    *ops;
+       struct kref                     ref;
+} __aligned(sizeof(u64));      /* alignment for other things alloc'd with */
+
+struct nvme_fc_rport {
+       struct nvme_fc_remote_port      remoteport;
+
+       struct list_head                endp_list; /* for lport->endp_list */
+       struct list_head                ctrl_list;
+       spinlock_t                      lock;
+       struct kref                     ref;
+} __aligned(sizeof(u64));      /* alignment for other things alloc'd with */
+
+enum nvme_fcctrl_state {
+       FCCTRL_INIT             = 0,
+       FCCTRL_ACTIVE           = 1,
+};
+
+struct nvme_fc_ctrl {
+       spinlock_t              lock;
+       struct nvme_fc_queue    *queues;
+       u32                     queue_count;
+
+       struct device           *dev;
+       struct nvme_fc_lport    *lport;
+       struct nvme_fc_rport    *rport;
+       u32                     cnum;
+
+       u64                     association_id;
+
+       u64                     cap;
+
+       struct list_head        ctrl_list;      /* rport->ctrl_list */
+       struct list_head        ls_req_list;
+
+       struct blk_mq_tag_set   admin_tag_set;
+       struct blk_mq_tag_set   tag_set;
+
+       struct work_struct      delete_work;
+       struct kref             ref;
+       int                     state;
+
+       struct nvme_fc_fcp_op   aen_ops[NVME_FC_NR_AEN_COMMANDS];
+
+       struct nvme_ctrl        ctrl;
+};
+
+static inline struct nvme_fc_ctrl *
+to_fc_ctrl(struct nvme_ctrl *ctrl)
+{
+       return container_of(ctrl, struct nvme_fc_ctrl, ctrl);
+}
+
+static inline struct nvme_fc_lport *
+localport_to_lport(struct nvme_fc_local_port *portptr)
+{
+       return container_of(portptr, struct nvme_fc_lport, localport);
+}
+
+static inline struct nvme_fc_rport *
+remoteport_to_rport(struct nvme_fc_remote_port *portptr)
+{
+       return container_of(portptr, struct nvme_fc_rport, remoteport);
+}
+
+static inline struct nvmefc_ls_req_op *
+ls_req_to_lsop(struct nvmefc_ls_req *lsreq)
+{
+       return container_of(lsreq, struct nvmefc_ls_req_op, ls_req);
+}
+
+static inline struct nvme_fc_fcp_op *
+fcp_req_to_fcp_op(struct nvmefc_fcp_req *fcpreq)
+{
+       return container_of(fcpreq, struct nvme_fc_fcp_op, fcp_req);
+}
+
+
+
+/* *************************** Globals **************************** */
+
+
+static DEFINE_SPINLOCK(nvme_fc_lock);
+
+static LIST_HEAD(nvme_fc_lport_list);
+static DEFINE_IDA(nvme_fc_local_port_cnt);
+static DEFINE_IDA(nvme_fc_ctrl_cnt);
+
+static struct workqueue_struct *nvme_fc_wq;
+
+
+
+/* *********************** FC-NVME Port Management ************************ */
+
+static int __nvme_fc_del_ctrl(struct nvme_fc_ctrl *);
+static void __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *,
+                       struct nvme_fc_queue *, unsigned int);
+
+
+/**
+ * nvme_fc_register_localport - transport entry point called by an
+ *                              LLDD to register the existence of a NVME
+ *                              host FC port.
+ * @pinfo:     pointer to information about the port to be registered
+ * @template:  LLDD entrypoints and operational parameters for the port
+ * @dev:       physical hardware device node port corresponds to. Will be
+ *             used for DMA mappings
+ * @lport_p:   pointer to a local port pointer. Upon success, the routine
+ *             will allocate a nvme_fc_local_port structure and place its
+ *             address in the local port pointer. Upon failure, local port
+ *             pointer will be set to 0.
+ *
+ * Returns:
+ * a completion status. Must be 0 upon success; a negative errno
+ * (ex: -ENXIO) upon failure.
+ */
+int
+nvme_fc_register_localport(struct nvme_fc_port_info *pinfo,
+                       struct nvme_fc_port_template *template,
+                       struct device *dev,
+                       struct nvme_fc_local_port **portptr)
+{
+       struct nvme_fc_lport *newrec;
+       unsigned long flags;
+       int ret, idx;
+
+       if (!template->localport_delete || !template->remoteport_delete ||
+           !template->ls_req || !template->fcp_io ||
+           !template->ls_abort || !template->fcp_abort ||
+           !template->max_hw_queues || !template->max_sgl_segments ||
+           !template->max_dif_sgl_segments || !template->dma_boundary) {
+               ret = -EINVAL;
+               goto out_reghost_failed;
+       }
+
+       newrec = kmalloc((sizeof(*newrec) + template->local_priv_sz),
+                        GFP_KERNEL);
+       if (!newrec) {
+               ret = -ENOMEM;
+               goto out_reghost_failed;
+       }
+
+       idx = ida_simple_get(&nvme_fc_local_port_cnt, 0, 0, GFP_KERNEL);
+       if (idx < 0) {
+               ret = -ENOSPC;
+               goto out_fail_kfree;
+       }
+
+       if (!get_device(dev) && dev) {
+               ret = -ENODEV;
+               goto out_ida_put;
+       }
+
+       INIT_LIST_HEAD(&newrec->port_list);
+       INIT_LIST_HEAD(&newrec->endp_list);
+       kref_init(&newrec->ref);
+       newrec->ops = template;
+       newrec->dev = dev;
+       ida_init(&newrec->endp_cnt);
+       newrec->localport.private = &newrec[1];
+       newrec->localport.node_name = pinfo->node_name;
+       newrec->localport.port_name = pinfo->port_name;
+       newrec->localport.port_role = pinfo->port_role;
+       newrec->localport.port_id = pinfo->port_id;
+       newrec->localport.port_state = FC_OBJSTATE_ONLINE;
+       newrec->localport.port_num = idx;
+
+       spin_lock_irqsave(&nvme_fc_lock, flags);
+       list_add_tail(&newrec->port_list, &nvme_fc_lport_list);
+       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+       if (dev)
+               dma_set_seg_boundary(dev, template->dma_boundary);
+
+       *portptr = &newrec->localport;
+       return 0;
+
+out_ida_put:
+       ida_simple_remove(&nvme_fc_local_port_cnt, idx);
+out_fail_kfree:
+       kfree(newrec);
+out_reghost_failed:
+       *portptr = NULL;
+
+       return ret;
+}
+EXPORT_SYMBOL_GPL(nvme_fc_register_localport);
+
+static void
+nvme_fc_free_lport(struct kref *ref)
+{
+       struct nvme_fc_lport *lport =
+               container_of(ref, struct nvme_fc_lport, ref);
+       unsigned long flags;
+
+       WARN_ON(lport->localport.port_state != FC_OBJSTATE_DELETED);
+       WARN_ON(!list_empty(&lport->endp_list));
+
+       /* remove from transport list */
+       spin_lock_irqsave(&nvme_fc_lock, flags);
+       list_del(&lport->port_list);
+       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+       /* let the LLDD know we've finished tearing it down */
+       lport->ops->localport_delete(&lport->localport);
+
+       ida_simple_remove(&nvme_fc_local_port_cnt, lport->localport.port_num);
+       ida_destroy(&lport->endp_cnt);
+
+       put_device(lport->dev);
+
+       kfree(lport);
+}
+
+static void
+nvme_fc_lport_put(struct nvme_fc_lport *lport)
+{
+       kref_put(&lport->ref, nvme_fc_free_lport);
+}
+
+static int
+nvme_fc_lport_get(struct nvme_fc_lport *lport)
+{
+       return kref_get_unless_zero(&lport->ref);
+}
+
+/**
+ * nvme_fc_unregister_localport - transport entry point called by an
+ *                              LLDD to deregister/remove a previously
+ *                              registered a NVME host FC port.
+ * @localport: pointer to the (registered) local port that is to be
+ *             deregistered.
+ *
+ * Returns:
+ * a completion status. Must be 0 upon success; a negative errno
+ * (ex: -ENXIO) upon failure.
+ */
+int
+nvme_fc_unregister_localport(struct nvme_fc_local_port *portptr)
+{
+       struct nvme_fc_lport *lport = localport_to_lport(portptr);
+       unsigned long flags;
+
+       if (!portptr)
+               return -EINVAL;
+
+       spin_lock_irqsave(&nvme_fc_lock, flags);
+
+       if (portptr->port_state != FC_OBJSTATE_ONLINE) {
+               spin_unlock_irqrestore(&nvme_fc_lock, flags);
+               return -EINVAL;
+       }
+       portptr->port_state = FC_OBJSTATE_DELETED;
+
+       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+       nvme_fc_lport_put(lport);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(nvme_fc_unregister_localport);
+
+/**
+ * nvme_fc_register_remoteport - transport entry point called by an
+ *                              LLDD to register the existence of a NVME
+ *                              subsystem FC port on its fabric.
+ * @localport: pointer to the (registered) local port that the remote
+ *             subsystem port is connected to.
+ * @pinfo:     pointer to information about the port to be registered
+ * @rport_p:   pointer to a remote port pointer. Upon success, the routine
+ *             will allocate a nvme_fc_remote_port structure and place its
+ *             address in the remote port pointer. Upon failure, remote port
+ *             pointer will be set to 0.
+ *
+ * Returns:
+ * a completion status. Must be 0 upon success; a negative errno
+ * (ex: -ENXIO) upon failure.
