]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/commitdiff
nvme-fabrics: Add target support for FC transport
authorJames Smart <jsmart2021@gmail.com>
Fri, 2 Dec 2016 08:28:43 +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 perform the FCP transactions
necessary to perform and FCP IO request for NVME.

The T11 definitions for FC-4 Link Services are implemented which create
NVMeOF connections.  Implements the hooks with nvmet layer to pass NVME
commands to it for processing and posting of data/response base to the
host via 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/target/Kconfig
drivers/nvme/target/Makefile
drivers/nvme/target/fc.c [new file with mode: 0644]

index 00505700b30a31889370e480c785fe0c49e7ae28..1c2c9f96dd5639b68c2d46ceac49bb41d48c5e2f 100644 (file)
@@ -8666,6 +8666,7 @@ S:        Supported
 F:     include/linux/nvme-fc.h
 F:     include/linux/nvme-fc-driver.h
 F:     drivers/nvme/host/fc.c
+F:     drivers/nvme/target/fc.c
 
 NVMEM FRAMEWORK
 M:     Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
index 3a5b9d0576cb0f71a5bf22a34c905ae8dfc634fb..746a63e4d54aaae49d5d460d2588f24d67e1c4b5 100644 (file)
@@ -34,3 +34,14 @@ config NVME_TARGET_RDMA
          devices over RDMA.
 
          If unsure, say N.
+
+config NVME_TARGET_FC
+       tristate "NVMe over Fabrics FC target driver"
+       depends on NVME_TARGET
+       depends on HAS_DMA
+       help
+         This enables the NVMe FC target support, which allows exporting NVMe
+         devices over FC.
+
+         If unsure, say N.
+
index b7a06232c9da79004120ddbb2d4f5f2a26de6bc8..80b128b9ac9f00de251656fe1e196def98e9c498 100644 (file)
@@ -2,8 +2,10 @@
 obj-$(CONFIG_NVME_TARGET)              += nvmet.o
 obj-$(CONFIG_NVME_TARGET_LOOP)         += nvme-loop.o
 obj-$(CONFIG_NVME_TARGET_RDMA)         += nvmet-rdma.o
+obj-$(CONFIG_NVME_TARGET_FC)           += nvmet-fc.o
 
 nvmet-y                += core.o configfs.o admin-cmd.o io-cmd.o fabrics-cmd.o \
                        discovery.o
 nvme-loop-y    += loop.o
 nvmet-rdma-y   += rdma.o
+nvmet-fc-y     += fc.o
diff --git a/drivers/nvme/target/fc.c b/drivers/nvme/target/fc.c
new file mode 100644 (file)
index 0000000..173e842
--- /dev/null
@@ -0,0 +1,2288 @@
+/*
+ * 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/slab.h>
+#include <linux/blk-mq.h>
+#include <linux/parser.h>
+#include <linux/random.h>
+#include <uapi/scsi/fc/fc_fs.h>
+#include <uapi/scsi/fc/fc_els.h>
+
+#include "nvmet.h"
+#include <linux/nvme-fc-driver.h>
+#include <linux/nvme-fc.h>
+
+
+/* *************************** Data Structures/Defines ****************** */
+
+
+#define NVMET_LS_CTX_COUNT             4
+
+/* for this implementation, assume small single frame rqst/rsp */
+#define NVME_FC_MAX_LS_BUFFER_SIZE             2048
+
+struct nvmet_fc_tgtport;
+struct nvmet_fc_tgt_assoc;
+
+struct nvmet_fc_ls_iod {
+       struct nvmefc_tgt_ls_req        *lsreq;
+       struct nvmefc_tgt_fcp_req       *fcpreq;        /* only if RS */
+
+       struct list_head                ls_list;        /* tgtport->ls_list */
+
+       struct nvmet_fc_tgtport         *tgtport;
+       struct nvmet_fc_tgt_assoc       *assoc;
+
+       u8                              *rqstbuf;
+       u8                              *rspbuf;
+       u16                             rqstdatalen;
+       dma_addr_t                      rspdma;
+
+       struct scatterlist              sg[2];
+
+       struct work_struct              work;
+} __aligned(sizeof(unsigned long long));
+
+#define NVMET_FC_MAX_KB_PER_XFR                256
+
+enum nvmet_fcp_datadir {
+       NVMET_FCP_NODATA,
+       NVMET_FCP_WRITE,
+       NVMET_FCP_READ,
+       NVMET_FCP_ABORTED,
+};
+
+struct nvmet_fc_fcp_iod {
+       struct nvmefc_tgt_fcp_req       *fcpreq;
+
+       struct nvme_fc_cmd_iu           cmdiubuf;
+       struct nvme_fc_ersp_iu          rspiubuf;
+       dma_addr_t                      rspdma;
+       struct scatterlist              *data_sg;
+       struct scatterlist              *next_sg;
+       int                             data_sg_cnt;
+       u32                             next_sg_offset;
+       u32                             total_length;
+       u32                             offset;
+       enum nvmet_fcp_datadir          io_dir;
+       bool                            active;
+       bool                            abort;
+       spinlock_t                      flock;
+
+       struct nvmet_req                req;
+       struct work_struct              work;
+
+       struct nvmet_fc_tgtport         *tgtport;
+       struct nvmet_fc_tgt_queue       *queue;
+
+       struct list_head                fcp_list;       /* tgtport->fcp_list */
+};
+
+struct nvmet_fc_tgtport {
+
+       struct nvmet_fc_target_port     fc_target_port;
+
+       struct list_head                tgt_list; /* nvmet_fc_target_list */
+       struct device                   *dev;   /* dev for dma mapping */
+       struct nvmet_fc_target_template *ops;
+
+       struct nvmet_fc_ls_iod          *iod;
+       spinlock_t                      lock;
+       struct list_head                ls_list;
+       struct list_head                ls_busylist;
+       struct list_head                assoc_list;
+       struct ida                      assoc_cnt;
+       struct nvmet_port               *port;
+       struct kref                     ref;
+};
+
+struct nvmet_fc_tgt_queue {
+       bool                            ninetypercent;
+       u16                             qid;
+       u16                             sqsize;
+       u16                             ersp_ratio;
+       u16                             sqhd;
+       int                             cpu;
+       atomic_t                        connected;
+       atomic_t                        sqtail;
+       atomic_t                        zrspcnt;
+       atomic_t                        rsn;
+       spinlock_t                      qlock;
+       struct nvmet_port               *port;
+       struct nvmet_cq                 nvme_cq;
+       struct nvmet_sq                 nvme_sq;
+       struct nvmet_fc_tgt_assoc       *assoc;
+       struct nvmet_fc_fcp_iod         *fod;           /* array of fcp_iods */
+       struct list_head                fod_list;
+       struct workqueue_struct         *work_q;
+       struct kref                     ref;
+} __aligned(sizeof(unsigned long long));
+
+struct nvmet_fc_tgt_assoc {
+       u64                             association_id;
+       u32                             a_id;
+       struct nvmet_fc_tgtport         *tgtport;
+       struct list_head                a_list;
+       struct nvmet_fc_tgt_queue       *queues[NVMET_NR_QUEUES];
+       struct kref                     ref;
+};
+
+
+static inline int
+nvmet_fc_iodnum(struct nvmet_fc_ls_iod *iodptr)
+{
+       return (iodptr - iodptr->tgtport->iod);
+}
+
+static inline int
+nvmet_fc_fodnum(struct nvmet_fc_fcp_iod *fodptr)
+{
+       return (fodptr - fodptr->queue->fod);
+}
+
+
+/*
+ * Association and Connection IDs:
+ *
+ * Association ID will have random number in upper 6 bytes and zero
+ *   in lower 2 bytes
+ *
+ * Connection IDs will be Association ID with QID or'd in lower 2 bytes
+ *
+ * note: Association ID = Connection ID for queue 0
+ */
+#define BYTES_FOR_QID                  sizeof(u16)
+#define BYTES_FOR_QID_SHIFT            (BYTES_FOR_QID * 8)
+#define NVMET_FC_QUEUEID_MASK          ((u64)((1 << BYTES_FOR_QID_SHIFT) - 1))
+
+static inline u64
+nvmet_fc_makeconnid(struct nvmet_fc_tgt_assoc *assoc, u16 qid)
+{
+       return (assoc->association_id | qid);
+}
+
+static inline u64
+nvmet_fc_getassociationid(u64 connectionid)
+{
+       return connectionid & ~NVMET_FC_QUEUEID_MASK;
+}
+
+static inline u16
+nvmet_fc_getqueueid(u64 connectionid)
+{
+       return (u16)(connectionid & NVMET_FC_QUEUEID_MASK);
+}
+
+static inline struct nvmet_fc_tgtport *
+targetport_to_tgtport(struct nvmet_fc_target_port *targetport)
+{
+       return container_of(targetport, struct nvmet_fc_tgtport,
+                                fc_target_port);
+}
+
+static inline struct nvmet_fc_fcp_iod *
+nvmet_req_to_fod(struct nvmet_req *nvme_req)
+{
+       return container_of(nvme_req, struct nvmet_fc_fcp_iod, req);
+}
+
+
+/* *************************** Globals **************************** */
+
+
+static DEFINE_SPINLOCK(nvmet_fc_tgtlock);
+
+static LIST_HEAD(nvmet_fc_target_list);
