generic NVMe over Fabrics library support

generic NVMe over Fabrics library support

[Christoph Hellwig]

This patch set adds the necessary infrastructure for the NVMe over
Fabrics functionality and the NVMe over Fabrics library itself.

First we add some needed parameters to NVMe request allocation such as flags
(for reserved commands - connect and keep-alive), also support tag
allocation of a given queue ID (for connect to be executed per-queue)
and allow request to be queued at the head of the request queue (so
reconnects can pass in flight I/O).

Second, we add support for additional sysfs attributes that are needed
or useful for the Fabrics driver.

Third we add the NVMe over Fabrics related header definitions and the
Fabrics library itself which is transport independent and handles
Fabrics specific commands and variables.

Last, we add support for periodic keep-alive mechanism which is mandatory
for Fabrics.

[PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Christoph Hellwig]

From: Ming Lin <ming.l@ssi.samsung.com>

For some protocols like NVMe over Fabrics we need to be able to send
initialization commands to a specific queue.

Based on an earlier patch from Christoph Hellwig <hch@lst.de>.

Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
 include/linux/blk-mq.h |  2 ++
 2 files changed, 35 insertions(+)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 29cbc1b..7bb45ed 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 }
 EXPORT_SYMBOL(blk_mq_alloc_request);
 
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+		unsigned int flags, unsigned int hctx_idx)
+{
+	struct blk_mq_hw_ctx *hctx;
+	struct blk_mq_ctx *ctx;
+	struct request *rq;
+	struct blk_mq_alloc_data alloc_data;
+	int ret;
+
+	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
+	if (ret)
+		return ERR_PTR(ret);
+
+	hctx = q->queue_hw_ctx[hctx_idx];
+	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
+
+	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
+
+	rq = __blk_mq_alloc_request(&alloc_data, rw);
+	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
+		__blk_mq_run_hw_queue(hctx);
+
+		rq =  __blk_mq_alloc_request(&alloc_data, rw);
+	}
+	if (!rq) {
+		blk_queue_exit(q);
+		return ERR_PTR(-EWOULDBLOCK);
+	}
+
+	return rq;
+}
+EXPORT_SYMBOL(blk_mq_alloc_request_hctx);
+
 static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx, struct request *rq)
 {
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 2498fdf..6bf8735 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -196,6 +196,8 @@ enum {
 
 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 		unsigned int flags);
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+		unsigned int flags, unsigned int hctx_idx);
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
 
-- 
2.1.4

[PATCH 2/8] nvme: allow transitioning from NEW to LIVE state

[Christoph Hellwig]

For Fabrics we're not going through an intermediate reset state
(at least for now).

Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/core.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 27cce17..025ac8f 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -85,6 +85,7 @@ bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 	switch (new_state) {
 	case NVME_CTRL_LIVE:
 		switch (old_state) {
+		case NVME_CTRL_NEW:
 		case NVME_CTRL_RESETTING:
 			changed = true;
 			/* FALLTHRU */
-- 
2.1.4

[PATCH 3/8] nvme: Modify and export sync command submission for fabrics

[Christoph Hellwig]

NVMe over fabrics will use __nvme_submit_sync_cmd in the the
transport and require a few tweaks to it.  For that we export it
and add a few more paramters:

1. allow passing a queue ID to the block layer

   For the NVMe over Fabrics connect command we need to able to specify a
   queue ID that we want to send the command on.  Add a qid parameter to
   the relevant functions to enable this behavior.

2. allow submitting at_head commands

   In cases where we want to (re)connect to a controller
   where we have inflight queued commands we want to first
   connect and only then allow the other queued commands to
   be kicked. This will prevents failures in controller resets
   and reconnects.

3. allow passing flags to blk_mq_allocate_request

   Both for Fabrics connect the the keep-alive feature in NVMe 1.2.1 we
   want to be able to use reserved requests.

Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Tested-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/core.c | 27 ++++++++++++++++++---------
 drivers/nvme/host/nvme.h |  5 +++--
 drivers/nvme/host/pci.c  |  4 ++--
 3 files changed, 23 insertions(+), 13 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 025ac8f..08d8b3b 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -192,11 +192,16 @@ void nvme_requeue_req(struct request *req)
 EXPORT_SYMBOL_GPL(nvme_requeue_req);
 
 struct request *nvme_alloc_request(struct request_queue *q,
-		struct nvme_command *cmd, unsigned int flags)
+		struct nvme_command *cmd, unsigned int flags, int qid)
 {
 	struct request *req;
 
-	req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
+	if (qid == NVME_QID_ANY) {
+		req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
+	} else {
+		req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
+				qid ? qid - 1 : 0);
+	}
 	if (IS_ERR(req))
 		return req;
 
@@ -324,12 +329,12 @@ EXPORT_SYMBOL_GPL(nvme_setup_cmd);
  */
 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 		struct nvme_completion *cqe, void *buffer, unsigned bufflen,
-		unsigned timeout)
+		unsigned timeout, int qid, int at_head, int flags)
 {
 	struct request *req;
 	int ret;
 
-	req = nvme_alloc_request(q, cmd, 0);
+	req = nvme_alloc_request(q, cmd, flags, qid);
 	if (IS_ERR(req))
 		return PTR_ERR(req);
 
@@ -342,17 +347,19 @@ int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 			goto out;
 	}
 
-	blk_execute_rq(req->q, NULL, req, 0);
+	blk_execute_rq(req->q, NULL, req, at_head);
 	ret = req->errors;
  out:
 	blk_mq_free_request(req);
 	return ret;
 }
+EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 
 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 		void *buffer, unsigned bufflen)
 {
-	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0);
+	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
+			NVME_QID_ANY, 0, 0);
 }
 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 
@@ -370,7 +377,7 @@ int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 	void *meta = NULL;
 	int ret;
 
-	req = nvme_alloc_request(q, cmd, 0);
+	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 	if (IS_ERR(req))
 		return PTR_ERR(req);
 
@@ -520,7 +527,8 @@ int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 	c.features.prp1 = cpu_to_le64(dma_addr);
 	c.features.fid = cpu_to_le32(fid);
 
-	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
+	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
+			NVME_QID_ANY, 0, 0);
 	if (ret >= 0)
 		*result = le32_to_cpu(cqe.result);
 	return ret;
@@ -539,7 +547,8 @@ int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 	c.features.fid = cpu_to_le32(fid);
 	c.features.dword11 = cpu_to_le32(dword11);
 
-	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
+	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
+			NVME_QID_ANY, 0, 0);
 	if (ret >= 0)
 		*result = le32_to_cpu(cqe.result);
 	return ret;
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index bb7f001..180669c 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -231,8 +231,9 @@ void nvme_stop_queues(struct nvme_ctrl *ctrl);
 void nvme_start_queues(struct nvme_ctrl *ctrl);
 void nvme_kill_queues(struct nvme_ctrl *ctrl);
 