+ */
+int
+nvme_fc_register_remoteport(struct nvme_fc_local_port *localport,
+                               struct nvme_fc_port_info *pinfo,
+                               struct nvme_fc_remote_port **portptr)
+{
+       struct nvme_fc_lport *lport = localport_to_lport(localport);
+       struct nvme_fc_rport *newrec;
+       unsigned long flags;
+       int ret, idx;
+
+       newrec = kmalloc((sizeof(*newrec) + lport->ops->remote_priv_sz),
+                        GFP_KERNEL);
+       if (!newrec) {
+               ret = -ENOMEM;
+               goto out_reghost_failed;
+       }
+
+       if (!nvme_fc_lport_get(lport)) {
+               ret = -ESHUTDOWN;
+               goto out_kfree_rport;
+       }
+
+       idx = ida_simple_get(&lport->endp_cnt, 0, 0, GFP_KERNEL);
+       if (idx < 0) {
+               ret = -ENOSPC;
+               goto out_lport_put;
+       }
+
+       INIT_LIST_HEAD(&newrec->endp_list);
+       INIT_LIST_HEAD(&newrec->ctrl_list);
+       kref_init(&newrec->ref);
+       spin_lock_init(&newrec->lock);
+       newrec->remoteport.localport = &lport->localport;
+       newrec->remoteport.private = &newrec[1];
+       newrec->remoteport.port_role = pinfo->port_role;
+       newrec->remoteport.node_name = pinfo->node_name;
+       newrec->remoteport.port_name = pinfo->port_name;
+       newrec->remoteport.port_id = pinfo->port_id;
+       newrec->remoteport.port_state = FC_OBJSTATE_ONLINE;
+       newrec->remoteport.port_num = idx;
+
+       spin_lock_irqsave(&nvme_fc_lock, flags);
+       list_add_tail(&newrec->endp_list, &lport->endp_list);
+       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+       *portptr = &newrec->remoteport;
+       return 0;
+
+out_lport_put:
+       nvme_fc_lport_put(lport);
+out_kfree_rport:
+       kfree(newrec);
+out_reghost_failed:
+       *portptr = NULL;
+       return ret;
+
+}
+EXPORT_SYMBOL_GPL(nvme_fc_register_remoteport);
+
+static void
+nvme_fc_free_rport(struct kref *ref)
+{
+       struct nvme_fc_rport *rport =
+               container_of(ref, struct nvme_fc_rport, ref);
+       struct nvme_fc_lport *lport =
+                       localport_to_lport(rport->remoteport.localport);
+       unsigned long flags;
+
+       WARN_ON(rport->remoteport.port_state != FC_OBJSTATE_DELETED);
+       WARN_ON(!list_empty(&rport->ctrl_list));
+
+       /* remove from lport list */
+       spin_lock_irqsave(&nvme_fc_lock, flags);
+       list_del(&rport->endp_list);
+       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+       /* let the LLDD know we've finished tearing it down */
+       lport->ops->remoteport_delete(&rport->remoteport);
+
+       ida_simple_remove(&lport->endp_cnt, rport->remoteport.port_num);
+
+       kfree(rport);
+
+       nvme_fc_lport_put(lport);
+}
+
+static void
+nvme_fc_rport_put(struct nvme_fc_rport *rport)
+{
+       kref_put(&rport->ref, nvme_fc_free_rport);
+}
+
+static int
+nvme_fc_rport_get(struct nvme_fc_rport *rport)
+{
+       return kref_get_unless_zero(&rport->ref);
+}
+
+/**
+ * nvme_fc_unregister_remoteport - transport entry point called by an
+ *                              LLDD to deregister/remove a previously
+ *                              registered a NVME subsystem FC port.
+ * @remoteport: pointer to the (registered) remote port that is to be
+ *              deregistered.
+ *
+ * Returns:
+ * a completion status. Must be 0 upon success; a negative errno
+ * (ex: -ENXIO) upon failure.
+ */
+int
+nvme_fc_unregister_remoteport(struct nvme_fc_remote_port *portptr)
+{
+       struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
+       struct nvme_fc_ctrl *ctrl;
+       unsigned long flags;
+
+       if (!portptr)
+               return -EINVAL;
+
+       spin_lock_irqsave(&rport->lock, flags);
+
+       if (portptr->port_state != FC_OBJSTATE_ONLINE) {
+               spin_unlock_irqrestore(&rport->lock, flags);
+               return -EINVAL;
+       }
+       portptr->port_state = FC_OBJSTATE_DELETED;
+
+       /* tear down all associations to the remote port */
+       list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list)
+               __nvme_fc_del_ctrl(ctrl);
+
+       spin_unlock_irqrestore(&rport->lock, flags);
+
+       nvme_fc_rport_put(rport);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(nvme_fc_unregister_remoteport);
+
+
+/* *********************** FC-NVME DMA Handling **************************** */
+
+/*
+ * The fcloop device passes in a NULL device pointer. Real LLD's will
+ * pass in a valid device pointer. If NULL is passed to the dma mapping
+ * routines, depending on the platform, it may or may not succeed, and
+ * may crash.
+ *
+ * As such:
+ * Wrapper all the dma routines and check the dev pointer.
+ *
+ * If simple mappings (return just a dma address, we'll noop them,
+ * returning a dma address of 0.
+ *
+ * On more complex mappings (dma_map_sg), a pseudo routine fills
+ * in the scatter list, setting all dma addresses to 0.
+ */
+
+static inline dma_addr_t
+fc_dma_map_single(struct device *dev, void *ptr, size_t size,
+               enum dma_data_direction dir)
+{
+       return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
+}
+
+static inline int
+fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
+{
+       return dev ? dma_mapping_error(dev, dma_addr) : 0;
+}
+
+static inline void
+fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
+       enum dma_data_direction dir)
+{
+       if (dev)
+               dma_unmap_single(dev, addr, size, dir);
+}
+
+static inline void
+fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
+               enum dma_data_direction dir)
+{
+       if (dev)
+               dma_sync_single_for_cpu(dev, addr, size, dir);
+}
+
+static inline void
+fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
+               enum dma_data_direction dir)
+{
+       if (dev)
+               dma_sync_single_for_device(dev, addr, size, dir);
+}
+
+/* pseudo dma_map_sg call */
+static int
+fc_map_sg(struct scatterlist *sg, int nents)
+{
+       struct scatterlist *s;
+       int i;
+
+       WARN_ON(nents == 0 || sg[0].length == 0);
+
+       for_each_sg(sg, s, nents, i) {
+               s->dma_address = 0L;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+               s->dma_length = s->length;
+#endif
+       }
+       return nents;
+}
+
+static inline int
+fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
+               enum dma_data_direction dir)
+{
+       return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
+}
+
+static inline void
+fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
+               enum dma_data_direction dir)
+{
+       if (dev)
+               dma_unmap_sg(dev, sg, nents, dir);
+}
+
+
+/* *********************** FC-NVME LS Handling **************************** */
+
+static void nvme_fc_ctrl_put(struct nvme_fc_ctrl *);
+static int nvme_fc_ctrl_get(struct nvme_fc_ctrl *);
+
+
+static void
+__nvme_fc_finish_ls_req(struct nvme_fc_ctrl *ctrl,
+               struct nvmefc_ls_req_op *lsop)
+{
+       struct nvmefc_ls_req *lsreq = &lsop->ls_req;
+       unsigned long flags;
+
+       spin_lock_irqsave(&ctrl->lock, flags);
+
+       if (!lsop->req_queued) {
+               spin_unlock_irqrestore(&ctrl->lock, flags);
+               return;
+       }
+
+       list_del(&lsop->lsreq_list);
+
+       lsop->req_queued = false;
+
+       spin_unlock_irqrestore(&ctrl->lock, flags);
+
+       fc_dma_unmap_single(ctrl->dev, lsreq->rqstdma,
+                                 (lsreq->rqstlen + lsreq->rsplen),
+                                 DMA_BIDIRECTIONAL);
+
+       nvme_fc_ctrl_put(ctrl);
+}
+
+static int
+__nvme_fc_send_ls_req(struct nvme_fc_ctrl *ctrl,
+               struct nvmefc_ls_req_op *lsop,
+               void (*done)(struct nvmefc_ls_req *req, int status))
+{
+       struct nvmefc_ls_req *lsreq = &lsop->ls_req;
+       unsigned long flags;
+       int ret;
+
+       if (!nvme_fc_ctrl_get(ctrl))
+               return -ESHUTDOWN;
+
+       lsreq->done = done;
+       lsop->ctrl = ctrl;
+       lsop->req_queued = false;
+       INIT_LIST_HEAD(&lsop->lsreq_list);
+       init_completion(&lsop->ls_done);
+
+       lsreq->rqstdma = fc_dma_map_single(ctrl->dev, lsreq->rqstaddr,
+                                 lsreq->rqstlen + lsreq->rsplen,
+                                 DMA_BIDIRECTIONAL);
+       if (fc_dma_mapping_error(ctrl->dev, lsreq->rqstdma)) {
+               nvme_fc_ctrl_put(ctrl);
+               dev_err(ctrl->dev,
+                       "els request command failed EFAULT.\n");
+               return -EFAULT;
+       }
+       lsreq->rspdma = lsreq->rqstdma + lsreq->rqstlen;
+
+       spin_lock_irqsave(&ctrl->lock, flags);
+
+       list_add_tail(&lsop->lsreq_list, &ctrl->ls_req_list);
+
+       lsop->req_queued = true;
+
+       spin_unlock_irqrestore(&ctrl->lock, flags);
+
+       ret = ctrl->lport->ops->ls_req(&ctrl->lport->localport,
+                                       &ctrl->rport->remoteport, lsreq);
+       if (ret)
+               lsop->ls_error = ret;
+
+       return ret;
+}
+
+static void
+nvme_fc_send_ls_req_done(struct nvmefc_ls_req *lsreq, int status)
+{
+       struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
+
+       lsop->ls_error = status;
+       complete(&lsop->ls_done);
+}
+
+static int
+nvme_fc_send_ls_req(struct nvme_fc_ctrl *ctrl, struct nvmefc_ls_req_op *lsop)
+{
+       struct nvmefc_ls_req *lsreq = &lsop->ls_req;
+       struct fcnvme_ls_rjt *rjt = lsreq->rspaddr;
+       int ret;
+
+       ret = __nvme_fc_send_ls_req(ctrl, lsop, nvme_fc_send_ls_req_done);
+
+       if (!ret)
+               /*
+                * No timeout/not interruptible as we need the struct
+                * to exist until the lldd calls us back. Thus mandate
+                * wait until driver calls back. lldd responsible for
+                * the timeout action
+                */
+               wait_for_completion(&lsop->ls_done);
+
+       __nvme_fc_finish_ls_req(ctrl, lsop);
+
+       if (ret) {
+               dev_err(ctrl->dev,
+                       "ls request command failed (%d).\n", ret);
+               return ret;
+       }
+
+       /* ACC or RJT payload ? */
+       if (rjt->w0.ls_cmd == FCNVME_LS_RJT)
+               return -ENXIO;
+
+       return 0;
+}
+
+static void
+nvme_fc_send_ls_req_async(struct nvme_fc_ctrl *ctrl,
+               struct nvmefc_ls_req_op *lsop,
+               void (*done)(struct nvmefc_ls_req *req, int status))
+{
+       int ret;
+
+       ret = __nvme_fc_send_ls_req(ctrl, lsop, done);
+
+       /* don't wait for completion */
+
+       if (ret)
+               done(&lsop->ls_req, ret);
+}
+
+/* Validation Error indexes into the string table below */