+static DEFINE_IDA(nvmet_fc_tgtport_cnt);
+
+
+static void nvmet_fc_handle_ls_rqst_work(struct work_struct *work);
+static void nvmet_fc_handle_fcp_rqst_work(struct work_struct *work);
+static void nvmet_fc_tgt_a_put(struct nvmet_fc_tgt_assoc *assoc);
+static int nvmet_fc_tgt_a_get(struct nvmet_fc_tgt_assoc *assoc);
+static void nvmet_fc_tgt_q_put(struct nvmet_fc_tgt_queue *queue);
+static int nvmet_fc_tgt_q_get(struct nvmet_fc_tgt_queue *queue);
+static void nvmet_fc_tgtport_put(struct nvmet_fc_tgtport *tgtport);
+static int nvmet_fc_tgtport_get(struct nvmet_fc_tgtport *tgtport);
+
+
+/* *********************** 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 Port Management ************************ */
+
+
+static int
+nvmet_fc_alloc_ls_iodlist(struct nvmet_fc_tgtport *tgtport)
+{
+       struct nvmet_fc_ls_iod *iod;
+       int i;
+
+       iod = kcalloc(NVMET_LS_CTX_COUNT, sizeof(struct nvmet_fc_ls_iod),
+                       GFP_KERNEL);
+       if (!iod)
+               return -ENOMEM;
+
+       tgtport->iod = iod;
+
+       for (i = 0; i < NVMET_LS_CTX_COUNT; iod++, i++) {
+               INIT_WORK(&iod->work, nvmet_fc_handle_ls_rqst_work);
+               iod->tgtport = tgtport;
+               list_add_tail(&iod->ls_list, &tgtport->ls_list);
+
+               iod->rqstbuf = kcalloc(2, NVME_FC_MAX_LS_BUFFER_SIZE,
+                       GFP_KERNEL);
+               if (!iod->rqstbuf)
+                       goto out_fail;
+
+               iod->rspbuf = iod->rqstbuf + NVME_FC_MAX_LS_BUFFER_SIZE;
+
+               iod->rspdma = fc_dma_map_single(tgtport->dev, iod->rspbuf,
+                                               NVME_FC_MAX_LS_BUFFER_SIZE,
+                                               DMA_TO_DEVICE);
+               if (fc_dma_mapping_error(tgtport->dev, iod->rspdma))
+                       goto out_fail;
+       }
+
+       return 0;
+
+out_fail:
+       kfree(iod->rqstbuf);
+       list_del(&iod->ls_list);
+       for (iod--, i--; i >= 0; iod--, i--) {
+               fc_dma_unmap_single(tgtport->dev, iod->rspdma,
+                               NVME_FC_MAX_LS_BUFFER_SIZE, DMA_TO_DEVICE);
+               kfree(iod->rqstbuf);
+               list_del(&iod->ls_list);
+       }
+
+       kfree(iod);
+
+       return -EFAULT;
+}
+
+static void
+nvmet_fc_free_ls_iodlist(struct nvmet_fc_tgtport *tgtport)
+{
+       struct nvmet_fc_ls_iod *iod = tgtport->iod;
+       int i;
+
+       for (i = 0; i < NVMET_LS_CTX_COUNT; iod++, i++) {
+               fc_dma_unmap_single(tgtport->dev,
+                               iod->rspdma, NVME_FC_MAX_LS_BUFFER_SIZE,
+                               DMA_TO_DEVICE);
+               kfree(iod->rqstbuf);
+               list_del(&iod->ls_list);
+       }
+       kfree(tgtport->iod);
+}
+
+static struct nvmet_fc_ls_iod *
+nvmet_fc_alloc_ls_iod(struct nvmet_fc_tgtport *tgtport)
+{
+       static struct nvmet_fc_ls_iod *iod;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       iod = list_first_entry_or_null(&tgtport->ls_list,
+                                       struct nvmet_fc_ls_iod, ls_list);
+       if (iod)
+               list_move_tail(&iod->ls_list, &tgtport->ls_busylist);
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+       return iod;
+}
+
+
+static void
+nvmet_fc_free_ls_iod(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_ls_iod *iod)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       list_move(&iod->ls_list, &tgtport->ls_list);
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+}
+
+static void
+nvmet_fc_prep_fcp_iodlist(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmet_fc_tgt_queue *queue)
+{
+       struct nvmet_fc_fcp_iod *fod = queue->fod;
+       int i;
+
+       for (i = 0; i < queue->sqsize; fod++, i++) {
+               INIT_WORK(&fod->work, nvmet_fc_handle_fcp_rqst_work);
+               fod->tgtport = tgtport;
+               fod->queue = queue;
+               fod->active = false;
+               list_add_tail(&fod->fcp_list, &queue->fod_list);
+               spin_lock_init(&fod->flock);
+
+               fod->rspdma = fc_dma_map_single(tgtport->dev, &fod->rspiubuf,
+                                       sizeof(fod->rspiubuf), DMA_TO_DEVICE);
+               if (fc_dma_mapping_error(tgtport->dev, fod->rspdma)) {
+                       list_del(&fod->fcp_list);
+                       for (fod--, i--; i >= 0; fod--, i--) {
+                               fc_dma_unmap_single(tgtport->dev, fod->rspdma,
+                                               sizeof(fod->rspiubuf),
+                                               DMA_TO_DEVICE);
+                               fod->rspdma = 0L;
+                               list_del(&fod->fcp_list);
+                       }
+
+                       return;
+               }
+       }
+}
+
+static void
+nvmet_fc_destroy_fcp_iodlist(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmet_fc_tgt_queue *queue)
+{
+       struct nvmet_fc_fcp_iod *fod = queue->fod;
+       int i;
+
+       for (i = 0; i < queue->sqsize; fod++, i++) {
+               if (fod->rspdma)
+                       fc_dma_unmap_single(tgtport->dev, fod->rspdma,
+                               sizeof(fod->rspiubuf), DMA_TO_DEVICE);
+       }
+}
+
+static struct nvmet_fc_fcp_iod *
+nvmet_fc_alloc_fcp_iod(struct nvmet_fc_tgt_queue *queue)
+{
+       static struct nvmet_fc_fcp_iod *fod;
+       unsigned long flags;
+
+       spin_lock_irqsave(&queue->qlock, flags);
+       fod = list_first_entry_or_null(&queue->fod_list,
+                                       struct nvmet_fc_fcp_iod, fcp_list);
+       if (fod) {
+               list_del(&fod->fcp_list);
+               fod->active = true;
+               fod->abort = false;
+               /*
+                * no queue reference is taken, as it was taken by the
+                * queue lookup just prior to the allocation. The iod
+                * will "inherit" that reference.
+                */
+       }
+       spin_unlock_irqrestore(&queue->qlock, flags);
+       return fod;
+}
+
+
+static void
+nvmet_fc_free_fcp_iod(struct nvmet_fc_tgt_queue *queue,
+                       struct nvmet_fc_fcp_iod *fod)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&queue->qlock, flags);
+       list_add_tail(&fod->fcp_list, &fod->queue->fod_list);
+       fod->active = false;
+       spin_unlock_irqrestore(&queue->qlock, flags);
+
+       /*
+        * release the reference taken at queue lookup and fod allocation
+        */
+       nvmet_fc_tgt_q_put(queue);
+}
+
+static int
+nvmet_fc_queue_to_cpu(struct nvmet_fc_tgtport *tgtport, int qid)
+{
+       int cpu, idx, cnt;
+
+       if (!(tgtport->ops->target_features &
+                       NVMET_FCTGTFEAT_NEEDS_CMD_CPUSCHED) ||
+           tgtport->ops->max_hw_queues == 1)
+               return WORK_CPU_UNBOUND;
+
+       /* Simple cpu selection based on qid modulo active cpu count */
+       idx = !qid ? 0 : (qid - 1) % num_active_cpus();
+
+       /* find the n'th active cpu */
+       for (cpu = 0, cnt = 0; ; ) {
+               if (cpu_active(cpu)) {
+                       if (cnt == idx)
+                               break;
+                       cnt++;
+               }
+               cpu = (cpu + 1) % num_possible_cpus();
+       }
+
+       return cpu;
+}
+
+static struct nvmet_fc_tgt_queue *
+nvmet_fc_alloc_target_queue(struct nvmet_fc_tgt_assoc *assoc,
+                       u16 qid, u16 sqsize)
+{
+       struct nvmet_fc_tgt_queue *queue;
+       unsigned long flags;
+       int ret;
+
+       if (qid >= NVMET_NR_QUEUES)
+               return NULL;
+
+       queue = kzalloc((sizeof(*queue) +
+                               (sizeof(struct nvmet_fc_fcp_iod) * sqsize)),
+                               GFP_KERNEL);
+       if (!queue)
+               return NULL;
+
+       if (!nvmet_fc_tgt_a_get(assoc))
+               goto out_free_queue;
+
+       queue->work_q = alloc_workqueue("ntfc%d.%d.%d", 0, 0,
+                               assoc->tgtport->fc_target_port.port_num,
+                               assoc->a_id, qid);
+       if (!queue->work_q)
+               goto out_a_put;
+
+       queue->fod = (struct nvmet_fc_fcp_iod *)&queue[1];
+       queue->qid = qid;
+       queue->sqsize = sqsize;
+       queue->assoc = assoc;
+       queue->port = assoc->tgtport->port;
+       queue->cpu = nvmet_fc_queue_to_cpu(assoc->tgtport, qid);