+#define NVME_QID_ANY -1
 struct request *nvme_alloc_request(struct request_queue *q,
-		struct nvme_command *cmd, unsigned int flags);
+		struct nvme_command *cmd, unsigned int flags, int qid);
 void nvme_requeue_req(struct request *req);
 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 		struct nvme_command *cmd);
@@ -240,7 +241,7 @@ int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 		void *buf, unsigned bufflen);
 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 		struct nvme_completion *cqe, void *buffer, unsigned bufflen,
-		unsigned timeout);
+		unsigned timeout, int qid, int at_head, int flags);
 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 		void __user *ubuffer, unsigned bufflen, u32 *result,
 		unsigned timeout);
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 2c3019a..571faa2 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -901,7 +901,7 @@ static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
 		 req->tag, nvmeq->qid);
 
 	abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
-			BLK_MQ_REQ_NOWAIT);
+			BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
 	if (IS_ERR(abort_req)) {
 		atomic_inc(&dev->ctrl.abort_limit);
 		return BLK_EH_RESET_TIMER;
@@ -1512,7 +1512,7 @@ static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
 	cmd.delete_queue.opcode = opcode;
 	cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
 
-	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT);
+	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
 	if (IS_ERR(req))
 		return PTR_ERR(req);
 
-- 
2.1.4

[PATCH 4/8] nvme: add fabrics sysfs attributes

[Christoph Hellwig]

- delete_controller: This attribute allows to delete a controller.
  A driver is not obligated to support it (pci doesn't) so it is
  created only if the driver supports it. The new fabrics drivers
  will support it (essentialy a disconnect operation).

  Usage:
  echo > /sys/class/nvme/nvme0/delete_controller

- subsysnqn: This attribute shows the subsystem nqn of the configured
  device. If a driver does not implement the get_subsysnqn method, the
  file will not appear in sysfs.

- transport: This attribute shows the transport name. Added a "name"
  field to struct nvme_ctrl_ops.

  For loop,
  cat /sys/class/nvme/nvme0/transport
  loop

  For RDMA,
  cat /sys/class/nvme/nvme0/transport
  rdma

  For PCIe,
  cat /sys/class/nvme/nvme0/transport
  pcie

- address: This attributes shows the controller address. The fabrics
  drivers that will implement get_address can show the address of the
  connected controller.

  example:
  cat /sys/class/nvme/nvme0/address
  traddr=192.168.2.2,trsvcid=1023

Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Reviewed-by: Jay Freyensee <james.p.freyensee@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/core.c | 76 ++++++++++++++++++++++++++++++++++++++++++++++--
 drivers/nvme/host/nvme.h |  4 +++
 drivers/nvme/host/pci.c  |  1 +
 3 files changed, 78 insertions(+), 3 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 08d8b3b..c272eb9 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1362,7 +1362,7 @@ static struct attribute *nvme_ns_attrs[] = {
 	NULL,
 };
 
-static umode_t nvme_attrs_are_visible(struct kobject *kobj,
+static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
 		struct attribute *a, int n)
 {
 	struct device *dev = container_of(kobj, struct device, kobj);
@@ -1381,7 +1381,7 @@ static umode_t nvme_attrs_are_visible(struct kobject *kobj,
 
 static const struct attribute_group nvme_ns_attr_group = {
 	.attrs		= nvme_ns_attrs,
-	.is_visible	= nvme_attrs_are_visible,
+	.is_visible	= nvme_ns_attrs_are_visible,
 };
 
 #define nvme_show_str_function(field)						\
@@ -1407,6 +1407,49 @@ nvme_show_str_function(serial);
 nvme_show_str_function(firmware_rev);
 nvme_show_int_function(cntlid);
 
+static ssize_t nvme_sysfs_delete(struct device *dev,
+				struct device_attribute *attr, const char *buf,
+				size_t count)
+{
+	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
+
+	if (device_remove_file_self(dev, attr))
+		ctrl->ops->delete_ctrl(ctrl);
+	return count;
+}
+static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
+
+static ssize_t nvme_sysfs_show_transport(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
+
+	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
+}
+static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
+
+static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
+
+	return snprintf(buf, PAGE_SIZE, "%s\n",
+			ctrl->ops->get_subsysnqn(ctrl));
+}
+static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
+
+static ssize_t nvme_sysfs_show_address(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
+
+	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
+}
+static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
+
 static struct attribute *nvme_dev_attrs[] = {
 	&dev_attr_reset_controller.attr,
 	&dev_attr_rescan_controller.attr,
@@ -1414,11 +1457,38 @@ static struct attribute *nvme_dev_attrs[] = {
 	&dev_attr_serial.attr,
 	&dev_attr_firmware_rev.attr,
 	&dev_attr_cntlid.attr,
+	&dev_attr_delete_controller.attr,
+	&dev_attr_transport.attr,
+	&dev_attr_subsysnqn.attr,
+	&dev_attr_address.attr,
 	NULL
 };
 
+#define CHECK_ATTR(ctrl, a, name)		\
+	if ((a) == &dev_attr_##name.attr &&	\
+	    !(ctrl)->ops->get_##name)		\
+		return 0
+
+static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
+		struct attribute *a, int n)
+{
+	struct device *dev = container_of(kobj, struct device, kobj);
+	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
+
+	if (a == &dev_attr_delete_controller.attr) {
+		if (!ctrl->ops->delete_ctrl)
+			return 0;
+	}
+
+	CHECK_ATTR(ctrl, a, subsysnqn);
+	CHECK_ATTR(ctrl, a, address);
+
+	return a->mode;
+}
+
 static struct attribute_group nvme_dev_attrs_group = {
-	.attrs = nvme_dev_attrs,
+	.attrs		= nvme_dev_attrs,
+	.is_visible	= nvme_dev_attrs_are_visible,
 };
 
 static const struct attribute_group *nvme_dev_attr_groups[] = {
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 180669c..b90e95a 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -144,6 +144,7 @@ struct nvme_ns {
 };
 
 struct nvme_ctrl_ops {
+	const char *name;
 	struct module *module;
 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
@@ -152,6 +153,9 @@ struct nvme_ctrl_ops {
 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
 	void (*post_scan)(struct nvme_ctrl *ctrl);
 	void (*submit_async_event)(struct nvme_ctrl *ctrl, int aer_idx);
+	int (*delete_ctrl)(struct nvme_ctrl *ctrl);
+	const char *(*get_subsysnqn)(struct nvme_ctrl *ctrl);
+	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
 };
 
 static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl)
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 571faa2..518147b 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1874,6 +1874,7 @@ static int nvme_pci_reset_ctrl(struct nvme_ctrl *ctrl)
 }
 
 static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
+	.name			= "pcie",
 	.module			= THIS_MODULE,
 	.reg_read32		= nvme_pci_reg_read32,
 	.reg_write32		= nvme_pci_reg_write32,
-- 
2.1.4

[PATCH 5/8] nvme.h: add NVMe over Fabrics definitions

[Christoph Hellwig]

The NVMe over Fabrics specification defines a protocol interface and
related extensions to NVMe that enable operation over network protocols.
The NVMe over Fabrics specification has an NVMe Transport binding for
each NVMe Transport.