+enum {
+       VERR_NO_ERROR           = 0,
+       VERR_LSACC              = 1,
+       VERR_LSDESC_RQST        = 2,
+       VERR_LSDESC_RQST_LEN    = 3,
+       VERR_ASSOC_ID           = 4,
+       VERR_ASSOC_ID_LEN       = 5,
+       VERR_CONN_ID            = 6,
+       VERR_CONN_ID_LEN        = 7,
+       VERR_CR_ASSOC           = 8,
+       VERR_CR_ASSOC_ACC_LEN   = 9,
+       VERR_CR_CONN            = 10,
+       VERR_CR_CONN_ACC_LEN    = 11,
+       VERR_DISCONN            = 12,
+       VERR_DISCONN_ACC_LEN    = 13,
+};
+
+static char *validation_errors[] = {
+       "OK",
+       "Not LS_ACC",
+       "Not LSDESC_RQST",
+       "Bad LSDESC_RQST Length",
+       "Not Association ID",
+       "Bad Association ID Length",
+       "Not Connection ID",
+       "Bad Connection ID Length",
+       "Not CR_ASSOC Rqst",
+       "Bad CR_ASSOC ACC Length",
+       "Not CR_CONN Rqst",
+       "Bad CR_CONN ACC Length",
+       "Not Disconnect Rqst",
+       "Bad Disconnect ACC Length",
+};
+
+static int
+nvme_fc_connect_admin_queue(struct nvme_fc_ctrl *ctrl,
+       struct nvme_fc_queue *queue, u16 qsize, u16 ersp_ratio)
+{
+       struct nvmefc_ls_req_op *lsop;
+       struct nvmefc_ls_req *lsreq;
+       struct fcnvme_ls_cr_assoc_rqst *assoc_rqst;
+       struct fcnvme_ls_cr_assoc_acc *assoc_acc;
+       int ret, fcret = 0;
+
+       lsop = kzalloc((sizeof(*lsop) +
+                        ctrl->lport->ops->lsrqst_priv_sz +
+                        sizeof(*assoc_rqst) + sizeof(*assoc_acc)), GFP_KERNEL);
+       if (!lsop) {
+               ret = -ENOMEM;
+               goto out_no_memory;
+       }
+       lsreq = &lsop->ls_req;
+
+       lsreq->private = (void *)&lsop[1];
+       assoc_rqst = (struct fcnvme_ls_cr_assoc_rqst *)
+                       (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
+       assoc_acc = (struct fcnvme_ls_cr_assoc_acc *)&assoc_rqst[1];
+
+       assoc_rqst->w0.ls_cmd = FCNVME_LS_CREATE_ASSOCIATION;
+       assoc_rqst->desc_list_len =
+                       cpu_to_be32(sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
+
+       assoc_rqst->assoc_cmd.desc_tag =
+                       cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD);
+       assoc_rqst->assoc_cmd.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
+
+       assoc_rqst->assoc_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
+       assoc_rqst->assoc_cmd.sqsize = cpu_to_be16(qsize);
+       /* Linux supports only Dynamic controllers */
+       assoc_rqst->assoc_cmd.cntlid = cpu_to_be16(0xffff);
+       memcpy(&assoc_rqst->assoc_cmd.hostid, &ctrl->ctrl.opts->host->id,
+               min_t(size_t, FCNVME_ASSOC_HOSTID_LEN, sizeof(uuid_be)));
+       strncpy(assoc_rqst->assoc_cmd.hostnqn, ctrl->ctrl.opts->host->nqn,
+               min(FCNVME_ASSOC_HOSTNQN_LEN, NVMF_NQN_SIZE));
+       strncpy(assoc_rqst->assoc_cmd.subnqn, ctrl->ctrl.opts->subsysnqn,
+               min(FCNVME_ASSOC_SUBNQN_LEN, NVMF_NQN_SIZE));
+
+       lsop->queue = queue;
+       lsreq->rqstaddr = assoc_rqst;
+       lsreq->rqstlen = sizeof(*assoc_rqst);
+       lsreq->rspaddr = assoc_acc;
+       lsreq->rsplen = sizeof(*assoc_acc);
+       lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
+
+       ret = nvme_fc_send_ls_req(ctrl, lsop);
+       if (ret)
+               goto out_free_buffer;
+
+       /* process connect LS completion */
+
+       /* validate the ACC response */
+       if (assoc_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
+               fcret = VERR_LSACC;
+       if (assoc_acc->hdr.desc_list_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_ls_cr_assoc_acc)))
+               fcret = VERR_CR_ASSOC_ACC_LEN;
+       if (assoc_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
+               fcret = VERR_LSDESC_RQST;
+       else if (assoc_acc->hdr.rqst.desc_len !=
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
+               fcret = VERR_LSDESC_RQST_LEN;
+       else if (assoc_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_ASSOCIATION)
+               fcret = VERR_CR_ASSOC;
+       else if (assoc_acc->associd.desc_tag !=
+                       cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
+               fcret = VERR_ASSOC_ID;
+       else if (assoc_acc->associd.desc_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_assoc_id)))
+               fcret = VERR_ASSOC_ID_LEN;
+       else if (assoc_acc->connectid.desc_tag !=
+                       cpu_to_be32(FCNVME_LSDESC_CONN_ID))
+               fcret = VERR_CONN_ID;
+       else if (assoc_acc->connectid.desc_len !=
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
+               fcret = VERR_CONN_ID_LEN;
+
+       if (fcret) {
+               ret = -EBADF;
+               dev_err(ctrl->dev,
+                       "q %d connect failed: %s\n",
+                       queue->qnum, validation_errors[fcret]);
+       } else {
+               ctrl->association_id =
+                       be64_to_cpu(assoc_acc->associd.association_id);
+               queue->connection_id =
+                       be64_to_cpu(assoc_acc->connectid.connection_id);
+               set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
+       }
+
+out_free_buffer:
+       kfree(lsop);
+out_no_memory:
+       if (ret)
+               dev_err(ctrl->dev,
+                       "queue %d connect admin queue failed (%d).\n",
+                       queue->qnum, ret);
+       return ret;
+}
+
+static int
+nvme_fc_connect_queue(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
+                       u16 qsize, u16 ersp_ratio)
+{
+       struct nvmefc_ls_req_op *lsop;
+       struct nvmefc_ls_req *lsreq;
+       struct fcnvme_ls_cr_conn_rqst *conn_rqst;
+       struct fcnvme_ls_cr_conn_acc *conn_acc;
+       int ret, fcret = 0;
+
+       lsop = kzalloc((sizeof(*lsop) +
+                        ctrl->lport->ops->lsrqst_priv_sz +
+                        sizeof(*conn_rqst) + sizeof(*conn_acc)), GFP_KERNEL);
+       if (!lsop) {
+               ret = -ENOMEM;
+               goto out_no_memory;
+       }
+       lsreq = &lsop->ls_req;
+
+       lsreq->private = (void *)&lsop[1];
+       conn_rqst = (struct fcnvme_ls_cr_conn_rqst *)
+                       (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
+       conn_acc = (struct fcnvme_ls_cr_conn_acc *)&conn_rqst[1];
+
+       conn_rqst->w0.ls_cmd = FCNVME_LS_CREATE_CONNECTION;
+       conn_rqst->desc_list_len = cpu_to_be32(
+                               sizeof(struct fcnvme_lsdesc_assoc_id) +
+                               sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
+
+       conn_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
+       conn_rqst->associd.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_assoc_id));
+       conn_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
+       conn_rqst->connect_cmd.desc_tag =
+                       cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD);
+       conn_rqst->connect_cmd.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
+       conn_rqst->connect_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
+       conn_rqst->connect_cmd.qid  = cpu_to_be16(queue->qnum);
+       conn_rqst->connect_cmd.sqsize = cpu_to_be16(qsize);
+
+       lsop->queue = queue;
+       lsreq->rqstaddr = conn_rqst;
+       lsreq->rqstlen = sizeof(*conn_rqst);
+       lsreq->rspaddr = conn_acc;
+       lsreq->rsplen = sizeof(*conn_acc);
+       lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
+
+       ret = nvme_fc_send_ls_req(ctrl, lsop);
+       if (ret)
+               goto out_free_buffer;
+
+       /* process connect LS completion */
+
+       /* validate the ACC response */
+       if (conn_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
+               fcret = VERR_LSACC;
+       if (conn_acc->hdr.desc_list_len !=
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)))
+               fcret = VERR_CR_CONN_ACC_LEN;
+       if (conn_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
+               fcret = VERR_LSDESC_RQST;
+       else if (conn_acc->hdr.rqst.desc_len !=
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
+               fcret = VERR_LSDESC_RQST_LEN;
+       else if (conn_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_CONNECTION)
+               fcret = VERR_CR_CONN;
+       else if (conn_acc->connectid.desc_tag !=
+                       cpu_to_be32(FCNVME_LSDESC_CONN_ID))
+               fcret = VERR_CONN_ID;
+       else if (conn_acc->connectid.desc_len !=
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
+               fcret = VERR_CONN_ID_LEN;
+
+       if (fcret) {
+               ret = -EBADF;
+               dev_err(ctrl->dev,
+                       "q %d connect failed: %s\n",
+                       queue->qnum, validation_errors[fcret]);
+       } else {
+               queue->connection_id =
+                       be64_to_cpu(conn_acc->connectid.connection_id);
+               set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
+       }
+
+out_free_buffer:
+       kfree(lsop);
+out_no_memory:
+       if (ret)
+               dev_err(ctrl->dev,
+                       "queue %d connect command failed (%d).\n",
+                       queue->qnum, ret);
+       return ret;
+}
+
+static void
+nvme_fc_disconnect_assoc_done(struct nvmefc_ls_req *lsreq, int status)
+{
+       struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
+       struct nvme_fc_ctrl *ctrl = lsop->ctrl;
+
+       __nvme_fc_finish_ls_req(ctrl, lsop);
+
+       if (status)
+               dev_err(ctrl->dev,
+                       "disconnect assoc ls request command failed (%d).\n",
+                       status);
+
+       /* fc-nvme iniator doesn't care about success or failure of cmd */
+
+       kfree(lsop);
+}
+
+/*
+ * This routine sends a FC-NVME LS to disconnect (aka terminate)
+ * the FC-NVME Association.  Terminating the association also
+ * terminates the FC-NVME connections (per queue, both admin and io
+ * queues) that are part of the association. E.g. things are torn
+ * down, and the related FC-NVME Association ID and Connection IDs
+ * become invalid.
+ *
+ * The behavior of the fc-nvme initiator is such that it's
+ * understanding of the association and connections will implicitly
+ * be torn down. The action is implicit as it may be due to a loss of
+ * connectivity with the fc-nvme target, so you may never get a
+ * response even if you tried.  As such, the action of this routine
+ * is to asynchronously send the LS, ignore any results of the LS, and
+ * continue on with terminating the association. If the fc-nvme target
+ * is present and receives the LS, it too can tear down.