+       INIT_LIST_HEAD(&queue->fod_list);
+       atomic_set(&queue->connected, 0);
+       atomic_set(&queue->sqtail, 0);
+       atomic_set(&queue->rsn, 1);
+       atomic_set(&queue->zrspcnt, 0);
+       spin_lock_init(&queue->qlock);
+       kref_init(&queue->ref);
+
+       nvmet_fc_prep_fcp_iodlist(assoc->tgtport, queue);
+
+       ret = nvmet_sq_init(&queue->nvme_sq);
+       if (ret)
+               goto out_fail_iodlist;
+
+       WARN_ON(assoc->queues[qid]);
+       spin_lock_irqsave(&assoc->tgtport->lock, flags);
+       assoc->queues[qid] = queue;
+       spin_unlock_irqrestore(&assoc->tgtport->lock, flags);
+
+       return queue;
+
+out_fail_iodlist:
+       nvmet_fc_destroy_fcp_iodlist(assoc->tgtport, queue);
+       destroy_workqueue(queue->work_q);
+out_a_put:
+       nvmet_fc_tgt_a_put(assoc);
+out_free_queue:
+       kfree(queue);
+       return NULL;
+}
+
+
+static void
+nvmet_fc_tgt_queue_free(struct kref *ref)
+{
+       struct nvmet_fc_tgt_queue *queue =
+               container_of(ref, struct nvmet_fc_tgt_queue, ref);
+       unsigned long flags;
+
+       spin_lock_irqsave(&queue->assoc->tgtport->lock, flags);
+       queue->assoc->queues[queue->qid] = NULL;
+       spin_unlock_irqrestore(&queue->assoc->tgtport->lock, flags);
+
+       nvmet_fc_destroy_fcp_iodlist(queue->assoc->tgtport, queue);
+
+       nvmet_fc_tgt_a_put(queue->assoc);
+
+       destroy_workqueue(queue->work_q);
+
+       kfree(queue);
+}
+
+static void
+nvmet_fc_tgt_q_put(struct nvmet_fc_tgt_queue *queue)
+{
+       kref_put(&queue->ref, nvmet_fc_tgt_queue_free);
+}
+
+static int
+nvmet_fc_tgt_q_get(struct nvmet_fc_tgt_queue *queue)
+{
+       return kref_get_unless_zero(&queue->ref);
+}
+
+
+static void
+nvmet_fc_abort_op(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmefc_tgt_fcp_req *fcpreq)
+{
+       int ret;
+
+       fcpreq->op = NVMET_FCOP_ABORT;
+       fcpreq->offset = 0;
+       fcpreq->timeout = 0;
+       fcpreq->transfer_length = 0;
+       fcpreq->transferred_length = 0;
+       fcpreq->fcp_error = 0;
+       fcpreq->sg_cnt = 0;
+
+       ret = tgtport->ops->fcp_op(&tgtport->fc_target_port, fcpreq);
+       if (ret)
+               /* should never reach here !! */
+               WARN_ON(1);
+}
+
+
+static void
+nvmet_fc_delete_target_queue(struct nvmet_fc_tgt_queue *queue)
+{
+       struct nvmet_fc_fcp_iod *fod = queue->fod;
+       unsigned long flags;
+       int i;
+       bool disconnect;
+
+       disconnect = atomic_xchg(&queue->connected, 0);
+
+       spin_lock_irqsave(&queue->qlock, flags);
+       /* about outstanding io's */
+       for (i = 0; i < queue->sqsize; fod++, i++) {
+               if (fod->active) {
+                       spin_lock(&fod->flock);
+                       fod->abort = true;
+                       spin_unlock(&fod->flock);
+               }
+       }
+       spin_unlock_irqrestore(&queue->qlock, flags);
+
+       flush_workqueue(queue->work_q);
+
+       if (disconnect)
+               nvmet_sq_destroy(&queue->nvme_sq);
+
+       nvmet_fc_tgt_q_put(queue);
+}
+
+static struct nvmet_fc_tgt_queue *
+nvmet_fc_find_target_queue(struct nvmet_fc_tgtport *tgtport,
+                               u64 connection_id)
+{
+       struct nvmet_fc_tgt_assoc *assoc;
+       struct nvmet_fc_tgt_queue *queue;
+       u64 association_id = nvmet_fc_getassociationid(connection_id);
+       u16 qid = nvmet_fc_getqueueid(connection_id);
+       unsigned long flags;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       list_for_each_entry(assoc, &tgtport->assoc_list, a_list) {
+               if (association_id == assoc->association_id) {
+                       queue = assoc->queues[qid];
+                       if (queue &&
+                           (!atomic_read(&queue->connected) ||
+                            !nvmet_fc_tgt_q_get(queue)))
+                               queue = NULL;
+                       spin_unlock_irqrestore(&tgtport->lock, flags);
+                       return queue;
+               }
+       }
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+       return NULL;
+}
+
+static struct nvmet_fc_tgt_assoc *
+nvmet_fc_alloc_target_assoc(struct nvmet_fc_tgtport *tgtport)
+{
+       struct nvmet_fc_tgt_assoc *assoc, *tmpassoc;
+       unsigned long flags;
+       u64 ran;
+       int idx;
+       bool needrandom = true;
+
+       assoc = kzalloc(sizeof(*assoc), GFP_KERNEL);
+       if (!assoc)
+               return NULL;
+
+       idx = ida_simple_get(&tgtport->assoc_cnt, 0, 0, GFP_KERNEL);
+       if (idx < 0)
+               goto out_free_assoc;
+
+       if (!nvmet_fc_tgtport_get(tgtport))
+               goto out_ida_put;
+
+       assoc->tgtport = tgtport;
+       assoc->a_id = idx;
+       INIT_LIST_HEAD(&assoc->a_list);
+       kref_init(&assoc->ref);
+
+       while (needrandom) {
+               get_random_bytes(&ran, sizeof(ran) - BYTES_FOR_QID);
+               ran = ran << BYTES_FOR_QID_SHIFT;
+
+               spin_lock_irqsave(&tgtport->lock, flags);
+               needrandom = false;
+               list_for_each_entry(tmpassoc, &tgtport->assoc_list, a_list)
+                       if (ran == tmpassoc->association_id) {
+                               needrandom = true;
+                               break;
+                       }
+               if (!needrandom) {
+                       assoc->association_id = ran;
+                       list_add_tail(&assoc->a_list, &tgtport->assoc_list);
+               }
+               spin_unlock_irqrestore(&tgtport->lock, flags);
+       }
+
+       return assoc;
+
+out_ida_put:
+       ida_simple_remove(&tgtport->assoc_cnt, idx);
+out_free_assoc:
+       kfree(assoc);
+       return NULL;
+}
+
+static void
+nvmet_fc_target_assoc_free(struct kref *ref)
+{
+       struct nvmet_fc_tgt_assoc *assoc =
+               container_of(ref, struct nvmet_fc_tgt_assoc, ref);
+       struct nvmet_fc_tgtport *tgtport = assoc->tgtport;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       list_del(&assoc->a_list);
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+       ida_simple_remove(&tgtport->assoc_cnt, assoc->a_id);
+       kfree(assoc);
+       nvmet_fc_tgtport_put(tgtport);
+}
+
+static void
+nvmet_fc_tgt_a_put(struct nvmet_fc_tgt_assoc *assoc)
+{
+       kref_put(&assoc->ref, nvmet_fc_target_assoc_free);
+}
+
+static int
+nvmet_fc_tgt_a_get(struct nvmet_fc_tgt_assoc *assoc)
+{
+       return kref_get_unless_zero(&assoc->ref);
+}
+
+static void
+nvmet_fc_delete_target_assoc(struct nvmet_fc_tgt_assoc *assoc)
+{
+       struct nvmet_fc_tgtport *tgtport = assoc->tgtport;
+       struct nvmet_fc_tgt_queue *queue;
+       unsigned long flags;
+       int i;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       for (i = NVMET_NR_QUEUES - 1; i >= 0; i--) {
+               queue = assoc->queues[i];
+               if (queue) {
+                       if (!nvmet_fc_tgt_q_get(queue))
+                               continue;
+                       spin_unlock_irqrestore(&tgtport->lock, flags);
+                       nvmet_fc_delete_target_queue(queue);
+                       nvmet_fc_tgt_q_put(queue);
+                       spin_lock_irqsave(&tgtport->lock, flags);
+               }
+       }
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+
+       nvmet_fc_tgt_a_put(assoc);
+}
+
+static struct nvmet_fc_tgt_assoc *
+nvmet_fc_find_target_assoc(struct nvmet_fc_tgtport *tgtport,
+                               u64 association_id)
+{
+       struct nvmet_fc_tgt_assoc *assoc;
+       struct nvmet_fc_tgt_assoc *ret = NULL;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       list_for_each_entry(assoc, &tgtport->assoc_list, a_list) {
+               if (association_id == assoc->association_id) {
+                       ret = assoc;
+                       nvmet_fc_tgt_a_get(assoc);
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+
+       return ret;
+}
+
+
+/**
+ * nvme_fc_register_targetport - transport entry point called by an
+ *                              LLDD to register the existence of a local
+ *                              NVME subystem 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
+ * @portptr:   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 NULL.
+ *
+ * Returns:
+ * a completion status. Must be 0 upon success; a negative errno
+ * (ex: -ENXIO) upon failure.