This patch adds the fabrics related definitions:
- fabric specific command set and error codes
- transport addressing and binding definitions
- fabrics sgl extensions
- controller identification fabrics enhancements
- discovery log page definition

Signed-off-by: Armen Baloyan <armenx.baloyan@intel.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com>
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/core.c |   4 +-
 drivers/nvme/host/pci.c  |   4 +-
 include/linux/nvme.h     | 337 ++++++++++++++++++++++++++++++++++++++++++++---
 3 files changed, 322 insertions(+), 23 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index c272eb9..0c1a9b5 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -524,7 +524,7 @@ int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 	memset(&c, 0, sizeof(c));
 	c.features.opcode = nvme_admin_get_features;
 	c.features.nsid = cpu_to_le32(nsid);
-	c.features.prp1 = cpu_to_le64(dma_addr);
+	c.features.dptr.prp1 = cpu_to_le64(dma_addr);
 	c.features.fid = cpu_to_le32(fid);
 
 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
@@ -543,7 +543,7 @@ int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 
 	memset(&c, 0, sizeof(c));
 	c.features.opcode = nvme_admin_set_features;
-	c.features.prp1 = cpu_to_le64(dma_addr);
+	c.features.dptr.prp1 = cpu_to_le64(dma_addr);
 	c.features.fid = cpu_to_le32(fid);
 	c.features.dword11 = cpu_to_le32(dword11);
 
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 518147b..90d9796 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -520,8 +520,8 @@ static int nvme_map_data(struct nvme_dev *dev, struct request *req,
 			goto out_unmap;
 	}
 
-	cmnd->rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
-	cmnd->rw.prp2 = cpu_to_le64(iod->first_dma);
+	cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
+	cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma);
 	if (blk_integrity_rq(req))
 		cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
 	return BLK_MQ_RQ_QUEUE_OK;
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index dc815cc..7525030 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -16,6 +16,78 @@
 #define _LINUX_NVME_H
 
 #include <linux/types.h>
+#include <linux/uuid.h>
+
+/* NQN names in commands fields specified one size */
+#define NVMF_NQN_FIELD_LEN	256
+
+/* However the max length of a qualified name is another size */
+#define NVMF_NQN_SIZE		223
+
+#define NVMF_TRSVCID_SIZE	32
+#define NVMF_TRADDR_SIZE	256
+#define NVMF_TSAS_SIZE		256
+
+#define NVME_DISC_SUBSYS_NAME	"nqn.2014-08.org.nvmexpress.discovery"
+
+#define NVME_RDMA_IP_PORT	4420
+
+enum nvme_subsys_type {
+	NVME_NQN_DISC	= 1,		/* Discovery type target subsystem */
+	NVME_NQN_NVME	= 2,		/* NVME type target subsystem */
+};
+
+/* Address Family codes for Discovery Log Page entry ADRFAM field */
+enum {
+	NVMF_ADDR_FAMILY_PCI	= 0,	/* PCIe */
+	NVMF_ADDR_FAMILY_IP4	= 1,	/* IP4 */
+	NVMF_ADDR_FAMILY_IP6	= 2,	/* IP6 */
+	NVMF_ADDR_FAMILY_IB	= 3,	/* InfiniBand */
+	NVMF_ADDR_FAMILY_FC	= 4,	/* Fibre Channel */
+};
+
+/* Transport Type codes for Discovery Log Page entry TRTYPE field */
+enum {
+	NVMF_TRTYPE_RDMA	= 1,	/* RDMA */
+	NVMF_TRTYPE_FC		= 2,	/* Fibre Channel */
+	NVMF_TRTYPE_LOOP	= 254,	/* Reserved for host usage */
+	NVMF_TRTYPE_MAX,
+};
+
+/* Transport Requirements codes for Discovery Log Page entry TREQ field */
+enum {
+	NVMF_TREQ_NOT_SPECIFIED	= 0,	/* Not specified */
+	NVMF_TREQ_REQUIRED	= 1,	/* Required */
+	NVMF_TREQ_NOT_REQUIRED	= 2,	/* Not Required */
+};
+
+/* RDMA QP Service Type codes for Discovery Log Page entry TSAS
+ * RDMA_QPTYPE field
+ */
+enum {
+	NVMF_RDMA_QPTYPE_CONNECTED	= 0, /* Reliable Connected */
+	NVMF_RDMA_QPTYPE_DATAGRAM	= 1, /* Reliable Datagram */
+};
+
+/* RDMA QP Service Type codes for Discovery Log Page entry TSAS
+ * RDMA_QPTYPE field
+ */
+enum {
+	NVMF_RDMA_PRTYPE_NOT_SPECIFIED	= 0, /* No Provider Specified */
+	NVMF_RDMA_PRTYPE_IB		= 1, /* InfiniBand */
+	NVMF_RDMA_PRTYPE_ROCE		= 2, /* InfiniBand RoCE */
+	NVMF_RDMA_PRTYPE_ROCEV2		= 3, /* InfiniBand RoCEV2 */
+	NVMF_RDMA_PRTYPE_IWARP		= 4, /* IWARP */
+};
+
+/* RDMA Connection Management Service Type codes for Discovery Log Page
+ * entry TSAS RDMA_CMS field
+ */
+enum {
+	NVMF_RDMA_CMS_RDMA_CM	= 0, /* Sockets based enpoint addressing */
+};
+
+#define NVMF_AQ_DEPTH		32
 
 enum {
 	NVME_REG_CAP	= 0x0000,	/* Controller Capabilities */
@@ -117,7 +189,8 @@ struct nvme_id_ctrl {
 	__le32			rtd3r;
 	__le32			rtd3e;
 	__le32			oaes;
-	__u8			rsvd96[160];
+	__le32			ctratt;
+	__u8			rsvd100[156];
 	__le16			oacs;
 	__u8			acl;
 	__u8			aerl;
@@ -132,7 +205,7 @@ struct nvme_id_ctrl {
 	__u8			rsvd270[242];
 	__u8			sqes;
 	__u8			cqes;
-	__u8			rsvd514[2];
+	__le16			maxcmd;
 	__le32			nn;
 	__le16			oncs;
 	__le16			fuses;
@@ -145,7 +218,15 @@ struct nvme_id_ctrl {
 	__le16			acwu;
 	__u8			rsvd534[2];
 	__le32			sgls;
-	__u8			rsvd540[1508];
+	__u8			rsvd540[228];
+	char			subnqn[256];
+	__u8			rsvd1024[768];
+	__le32			ioccsz;
+	__le32			iorcsz;
+	__le16			icdoff;
+	__u8			ctrattr;
+	__u8			msdbd;
+	__u8			rsvd1804[244];
 	struct nvme_id_power_state	psd[32];
 	__u8			vs[1024];
 };
@@ -307,6 +388,61 @@ enum nvme_opcode {
 };
 