+ */
+static void
+nvme_fc_xmt_disconnect_assoc(struct nvme_fc_ctrl *ctrl)
+{
+       struct fcnvme_ls_disconnect_rqst *discon_rqst;
+       struct fcnvme_ls_disconnect_acc *discon_acc;
+       struct nvmefc_ls_req_op *lsop;
+       struct nvmefc_ls_req *lsreq;
+
+       lsop = kzalloc((sizeof(*lsop) +
+                        ctrl->lport->ops->lsrqst_priv_sz +
+                        sizeof(*discon_rqst) + sizeof(*discon_acc)),
+                       GFP_KERNEL);
+       if (!lsop)
+               /* couldn't sent it... too bad */
+               return;
+
+       lsreq = &lsop->ls_req;
+
+       lsreq->private = (void *)&lsop[1];
+       discon_rqst = (struct fcnvme_ls_disconnect_rqst *)
+                       (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
+       discon_acc = (struct fcnvme_ls_disconnect_acc *)&discon_rqst[1];
+
+       discon_rqst->w0.ls_cmd = FCNVME_LS_DISCONNECT;
+       discon_rqst->desc_list_len = cpu_to_be32(
+                               sizeof(struct fcnvme_lsdesc_assoc_id) +
+                               sizeof(struct fcnvme_lsdesc_disconn_cmd));
+
+       discon_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
+       discon_rqst->associd.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_assoc_id));
+
+       discon_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
+
+       discon_rqst->discon_cmd.desc_tag = cpu_to_be32(
+                                               FCNVME_LSDESC_DISCONN_CMD);
+       discon_rqst->discon_cmd.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_disconn_cmd));
+       discon_rqst->discon_cmd.scope = FCNVME_DISCONN_ASSOCIATION;
+       discon_rqst->discon_cmd.id = cpu_to_be64(ctrl->association_id);
+
+       lsreq->rqstaddr = discon_rqst;
+       lsreq->rqstlen = sizeof(*discon_rqst);
+       lsreq->rspaddr = discon_acc;
+       lsreq->rsplen = sizeof(*discon_acc);
+       lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
+
+       nvme_fc_send_ls_req_async(ctrl, lsop, nvme_fc_disconnect_assoc_done);
+
+       /* only meaningful part to terminating the association */
+       ctrl->association_id = 0;
+}
+
+
+/* *********************** NVME Ctrl Routines **************************** */
+
+
+static int
+nvme_fc_reinit_request(void *data, struct request *rq)
+{
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+       struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
+
+       memset(cmdiu, 0, sizeof(*cmdiu));
+       cmdiu->scsi_id = NVME_CMD_SCSI_ID;
+       cmdiu->fc_id = NVME_CMD_FC_ID;
+       cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
+       memset(&op->rsp_iu, 0, sizeof(op->rsp_iu));
+
+       return 0;
+}
+
+static void
+__nvme_fc_exit_request(struct nvme_fc_ctrl *ctrl,
+               struct nvme_fc_fcp_op *op)
+{
+       fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.rspdma,
+                               sizeof(op->rsp_iu), DMA_FROM_DEVICE);
+       fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.cmddma,
+                               sizeof(op->cmd_iu), DMA_TO_DEVICE);
+
+       atomic_set(&op->state, FCPOP_STATE_UNINIT);
+}
+
+static void
+nvme_fc_exit_request(void *data, struct request *rq,
+                               unsigned int hctx_idx, unsigned int rq_idx)
+{
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+
+       return __nvme_fc_exit_request(data, op);
+}
+
+static void
+nvme_fc_exit_aen_ops(struct nvme_fc_ctrl *ctrl)
+{
+       struct nvme_fc_fcp_op *aen_op = ctrl->aen_ops;
+       int i;
+
+       for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
+               if (atomic_read(&aen_op->state) == FCPOP_STATE_UNINIT)
+                       continue;
+               __nvme_fc_exit_request(ctrl, aen_op);
+               nvme_fc_ctrl_put(ctrl);
+       }
+}
+
+void
+nvme_fc_fcpio_done(struct nvmefc_fcp_req *req)
+{
+       struct nvme_fc_fcp_op *op = fcp_req_to_fcp_op(req);
+       struct request *rq = op->rq;
+       struct nvmefc_fcp_req *freq = &op->fcp_req;
+       struct nvme_fc_ctrl *ctrl = op->ctrl;
+       struct nvme_fc_queue *queue = op->queue;
+       struct nvme_completion *cqe = &op->rsp_iu.cqe;
+       u16 status;
+
+       /*
+        * WARNING:
+        * The current linux implementation of a nvme controller
+        * allocates a single tag set for all io queues and sizes
+        * the io queues to fully hold all possible tags. Thus, the
+        * implementation does not reference or care about the sqhd
+        * value as it never needs to use the sqhd/sqtail pointers
+        * for submission pacing.
+        *
+        * This affects the FC-NVME implementation in two ways:
+        * 1) As the value doesn't matter, we don't need to waste
+        *    cycles extracting it from ERSPs and stamping it in the
+        *    cases where the transport fabricates CQEs on successful
+        *    completions.
+        * 2) The FC-NVME implementation requires that delivery of
+        *    ERSP completions are to go back to the nvme layer in order
+        *    relative to the rsn, such that the sqhd value will always
+        *    be "in order" for the nvme layer. As the nvme layer in
+        *    linux doesn't care about sqhd, there's no need to return
+        *    them in order.
+        *
+        * Additionally:
+        * As the core nvme layer in linux currently does not look at
+        * every field in the cqe - in cases where the FC transport must
+        * fabricate a CQE, the following fields will not be set as they
+        * are not referenced:
+        *      cqe.sqid,  cqe.sqhd,  cqe.command_id
+        */
+
+       fc_dma_sync_single_for_cpu(ctrl->lport->dev, op->fcp_req.rspdma,
+                               sizeof(op->rsp_iu), DMA_FROM_DEVICE);
+
+       if (atomic_read(&op->state) == FCPOP_STATE_ABORTED)
+               status = NVME_SC_ABORT_REQ | NVME_SC_DNR;
+       else
+               status = freq->status;
+
+       /*
+        * For the linux implementation, if we have an unsuccesful
+        * status, they blk-mq layer can typically be called with the
+        * non-zero status and the content of the cqe isn't important.
+        */
+       if (status)
+               goto done;
+
+       /*
+        * command completed successfully relative to the wire
+        * protocol. However, validate anything received and
+        * extract the status and result from the cqe (create it
+        * where necessary).
+        */
+
+       switch (freq->rcv_rsplen) {
+
+       case 0:
+       case NVME_FC_SIZEOF_ZEROS_RSP:
+               /*
+                * No response payload or 12 bytes of payload (which
+                * should all be zeros) are considered successful and
+                * no payload in the CQE by the transport.
+                */
+               if (freq->transferred_length !=
+                       be32_to_cpu(op->cmd_iu.data_len)) {
+                       status = -EIO;
+                       goto done;
+               }
+               op->nreq.result.u64 = 0;
+               break;
+
+       case sizeof(struct nvme_fc_ersp_iu):
+               /*
+                * The ERSP IU contains a full completion with CQE.
+                * Validate ERSP IU and look at cqe.
+                */
+               if (unlikely(be16_to_cpu(op->rsp_iu.iu_len) !=
+                                       (freq->rcv_rsplen / 4) ||
+                            be32_to_cpu(op->rsp_iu.xfrd_len) !=
+                                       freq->transferred_length ||
+                            op->rqno != le16_to_cpu(cqe->command_id))) {
+                       status = -EIO;
+                       goto done;
+               }
+               op->nreq.result = cqe->result;
+               status = le16_to_cpu(cqe->status) >> 1;
+               break;
+
+       default:
+               status = -EIO;
+               goto done;
+       }
+
+done:
+       if (!queue->qnum && op->rqno >= AEN_CMDID_BASE) {
+               nvme_complete_async_event(&queue->ctrl->ctrl, status,
+                                       &op->nreq.result);
+               nvme_fc_ctrl_put(ctrl);
+               return;
+       }
+
+       blk_mq_complete_request(rq, status);
+}
+
+static int
+__nvme_fc_init_request(struct nvme_fc_ctrl *ctrl,
+               struct nvme_fc_queue *queue, struct nvme_fc_fcp_op *op,
+               struct request *rq, u32 rqno)
+{
+       struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
+       int ret = 0;
+
+       memset(op, 0, sizeof(*op));
+       op->fcp_req.cmdaddr = &op->cmd_iu;
+       op->fcp_req.cmdlen = sizeof(op->cmd_iu);
+       op->fcp_req.rspaddr = &op->rsp_iu;
+       op->fcp_req.rsplen = sizeof(op->rsp_iu);
+       op->fcp_req.done = nvme_fc_fcpio_done;
+       op->fcp_req.first_sgl = (struct scatterlist *)&op[1];
+       op->fcp_req.private = &op->fcp_req.first_sgl[SG_CHUNK_SIZE];
+       op->ctrl = ctrl;
+       op->queue = queue;
+       op->rq = rq;
+       op->rqno = rqno;
+
+       cmdiu->scsi_id = NVME_CMD_SCSI_ID;
+       cmdiu->fc_id = NVME_CMD_FC_ID;
+       cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
+
+       op->fcp_req.cmddma = fc_dma_map_single(ctrl->lport->dev,
+                               &op->cmd_iu, sizeof(op->cmd_iu), DMA_TO_DEVICE);
+       if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.cmddma)) {
+               dev_err(ctrl->dev,
+                       "FCP Op failed - cmdiu dma mapping failed.\n");
+               ret = EFAULT;
+               goto out_on_error;
+       }
+
+       op->fcp_req.rspdma = fc_dma_map_single(ctrl->lport->dev,
+                               &op->rsp_iu, sizeof(op->rsp_iu),
+                               DMA_FROM_DEVICE);
+       if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.rspdma)) {
+               dev_err(ctrl->dev,
+                       "FCP Op failed - rspiu dma mapping failed.\n");
+               ret = EFAULT;
+       }
+
+       atomic_set(&op->state, FCPOP_STATE_IDLE);
+out_on_error:
+       return ret;
+}
+
+static int
+nvme_fc_init_request(void *data, struct request *rq,
+                               unsigned int hctx_idx, unsigned int rq_idx,
+                               unsigned int numa_node)
+{
+       struct nvme_fc_ctrl *ctrl = data;
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+       struct nvme_fc_queue *queue = &ctrl->queues[hctx_idx+1];
+
+       return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
+}
+
+static int
+nvme_fc_init_admin_request(void *data, struct request *rq,
+                               unsigned int hctx_idx, unsigned int rq_idx,
+                               unsigned int numa_node)
+{
+       struct nvme_fc_ctrl *ctrl = data;
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+       struct nvme_fc_queue *queue = &ctrl->queues[0];
+
+       return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
+}
+
+static int
+nvme_fc_init_aen_ops(struct nvme_fc_ctrl *ctrl)
+{
+       struct nvme_fc_fcp_op *aen_op;
+       struct nvme_fc_cmd_iu *cmdiu;
+       struct nvme_command *sqe;
+       int i, ret;
+
+       aen_op = ctrl->aen_ops;
+       for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
+               cmdiu = &aen_op->cmd_iu;
+               sqe = &cmdiu->sqe;
+               ret = __nvme_fc_init_request(ctrl, &ctrl->queues[0],
+                               aen_op, (struct request *)NULL,
+                               (AEN_CMDID_BASE + i));
+               if (ret)
+                       return ret;
+
+               memset(sqe, 0, sizeof(*sqe));
+               sqe->common.opcode = nvme_admin_async_event;
+               sqe->common.command_id = AEN_CMDID_BASE + i;
+       }
+       return 0;
+}
+
+
+static inline void
+__nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, struct nvme_fc_ctrl *ctrl,
+               unsigned int qidx)
+{
+       struct nvme_fc_queue *queue = &ctrl->queues[qidx];
+
+       hctx->driver_data = queue;
+       queue->hctx = hctx;
+}
+
+static int
+nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
+               unsigned int hctx_idx)
+{
+       struct nvme_fc_ctrl *ctrl = data;
+
+       __nvme_fc_init_hctx(hctx, ctrl, hctx_idx + 1);
+
+       return 0;
+}
+
+static int
+nvme_fc_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
+               unsigned int hctx_idx)
+{
+       struct nvme_fc_ctrl *ctrl = data;
+
+       __nvme_fc_init_hctx(hctx, ctrl, hctx_idx);
+
+       return 0;
+}
+
+static void
+nvme_fc_init_queue(struct nvme_fc_ctrl *ctrl, int idx, size_t queue_size)
+{
+       struct nvme_fc_queue *queue;
+
+       queue = &ctrl->queues[idx];
+       memset(queue, 0, sizeof(*queue));
+       queue->ctrl = ctrl;
+       queue->qnum = idx;
+       atomic_set(&queue->csn, 1);
+       queue->dev = ctrl->dev;
+
+       if (idx > 0)
+               queue->cmnd_capsule_len = ctrl->ctrl.ioccsz * 16;
+       else
+               queue->cmnd_capsule_len = sizeof(struct nvme_command);
+
+       queue->queue_size = queue_size;
+
+       /*
+        * Considered whether we should allocate buffers for all SQEs
+        * and CQEs and dma map them - mapping their respective entries
+        * into the request structures (kernel vm addr and dma address)
+        * thus the driver could use the buffers/mappings directly.