+ */
+int
+nvmet_fc_register_targetport(struct nvmet_fc_port_info *pinfo,
+                       struct nvmet_fc_target_template *template,
+                       struct device *dev,
+                       struct nvmet_fc_target_port **portptr)
+{
+       struct nvmet_fc_tgtport *newrec;
+       unsigned long flags;
+       int ret, idx;
+
+       if (!template->xmt_ls_rsp || !template->fcp_op ||
+           !template->targetport_delete ||
+           !template->max_hw_queues || !template->max_sgl_segments ||
+           !template->max_dif_sgl_segments || !template->dma_boundary) {
+               ret = -EINVAL;
+               goto out_regtgt_failed;
+       }
+
+       newrec = kzalloc((sizeof(*newrec) + template->target_priv_sz),
+                        GFP_KERNEL);
+       if (!newrec) {
+               ret = -ENOMEM;
+               goto out_regtgt_failed;
+       }
+
+       idx = ida_simple_get(&nvmet_fc_tgtport_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;
+       }
+
+       newrec->fc_target_port.node_name = pinfo->node_name;
+       newrec->fc_target_port.port_name = pinfo->port_name;
+       newrec->fc_target_port.private = &newrec[1];
+       newrec->fc_target_port.port_id = pinfo->port_id;
+       newrec->fc_target_port.port_num = idx;
+       INIT_LIST_HEAD(&newrec->tgt_list);
+       newrec->dev = dev;
+       newrec->ops = template;
+       spin_lock_init(&newrec->lock);
+       INIT_LIST_HEAD(&newrec->ls_list);
+       INIT_LIST_HEAD(&newrec->ls_busylist);
+       INIT_LIST_HEAD(&newrec->assoc_list);
+       kref_init(&newrec->ref);
+       ida_init(&newrec->assoc_cnt);
+
+       ret = nvmet_fc_alloc_ls_iodlist(newrec);
+       if (ret) {
+               ret = -ENOMEM;
+               goto out_free_newrec;
+       }
+
+       spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
+       list_add_tail(&newrec->tgt_list, &nvmet_fc_target_list);
+       spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
+
+       *portptr = &newrec->fc_target_port;
+       return 0;
+
+out_free_newrec:
+       put_device(dev);
+out_ida_put:
+       ida_simple_remove(&nvmet_fc_tgtport_cnt, idx);
+out_fail_kfree:
+       kfree(newrec);
+out_regtgt_failed:
+       *portptr = NULL;
+       return ret;
+}
+EXPORT_SYMBOL_GPL(nvmet_fc_register_targetport);
+
+
+static void
+nvmet_fc_free_tgtport(struct kref *ref)
+{
+       struct nvmet_fc_tgtport *tgtport =
+               container_of(ref, struct nvmet_fc_tgtport, ref);
+       struct device *dev = tgtport->dev;
+       unsigned long flags;
+
+       spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
+       list_del(&tgtport->tgt_list);
+       spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
+
+       nvmet_fc_free_ls_iodlist(tgtport);
+
+       /* let the LLDD know we've finished tearing it down */
+       tgtport->ops->targetport_delete(&tgtport->fc_target_port);
+
+       ida_simple_remove(&nvmet_fc_tgtport_cnt,
+                       tgtport->fc_target_port.port_num);
+
+       ida_destroy(&tgtport->assoc_cnt);
+
+       kfree(tgtport);
+
+       put_device(dev);
+}
+
+static void
+nvmet_fc_tgtport_put(struct nvmet_fc_tgtport *tgtport)
+{
+       kref_put(&tgtport->ref, nvmet_fc_free_tgtport);
+}
+
+static int
+nvmet_fc_tgtport_get(struct nvmet_fc_tgtport *tgtport)
+{
+       return kref_get_unless_zero(&tgtport->ref);
+}
+
+static void
+__nvmet_fc_free_assocs(struct nvmet_fc_tgtport *tgtport)
+{
+       struct nvmet_fc_tgt_assoc *assoc, *next;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tgtport->lock, flags);
+       list_for_each_entry_safe(assoc, next,
+                               &tgtport->assoc_list, a_list) {
+               if (!nvmet_fc_tgt_a_get(assoc))
+                       continue;
+               spin_unlock_irqrestore(&tgtport->lock, flags);
+               nvmet_fc_delete_target_assoc(assoc);
+               nvmet_fc_tgt_a_put(assoc);
+               spin_lock_irqsave(&tgtport->lock, flags);
+       }
+       spin_unlock_irqrestore(&tgtport->lock, flags);
+}
+
+/*
+ * nvmet layer has called to terminate an association
+ */
+static void
+nvmet_fc_delete_ctrl(struct nvmet_ctrl *ctrl)
+{
+       struct nvmet_fc_tgtport *tgtport, *next;
+       struct nvmet_fc_tgt_assoc *assoc;
+       struct nvmet_fc_tgt_queue *queue;
+       unsigned long flags;
+       bool found_ctrl = false;
+
+       /* this is a bit ugly, but don't want to make locks layered */
+       spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
+       list_for_each_entry_safe(tgtport, next, &nvmet_fc_target_list,
+                       tgt_list) {
+               if (!nvmet_fc_tgtport_get(tgtport))
+                       continue;
+               spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
+
+               spin_lock_irqsave(&tgtport->lock, flags);
+               list_for_each_entry(assoc, &tgtport->assoc_list, a_list) {
+                       queue = assoc->queues[0];
+                       if (queue && queue->nvme_sq.ctrl == ctrl) {
+                               if (nvmet_fc_tgt_a_get(assoc))
+                                       found_ctrl = true;
+                               break;
+                       }
+               }
+               spin_unlock_irqrestore(&tgtport->lock, flags);
+
+               nvmet_fc_tgtport_put(tgtport);
+
+               if (found_ctrl) {
+                       nvmet_fc_delete_target_assoc(assoc);
+                       nvmet_fc_tgt_a_put(assoc);
+                       return;
+               }
+
+               spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
+       }
+       spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
+}
+
+/**
+ * nvme_fc_unregister_targetport - transport entry point called by an
+ *                              LLDD to deregister/remove a previously
+ *                              registered a local NVME subsystem FC port.
+ * @tgtport: pointer to the (registered) target port that is to be
+ *           deregistered.
+ *
+ * Returns:
+ * a completion status. Must be 0 upon success; a negative errno
+ * (ex: -ENXIO) upon failure.
+ */
+int
+nvmet_fc_unregister_targetport(struct nvmet_fc_target_port *target_port)
+{
+       struct nvmet_fc_tgtport *tgtport = targetport_to_tgtport(target_port);
+
+       /* terminate any outstanding associations */
+       __nvmet_fc_free_assocs(tgtport);
+
+       nvmet_fc_tgtport_put(tgtport);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(nvmet_fc_unregister_targetport);
+
+
+/* *********************** FC-NVME LS Handling **************************** */
+
+
+static void
+nvmet_fc_format_rsp_hdr(void *buf, u8 ls_cmd, u32 desc_len, u8 rqst_ls_cmd)
+{
+       struct fcnvme_ls_acc_hdr *acc = buf;
+
+       acc->w0.ls_cmd = ls_cmd;
+       acc->desc_list_len = desc_len;
+       acc->rqst.desc_tag = cpu_to_be32(FCNVME_LSDESC_RQST);
+       acc->rqst.desc_len =
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst));
+       acc->rqst.w0.ls_cmd = rqst_ls_cmd;
+}
+
+static int
+nvmet_fc_format_rjt(void *buf, u16 buflen, u8 ls_cmd,
+                       u8 reason, u8 explanation, u8 vendor)
+{
+       struct fcnvme_ls_rjt *rjt = buf;
+
+       nvmet_fc_format_rsp_hdr(buf, FCNVME_LSDESC_RQST,
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_rjt)),
+                       ls_cmd);
+       rjt->rjt.desc_tag = cpu_to_be32(FCNVME_LSDESC_RJT);
+       rjt->rjt.desc_len = fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rjt));
+       rjt->rjt.reason_code = reason;
+       rjt->rjt.reason_explanation = explanation;
+       rjt->rjt.vendor = vendor;
+
+       return sizeof(struct fcnvme_ls_rjt);
+}
+
+/* Validation Error indexes into the string table below */
+enum {
+       VERR_NO_ERROR           = 0,
+       VERR_CR_ASSOC_LEN       = 1,
+       VERR_CR_ASSOC_RQST_LEN  = 2,
+       VERR_CR_ASSOC_CMD       = 3,
+       VERR_CR_ASSOC_CMD_LEN   = 4,
+       VERR_ERSP_RATIO         = 5,
+       VERR_ASSOC_ALLOC_FAIL   = 6,
+       VERR_QUEUE_ALLOC_FAIL   = 7,
+       VERR_CR_CONN_LEN        = 8,
+       VERR_CR_CONN_RQST_LEN   = 9,
+       VERR_ASSOC_ID           = 10,
+       VERR_ASSOC_ID_LEN       = 11,
+       VERR_NO_ASSOC           = 12,
+       VERR_CONN_ID            = 13,
+       VERR_CONN_ID_LEN        = 14,
+       VERR_NO_CONN            = 15,
+       VERR_CR_CONN_CMD        = 16,
+       VERR_CR_CONN_CMD_LEN    = 17,
+       VERR_DISCONN_LEN        = 18,
+       VERR_DISCONN_RQST_LEN   = 19,
+       VERR_DISCONN_CMD        = 20,
+       VERR_DISCONN_CMD_LEN    = 21,
+       VERR_DISCONN_SCOPE      = 22,
+       VERR_RS_LEN             = 23,
+       VERR_RS_RQST_LEN        = 24,
+       VERR_RS_CMD             = 25,
+       VERR_RS_CMD_LEN         = 26,
+       VERR_RS_RCTL            = 27,
+       VERR_RS_RO              = 28,
+};
+
+static char *validation_errors[] = {
+       "OK",
+       "Bad CR_ASSOC Length",
+       "Bad CR_ASSOC Rqst Length",
+       "Not CR_ASSOC Cmd",
+       "Bad CR_ASSOC Cmd Length",
+       "Bad Ersp Ratio",
+       "Association Allocation Failed",
+       "Queue Allocation Failed",
+       "Bad CR_CONN Length",
+       "Bad CR_CONN Rqst Length",
+       "Not Association ID",
+       "Bad Association ID Length",
+       "No Association",
+       "Not Connection ID",
+       "Bad Connection ID Length",
+       "No Connection",
+       "Not CR_CONN Cmd",
+       "Bad CR_CONN Cmd Length",
+       "Bad DISCONN Length",
+       "Bad DISCONN Rqst Length",
+       "Not DISCONN Cmd",
+       "Bad DISCONN Cmd Length",
+       "Bad Disconnect Scope",
+       "Bad RS Length",
+       "Bad RS Rqst Length",
+       "Not RS Cmd",
+       "Bad RS Cmd Length",
+       "Bad RS R_CTL",
+       "Bad RS Relative Offset",
+};
+
+static void
+nvmet_fc_ls_create_association(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_ls_iod *iod)
+{
+       struct fcnvme_ls_cr_assoc_rqst *rqst =
+                               (struct fcnvme_ls_cr_assoc_rqst *)iod->rqstbuf;
+       struct fcnvme_ls_cr_assoc_acc *acc =
+                               (struct fcnvme_ls_cr_assoc_acc *)iod->rspbuf;
+       struct nvmet_fc_tgt_queue *queue;
+       int ret = 0;
+
+       memset(acc, 0, sizeof(*acc));
+
+       if (iod->rqstdatalen < sizeof(struct fcnvme_ls_cr_assoc_rqst))
+               ret = VERR_CR_ASSOC_LEN;
+       else if (rqst->desc_list_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_ls_cr_assoc_rqst)))
+               ret = VERR_CR_ASSOC_RQST_LEN;
+       else if (rqst->assoc_cmd.desc_tag !=
+                       cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD))
+               ret = VERR_CR_ASSOC_CMD;
+       else if (rqst->assoc_cmd.desc_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_cr_assoc_cmd)))
+               ret = VERR_CR_ASSOC_CMD_LEN;
+       else if (!rqst->assoc_cmd.ersp_ratio ||
+                (be16_to_cpu(rqst->assoc_cmd.ersp_ratio) >=
+                               be16_to_cpu(rqst->assoc_cmd.sqsize)))
+               ret = VERR_ERSP_RATIO;
+
+       else {
+               /* new association w/ admin queue */
+               iod->assoc = nvmet_fc_alloc_target_assoc(tgtport);
+               if (!iod->assoc)
+                       ret = VERR_ASSOC_ALLOC_FAIL;
+               else {
+                       queue = nvmet_fc_alloc_target_queue(iod->assoc, 0,
+                                       be16_to_cpu(rqst->assoc_cmd.sqsize));
+                       if (!queue)
+                               ret = VERR_QUEUE_ALLOC_FAIL;
+               }
+       }
+
+       if (ret) {
+               dev_err(tgtport->dev,
+                       "Create Association LS failed: %s\n",