 /*
+ * Descriptor subtype - lower 4 bits of nvme_(keyed_)sgl_desc identifier
+ *
+ * @NVME_SGL_FMT_ADDRESS:     absolute address of the data block
+ * @NVME_SGL_FMT_OFFSET:      relative offset of the in-capsule data block
+ * @NVME_SGL_FMT_INVALIDATE:  RDMA transport specific remote invalidation
+ *                            request subtype
+ */
+enum {
+	NVME_SGL_FMT_ADDRESS		= 0x00,
+	NVME_SGL_FMT_OFFSET		= 0x01,
+	NVME_SGL_FMT_INVALIDATE		= 0x0f,
+};
+
+/*
+ * Descriptor type - upper 4 bits of nvme_(keyed_)sgl_desc identifier
+ *
+ * For struct nvme_sgl_desc:
+ *   @NVME_SGL_FMT_DATA_DESC:		data block descriptor
+ *   @NVME_SGL_FMT_SEG_DESC:		sgl segment descriptor
+ *   @NVME_SGL_FMT_LAST_SEG_DESC:	last sgl segment descriptor
+ *
+ * For struct nvme_keyed_sgl_desc:
+ *   @NVME_KEY_SGL_FMT_DATA_DESC:	keyed data block descriptor
+ */
+enum {
+	NVME_SGL_FMT_DATA_DESC		= 0x00,
+	NVME_SGL_FMT_SEG_DESC		= 0x02,
+	NVME_SGL_FMT_LAST_SEG_DESC	= 0x03,
+	NVME_KEY_SGL_FMT_DATA_DESC	= 0x04,
+};
+
+struct nvme_sgl_desc {
+	__le64	addr;
+	__le32	length;
+	__u8	rsvd[3];
+	__u8	type;
+};
+
+struct nvme_keyed_sgl_desc {
+	__le64	addr;
+	__u8	length[3];
+	__u8	key[4];
+	__u8	type;
+};
+
+union nvme_data_ptr {
+	struct {
+		__le64	prp1;
+		__le64	prp2;
+	};
+	struct nvme_sgl_desc	sgl;
+	struct nvme_keyed_sgl_desc ksgl;
+};
+
+/*
  * Lowest two bits of our flags field (FUSE field in the spec):
  *
  * @NVME_CMD_FUSE_FIRST:   Fused Operation, first command
@@ -336,8 +472,7 @@ struct nvme_common_command {
 	__le32			nsid;
 	__le32			cdw2[2];
 	__le64			metadata;
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__le32			cdw10[6];
 };
 
@@ -348,8 +483,7 @@ struct nvme_rw_command {
 	__le32			nsid;
 	__u64			rsvd2;
 	__le64			metadata;
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__le64			slba;
 	__le16			length;
 	__le16			control;
@@ -389,8 +523,7 @@ struct nvme_dsm_cmd {
 	__u16			command_id;
 	__le32			nsid;
 	__u64			rsvd2[2];
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__le32			nr;
 	__le32			attributes;
 	__u32			rsvd12[4];
@@ -454,6 +587,7 @@ enum {
 	NVME_LOG_ERROR		= 0x01,
 	NVME_LOG_SMART		= 0x02,
 	NVME_LOG_FW_SLOT	= 0x03,
+	NVME_LOG_DISC		= 0x70,
 	NVME_LOG_RESERVATION	= 0x80,
 	NVME_FWACT_REPL		= (0 << 3),
 	NVME_FWACT_REPL_ACTV	= (1 << 3),
@@ -466,8 +600,7 @@ struct nvme_identify {
 	__u16			command_id;
 	__le32			nsid;
 	__u64			rsvd2[2];
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__le32			cns;
 	__u32			rsvd11[5];
 };
@@ -478,8 +611,7 @@ struct nvme_features {
 	__u16			command_id;
 	__le32			nsid;
 	__u64			rsvd2[2];
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__le32			fid;
 	__le32			dword11;
 	__u32			rsvd12[4];
@@ -538,8 +670,7 @@ struct nvme_download_firmware {
 	__u8			flags;
 	__u16			command_id;
 	__u32			rsvd1[5];
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__le32			numd;
 	__le32			offset;
 	__u32			rsvd12[4];
@@ -561,8 +692,7 @@ struct nvme_get_log_page_command {
 	__u16			command_id;
 	__le32			nsid;
 	__u64			rsvd2[2];
-	__le64			prp1;
-	__le64			prp2;
+	union nvme_data_ptr	dptr;
 	__u8			lid;
 	__u8			rsvd10;
 	__le16			numdl;
@@ -573,6 +703,126 @@ struct nvme_get_log_page_command {
 	__u32			rsvd14[2];
 };
 