+        * It only makes sense if the LLDD would use them for its
+        * messaging api. It's very unlikely most adapter api's would use
+        * a native NVME sqe/cqe. More reasonable if FC-NVME IU payload
+        * structures were used instead.
+        */
+}
+
+/*
+ * This routine terminates a queue at the transport level.
+ * The transport has already ensured that all outstanding ios on
+ * the queue have been terminated.
+ * The transport will send a Disconnect LS request to terminate
+ * the queue's connection. Termination of the admin queue will also
+ * terminate the association at the target.
+ */
+static void
+nvme_fc_free_queue(struct nvme_fc_queue *queue)
+{
+       if (!test_and_clear_bit(NVME_FC_Q_CONNECTED, &queue->flags))
+               return;
+
+       /*
+        * Current implementation never disconnects a single queue.
+        * It always terminates a whole association. So there is never
+        * a disconnect(queue) LS sent to the target.
+        */
+
+       queue->connection_id = 0;
+       clear_bit(NVME_FC_Q_CONNECTED, &queue->flags);
+}
+
+static void
+__nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *ctrl,
+       struct nvme_fc_queue *queue, unsigned int qidx)
+{
+       if (ctrl->lport->ops->delete_queue)
+               ctrl->lport->ops->delete_queue(&ctrl->lport->localport, qidx,
+                               queue->lldd_handle);
+       queue->lldd_handle = NULL;
+}
+
+static void
+nvme_fc_destroy_admin_queue(struct nvme_fc_ctrl *ctrl)
+{
+       __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
+       blk_cleanup_queue(ctrl->ctrl.admin_q);
+       blk_mq_free_tag_set(&ctrl->admin_tag_set);
+       nvme_fc_free_queue(&ctrl->queues[0]);
+}
+
+static void
+nvme_fc_free_io_queues(struct nvme_fc_ctrl *ctrl)
+{
+       int i;
+
+       for (i = 1; i < ctrl->queue_count; i++)
+               nvme_fc_free_queue(&ctrl->queues[i]);
+}
+
+static int
+__nvme_fc_create_hw_queue(struct nvme_fc_ctrl *ctrl,
+       struct nvme_fc_queue *queue, unsigned int qidx, u16 qsize)
+{
+       int ret = 0;
+
+       queue->lldd_handle = NULL;
+       if (ctrl->lport->ops->create_queue)
+               ret = ctrl->lport->ops->create_queue(&ctrl->lport->localport,
+                               qidx, qsize, &queue->lldd_handle);
+
+       return ret;
+}
+
+static void
+nvme_fc_delete_hw_io_queues(struct nvme_fc_ctrl *ctrl)
+{
+       struct nvme_fc_queue *queue = &ctrl->queues[ctrl->queue_count - 1];
+       int i;
+
+       for (i = ctrl->queue_count - 1; i >= 1; i--, queue--)
+               __nvme_fc_delete_hw_queue(ctrl, queue, i);
+}
+
+static int
+nvme_fc_create_hw_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
+{
+       struct nvme_fc_queue *queue = &ctrl->queues[1];
+       int i, j, ret;
+
+       for (i = 1; i < ctrl->queue_count; i++, queue++) {
+               ret = __nvme_fc_create_hw_queue(ctrl, queue, i, qsize);
+               if (ret) {
+                       for (j = i-1; j >= 0; j--)
+                               __nvme_fc_delete_hw_queue(ctrl,
+                                               &ctrl->queues[j], j);
+                       return ret;
+               }
+       }
+
+       return 0;
+}
+
+static int
+nvme_fc_connect_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
+{
+       int i, ret = 0;
+
+       for (i = 1; i < ctrl->queue_count; i++) {
+               ret = nvme_fc_connect_queue(ctrl, &ctrl->queues[i], qsize,
+                                       (qsize / 5));
+               if (ret)
+                       break;
+               ret = nvmf_connect_io_queue(&ctrl->ctrl, i);
+               if (ret)
+                       break;
+       }
+
+       return ret;
+}
+
+static void
+nvme_fc_init_io_queues(struct nvme_fc_ctrl *ctrl)
+{
+       int i;
+
+       for (i = 1; i < ctrl->queue_count; i++)
+               nvme_fc_init_queue(ctrl, i, ctrl->ctrl.sqsize);
+}
+
+static void
+nvme_fc_ctrl_free(struct kref *ref)
+{
+       struct nvme_fc_ctrl *ctrl =
+               container_of(ref, struct nvme_fc_ctrl, ref);
+       unsigned long flags;
+
+       if (ctrl->state != FCCTRL_INIT) {
+               /* remove from rport list */
+               spin_lock_irqsave(&ctrl->rport->lock, flags);
+               list_del(&ctrl->ctrl_list);
+               spin_unlock_irqrestore(&ctrl->rport->lock, flags);
+       }
+
+       put_device(ctrl->dev);
+       nvme_fc_rport_put(ctrl->rport);
+
+       kfree(ctrl->queues);
+       ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
+       nvmf_free_options(ctrl->ctrl.opts);
+       kfree(ctrl);
+}
+
+static void
+nvme_fc_ctrl_put(struct nvme_fc_ctrl *ctrl)
+{
+       kref_put(&ctrl->ref, nvme_fc_ctrl_free);
+}
+
+static int
+nvme_fc_ctrl_get(struct nvme_fc_ctrl *ctrl)
+{
+       return kref_get_unless_zero(&ctrl->ref);
+}
+
+/*
+ * All accesses from nvme core layer done - can now free the
+ * controller. Called after last nvme_put_ctrl() call
+ */
+static void
+nvme_fc_free_nvme_ctrl(struct nvme_ctrl *nctrl)
+{
+       struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
+
+       WARN_ON(nctrl != &ctrl->ctrl);
+
+       /*
+        * Tear down the association, which will generate link
+        * traffic to terminate connections
+        */
+
+       if (ctrl->state != FCCTRL_INIT) {
+               /* send a Disconnect(association) LS to fc-nvme target */
+               nvme_fc_xmt_disconnect_assoc(ctrl);
+
+               if (ctrl->ctrl.tagset) {
+                       blk_cleanup_queue(ctrl->ctrl.connect_q);
+                       blk_mq_free_tag_set(&ctrl->tag_set);
+                       nvme_fc_delete_hw_io_queues(ctrl);
+                       nvme_fc_free_io_queues(ctrl);
+               }
+
+               nvme_fc_exit_aen_ops(ctrl);
+
+               nvme_fc_destroy_admin_queue(ctrl);
+       }
+
+       nvme_fc_ctrl_put(ctrl);
+}
+
+
+static int
+__nvme_fc_abort_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_fcp_op *op)
+{
+       int state;
+
+       state = atomic_xchg(&op->state, FCPOP_STATE_ABORTED);
+       if (state != FCPOP_STATE_ACTIVE) {
+               atomic_set(&op->state, state);
+               return -ECANCELED; /* fail */
+       }
+
+       ctrl->lport->ops->fcp_abort(&ctrl->lport->localport,
+                                       &ctrl->rport->remoteport,
+                                       op->queue->lldd_handle,
+                                       &op->fcp_req);
+
+       return 0;
+}
+
+enum blk_eh_timer_return
+nvme_fc_timeout(struct request *rq, bool reserved)
+{
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+       struct nvme_fc_ctrl *ctrl = op->ctrl;
+       int ret;
+
+       if (reserved)
+               return BLK_EH_RESET_TIMER;
+
+       ret = __nvme_fc_abort_op(ctrl, op);
+       if (ret)
+               /* io wasn't active to abort consider it done */
+               return BLK_EH_HANDLED;
+
+       /*
+        * TODO: force a controller reset
+        *   when that happens, queues will be torn down and outstanding
+        *   ios will be terminated, and the above abort, on a single io
+        *   will no longer be needed.
+        */
+
+       return BLK_EH_HANDLED;
+}
+
+static int
+nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
+               struct nvme_fc_fcp_op *op)
+{
+       struct nvmefc_fcp_req *freq = &op->fcp_req;
+       u32 map_len = nvme_map_len(rq);
+       enum dma_data_direction dir;
+       int ret;
+
+       freq->sg_cnt = 0;
+
+       if (!map_len)
+               return 0;
+
+       freq->sg_table.sgl = freq->first_sgl;
+       ret = sg_alloc_table_chained(&freq->sg_table, rq->nr_phys_segments,
+                       freq->sg_table.sgl);
+       if (ret)
+               return -ENOMEM;
+
+       op->nents = blk_rq_map_sg(rq->q, rq, freq->sg_table.sgl);
+       WARN_ON(op->nents > rq->nr_phys_segments);
+       dir = (rq_data_dir(rq) == WRITE) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
+       freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl,
+                               op->nents, dir);
+       if (unlikely(freq->sg_cnt <= 0)) {
+               sg_free_table_chained(&freq->sg_table, true);
+               freq->sg_cnt = 0;
+               return -EFAULT;
+       }
+
+       /*
+        * TODO: blk_integrity_rq(rq)  for DIF
+        */
+       return 0;
+}
+
+static void
+nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
+               struct nvme_fc_fcp_op *op)
+{
+       struct nvmefc_fcp_req *freq = &op->fcp_req;
+
+       if (!freq->sg_cnt)
+               return;
+
+       fc_dma_unmap_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents,
+                               ((rq_data_dir(rq) == WRITE) ?