+                       validation_errors[ret]);
+               iod->lsreq->rsplen = nvmet_fc_format_rjt(acc,
+                               NVME_FC_MAX_LS_BUFFER_SIZE, rqst->w0.ls_cmd,
+                               ELS_RJT_LOGIC,
+                               ELS_EXPL_NONE, 0);
+               return;
+       }
+
+       queue->ersp_ratio = be16_to_cpu(rqst->assoc_cmd.ersp_ratio);
+       atomic_set(&queue->connected, 1);
+       queue->sqhd = 0;        /* best place to init value */
+
+       /* format a response */
+
+       iod->lsreq->rsplen = sizeof(*acc);
+
+       nvmet_fc_format_rsp_hdr(acc, FCNVME_LS_ACC,
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_ls_cr_assoc_acc)),
+                       FCNVME_LS_CREATE_ASSOCIATION);
+       acc->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
+       acc->associd.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_assoc_id));
+       acc->associd.association_id =
+                       cpu_to_be64(nvmet_fc_makeconnid(iod->assoc, 0));
+       acc->connectid.desc_tag = cpu_to_be32(FCNVME_LSDESC_CONN_ID);
+       acc->connectid.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_conn_id));
+       acc->connectid.connection_id = acc->associd.association_id;
+}
+
+static void
+nvmet_fc_ls_create_connection(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_ls_iod *iod)
+{
+       struct fcnvme_ls_cr_conn_rqst *rqst =
+                               (struct fcnvme_ls_cr_conn_rqst *)iod->rqstbuf;
+       struct fcnvme_ls_cr_conn_acc *acc =
+                               (struct fcnvme_ls_cr_conn_acc *)iod->rspbuf;
+       struct nvmet_fc_tgt_queue *queue;
+       int ret = 0;
+
+       memset(acc, 0, sizeof(*acc));
+
+       if (iod->rqstdatalen < sizeof(struct fcnvme_ls_cr_conn_rqst))
+               ret = VERR_CR_CONN_LEN;
+       else if (rqst->desc_list_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_ls_cr_conn_rqst)))
+               ret = VERR_CR_CONN_RQST_LEN;
+       else if (rqst->associd.desc_tag != cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
+               ret = VERR_ASSOC_ID;
+       else if (rqst->associd.desc_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_assoc_id)))
+               ret = VERR_ASSOC_ID_LEN;
+       else if (rqst->connect_cmd.desc_tag !=
+                       cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD))
+               ret = VERR_CR_CONN_CMD;
+       else if (rqst->connect_cmd.desc_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_cr_conn_cmd)))
+               ret = VERR_CR_CONN_CMD_LEN;
+       else if (!rqst->connect_cmd.ersp_ratio ||
+                (be16_to_cpu(rqst->connect_cmd.ersp_ratio) >=
+                               be16_to_cpu(rqst->connect_cmd.sqsize)))
+               ret = VERR_ERSP_RATIO;
+
+       else {
+               /* new io queue */
+               iod->assoc = nvmet_fc_find_target_assoc(tgtport,
+                               be64_to_cpu(rqst->associd.association_id));
+               if (!iod->assoc)
+                       ret = VERR_NO_ASSOC;
+               else {
+                       queue = nvmet_fc_alloc_target_queue(iod->assoc,
+                                       be16_to_cpu(rqst->connect_cmd.qid),
+                                       be16_to_cpu(rqst->connect_cmd.sqsize));
+                       if (!queue)
+                               ret = VERR_QUEUE_ALLOC_FAIL;
+
+                       /* release get taken in nvmet_fc_find_target_assoc */
+                       nvmet_fc_tgt_a_put(iod->assoc);
+               }
+       }
+
+       if (ret) {
+               dev_err(tgtport->dev,
+                       "Create Connection LS failed: %s\n",
+                       validation_errors[ret]);
+               iod->lsreq->rsplen = nvmet_fc_format_rjt(acc,
+                               NVME_FC_MAX_LS_BUFFER_SIZE, rqst->w0.ls_cmd,
+                               (ret == VERR_NO_ASSOC) ?
+                                               ELS_RJT_PROT : ELS_RJT_LOGIC,
+                               ELS_EXPL_NONE, 0);
+               return;
+       }
+
+       queue->ersp_ratio = be16_to_cpu(rqst->connect_cmd.ersp_ratio);
+       atomic_set(&queue->connected, 1);
+       queue->sqhd = 0;        /* best place to init value */
+
+       /* format a response */
+
+       iod->lsreq->rsplen = sizeof(*acc);
+
+       nvmet_fc_format_rsp_hdr(acc, FCNVME_LS_ACC,
+                       fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)),
+                       FCNVME_LS_CREATE_CONNECTION);
+       acc->connectid.desc_tag = cpu_to_be32(FCNVME_LSDESC_CONN_ID);
+       acc->connectid.desc_len =
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_conn_id));
+       acc->connectid.connection_id =
+                       cpu_to_be64(nvmet_fc_makeconnid(iod->assoc,
+                               be16_to_cpu(rqst->connect_cmd.qid)));
+}
+
+static void
+nvmet_fc_ls_disconnect(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_ls_iod *iod)
+{
+       struct fcnvme_ls_disconnect_rqst *rqst =
+                       (struct fcnvme_ls_disconnect_rqst *)iod->rqstbuf;
+       struct fcnvme_ls_disconnect_acc *acc =
+                       (struct fcnvme_ls_disconnect_acc *)iod->rspbuf;
+       struct nvmet_fc_tgt_queue *queue;
+       struct nvmet_fc_tgt_assoc *assoc;
+       int ret = 0;
+       bool del_assoc = false;
+
+       memset(acc, 0, sizeof(*acc));
+
+       if (iod->rqstdatalen < sizeof(struct fcnvme_ls_disconnect_rqst))
+               ret = VERR_DISCONN_LEN;
+       else if (rqst->desc_list_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_ls_disconnect_rqst)))
+               ret = VERR_DISCONN_RQST_LEN;
+       else if (rqst->associd.desc_tag != cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
+               ret = VERR_ASSOC_ID;
+       else if (rqst->associd.desc_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_assoc_id)))
+               ret = VERR_ASSOC_ID_LEN;
+       else if (rqst->discon_cmd.desc_tag !=
+                       cpu_to_be32(FCNVME_LSDESC_DISCONN_CMD))
+               ret = VERR_DISCONN_CMD;
+       else if (rqst->discon_cmd.desc_len !=
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_lsdesc_disconn_cmd)))
+               ret = VERR_DISCONN_CMD_LEN;
+       else if ((rqst->discon_cmd.scope != FCNVME_DISCONN_ASSOCIATION) &&
+                       (rqst->discon_cmd.scope != FCNVME_DISCONN_CONNECTION))
+               ret = VERR_DISCONN_SCOPE;
+       else {
+               /* match an active association */
+               assoc = nvmet_fc_find_target_assoc(tgtport,
+                               be64_to_cpu(rqst->associd.association_id));
+               iod->assoc = assoc;
+               if (!assoc)
+                       ret = VERR_NO_ASSOC;
+       }
+
+       if (ret) {
+               dev_err(tgtport->dev,
+                       "Disconnect LS failed: %s\n",
+                       validation_errors[ret]);
+               iod->lsreq->rsplen = nvmet_fc_format_rjt(acc,
+                               NVME_FC_MAX_LS_BUFFER_SIZE, rqst->w0.ls_cmd,
+                               (ret == 8) ? ELS_RJT_PROT : ELS_RJT_LOGIC,
+                               ELS_EXPL_NONE, 0);
+               return;
+       }
+
+       /* format a response */
+
+       iod->lsreq->rsplen = sizeof(*acc);
+
+       nvmet_fc_format_rsp_hdr(acc, FCNVME_LS_ACC,
+                       fcnvme_lsdesc_len(
+                               sizeof(struct fcnvme_ls_disconnect_acc)),
+                       FCNVME_LS_DISCONNECT);
+
+
+       if (rqst->discon_cmd.scope == FCNVME_DISCONN_CONNECTION) {
+               queue = nvmet_fc_find_target_queue(tgtport,
+                                       be64_to_cpu(rqst->discon_cmd.id));
+               if (queue) {
+                       int qid = queue->qid;
+
+                       nvmet_fc_delete_target_queue(queue);
+
+                       /* release the get taken by find_target_queue */
+                       nvmet_fc_tgt_q_put(queue);
+
+                       /* tear association down if io queue terminated */
+                       if (!qid)
+                               del_assoc = true;
+               }
+       }
+
+       /* release get taken in nvmet_fc_find_target_assoc */
+       nvmet_fc_tgt_a_put(iod->assoc);
+
+       if (del_assoc)
+               nvmet_fc_delete_target_assoc(iod->assoc);
+}
+
+
+/* *********************** NVME Ctrl Routines **************************** */
+
+
+static void nvmet_fc_fcp_nvme_cmd_done(struct nvmet_req *nvme_req);
+
+static struct nvmet_fabrics_ops nvmet_fc_tgt_fcp_ops;
+
+static void
+nvmet_fc_xmt_ls_rsp_done(struct nvmefc_tgt_ls_req *lsreq)
+{
+       struct nvmet_fc_ls_iod *iod = lsreq->nvmet_fc_private;
+       struct nvmet_fc_tgtport *tgtport = iod->tgtport;
+
+       fc_dma_sync_single_for_cpu(tgtport->dev, iod->rspdma,
+                               NVME_FC_MAX_LS_BUFFER_SIZE, DMA_TO_DEVICE);
+       nvmet_fc_free_ls_iod(tgtport, iod);
+       nvmet_fc_tgtport_put(tgtport);
+}
+
+static void
+nvmet_fc_xmt_ls_rsp(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmet_fc_ls_iod *iod)
+{
+       int ret;
+
+       fc_dma_sync_single_for_device(tgtport->dev, iod->rspdma,
+                                 NVME_FC_MAX_LS_BUFFER_SIZE, DMA_TO_DEVICE);
+
+       ret = tgtport->ops->xmt_ls_rsp(&tgtport->fc_target_port, iod->lsreq);
+       if (ret)
+               nvmet_fc_xmt_ls_rsp_done(iod->lsreq);
+}
+
+/*
+ * Actual processing routine for received FC-NVME LS Requests from the LLD
+ */
+static void
+nvmet_fc_handle_ls_rqst(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_ls_iod *iod)
+{
+       struct fcnvme_ls_rqst_w0 *w0 =
+                       (struct fcnvme_ls_rqst_w0 *)iod->rqstbuf;
+
+       iod->lsreq->nvmet_fc_private = iod;
+       iod->lsreq->rspbuf = iod->rspbuf;
+       iod->lsreq->rspdma = iod->rspdma;
+       iod->lsreq->done = nvmet_fc_xmt_ls_rsp_done;
+       /* Be preventative. handlers will later set to valid length */
+       iod->lsreq->rsplen = 0;
+
+       iod->assoc = NULL;
+
+       /*
+        * handlers:
+        *   parse request input, execute the request, and format the
+        *   LS response
+        */
+       switch (w0->ls_cmd) {
+       case FCNVME_LS_CREATE_ASSOCIATION:
+               /* Creates Association and initial Admin Queue/Connection */
+               nvmet_fc_ls_create_association(tgtport, iod);
+               break;
+       case FCNVME_LS_CREATE_CONNECTION:
+               /* Creates an IO Queue/Connection */
+               nvmet_fc_ls_create_connection(tgtport, iod);
+               break;
+       case FCNVME_LS_DISCONNECT:
+               /* Terminate a Queue/Connection or the Association */
+               nvmet_fc_ls_disconnect(tgtport, iod);
+               break;
+       default:
+               iod->lsreq->rsplen = nvmet_fc_format_rjt(iod->rspbuf,
+                               NVME_FC_MAX_LS_BUFFER_SIZE, w0->ls_cmd,
+                               ELS_RJT_INVAL, ELS_EXPL_NONE, 0);
+       }
+
+       nvmet_fc_xmt_ls_rsp(tgtport, iod);
+}
+
+/*
+ * Actual processing routine for received FC-NVME LS Requests from the LLD
+ */
+static void
+nvmet_fc_handle_ls_rqst_work(struct work_struct *work)
+{
+       struct nvmet_fc_ls_iod *iod =
+               container_of(work, struct nvmet_fc_ls_iod, work);
+       struct nvmet_fc_tgtport *tgtport = iod->tgtport;
+
+       nvmet_fc_handle_ls_rqst(tgtport, iod);
+}
+
+
+/**
+ * nvmet_fc_rcv_ls_req - transport entry point called by an LLDD
+ *                       upon the reception of a NVME LS request.