+/*
+ * Fabrics subcommands.
+ */
+enum nvmf_fabrics_opcode {
+	nvme_fabrics_command		= 0x7f,
+};
+
+enum nvmf_capsule_command {
+	nvme_fabrics_type_property_set	= 0x00,
+	nvme_fabrics_type_connect	= 0x01,
+	nvme_fabrics_type_property_get	= 0x04,
+};
+
+struct nvmf_common_command {
+	__u8	opcode;
+	__u8	resv1;
+	__u16	command_id;
+	__u8	fctype;
+	__u8	resv2[35];
+	__u8	ts[24];
+};
+
+/*
+ * The legal cntlid range a NVMe Target will provide.
+ * Note that cntlid of value 0 is considered illegal in the fabrics world.
+ * Devices based on earlier specs did not have the subsystem concept;
+ * therefore, those devices had their cntlid value set to 0 as a result.
+ */
+#define NVME_CNTLID_MIN		1
+#define NVME_CNTLID_MAX		0xffef
+#define NVME_CNTLID_DYNAMIC	0xffff
+
+#define MAX_DISC_LOGS	255
+
+/* Discovery log page entry */
+struct nvmf_disc_rsp_page_entry {
+	__u8		trtype;
+	__u8		adrfam;
+	__u8		nqntype;
+	__u8		treq;
+	__le16		portid;
+	__le16		cntlid;
+	__le16		asqsz;
+	__u8		resv8[22];
+	char		trsvcid[NVMF_TRSVCID_SIZE];
+	__u8		resv64[192];
+	char		subnqn[NVMF_NQN_FIELD_LEN];
+	char		traddr[NVMF_TRADDR_SIZE];
+	union tsas {
+		char		common[NVMF_TSAS_SIZE];
+		struct rdma {
+			__u8	qptype;
+			__u8	prtype;
+			__u8	cms;
+			__u8	resv3[5];
+			__u16	pkey;
+			__u8	resv10[246];
+		} rdma;
+	} tsas;
+};
+
+/* Discovery log page header */
+struct nvmf_disc_rsp_page_hdr {
+	__le64		genctr;
+	__le64		numrec;
+	__le16		recfmt;
+	__u8		resv14[1006];
+	struct nvmf_disc_rsp_page_entry entries[0];
+};
+
+struct nvmf_connect_command {
+	__u8		opcode;
+	__u8		resv1;
+	__u16		command_id;
+	__u8		fctype;
+	__u8		resv2[19];
+	union nvme_data_ptr dptr;
+	__le16		recfmt;
+	__le16		qid;
+	__le16		sqsize;
+	__u8		cattr;
+	__u8		resv3;
+	__le32		kato;
+	__u8		resv4[12];
+};
+
+struct nvmf_connect_data {
+	uuid_le		hostid;
+	__le16		cntlid;
+	char		resv4[238];
+	char		subsysnqn[NVMF_NQN_FIELD_LEN];
+	char		hostnqn[NVMF_NQN_FIELD_LEN];
+	char		resv5[256];
+};
+
+struct nvmf_property_set_command {
+	__u8		opcode;
+	__u8		resv1;
+	__u16		command_id;
+	__u8		fctype;
+	__u8		resv2[35];
+	__u8		attrib;
+	__u8		resv3[3];
+	__le32		offset;
+	__le64		value;
+	__u8		resv4[8];
+};
+
+struct nvmf_property_get_command {
+	__u8		opcode;
+	__u8		resv1;
+	__u16		command_id;
+	__u8		fctype;
+	__u8		resv2[35];
+	__u8		attrib;
+	__u8		resv3[3];
+	__le32		offset;
+	__u8		resv4[16];
+};
+
 struct nvme_command {
 	union {
 		struct nvme_common_command common;
@@ -587,15 +837,29 @@ struct nvme_command {
 		struct nvme_dsm_cmd dsm;
 		struct nvme_abort_cmd abort;
 		struct nvme_get_log_page_command get_log_page;
+		struct nvmf_common_command fabrics;
+		struct nvmf_connect_command connect;
+		struct nvmf_property_set_command prop_set;
+		struct nvmf_property_get_command prop_get;
 	};
 };
 
 static inline bool nvme_is_write(struct nvme_command *cmd)
 {
+	/*
+	 * What a mess...
+	 *
+	 * Why can't we simply have a Fabrics In and Fabrics out command?
+	 */
+	if (unlikely(cmd->common.opcode == nvme_fabrics_command))
+		return cmd->fabrics.opcode & 1;
 	return cmd->common.opcode & 1;
 }
 
 enum {
+	/*
+	 * Generic Command Status:
+	 */
 	NVME_SC_SUCCESS			= 0x0,
 	NVME_SC_INVALID_OPCODE		= 0x1,
 	NVME_SC_INVALID_FIELD		= 0x2,
@@ -614,10 +878,18 @@ enum {
 	NVME_SC_SGL_INVALID_DATA	= 0xf,
 	NVME_SC_SGL_INVALID_METADATA	= 0x10,
 	NVME_SC_SGL_INVALID_TYPE	= 0x11,
+
+	NVME_SC_SGL_INVALID_OFFSET	= 0x16,
+	NVME_SC_SGL_INVALID_SUBTYPE	= 0x17,
+
 	NVME_SC_LBA_RANGE		= 0x80,
 	NVME_SC_CAP_EXCEEDED		= 0x81,
 	NVME_SC_NS_NOT_READY		= 0x82,
 	NVME_SC_RESERVATION_CONFLICT	= 0x83,
+
+	/*
+	 * Command Specific Status:
+	 */
 	NVME_SC_CQ_INVALID		= 0x100,
 	NVME_SC_QID_INVALID		= 0x101,
 	NVME_SC_QUEUE_SIZE		= 0x102,
@@ -635,9 +907,29 @@ enum {
 	NVME_SC_FEATURE_NOT_CHANGEABLE	= 0x10e,
 	NVME_SC_FEATURE_NOT_PER_NS	= 0x10f,
 	NVME_SC_FW_NEEDS_RESET_SUBSYS	= 0x110,
+
+	/*
+	 * I/O Command Set Specific - NVM commands:
+	 */
 	NVME_SC_BAD_ATTRIBUTES		= 0x180,
 	NVME_SC_INVALID_PI		= 0x181,
 	NVME_SC_READ_ONLY		= 0x182,
+
+	/*
+	 * I/O Command Set Specific - Fabrics commands:
+	 */
+	NVME_SC_CONNECT_FORMAT		= 0x180,
+	NVME_SC_CONNECT_CTRL_BUSY	= 0x181,
+	NVME_SC_CONNECT_INVALID_PARAM	= 0x182,
+	NVME_SC_CONNECT_RESTART_DISC	= 0x183,
+	NVME_SC_CONNECT_INVALID_HOST	= 0x184,
+
+	NVME_SC_DISCOVERY_RESTART	= 0x190,
+	NVME_SC_AUTH_REQUIRED		= 0x191,
+
+	/*
+	 * Media and Data Integrity Errors:
+	 */
 	NVME_SC_WRITE_FAULT		= 0x280,
 	NVME_SC_READ_ERROR		= 0x281,
 	NVME_SC_GUARD_CHECK		= 0x282,
@@ -645,12 +937,19 @@ enum {
 	NVME_SC_REFTAG_CHECK		= 0x284,
 	NVME_SC_COMPARE_FAILED		= 0x285,
 	NVME_SC_ACCESS_DENIED		= 0x286,
+
 	NVME_SC_DNR			= 0x4000,
 };
 
 struct nvme_completion {
-	__le32	result;		/* Used by admin commands to return data */
-	__u32	rsvd;
+	/*
+	 * Used by Admin and Fabrics commands to return data:
+	 */
+	union {
+		__le16	result16;
+		__le32	result;
+		__le64	result64;
+	};
 	__le16	sq_head;	/* how much of this queue may be reclaimed */
 	__le16	sq_id;		/* submission queue that generated this entry */
 	__u16	command_id;	/* of the command which completed */
-- 
2.1.4

[PATCH 6/8] nvme-fabrics: add a generic NVMe over Fabrics library

[Christoph Hellwig]

The NVMe over Fabrics library provides an interface for both transports
and the nvme core to handle fabrics specific commands and attributes
independent of the underlying transport.