+                                       DMA_TO_DEVICE : DMA_FROM_DEVICE));
+
+       nvme_cleanup_cmd(rq);
+
+       sg_free_table_chained(&freq->sg_table, true);
+
+       freq->sg_cnt = 0;
+}
+
+/*
+ * In FC, the queue is a logical thing. At transport connect, the target
+ * creates its "queue" and returns a handle that is to be given to the
+ * target whenever it posts something to the corresponding SQ.  When an
+ * SQE is sent on a SQ, FC effectively considers the SQE, or rather the
+ * command contained within the SQE, an io, and assigns a FC exchange
+ * to it. The SQE and the associated SQ handle are sent in the initial
+ * CMD IU sents on the exchange. All transfers relative to the io occur
+ * as part of the exchange.  The CQE is the last thing for the io,
+ * which is transferred (explicitly or implicitly) with the RSP IU
+ * sent on the exchange. After the CQE is received, the FC exchange is
+ * terminaed and the Exchange may be used on a different io.
+ *
+ * The transport to LLDD api has the transport making a request for a
+ * new fcp io request to the LLDD. The LLDD then allocates a FC exchange
+ * resource and transfers the command. The LLDD will then process all
+ * steps to complete the io. Upon completion, the transport done routine
+ * is called.
+ *
+ * So - while the operation is outstanding to the LLDD, there is a link
+ * level FC exchange resource that is also outstanding. This must be
+ * considered in all cleanup operations.
+ */
+static int
+nvme_fc_start_fcp_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
+       struct nvme_fc_fcp_op *op, u32 data_len,
+       enum nvmefc_fcp_datadir io_dir)
+{
+       struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
+       struct nvme_command *sqe = &cmdiu->sqe;
+       u32 csn;
+       int ret;
+
+       if (!nvme_fc_ctrl_get(ctrl))
+               return BLK_MQ_RQ_QUEUE_ERROR;
+
+       /* format the FC-NVME CMD IU and fcp_req */
+       cmdiu->connection_id = cpu_to_be64(queue->connection_id);
+       csn = atomic_inc_return(&queue->csn);
+       cmdiu->csn = cpu_to_be32(csn);
+       cmdiu->data_len = cpu_to_be32(data_len);
+       switch (io_dir) {
+       case NVMEFC_FCP_WRITE:
+               cmdiu->flags = FCNVME_CMD_FLAGS_WRITE;
+               break;
+       case NVMEFC_FCP_READ:
+               cmdiu->flags = FCNVME_CMD_FLAGS_READ;
+               break;
+       case NVMEFC_FCP_NODATA:
+               cmdiu->flags = 0;
+               break;
+       }
+       op->fcp_req.payload_length = data_len;
+       op->fcp_req.io_dir = io_dir;
+       op->fcp_req.transferred_length = 0;
+       op->fcp_req.rcv_rsplen = 0;
+       op->fcp_req.status = 0;
+       op->fcp_req.sqid = cpu_to_le16(queue->qnum);
+
+       /*
+        * validate per fabric rules, set fields mandated by fabric spec
+        * as well as those by FC-NVME spec.
+        */
+       WARN_ON_ONCE(sqe->common.metadata);
+       WARN_ON_ONCE(sqe->common.dptr.prp1);
+       WARN_ON_ONCE(sqe->common.dptr.prp2);
+       sqe->common.flags |= NVME_CMD_SGL_METABUF;
+
+       /*
+        * format SQE DPTR field per FC-NVME rules
+        *    type=data block descr; subtype=offset;
+        *    offset is currently 0.
+        */
+       sqe->rw.dptr.sgl.type = NVME_SGL_FMT_OFFSET;
+       sqe->rw.dptr.sgl.length = cpu_to_le32(data_len);
+       sqe->rw.dptr.sgl.addr = 0;
+
+       /* odd that we set the command_id - should come from nvme-fabrics */
+       WARN_ON_ONCE(sqe->common.command_id != cpu_to_le16(op->rqno));
+
+       if (op->rq) {                           /* skipped on aens */
+               ret = nvme_fc_map_data(ctrl, op->rq, op);
+               if (ret < 0) {
+                       dev_err(queue->ctrl->ctrl.device,
+                            "Failed to map data (%d)\n", ret);
+                       nvme_cleanup_cmd(op->rq);
+                       nvme_fc_ctrl_put(ctrl);
+                       return (ret == -ENOMEM || ret == -EAGAIN) ?
+                               BLK_MQ_RQ_QUEUE_BUSY : BLK_MQ_RQ_QUEUE_ERROR;
+               }
+       }
+
+       fc_dma_sync_single_for_device(ctrl->lport->dev, op->fcp_req.cmddma,
+                                 sizeof(op->cmd_iu), DMA_TO_DEVICE);
+
+       atomic_set(&op->state, FCPOP_STATE_ACTIVE);
+
+       if (op->rq)
+               blk_mq_start_request(op->rq);
+
+       ret = ctrl->lport->ops->fcp_io(&ctrl->lport->localport,
+                                       &ctrl->rport->remoteport,
+                                       queue->lldd_handle, &op->fcp_req);
+
+       if (ret) {
+               dev_err(ctrl->dev,
+                       "Send nvme command failed - lldd returned %d.\n", ret);
+
+               if (op->rq) {                   /* normal request */
+                       nvme_fc_unmap_data(ctrl, op->rq, op);
+                       nvme_cleanup_cmd(op->rq);
+               }
+               /* else - aen. no cleanup needed */
+
+               nvme_fc_ctrl_put(ctrl);
+
+               if (ret != -EBUSY)
+                       return BLK_MQ_RQ_QUEUE_ERROR;
+
+               if (op->rq) {
+                       blk_mq_stop_hw_queues(op->rq->q);
+                       blk_mq_delay_queue(queue->hctx, NVMEFC_QUEUE_DELAY);
+               }
+               return BLK_MQ_RQ_QUEUE_BUSY;
+       }
+
+       return BLK_MQ_RQ_QUEUE_OK;
+}
+
+static int
+nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,
+                       const struct blk_mq_queue_data *bd)
+{
+       struct nvme_ns *ns = hctx->queue->queuedata;
+       struct nvme_fc_queue *queue = hctx->driver_data;
+       struct nvme_fc_ctrl *ctrl = queue->ctrl;
+       struct request *rq = bd->rq;
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+       struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
+       struct nvme_command *sqe = &cmdiu->sqe;
+       enum nvmefc_fcp_datadir io_dir;
+       u32 data_len;
+       int ret;
+
+       ret = nvme_setup_cmd(ns, rq, sqe);
+       if (ret)
+               return ret;
+
+       data_len = nvme_map_len(rq);
+       if (data_len)
+               io_dir = ((rq_data_dir(rq) == WRITE) ?
+                                       NVMEFC_FCP_WRITE : NVMEFC_FCP_READ);
+       else
+               io_dir = NVMEFC_FCP_NODATA;
+
+       return nvme_fc_start_fcp_op(ctrl, queue, op, data_len, io_dir);
+}
+
+static struct blk_mq_tags *
+nvme_fc_tagset(struct nvme_fc_queue *queue)
+{
+       if (queue->qnum == 0)
+               return queue->ctrl->admin_tag_set.tags[queue->qnum];
+
+       return queue->ctrl->tag_set.tags[queue->qnum - 1];
+}
+
+static int
+nvme_fc_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
+
+{
+       struct nvme_fc_queue *queue = hctx->driver_data;
+       struct nvme_fc_ctrl *ctrl = queue->ctrl;
+       struct request *req;
+       struct nvme_fc_fcp_op *op;
+
+       req = blk_mq_tag_to_rq(nvme_fc_tagset(queue), tag);
+       if (!req) {
+               dev_err(queue->ctrl->ctrl.device,
+                        "tag 0x%x on QNum %#x not found\n",
+                       tag, queue->qnum);
+               return 0;
+       }
+
+       op = blk_mq_rq_to_pdu(req);
+
+       if ((atomic_read(&op->state) == FCPOP_STATE_ACTIVE) &&
+                (ctrl->lport->ops->poll_queue))
+               ctrl->lport->ops->poll_queue(&ctrl->lport->localport,
+                                                queue->lldd_handle);
+
+       return ((atomic_read(&op->state) != FCPOP_STATE_ACTIVE));
+}
+
+static void
+nvme_fc_submit_async_event(struct nvme_ctrl *arg, int aer_idx)
+{
+       struct nvme_fc_ctrl *ctrl = to_fc_ctrl(arg);
+       struct nvme_fc_fcp_op *aen_op;
+       int ret;
+
+       if (aer_idx > NVME_FC_NR_AEN_COMMANDS)
+               return;
+
+       aen_op = &ctrl->aen_ops[aer_idx];
+
+       ret = nvme_fc_start_fcp_op(ctrl, aen_op->queue, aen_op, 0,
+                                       NVMEFC_FCP_NODATA);
+       if (ret)
+               dev_err(ctrl->ctrl.device,
+                       "failed async event work [%d]\n", aer_idx);
+}
+
+static void
+nvme_fc_complete_rq(struct request *rq)
+{
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
+       struct nvme_fc_ctrl *ctrl = op->ctrl;
+       int error = 0, state;
+
+       state = atomic_xchg(&op->state, FCPOP_STATE_IDLE);
+
+       nvme_cleanup_cmd(rq);
+
+       nvme_fc_unmap_data(ctrl, rq, op);
+
+       if (unlikely(rq->errors)) {
+               if (nvme_req_needs_retry(rq, rq->errors)) {
+                       nvme_requeue_req(rq);
+                       return;
+               }
+
+               if (rq->cmd_type == REQ_TYPE_DRV_PRIV)
+                       error = rq->errors;
+               else
+                       error = nvme_error_status(rq->errors);
+       }
+
+       nvme_fc_ctrl_put(ctrl);
+
+       blk_mq_end_request(rq, error);
+}
+
+static struct blk_mq_ops nvme_fc_mq_ops = {
+       .queue_rq       = nvme_fc_queue_rq,
+       .complete       = nvme_fc_complete_rq,
+       .init_request   = nvme_fc_init_request,
+       .exit_request   = nvme_fc_exit_request,
+       .reinit_request = nvme_fc_reinit_request,
+       .init_hctx      = nvme_fc_init_hctx,
+       .poll           = nvme_fc_poll,
+       .timeout        = nvme_fc_timeout,
+};
+
+static struct blk_mq_ops nvme_fc_admin_mq_ops = {
+       .queue_rq       = nvme_fc_queue_rq,
+       .complete       = nvme_fc_complete_rq,
+       .init_request   = nvme_fc_init_admin_request,
+       .exit_request   = nvme_fc_exit_request,
+       .reinit_request = nvme_fc_reinit_request,
+       .init_hctx      = nvme_fc_init_admin_hctx,
+       .timeout        = nvme_fc_timeout,
+};
+
+static int
+nvme_fc_configure_admin_queue(struct nvme_fc_ctrl *ctrl)
+{
+       u32 segs;
+       int error;
+
+       nvme_fc_init_queue(ctrl, 0, NVME_FC_AQ_BLKMQ_DEPTH);
+
+       error = nvme_fc_connect_admin_queue(ctrl, &ctrl->queues[0],
+                               NVME_FC_AQ_BLKMQ_DEPTH,
+                               (NVME_FC_AQ_BLKMQ_DEPTH / 4));
+       if (error)
+               return error;
+
+       memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