+ *
+ * The nvmet-fc layer will copy payload to an internal structure for
+ * processing.  As such, upon completion of the routine, the LLDD may
+ * immediately free/reuse the LS request buffer passed in the call.
+ *
+ * If this routine returns error, the LLDD should abort the exchange.
+ *
+ * @tgtport:    pointer to the (registered) target port the LS was
+ *              received on.
+ * @lsreq:      pointer to a lsreq request structure to be used to reference
+ *              the exchange corresponding to the LS.
+ * @lsreqbuf:   pointer to the buffer containing the LS Request
+ * @lsreqbuf_len: length, in bytes, of the received LS request
+ */
+int
+nvmet_fc_rcv_ls_req(struct nvmet_fc_target_port *target_port,
+                       struct nvmefc_tgt_ls_req *lsreq,
+                       void *lsreqbuf, u32 lsreqbuf_len)
+{
+       struct nvmet_fc_tgtport *tgtport = targetport_to_tgtport(target_port);
+       struct nvmet_fc_ls_iod *iod;
+
+       if (lsreqbuf_len > NVME_FC_MAX_LS_BUFFER_SIZE)
+               return -E2BIG;
+
+       if (!nvmet_fc_tgtport_get(tgtport))
+               return -ESHUTDOWN;
+
+       iod = nvmet_fc_alloc_ls_iod(tgtport);
+       if (!iod) {
+               nvmet_fc_tgtport_put(tgtport);
+               return -ENOENT;
+       }
+
+       iod->lsreq = lsreq;
+       iod->fcpreq = NULL;
+       memcpy(iod->rqstbuf, lsreqbuf, lsreqbuf_len);
+       iod->rqstdatalen = lsreqbuf_len;
+
+       schedule_work(&iod->work);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(nvmet_fc_rcv_ls_req);
+
+
+/*
+ * **********************
+ * Start of FCP handling
+ * **********************
+ */
+
+static int
+nvmet_fc_alloc_tgt_pgs(struct nvmet_fc_fcp_iod *fod)
+{
+       struct scatterlist *sg;
+       struct page *page;
+       unsigned int nent;
+       u32 page_len, length;
+       int i = 0;
+
+       length = fod->total_length;
+       nent = DIV_ROUND_UP(length, PAGE_SIZE);
+       sg = kmalloc_array(nent, sizeof(struct scatterlist), GFP_KERNEL);
+       if (!sg)
+               goto out;
+
+       sg_init_table(sg, nent);
+
+       while (length) {
+               page_len = min_t(u32, length, PAGE_SIZE);
+
+               page = alloc_page(GFP_KERNEL);
+               if (!page)
+                       goto out_free_pages;
+
+               sg_set_page(&sg[i], page, page_len, 0);
+               length -= page_len;
+               i++;
+       }
+
+       fod->data_sg = sg;
+       fod->data_sg_cnt = nent;
+       fod->data_sg_cnt = fc_dma_map_sg(fod->tgtport->dev, sg, nent,
+                               ((fod->io_dir == NVMET_FCP_WRITE) ?
+                                       DMA_FROM_DEVICE : DMA_TO_DEVICE));
+                               /* note: write from initiator perspective */
+
+       return 0;
+
+out_free_pages:
+       while (i > 0) {
+               i--;
+               __free_page(sg_page(&sg[i]));
+       }
+       kfree(sg);
+       fod->data_sg = NULL;
+       fod->data_sg_cnt = 0;
+out:
+       return NVME_SC_INTERNAL;
+}
+
+static void
+nvmet_fc_free_tgt_pgs(struct nvmet_fc_fcp_iod *fod)
+{
+       struct scatterlist *sg;
+       int count;
+
+       if (!fod->data_sg || !fod->data_sg_cnt)
+               return;
+
+       fc_dma_unmap_sg(fod->tgtport->dev, fod->data_sg, fod->data_sg_cnt,
+                               ((fod->io_dir == NVMET_FCP_WRITE) ?
+                                       DMA_FROM_DEVICE : DMA_TO_DEVICE));
+       for_each_sg(fod->data_sg, sg, fod->data_sg_cnt, count)
+               __free_page(sg_page(sg));
+       kfree(fod->data_sg);
+}
+
+
+static bool
+queue_90percent_full(struct nvmet_fc_tgt_queue *q, u32 sqhd)
+{
+       u32 sqtail, used;
+
+       /* egad, this is ugly. And sqtail is just a best guess */
+       sqtail = atomic_read(&q->sqtail) % q->sqsize;
+
+       used = (sqtail < sqhd) ? (sqtail + q->sqsize - sqhd) : (sqtail - sqhd);
+       return ((used * 10) >= (((u32)(q->sqsize - 1) * 9)));
+}
+
+/*
+ * Prep RSP payload.
+ * May be a NVMET_FCOP_RSP or NVMET_FCOP_READDATA_RSP op
+ */
+static void
+nvmet_fc_prep_fcp_rsp(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmet_fc_fcp_iod *fod)
+{
+       struct nvme_fc_ersp_iu *ersp = &fod->rspiubuf;
+       struct nvme_common_command *sqe = &fod->cmdiubuf.sqe.common;
+       struct nvme_completion *cqe = &ersp->cqe;
+       u32 *cqewd = (u32 *)cqe;
+       bool send_ersp = false;
+       u32 rsn, rspcnt, xfr_length;
+
+       if (fod->fcpreq->op == NVMET_FCOP_READDATA_RSP)
+               xfr_length = fod->total_length;
+       else
+               xfr_length = fod->offset;
+
+       /*
+        * check to see if we can send a 0's rsp.
+        *   Note: to send a 0's response, the NVME-FC host transport will
+        *   recreate the CQE. The host transport knows: sq id, SQHD (last
+        *   seen in an ersp), and command_id. Thus it will create a
+        *   zero-filled CQE with those known fields filled in. Transport
+        *   must send an ersp for any condition where the cqe won't match
+        *   this.
+        *
+        * Here are the FC-NVME mandated cases where we must send an ersp:
+        *  every N responses, where N=ersp_ratio
+        *  force fabric commands to send ersp's (not in FC-NVME but good
+        *    practice)
+        *  normal cmds: any time status is non-zero, or status is zero
+        *     but words 0 or 1 are non-zero.