In addition, the fabrics library adds a misc device interface that allow
actually creating a fabrics controller, as we can't just autodiscover
it like in the PCI case.  The nvme-cli utility has been enhanced to use
this interface to support fabric connect and discovery.

Signed-off-by: Armen Baloyan <armenx.baloyan@intel.com>,
Signed-off-by: Jay Freyensee <james.p.freyensee@intel.com>,
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/Kconfig   |   3 +
 drivers/nvme/host/Makefile  |   3 +
 drivers/nvme/host/core.c    |  19 +-
 drivers/nvme/host/fabrics.c | 932 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/nvme/host/fabrics.h | 132 +++++++
 drivers/nvme/host/nvme.h    |  11 +
 6 files changed, 1099 insertions(+), 1 deletion(-)
 create mode 100644 drivers/nvme/host/fabrics.c
 create mode 100644 drivers/nvme/host/fabrics.h

diff --git a/drivers/nvme/host/Kconfig b/drivers/nvme/host/Kconfig
index d296fc3..3397651 100644
--- a/drivers/nvme/host/Kconfig
+++ b/drivers/nvme/host/Kconfig
@@ -24,3 +24,6 @@ config BLK_DEV_NVME_SCSI
 	  to say N here, unless you run a distro that abuses the SCSI
 	  emulation to provide stable device names for mount by id, like
 	  some OpenSuSE and SLES versions.
+
+config NVME_FABRICS
+	tristate
diff --git a/drivers/nvme/host/Makefile b/drivers/nvme/host/Makefile
index 9a3ca89..5f8648f 100644
--- a/drivers/nvme/host/Makefile
+++ b/drivers/nvme/host/Makefile
@@ -1,8 +1,11 @@
 obj-$(CONFIG_NVME_CORE)			+= nvme-core.o
 obj-$(CONFIG_BLK_DEV_NVME)		+= nvme.o
+obj-$(CONFIG_NVME_FABRICS)		+= nvme-fabrics.o
 
 nvme-core-y				:= core.o
 nvme-core-$(CONFIG_BLK_DEV_NVME_SCSI)	+= scsi.o
 nvme-core-$(CONFIG_NVM)			+= lightnvm.o
 
 nvme-y					+= pci.o
+
+nvme-fabrics-y				+= fabrics.o
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 0c1a9b5..1f70393 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1178,9 +1178,26 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
 	}
 
 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
+	ctrl->sgls = le32_to_cpu(id->sgls);
+
+	if (ctrl->ops->is_fabrics) {
+		ctrl->icdoff = le16_to_cpu(id->icdoff);
+		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
+		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
+		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
+
+		/*
+		 * In fabrics we need to verify the cntlid matches the
+		 * admin connect
+		 */
+		if (ctrl->cntlid != le16_to_cpu(id->cntlid))
+			ret = -EINVAL;
+	} else {
+		ctrl->cntlid = le16_to_cpu(id->cntlid);
+	}
 
 	kfree(id);
-	return 0;
+	return ret;
 }
 EXPORT_SYMBOL_GPL(nvme_init_identify);
 