+       ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops;
+       ctrl->admin_tag_set.queue_depth = NVME_FC_AQ_BLKMQ_DEPTH;
+       ctrl->admin_tag_set.reserved_tags = 2; /* fabric connect + Keep-Alive */
+       ctrl->admin_tag_set.numa_node = NUMA_NO_NODE;
+       ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
+                                       (SG_CHUNK_SIZE *
+                                               sizeof(struct scatterlist)) +
+                                       ctrl->lport->ops->fcprqst_priv_sz;
+       ctrl->admin_tag_set.driver_data = ctrl;
+       ctrl->admin_tag_set.nr_hw_queues = 1;
+       ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
+
+       error = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
+       if (error)
+               goto out_free_queue;
+
+       ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
+       if (IS_ERR(ctrl->ctrl.admin_q)) {
+               error = PTR_ERR(ctrl->ctrl.admin_q);
+               goto out_free_tagset;
+       }
+
+       error = __nvme_fc_create_hw_queue(ctrl, &ctrl->queues[0], 0,
+                               NVME_FC_AQ_BLKMQ_DEPTH);
+       if (error)
+               goto out_cleanup_queue;
+
+       error = nvmf_connect_admin_queue(&ctrl->ctrl);
+       if (error)
+               goto out_delete_hw_queue;
+
+       error = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->cap);
+       if (error) {
+               dev_err(ctrl->ctrl.device,
+                       "prop_get NVME_REG_CAP failed\n");
+               goto out_delete_hw_queue;
+       }
+
+       ctrl->ctrl.sqsize =
+               min_t(int, NVME_CAP_MQES(ctrl->cap) + 1, ctrl->ctrl.sqsize);
+
+       error = nvme_enable_ctrl(&ctrl->ctrl, ctrl->cap);
+       if (error)
+               goto out_delete_hw_queue;
+
+       segs = min_t(u32, NVME_FC_MAX_SEGMENTS,
+                       ctrl->lport->ops->max_sgl_segments);
+       ctrl->ctrl.max_hw_sectors = (segs - 1) << (PAGE_SHIFT - 9);
+
+       error = nvme_init_identify(&ctrl->ctrl);
+       if (error)
+               goto out_delete_hw_queue;
+
+       nvme_start_keep_alive(&ctrl->ctrl);
+
+       return 0;
+
+out_delete_hw_queue:
+       __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
+out_cleanup_queue:
+       blk_cleanup_queue(ctrl->ctrl.admin_q);
+out_free_tagset:
+       blk_mq_free_tag_set(&ctrl->admin_tag_set);
+out_free_queue:
+       nvme_fc_free_queue(&ctrl->queues[0]);
+       return error;
+}
+
+/*
+ * This routine is used by the transport when it needs to find active
+ * io on a queue that is to be terminated. The transport uses
+ * blk_mq_tagset_busy_itr() to find the busy requests, which then invoke
+ * this routine to kill them on a 1 by 1 basis.
+ *
+ * As FC allocates FC exchange for each io, the transport must contact
+ * the LLDD to terminate the exchange, thus releasing the FC exchange.
+ * After terminating the exchange the LLDD will call the transport's
+ * normal io done path for the request, but it will have an aborted
+ * status. The done path will return the io request back to the block
+ * layer with an error status.
+ */
+static void
+nvme_fc_terminate_exchange(struct request *req, void *data, bool reserved)
+{
+       struct nvme_ctrl *nctrl = data;
+       struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
+       struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(req);
+int status;
+
+       if (!blk_mq_request_started(req))
+               return;
+
+       /* this performs an ABTS-LS on the FC exchange for the io */
+       status = __nvme_fc_abort_op(ctrl, op);
+       /*
+        * if __nvme_fc_abort_op failed: io wasn't active to abort
+        * consider it done. Assume completion path already completing
+        * in parallel
+        */
+       if (status)
+               /* io wasn't active to abort consider it done */
+               /* assume completion path already completing in parallel */
+               return;
+}
+
+
+/*
+ * This routine stops operation of the controller. Admin and IO queues
+ * are stopped, outstanding ios on them terminated, and the nvme ctrl
+ * is shutdown.
+ */
+static void
+nvme_fc_shutdown_ctrl(struct nvme_fc_ctrl *ctrl)
+{
+       /*
+        * If io queues are present, stop them and terminate all outstanding
+        * ios on them. As FC allocates FC exchange for each io, the
+        * transport must contact the LLDD to terminate the exchange,
+        * thus releasing the FC exchange. We use blk_mq_tagset_busy_itr()
+        * to tell us what io's are busy and invoke a transport routine
+        * to kill them with the LLDD.  After terminating the exchange
+        * the LLDD will call the transport's normal io done path, but it
+        * will have an aborted status. The done path will return the
+        * io requests back to the block layer as part of normal completions
+        * (but with error status).
+        */
+       if (ctrl->queue_count > 1) {
+               nvme_stop_queues(&ctrl->ctrl);
+               blk_mq_tagset_busy_iter(&ctrl->tag_set,
+                               nvme_fc_terminate_exchange, &ctrl->ctrl);
+       }
+
+       if (ctrl->ctrl.state == NVME_CTRL_LIVE)
+               nvme_shutdown_ctrl(&ctrl->ctrl);
+
+       /*
+        * now clean up the admin queue. Same thing as above.
+        * use blk_mq_tagset_busy_itr() and the transport routine to
+        * terminate the exchanges.
+        */
+       blk_mq_stop_hw_queues(ctrl->ctrl.admin_q);
+       blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
+                               nvme_fc_terminate_exchange, &ctrl->ctrl);
+}
+
+/*
+ * Called to teardown an association.
+ * May be called with association fully in place or partially in place.
+ */
+static void
+__nvme_fc_remove_ctrl(struct nvme_fc_ctrl *ctrl)
+{
+       nvme_stop_keep_alive(&ctrl->ctrl);
+
+       /* stop and terminate ios on admin and io queues */
+       nvme_fc_shutdown_ctrl(ctrl);
+
+       /*
+        * tear down the controller
+        * This will result in the last reference on the nvme ctrl to
+        * expire, calling the transport nvme_fc_free_nvme_ctrl() callback.
+        * From there, the transport will tear down it's logical queues and
+        * association.
+        */
+       nvme_uninit_ctrl(&ctrl->ctrl);
+
+       nvme_put_ctrl(&ctrl->ctrl);
+}
+
+static void
+nvme_fc_del_ctrl_work(struct work_struct *work)
+{
+       struct nvme_fc_ctrl *ctrl =
+                       container_of(work, struct nvme_fc_ctrl, delete_work);
+
+       __nvme_fc_remove_ctrl(ctrl);
+}
+
+static int
+__nvme_fc_del_ctrl(struct nvme_fc_ctrl *ctrl)
+{
+       if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING))
+               return -EBUSY;
+
+       if (!queue_work(nvme_fc_wq, &ctrl->delete_work))
+               return -EBUSY;
+
+       return 0;
+}
+
+/*
+ * Request from nvme core layer to delete the controller
+ */
+static int
+nvme_fc_del_nvme_ctrl(struct nvme_ctrl *nctrl)
+{
+       struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
+       struct nvme_fc_rport *rport = ctrl->rport;
+       unsigned long flags;
+       int ret;
+
+       spin_lock_irqsave(&rport->lock, flags);
+       ret = __nvme_fc_del_ctrl(ctrl);
+       spin_unlock_irqrestore(&rport->lock, flags);
+       if (ret)
+               return ret;
+
+       flush_work(&ctrl->delete_work);
+
+       return 0;
+}
+
+static int
+nvme_fc_reset_nvme_ctrl(struct nvme_ctrl *nctrl)
+{
+       return -EIO;
+}
+
+static const struct nvme_ctrl_ops nvme_fc_ctrl_ops = {
+       .name                   = "fc",
+       .module                 = THIS_MODULE,
+       .is_fabrics             = true,
+       .reg_read32             = nvmf_reg_read32,
+       .reg_read64             = nvmf_reg_read64,
+       .reg_write32            = nvmf_reg_write32,
+       .reset_ctrl             = nvme_fc_reset_nvme_ctrl,
+       .free_ctrl              = nvme_fc_free_nvme_ctrl,
+       .submit_async_event     = nvme_fc_submit_async_event,
+       .delete_ctrl            = nvme_fc_del_nvme_ctrl,
+       .get_subsysnqn          = nvmf_get_subsysnqn,
+       .get_address            = nvmf_get_address,
+};
+
+static int
+nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
+{
+       struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
+       int ret;
+
+       ret = nvme_set_queue_count(&ctrl->ctrl, &opts->nr_io_queues);
+       if (ret) {
+               dev_info(ctrl->ctrl.device,
+                       "set_queue_count failed: %d\n", ret);
+               return ret;
+       }
+
+       ctrl->queue_count = opts->nr_io_queues + 1;
+       if (!opts->nr_io_queues)
+               return 0;
+
+       dev_info(ctrl->ctrl.device, "creating %d I/O queues.\n",
+                       opts->nr_io_queues);
+
+       nvme_fc_init_io_queues(ctrl);
+
+       memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
+       ctrl->tag_set.ops = &nvme_fc_mq_ops;
+       ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
+       ctrl->tag_set.reserved_tags = 1; /* fabric connect */
+       ctrl->tag_set.numa_node = NUMA_NO_NODE;
+       ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
+       ctrl->tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
+                                       (SG_CHUNK_SIZE *
+                                               sizeof(struct scatterlist)) +
+                                       ctrl->lport->ops->fcprqst_priv_sz;
+       ctrl->tag_set.driver_data = ctrl;
+       ctrl->tag_set.nr_hw_queues = ctrl->queue_count - 1;
+       ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
+
+       ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
+       if (ret)
+               return ret;
+
+       ctrl->ctrl.tagset = &ctrl->tag_set;
+
+       ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
+       if (IS_ERR(ctrl->ctrl.connect_q)) {
+               ret = PTR_ERR(ctrl->ctrl.connect_q);
+               goto out_free_tag_set;
+       }
+
+       ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
+       if (ret)