+        *  the SQ is 90% or more full
+        *  the cmd is a fused command
+        *  transferred data length not equal to cmd iu length
+        */
+       rspcnt = atomic_inc_return(&fod->queue->zrspcnt);
+       if (!(rspcnt % fod->queue->ersp_ratio) ||
+           sqe->opcode == nvme_fabrics_command ||
+           xfr_length != fod->total_length ||
+           (le16_to_cpu(cqe->status) & 0xFFFE) || cqewd[0] || cqewd[1] ||
+           (sqe->flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND)) ||
+           queue_90percent_full(fod->queue, cqe->sq_head))
+               send_ersp = true;
+
+       /* re-set the fields */
+       fod->fcpreq->rspaddr = ersp;
+       fod->fcpreq->rspdma = fod->rspdma;
+
+       if (!send_ersp) {
+               memset(ersp, 0, NVME_FC_SIZEOF_ZEROS_RSP);
+               fod->fcpreq->rsplen = NVME_FC_SIZEOF_ZEROS_RSP;
+       } else {
+               ersp->iu_len = cpu_to_be16(sizeof(*ersp)/sizeof(u32));
+               rsn = atomic_inc_return(&fod->queue->rsn);
+               ersp->rsn = cpu_to_be32(rsn);
+               ersp->xfrd_len = cpu_to_be32(xfr_length);
+               fod->fcpreq->rsplen = sizeof(*ersp);
+       }
+
+       fc_dma_sync_single_for_device(tgtport->dev, fod->rspdma,
+                                 sizeof(fod->rspiubuf), DMA_TO_DEVICE);
+}
+
+static void nvmet_fc_xmt_fcp_op_done(struct nvmefc_tgt_fcp_req *fcpreq);
+
+static void
+nvmet_fc_xmt_fcp_rsp(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmet_fc_fcp_iod *fod)
+{
+       int ret;
+
+       fod->fcpreq->op = NVMET_FCOP_RSP;
+       fod->fcpreq->timeout = 0;
+
+       nvmet_fc_prep_fcp_rsp(tgtport, fod);
+
+       ret = tgtport->ops->fcp_op(&tgtport->fc_target_port, fod->fcpreq);
+       if (ret)
+               nvmet_fc_abort_op(tgtport, fod->fcpreq);
+}
+
+static void
+nvmet_fc_transfer_fcp_data(struct nvmet_fc_tgtport *tgtport,
+                               struct nvmet_fc_fcp_iod *fod, u8 op)
+{
+       struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq;
+       struct scatterlist *sg, *datasg;
+       u32 tlen, sg_off;
+       int ret;
+
+       fcpreq->op = op;
+       fcpreq->offset = fod->offset;
+       fcpreq->timeout = NVME_FC_TGTOP_TIMEOUT_SEC;
+       tlen = min_t(u32, (NVMET_FC_MAX_KB_PER_XFR * 1024),
+                       (fod->total_length - fod->offset));
+       tlen = min_t(u32, tlen, NVME_FC_MAX_SEGMENTS * PAGE_SIZE);
+       tlen = min_t(u32, tlen, fod->tgtport->ops->max_sgl_segments
+                                       * PAGE_SIZE);
+       fcpreq->transfer_length = tlen;
+       fcpreq->transferred_length = 0;
+       fcpreq->fcp_error = 0;
+       fcpreq->rsplen = 0;
+
+       fcpreq->sg_cnt = 0;
+
+       datasg = fod->next_sg;
+       sg_off = fod->next_sg_offset;
+
+       for (sg = fcpreq->sg ; tlen; sg++) {
+               *sg = *datasg;
+               if (sg_off) {
+                       sg->offset += sg_off;
+                       sg->length -= sg_off;
+                       sg->dma_address += sg_off;
+                       sg_off = 0;
+               }
+               if (tlen < sg->length) {
+                       sg->length = tlen;
+                       fod->next_sg = datasg;
+                       fod->next_sg_offset += tlen;
+               } else if (tlen == sg->length) {
+                       fod->next_sg_offset = 0;
+                       fod->next_sg = sg_next(datasg);
+               } else {
+                       fod->next_sg_offset = 0;
+                       datasg = sg_next(datasg);
+               }
+               tlen -= sg->length;
+               fcpreq->sg_cnt++;
+       }
+
+       /*
+        * If the last READDATA request: check if LLDD supports
+        * combined xfr with response.
+        */
+       if ((op == NVMET_FCOP_READDATA) &&
+           ((fod->offset + fcpreq->transfer_length) == fod->total_length) &&
+           (tgtport->ops->target_features & NVMET_FCTGTFEAT_READDATA_RSP)) {
+               fcpreq->op = NVMET_FCOP_READDATA_RSP;
+               nvmet_fc_prep_fcp_rsp(tgtport, fod);
+       }
+
+       ret = tgtport->ops->fcp_op(&tgtport->fc_target_port, fod->fcpreq);
+       if (ret) {
+               /*
+                * should be ok to set w/o lock as its in the thread of
+                * execution (not an async timer routine) and doesn't
+                * contend with any clearing action
+                */
+               fod->abort = true;
+
+               if (op == NVMET_FCOP_WRITEDATA)
+                       nvmet_req_complete(&fod->req,
+                                       NVME_SC_FC_TRANSPORT_ERROR);
+               else /* NVMET_FCOP_READDATA or NVMET_FCOP_READDATA_RSP */ {
+                       fcpreq->fcp_error = ret;
+                       fcpreq->transferred_length = 0;
+                       nvmet_fc_xmt_fcp_op_done(fod->fcpreq);
+               }
+       }
+}
+
+static void
+nvmet_fc_xmt_fcp_op_done(struct nvmefc_tgt_fcp_req *fcpreq)
+{
+       struct nvmet_fc_fcp_iod *fod = fcpreq->nvmet_fc_private;
+       struct nvmet_fc_tgtport *tgtport = fod->tgtport;
+       unsigned long flags;
+       bool abort;
+
+       spin_lock_irqsave(&fod->flock, flags);
+       abort = fod->abort;
+       spin_unlock_irqrestore(&fod->flock, flags);
+
+       /* if in the middle of an io and we need to tear down */
+       if (abort && fcpreq->op != NVMET_FCOP_ABORT) {
+               /* data no longer needed */
+               nvmet_fc_free_tgt_pgs(fod);
+
+               if (fcpreq->fcp_error || abort)
+                       nvmet_req_complete(&fod->req, fcpreq->fcp_error);
+
+               return;
+       }
+
+       switch (fcpreq->op) {
+
+       case NVMET_FCOP_WRITEDATA:
+               if (abort || fcpreq->fcp_error ||
+                   fcpreq->transferred_length != fcpreq->transfer_length) {
+                       nvmet_req_complete(&fod->req,
+                                       NVME_SC_FC_TRANSPORT_ERROR);
+                       return;
+               }
+
+               fod->offset += fcpreq->transferred_length;
+               if (fod->offset != fod->total_length) {
+                       /* transfer the next chunk */
+                       nvmet_fc_transfer_fcp_data(tgtport, fod,
+                                               NVMET_FCOP_WRITEDATA);
+                       return;
+               }
+
+               /* data transfer complete, resume with nvmet layer */
+
+               fod->req.execute(&fod->req);
+
+               break;
+
+       case NVMET_FCOP_READDATA:
+       case NVMET_FCOP_READDATA_RSP:
+               if (abort || fcpreq->fcp_error ||
+                   fcpreq->transferred_length != fcpreq->transfer_length) {
+                       /* data no longer needed */
+                       nvmet_fc_free_tgt_pgs(fod);
+
+                       nvmet_fc_abort_op(tgtport, fod->fcpreq);
+                       return;
+               }
+
+               /* success */
+
+               if (fcpreq->op == NVMET_FCOP_READDATA_RSP) {
+                       /* data no longer needed */
+                       nvmet_fc_free_tgt_pgs(fod);
+                       fc_dma_sync_single_for_cpu(tgtport->dev, fod->rspdma,
+                                       sizeof(fod->rspiubuf), DMA_TO_DEVICE);
+                       nvmet_fc_free_fcp_iod(fod->queue, fod);
+                       return;
+               }
+
+               fod->offset += fcpreq->transferred_length;
+               if (fod->offset != fod->total_length) {
+                       /* transfer the next chunk */
+                       nvmet_fc_transfer_fcp_data(tgtport, fod,
+                                               NVMET_FCOP_READDATA);
+                       return;
+               }
+
+               /* data transfer complete, send response */
+
+               /* data no longer needed */
+               nvmet_fc_free_tgt_pgs(fod);
+
+               nvmet_fc_xmt_fcp_rsp(tgtport, fod);
+
+               break;
+
+       case NVMET_FCOP_RSP:
+       case NVMET_FCOP_ABORT:
+               fc_dma_sync_single_for_cpu(tgtport->dev, fod->rspdma,
+                               sizeof(fod->rspiubuf), DMA_TO_DEVICE);
+               nvmet_fc_free_fcp_iod(fod->queue, fod);
+               break;
+
+       default:
+               nvmet_fc_free_tgt_pgs(fod);
+               nvmet_fc_abort_op(tgtport, fod->fcpreq);
+               break;
+       }
+}
+
+/*
+ * actual completion handler after execution by the nvmet layer
+ */
+static void
+__nvmet_fc_fcp_nvme_cmd_done(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_fcp_iod *fod, int status)
+{
+       struct nvme_common_command *sqe = &fod->cmdiubuf.sqe.common;
+       struct nvme_completion *cqe = &fod->rspiubuf.cqe;
+       unsigned long flags;
+       bool abort;
+
+       spin_lock_irqsave(&fod->flock, flags);
+       abort = fod->abort;
+       spin_unlock_irqrestore(&fod->flock, flags);
+
+       /* if we have a CQE, snoop the last sq_head value */
+       if (!status)
+               fod->queue->sqhd = cqe->sq_head;
+
+       if (abort) {
+               /* data no longer needed */
+               nvmet_fc_free_tgt_pgs(fod);
+
+               nvmet_fc_abort_op(tgtport, fod->fcpreq);
+               return;
+       }
+
+       /* if an error handling the cmd post initial parsing */
+       if (status) {
+               /* fudge up a failed CQE status for our transport error */
+               memset(cqe, 0, sizeof(*cqe));
+               cqe->sq_head = fod->queue->sqhd;        /* echo last cqe sqhd */
+               cqe->sq_id = cpu_to_le16(fod->queue->qid);
+               cqe->command_id = sqe->command_id;
+               cqe->status = cpu_to_le16(status);
+       } else {
+
+               /*
+                * try to push the data even if the SQE status is non-zero.