diff --git a/drivers/nvme/host/fabrics.c b/drivers/nvme/host/fabrics.c
new file mode 100644
index 0000000..8372eb6
--- /dev/null
+++ b/drivers/nvme/host/fabrics.c
@@ -0,0 +1,932 @@
+/*
+ * NVMe over Fabrics common host code.
+ * Copyright (c) 2015-2016 HGST, a Western Digital Company.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ */
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+#include <linux/init.h>
+#include <linux/miscdevice.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/parser.h>
+#include <linux/seq_file.h>
+#include "nvme.h"
+#include "fabrics.h"
+
+static LIST_HEAD(nvmf_transports);
+static DEFINE_MUTEX(nvmf_transports_mutex);
+
+static LIST_HEAD(nvmf_hosts);
+static DEFINE_MUTEX(nvmf_hosts_mutex);
+
+static struct nvmf_host *nvmf_default_host;
+
+static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
+{
+	struct nvmf_host *host;
+
+	list_for_each_entry(host, &nvmf_hosts, list) {
+		if (!strcmp(host->nqn, hostnqn))
+			return host;
+	}
+
+	return NULL;
+}
+
+static struct nvmf_host *nvmf_host_add(const char *hostnqn)
+{
+	struct nvmf_host *host;
+
+	mutex_lock(&nvmf_hosts_mutex);
+	host = __nvmf_host_find(hostnqn);
+	if (host)
+		goto out_unlock;
+
+	host = kmalloc(sizeof(*host), GFP_KERNEL);
+	if (!host)
+		goto out_unlock;
+
+	kref_init(&host->ref);
+	memcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
+	uuid_le_gen(&host->id);
+
+	list_add_tail(&host->list, &nvmf_hosts);
+out_unlock:
+	mutex_unlock(&nvmf_hosts_mutex);
+	return host;
+}
+
+static struct nvmf_host *nvmf_host_default(void)
+{
+	struct nvmf_host *host;
+
+	host = kmalloc(sizeof(*host), GFP_KERNEL);
+	if (!host)
+		return NULL;
+
+	kref_init(&host->ref);
+	uuid_le_gen(&host->id);
+	snprintf(host->nqn, NVMF_NQN_SIZE,
+		"nqn.2014-08.org.nvmexpress:NVMf:uuid:%pUl", &host->id);
+
+	mutex_lock(&nvmf_hosts_mutex);
+	list_add_tail(&host->list, &nvmf_hosts);
+	mutex_unlock(&nvmf_hosts_mutex);
+
+	return host;
+}
+
+static void nvmf_host_destroy(struct kref *ref)
+{
+	struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
+
+	kfree(host);
+}
+
+static void nvmf_host_put(struct nvmf_host *host)
+{
+	if (host)
+		kref_put(&host->ref, nvmf_host_destroy);
+}
+
+/**
+ * nvmf_get_address() -  Get address/port
+ * @ctrl:	Host NVMe controller instance which we got the address
+ * @buf:	OUTPUT parameter that will contain the address/port
+ * @size:	buffer size
+ */
+int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
+{
+	return snprintf(buf, size, "traddr=%s,trsvcid=%s\n",
+			ctrl->opts->traddr, ctrl->opts->trsvcid);
+}
+EXPORT_SYMBOL_GPL(nvmf_get_address);
+
+/**
+ * nvmf_get_subsysnqn() - Get subsystem NQN
+ * @ctrl:	Host NVMe controller instance which we got the NQN
+ */
+const char *nvmf_get_subsysnqn(struct nvme_ctrl *ctrl)
+{
+	return ctrl->opts->subsysnqn;
+}
+EXPORT_SYMBOL_GPL(nvmf_get_subsysnqn);
+
+/**
+ * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
+ * @ctrl:	Host NVMe controller instance maintaining the admin
+ *		queue used to submit the property read command to
+ *		the allocated NVMe controller resource on the target system.
+ * @off:	Starting offset value of the targeted property
+ *		register (see the fabrics section of the NVMe standard).
+ * @val:	OUTPUT parameter that will contain the value of
+ *		the property after a successful read.
+ *
+ * Used by the host system to retrieve a 32-bit capsule property value
+ * from an NVMe controller on the target system.
+ *
+ * ("Capsule property" is an "PCIe register concept" applied to the
+ * NVMe fabrics space.)
+ *
+ * Return:
+ *	0: successful read
+ *	> 0: NVMe error status code
+ *	< 0: Linux errno error code
+ */
+int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
+{
+	struct nvme_command cmd;
+	struct nvme_completion cqe;
+	int ret;
+
+	memset(&cmd, 0, sizeof(cmd));
+	cmd.prop_get.opcode = nvme_fabrics_command;
+	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
+	cmd.prop_get.offset = cpu_to_le32(off);
+
+	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
+			NVME_QID_ANY, 0, 0);
+
+	if (ret >= 0)
+		*val = le64_to_cpu(cqe.result64);
+	if (unlikely(ret != 0))
+		dev_err(ctrl->device,
+			"Property Get error: %d, offset %#x\n",
+			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvmf_reg_read32);
+
+/**
+ * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
+ * @ctrl:	Host NVMe controller instance maintaining the admin
+ *		queue used to submit the property read command to
+ *		the allocated controller resource on the target system.
+ * @off:	Starting offset value of the targeted property
+ *		register (see the fabrics section of the NVMe standard).
+ * @val:	OUTPUT parameter that will contain the value of
+ *		the property after a successful read.
+ *
+ * Used by the host system to retrieve a 64-bit capsule property value
+ * from an NVMe controller on the target system.
+ *
+ * ("Capsule property" is an "PCIe register concept" applied to the
+ * NVMe fabrics space.)
+ *
+ * Return:
+ *	0: successful read
+ *	> 0: NVMe error status code
+ *	< 0: Linux errno error code
+ */
+int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
+{
+	struct nvme_command cmd;
+	struct nvme_completion cqe;
+	int ret;
+
+	memset(&cmd, 0, sizeof(cmd));
+	cmd.prop_get.opcode = nvme_fabrics_command;
+	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
+	cmd.prop_get.attrib = 1;
+	cmd.prop_get.offset = cpu_to_le32(off);
+
+	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
+			NVME_QID_ANY, 0, 0);
+
+	if (ret >= 0)
+		*val = le64_to_cpu(cqe.result64);
+	if (unlikely(ret != 0))
+		dev_err(ctrl->device,
+			"Property Get error: %d, offset %#x\n",
+			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvmf_reg_read64);
+
+/**
+ * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
+ * @ctrl:	Host NVMe controller instance maintaining the admin
+ *		queue used to submit the property read command to
+ *		the allocated NVMe controller resource on the target system.
+ * @off:	Starting offset value of the targeted property
+ *		register (see the fabrics section of the NVMe standard).
+ * @val:	Input parameter that contains the value to be
+ *		written to the property.
+ *
+ * Used by the NVMe host system to write a 32-bit capsule property value
+ * to an NVMe controller on the target system.
+ *
+ * ("Capsule property" is an "PCIe register concept" applied to the
+ * NVMe fabrics space.)
+ *
+ * Return:
+ *	0: successful write
+ *	> 0: NVMe error status code
+ *	< 0: Linux errno error code
+ */
+int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
+{
+	struct nvme_command cmd;
+	int ret;
+
+	memset(&cmd, 0, sizeof(cmd));
+	cmd.prop_set.opcode = nvme_fabrics_command;
+	cmd.prop_set.fctype = nvme_fabrics_type_property_set;
+	cmd.prop_set.attrib = 0;
+	cmd.prop_set.offset = cpu_to_le32(off);
+	cmd.prop_set.value = cpu_to_le64(val);
+
+	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
+			NVME_QID_ANY, 0, 0);
+	if (unlikely(ret))
+		dev_err(ctrl->device,
+			"Property Set error: %d, offset %#x\n",
+			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvmf_reg_write32);
+
+/**
+ * nvmf_log_connect_error() - Error-parsing-diagnostic print
+ * out function for connect() errors.
+ *
+ * @ctrl: the specific /dev/nvmeX device that had the error.
+ *
+ * @errval: Error code to be decoded in a more human-friendly
+ *	    printout.
+ *
+ * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
+ *
+ * @cmd: This is the SQE portion of a submission capsule.
+ *
+ * @data: This is the "Data" portion of a submission capsule.
+ */
+static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
+		int errval, int offset, struct nvme_command *cmd,
+		struct nvmf_connect_data *data)
+{
+	int err_sctype = errval & (~NVME_SC_DNR);
+
+	switch (err_sctype) {
+
+	case (NVME_SC_CONNECT_INVALID_PARAM):