+               goto out_cleanup_blk_queue;
+
+       ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
+       if (ret)
+               goto out_delete_hw_queues;
+
+       return 0;
+
+out_delete_hw_queues:
+       nvme_fc_delete_hw_io_queues(ctrl);
+out_cleanup_blk_queue:
+       nvme_stop_keep_alive(&ctrl->ctrl);
+       blk_cleanup_queue(ctrl->ctrl.connect_q);
+out_free_tag_set:
+       blk_mq_free_tag_set(&ctrl->tag_set);
+       nvme_fc_free_io_queues(ctrl);
+
+       /* force put free routine to ignore io queues */
+       ctrl->ctrl.tagset = NULL;
+
+       return ret;
+}
+
+
+static struct nvme_ctrl *
+__nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,
+       struct nvme_fc_lport *lport, struct nvme_fc_rport *rport)
+{
+       struct nvme_fc_ctrl *ctrl;
+       unsigned long flags;
+       int ret, idx;
+       bool changed;
+
+       ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
+       if (!ctrl) {
+               ret = -ENOMEM;
+               goto out_fail;
+       }
+
+       idx = ida_simple_get(&nvme_fc_ctrl_cnt, 0, 0, GFP_KERNEL);
+       if (idx < 0) {
+               ret = -ENOSPC;
+               goto out_free_ctrl;
+       }
+
+       ctrl->ctrl.opts = opts;
+       INIT_LIST_HEAD(&ctrl->ctrl_list);
+       INIT_LIST_HEAD(&ctrl->ls_req_list);
+       ctrl->lport = lport;
+       ctrl->rport = rport;
+       ctrl->dev = lport->dev;
+       ctrl->state = FCCTRL_INIT;
+       ctrl->cnum = idx;
+
+       ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_fc_ctrl_ops, 0);
+       if (ret)
+               goto out_free_ida;
+
+       get_device(ctrl->dev);
+       kref_init(&ctrl->ref);
+
+       INIT_WORK(&ctrl->delete_work, nvme_fc_del_ctrl_work);
+       spin_lock_init(&ctrl->lock);
+
+       /* io queue count */
+       ctrl->queue_count = min_t(unsigned int,
+                               opts->nr_io_queues,
+                               lport->ops->max_hw_queues);
+       opts->nr_io_queues = ctrl->queue_count; /* so opts has valid value */
+       ctrl->queue_count++;    /* +1 for admin queue */
+
+       ctrl->ctrl.sqsize = opts->queue_size - 1;
+       ctrl->ctrl.kato = opts->kato;
+
+       ret = -ENOMEM;
+       ctrl->queues = kcalloc(ctrl->queue_count, sizeof(struct nvme_fc_queue),
+                               GFP_KERNEL);
+       if (!ctrl->queues)
+               goto out_uninit_ctrl;
+
+       ret = nvme_fc_configure_admin_queue(ctrl);
+       if (ret)
+               goto out_uninit_ctrl;
+
+       /* sanity checks */
+
+       /* FC-NVME supports 64-byte SQE only */
+       if (ctrl->ctrl.ioccsz != 4) {
+               dev_err(ctrl->ctrl.device, "ioccsz %d is not supported!\n",
+                               ctrl->ctrl.ioccsz);
+               goto out_remove_admin_queue;
+       }
+       /* FC-NVME supports 16-byte CQE only */
+       if (ctrl->ctrl.iorcsz != 1) {
+               dev_err(ctrl->ctrl.device, "iorcsz %d is not supported!\n",
+                               ctrl->ctrl.iorcsz);
+               goto out_remove_admin_queue;
+       }
+       /* FC-NVME does not have other data in the capsule */
+       if (ctrl->ctrl.icdoff) {
+               dev_err(ctrl->ctrl.device, "icdoff %d is not supported!\n",
+                               ctrl->ctrl.icdoff);
+               goto out_remove_admin_queue;
+       }
+
+       /* FC-NVME supports normal SGL Data Block Descriptors */
+
+       if (opts->queue_size > ctrl->ctrl.maxcmd) {
+               /* warn if maxcmd is lower than queue_size */
+               dev_warn(ctrl->ctrl.device,
+                       "queue_size %zu > ctrl maxcmd %u, reducing "
+                       "to queue_size\n",
+                       opts->queue_size, ctrl->ctrl.maxcmd);
+               opts->queue_size = ctrl->ctrl.maxcmd;
+       }
+
+       ret = nvme_fc_init_aen_ops(ctrl);
+       if (ret)
+               goto out_exit_aen_ops;
+
+       if (ctrl->queue_count > 1) {
+               ret = nvme_fc_create_io_queues(ctrl);
+               if (ret)
+                       goto out_exit_aen_ops;
+       }
+
+       spin_lock_irqsave(&ctrl->lock, flags);
+       ctrl->state = FCCTRL_ACTIVE;
+       spin_unlock_irqrestore(&ctrl->lock, flags);
+
+       changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
+       WARN_ON_ONCE(!changed);
+
+       dev_info(ctrl->ctrl.device,
+               "NVME-FC{%d}: new ctrl: NQN \"%s\" (%p)\n",
+               ctrl->cnum, ctrl->ctrl.opts->subsysnqn, &ctrl);
+
+       kref_get(&ctrl->ctrl.kref);
+
+       spin_lock_irqsave(&rport->lock, flags);
+       list_add_tail(&ctrl->ctrl_list, &rport->ctrl_list);
+       spin_unlock_irqrestore(&rport->lock, flags);
+
+       if (opts->nr_io_queues) {
+               nvme_queue_scan(&ctrl->ctrl);
+               nvme_queue_async_events(&ctrl->ctrl);
+       }
+
+       return &ctrl->ctrl;
+
+out_exit_aen_ops:
+       nvme_fc_exit_aen_ops(ctrl);
+out_remove_admin_queue:
+       /* send a Disconnect(association) LS to fc-nvme target */
+       nvme_fc_xmt_disconnect_assoc(ctrl);
+       nvme_stop_keep_alive(&ctrl->ctrl);
+       nvme_fc_destroy_admin_queue(ctrl);
+out_uninit_ctrl:
+       nvme_uninit_ctrl(&ctrl->ctrl);
+       nvme_put_ctrl(&ctrl->ctrl);
+       if (ret > 0)
+               ret = -EIO;
+       /* exit via here will follow ctlr ref point callbacks to free */
+       return ERR_PTR(ret);
+
+out_free_ida:
+       ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
+out_free_ctrl:
+       kfree(ctrl);
+out_fail:
+       nvme_fc_rport_put(rport);
+       /* exit via here doesn't follow ctlr ref points */
+       return ERR_PTR(ret);
+}
+
+enum {
+       FCT_TRADDR_ERR          = 0,
+       FCT_TRADDR_WWNN         = 1 << 0,
+       FCT_TRADDR_WWPN         = 1 << 1,
+};
+
+struct nvmet_fc_traddr {
+       u64     nn;
+       u64     pn;
+};
+
+static const match_table_t traddr_opt_tokens = {
+       { FCT_TRADDR_WWNN,      "nn-%s"         },
+       { FCT_TRADDR_WWPN,      "pn-%s"         },
+       { FCT_TRADDR_ERR,       NULL            }
+};
+
+static int
+nvme_fc_parse_address(struct nvmet_fc_traddr *traddr, char *buf)
+{
+       substring_t args[MAX_OPT_ARGS];
+       char *options, *o, *p;
+       int token, ret = 0;
+       u64 token64;
+
+       options = o = kstrdup(buf, GFP_KERNEL);
+       if (!options)
+               return -ENOMEM;
+
+       while ((p = strsep(&o, ":\n")) != NULL) {
+               if (!*p)
+                       continue;
+
+               token = match_token(p, traddr_opt_tokens, args);
+               switch (token) {
+               case FCT_TRADDR_WWNN:
+                       if (match_u64(args, &token64)) {
+                               ret = -EINVAL;
+                               goto out;
+                       }
+                       traddr->nn = token64;
+                       break;
+               case FCT_TRADDR_WWPN:
+                       if (match_u64(args, &token64)) {
+                               ret = -EINVAL;
+                               goto out;
+                       }
+                       traddr->pn = token64;
+                       break;
+               default:
+                       pr_warn("unknown traddr token or missing value '%s'\n",
+                                       p);
+                       ret = -EINVAL;
+                       goto out;
+               }
+       }
+
+out:
+       kfree(options);
+       return ret;
+}
+
+static struct nvme_ctrl *
+nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts)
+{
+       struct nvme_fc_lport *lport;
+       struct nvme_fc_rport *rport;
+       struct nvmet_fc_traddr laddr = { 0L, 0L };
+       struct nvmet_fc_traddr raddr = { 0L, 0L };
+       unsigned long flags;
+       int ret;
+
+       ret = nvme_fc_parse_address(&raddr, opts->traddr);
+       if (ret || !raddr.nn || !raddr.pn)
+               return ERR_PTR(-EINVAL);
+
+       ret = nvme_fc_parse_address(&laddr, opts->host_traddr);
+       if (ret || !laddr.nn || !laddr.pn)
+               return ERR_PTR(-EINVAL);
+
+       /* find the host and remote ports to connect together */
+       spin_lock_irqsave(&nvme_fc_lock, flags);
+       list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
+               if (lport->localport.node_name != laddr.nn ||
+                   lport->localport.port_name != laddr.pn)
+                       continue;
+
+               list_for_each_entry(rport, &lport->endp_list, endp_list) {
+                       if (rport->remoteport.node_name != raddr.nn ||
+                           rport->remoteport.port_name != raddr.pn)
+                               continue;
+
+                       /* if fail to get reference fall through. Will error */
+                       if (!nvme_fc_rport_get(rport))
+                               break;
+
+                       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+                       return __nvme_fc_create_ctrl(dev, opts, lport, rport);
+               }
+       }
+       spin_unlock_irqrestore(&nvme_fc_lock, flags);
+
+       return ERR_PTR(-ENOENT);
+}
+
+
+static struct nvmf_transport_ops nvme_fc_transport = {
+       .name           = "fc",
+       .required_opts  = NVMF_OPT_TRADDR | NVMF_OPT_HOST_TRADDR,
+       .allowed_opts   = NVMF_OPT_RECONNECT_DELAY,
+       .create_ctrl    = nvme_fc_create_ctrl,
+};
+
+static int __init nvme_fc_init_module(void)
+{
+       nvme_fc_wq = create_workqueue("nvme_fc_wq");
+       if (!nvme_fc_wq)
+               return -ENOMEM;
+
+       nvmf_register_transport(&nvme_fc_transport);
+       return 0;
+}
+
+static void __exit nvme_fc_exit_module(void)
+{
+       /* sanity check - all lports should be removed */
+       if (!list_empty(&nvme_fc_lport_list))
+               pr_warn("%s: localport list not empty\n", __func__);
+
+       nvmf_unregister_transport(&nvme_fc_transport);
+
+       destroy_workqueue(nvme_fc_wq);
+
+       ida_destroy(&nvme_fc_local_port_cnt);
+       ida_destroy(&nvme_fc_ctrl_cnt);
+}
+
+module_init(nvme_fc_init_module);
+module_exit(nvme_fc_exit_module);
+
+MODULE_LICENSE("GPL v2");