+                * There may be a status where data still was intended to
+                * be moved
+                */
+               if ((fod->io_dir == NVMET_FCP_READ) && (fod->data_sg_cnt)) {
+                       /* push the data over before sending rsp */
+                       nvmet_fc_transfer_fcp_data(tgtport, fod,
+                                               NVMET_FCOP_READDATA);
+                       return;
+               }
+
+               /* writes & no data - fall thru */
+       }
+
+       /* data no longer needed */
+       nvmet_fc_free_tgt_pgs(fod);
+
+       nvmet_fc_xmt_fcp_rsp(tgtport, fod);
+}
+
+
+static void
+nvmet_fc_fcp_nvme_cmd_done(struct nvmet_req *nvme_req)
+{
+       struct nvmet_fc_fcp_iod *fod = nvmet_req_to_fod(nvme_req);
+       struct nvmet_fc_tgtport *tgtport = fod->tgtport;
+
+       __nvmet_fc_fcp_nvme_cmd_done(tgtport, fod, 0);
+}
+
+
+/*
+ * Actual processing routine for received FC-NVME LS Requests from the LLD
+ */
+void
+nvmet_fc_handle_fcp_rqst(struct nvmet_fc_tgtport *tgtport,
+                       struct nvmet_fc_fcp_iod *fod)
+{
+       struct nvme_fc_cmd_iu *cmdiu = &fod->cmdiubuf;
+       int ret;
+
+       /*
+        * Fused commands are currently not supported in the linux
+        * implementation.
+        *
+        * As such, the implementation of the FC transport does not
+        * look at the fused commands and order delivery to the upper
+        * layer until we have both based on csn.
+        */
+
+       fod->fcpreq->done = nvmet_fc_xmt_fcp_op_done;
+
+       fod->total_length = be32_to_cpu(cmdiu->data_len);
+       if (cmdiu->flags & FCNVME_CMD_FLAGS_WRITE) {
+               fod->io_dir = NVMET_FCP_WRITE;
+               if (!nvme_is_write(&cmdiu->sqe))
+                       goto transport_error;
+       } else if (cmdiu->flags & FCNVME_CMD_FLAGS_READ) {
+               fod->io_dir = NVMET_FCP_READ;
+               if (nvme_is_write(&cmdiu->sqe))
+                       goto transport_error;
+       } else {
+               fod->io_dir = NVMET_FCP_NODATA;
+               if (fod->total_length)
+                       goto transport_error;
+       }
+
+       fod->req.cmd = &fod->cmdiubuf.sqe;
+       fod->req.rsp = &fod->rspiubuf.cqe;
+       fod->req.port = fod->queue->port;
+
+       /* ensure nvmet handlers will set cmd handler callback */
+       fod->req.execute = NULL;
+
+       /* clear any response payload */
+       memset(&fod->rspiubuf, 0, sizeof(fod->rspiubuf));
+
+       ret = nvmet_req_init(&fod->req,
+                               &fod->queue->nvme_cq,
+                               &fod->queue->nvme_sq,
+                               &nvmet_fc_tgt_fcp_ops);
+       if (!ret) {     /* bad SQE content */
+               nvmet_fc_abort_op(tgtport, fod->fcpreq);
+               return;
+       }
+
+       /* keep a running counter of tail position */
+       atomic_inc(&fod->queue->sqtail);
+
+       fod->data_sg = NULL;
+       fod->data_sg_cnt = 0;
+       if (fod->total_length) {
+               ret = nvmet_fc_alloc_tgt_pgs(fod);
+               if (ret) {
+                       nvmet_req_complete(&fod->req, ret);
+                       return;
+               }
+       }
+       fod->req.sg = fod->data_sg;
+       fod->req.sg_cnt = fod->data_sg_cnt;
+       fod->offset = 0;
+       fod->next_sg = fod->data_sg;
+       fod->next_sg_offset = 0;
+
+       if (fod->io_dir == NVMET_FCP_WRITE) {
+               /* pull the data over before invoking nvmet layer */
+               nvmet_fc_transfer_fcp_data(tgtport, fod, NVMET_FCOP_WRITEDATA);
+               return;
+       }
+
+       /*
+        * Reads or no data:
+        *
+        * can invoke the nvmet_layer now. If read data, cmd completion will
+        * push the data
+        */
+
+       fod->req.execute(&fod->req);
+
+       return;
+
+transport_error:
+       nvmet_fc_abort_op(tgtport, fod->fcpreq);
+}
+
+/*
+ * Actual processing routine for received FC-NVME LS Requests from the LLD
+ */
+static void
+nvmet_fc_handle_fcp_rqst_work(struct work_struct *work)
+{
+       struct nvmet_fc_fcp_iod *fod =
+               container_of(work, struct nvmet_fc_fcp_iod, work);
+       struct nvmet_fc_tgtport *tgtport = fod->tgtport;
+
+       nvmet_fc_handle_fcp_rqst(tgtport, fod);
+}
+
+/**
+ * nvmet_fc_rcv_fcp_req - transport entry point called by an LLDD
+ *                       upon the reception of a NVME FCP CMD IU.
+ *
+ * Pass a FC-NVME FCP CMD IU received from the FC link to the nvmet-fc
+ * layer for processing.
+ *
+ * The nvmet-fc layer will copy cmd payload to an internal structure for
+ * processing.  As such, upon completion of the routine, the LLDD may
+ * immediately free/reuse the CMD IU buffer passed in the call.
+ *
+ * If this routine returns error, the lldd should abort the exchange.
+ *
+ * @target_port: pointer to the (registered) target port the FCP CMD IU
+ *              was receive on.
+ * @fcpreq:     pointer to a fcpreq request structure to be used to reference
+ *              the exchange corresponding to the FCP Exchange.
+ * @cmdiubuf:   pointer to the buffer containing the FCP CMD IU
+ * @cmdiubuf_len: length, in bytes, of the received FCP CMD IU
+ */
+int
+nvmet_fc_rcv_fcp_req(struct nvmet_fc_target_port *target_port,
+                       struct nvmefc_tgt_fcp_req *fcpreq,
+                       void *cmdiubuf, u32 cmdiubuf_len)
+{
+       struct nvmet_fc_tgtport *tgtport = targetport_to_tgtport(target_port);
+       struct nvme_fc_cmd_iu *cmdiu = cmdiubuf;
+       struct nvmet_fc_tgt_queue *queue;
+       struct nvmet_fc_fcp_iod *fod;
+
+       /* validate iu, so the connection id can be used to find the queue */
+       if ((cmdiubuf_len != sizeof(*cmdiu)) ||
+                       (cmdiu->scsi_id != NVME_CMD_SCSI_ID) ||
+                       (cmdiu->fc_id != NVME_CMD_FC_ID) ||
+                       (be16_to_cpu(cmdiu->iu_len) != (sizeof(*cmdiu)/4)))
+               return -EIO;
+
+
+       queue = nvmet_fc_find_target_queue(tgtport,
+                               be64_to_cpu(cmdiu->connection_id));
+       if (!queue)
+               return -ENOTCONN;
+
+       /*
+        * note: reference taken by find_target_queue
+        * After successful fod allocation, the fod will inherit the
+        * ownership of that reference and will remove the reference
+        * when the fod is freed.
+        */
+
+       fod = nvmet_fc_alloc_fcp_iod(queue);
+       if (!fod) {
+               /* release the queue lookup reference */
+               nvmet_fc_tgt_q_put(queue);
+               return -ENOENT;
+       }
+
+       fcpreq->nvmet_fc_private = fod;
+       fod->fcpreq = fcpreq;
+       /*
+        * put all admin cmds on hw queue id 0. All io commands go to
+        * the respective hw queue based on a modulo basis
+        */
+       fcpreq->hwqid = queue->qid ?
+                       ((queue->qid - 1) % tgtport->ops->max_hw_queues) : 0;
+       memcpy(&fod->cmdiubuf, cmdiubuf, cmdiubuf_len);
+
+       queue_work_on(queue->cpu, queue->work_q, &fod->work);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(nvmet_fc_rcv_fcp_req);
+
+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
+nvmet_fc_parse_traddr(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 int
+nvmet_fc_add_port(struct nvmet_port *port)
+{
+       struct nvmet_fc_tgtport *tgtport;
+       struct nvmet_fc_traddr traddr = { 0L, 0L };
+       unsigned long flags;
+       int ret;
+
+       /* validate the address info */
+       if ((port->disc_addr.trtype != NVMF_TRTYPE_FC) ||
+           (port->disc_addr.adrfam != NVMF_ADDR_FAMILY_FC))
+               return -EINVAL;
+
+       /* map the traddr address info to a target port */
+
+       ret = nvmet_fc_parse_traddr(&traddr, port->disc_addr.traddr);
+       if (ret)
+               return ret;
+
+       ret = -ENXIO;
+       spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
+       list_for_each_entry(tgtport, &nvmet_fc_target_list, tgt_list) {
+               if ((tgtport->fc_target_port.node_name == traddr.nn) &&
+                   (tgtport->fc_target_port.port_name == traddr.pn)) {
+                       /* a FC port can only be 1 nvmet port id */
+                       if (!tgtport->port) {
+                               tgtport->port = port;
+                               port->priv = tgtport;
+                               ret = 0;
+                       } else
+                               ret = -EALREADY;
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
+       return ret;
+}
+
+static void
+nvmet_fc_remove_port(struct nvmet_port *port)
+{
+       struct nvmet_fc_tgtport *tgtport = port->priv;
+       unsigned long flags;
+
+       spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
+       if (tgtport->port == port) {
+               nvmet_fc_tgtport_put(tgtport);
+               tgtport->port = NULL;
+       }
+       spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
+}
+
+static struct nvmet_fabrics_ops nvmet_fc_tgt_fcp_ops = {
+       .owner                  = THIS_MODULE,
+       .type                   = NVMF_TRTYPE_FC,
+       .msdbd                  = 1,
+       .add_port               = nvmet_fc_add_port,
+       .remove_port            = nvmet_fc_remove_port,
+       .queue_response         = nvmet_fc_fcp_nvme_cmd_done,
+       .delete_ctrl            = nvmet_fc_delete_ctrl,
+};
+
+static int __init nvmet_fc_init_module(void)
+{
+       return nvmet_register_transport(&nvmet_fc_tgt_fcp_ops);
+}
+
+static void __exit nvmet_fc_exit_module(void)
+{
+       /* sanity check - all lports should be removed */
+       if (!list_empty(&nvmet_fc_target_list))
+               pr_warn("%s: targetport list not empty\n", __func__);
+
+       nvmet_unregister_transport(&nvmet_fc_tgt_fcp_ops);
+
+       ida_destroy(&nvmet_fc_tgtport_cnt);
+}
+
+module_init(nvmet_fc_init_module);
+module_exit(nvmet_fc_exit_module);
+
+MODULE_LICENSE("GPL v2");