+		if (offset >> 16) {
+			char *inv_data = "Connect Invalid Data Parameter";
+
+			switch (offset & 0xffff) {
+			case (offsetof(struct nvmf_connect_data, cntlid)):
+				dev_err(ctrl->device,
+					"%s, cntlid: %d\n",
+					inv_data, data->cntlid);
+				break;
+			case (offsetof(struct nvmf_connect_data, hostnqn)):
+				dev_err(ctrl->device,
+					"%s, hostnqn \"%s\"\n",
+					inv_data, data->hostnqn);
+				break;
+			case (offsetof(struct nvmf_connect_data, subsysnqn)):
+				dev_err(ctrl->device,
+					"%s, subsysnqn \"%s\"\n",
+					inv_data, data->subsysnqn);
+				break;
+			default:
+				dev_err(ctrl->device,
+					"%s, starting byte offset: %d\n",
+				       inv_data, offset & 0xffff);
+				break;
+			}
+		} else {
+			char *inv_sqe = "Connect Invalid SQE Parameter";
+
+			switch (offset) {
+			case (offsetof(struct nvmf_connect_command, qid)):
+				dev_err(ctrl->device,
+				       "%s, qid %d\n",
+					inv_sqe, cmd->connect.qid);
+				break;
+			default:
+				dev_err(ctrl->device,
+					"%s, starting byte offset: %d\n",
+					inv_sqe, offset);
+			}
+		}
+		break;
+	default:
+		dev_err(ctrl->device,
+			"Connect command failed, error wo/DNR bit: %d\n",
+			err_sctype);
+		break;
+	} /* switch (err_sctype) */
+}
+
+/**
+ * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
+ *				API function.
+ * @ctrl:	Host nvme controller instance used to request
+ *              a new NVMe controller allocation on the target
+ *              system and  establish an NVMe Admin connection to
+ *              that controller.
+ *
+ * This function enables an NVMe host device to request a new allocation of
+ * an NVMe controller resource on a target system as well establish a
+ * fabrics-protocol connection of the NVMe Admin queue between the
+ * host system device and the allocated NVMe controller on the
+ * target system via a NVMe Fabrics "Connect" command.
+ *
+ * Return:
+ *	0: success
+ *	> 0: NVMe error status code
+ *	< 0: Linux errno error code
+ *
+ */
+int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
+{
+	struct nvme_command cmd;
+	struct nvme_completion cqe;
+	struct nvmf_connect_data *data;
+	int ret;
+
+	memset(&cmd, 0, sizeof(cmd));
+	cmd.connect.opcode = nvme_fabrics_command;
+	cmd.connect.fctype = nvme_fabrics_type_connect;
+	cmd.connect.qid = 0;
+	cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
+
+	data = kzalloc(sizeof(*data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_le));
+	data->cntlid = cpu_to_le16(0xffff);
+	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
+	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
+
+	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe,
+			data, sizeof(*data), 0, NVME_QID_ANY, 1,
+			BLK_MQ_REQ_RESERVED);
+	if (ret) {
+		nvmf_log_connect_error(ctrl,
+					 ret, le32_to_cpu(cqe.result),
+					 &cmd, data);
+		goto out_free_data;
+	}
+
+	ctrl->cntlid = le16_to_cpu(cqe.result16);
+
+out_free_data:
+	kfree(data);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
+
+/**
+ * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
+ *			     API function.
+ * @ctrl:	Host nvme controller instance used to establish an
+ *		NVMe I/O queue connection to the already allocated NVMe
+ *		controller on the target system.
+ * @qid:	NVMe I/O queue number for the new I/O connection between
+ *		host and target (note qid == 0 is illegal as this is
+ *		the Admin queue, per NVMe standard).
+ *
+ * This function issues a fabrics-protocol connection
+ * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
+ * between the host system device and the allocated NVMe controller
+ * on the target system.
+ *
+ * Return:
+ *	0: success
+ *	> 0: NVMe error status code
+ *	< 0: Linux errno error code
+ */
+int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
+{
+	struct nvme_command cmd;
+	struct nvmf_connect_data *data;
+	struct nvme_completion cqe;
+	int ret;
+
+	memset(&cmd, 0, sizeof(cmd));
+	cmd.connect.opcode = nvme_fabrics_command;
+	cmd.connect.fctype = nvme_fabrics_type_connect;
+	cmd.connect.qid = cpu_to_le16(qid);
+	cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
+
+	data = kzalloc(sizeof(*data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_le));
+	data->cntlid = cpu_to_le16(ctrl->cntlid);
+	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
+	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
+
+	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &cqe,
+			data, sizeof(*data), 0, qid, 1, BLK_MQ_REQ_RESERVED);
+	if (ret)
+		nvmf_log_connect_error(ctrl,
+					 ret, le32_to_cpu(cqe.result),
+					 &cmd, data);
+
+	kfree(data);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
+
+/**
+ * nvmf_register_transport() - NVMe Fabrics Library registration function.
+ * @ops:	Transport ops instance to be registered to the
+ *		common fabrics library.
+ *
+ * API function that registers the type of specific transport fabric
+ * being implemented to the common NVMe fabrics library. Part of
+ * the overall init sequence of starting up a fabrics driver.
+ */
+void nvmf_register_transport(struct nvmf_transport_ops *ops)
+{
+	mutex_lock(&nvmf_transports_mutex);
+	list_add_tail(&ops->entry, &nvmf_transports);
+	mutex_unlock(&nvmf_transports_mutex);
+}
+EXPORT_SYMBOL_GPL(nvmf_register_transport);
+
+/**
+ * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
+ * @ops:	Transport ops instance to be unregistered from the
+ *		common fabrics library.
+ *
+ * Fabrics API function that unregisters the type of specific transport
+ * fabric being implemented from the common NVMe fabrics library.
+ * Part of the overall exit sequence of unloading the implemented driver.
+ */
+void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
+{
+	mutex_lock(&nvmf_transports_mutex);
+	list_del(&ops->entry);
+	mutex_unlock(&nvmf_transports_mutex);
+}
+EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
+
+static struct nvmf_transport_ops *nvmf_lookup_transport(
+		struct nvmf_ctrl_options *opts)
+{
+	struct nvmf_transport_ops *ops;
+
+	lockdep_assert_held(&nvmf_transports_mutex);
+
+	list_for_each_entry(ops, &nvmf_transports, entry) {
+		if (strcmp(ops->name, opts->transport) == 0)
+			return ops;
+	}
+
+	return NULL;
+}
+
+static const match_table_t opt_tokens = {
+	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
+	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
+	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
+	{ NVMF_OPT_NQN,			"nqn=%s"		},
+	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
+	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
+	{ NVMF_OPT_TL_RETRY_COUNT,	"tl_retry_count=%d"	},
+	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
+	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
+	{ NVMF_OPT_ERR,			NULL			}
+};
+
+static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
+		const char *buf)
+{
+	substring_t args[MAX_OPT_ARGS];
+	char *options, *o, *p;
+	int token, ret = 0;
+	size_t nqnlen  = 0;
+
+	/* Set defaults */
+	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
+	opts->nr_io_queues = num_online_cpus();
+	opts->tl_retry_count = 2;
+	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
+
+	options = o = kstrdup(buf, GFP_KERNEL);
+	if (!options)
+		return -ENOMEM;
+
+	while ((p = strsep(&o, ",\n")) != NULL) {
+		if (!*p)
+			continue;
+
+		token = match_token(p, opt_tokens, args);
+		opts->mask |= token;
+		switch (token) {
+		case NVMF_OPT_TRANSPORT:
+			p = match_strdup(args);
+			if (!p) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			opts->transport = p;
+			break;
+		case NVMF_OPT_NQN:
+			p = match_strdup(args);
+			if (!p) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			opts->subsysnqn = p;
+			nqnlen = strlen(opts->subsysnqn);
+			if (nqnlen >= NVMF_NQN_SIZE) {
+				pr_err("%s needs to be < %d bytes\n",
+				opts->subsysnqn, NVMF_NQN_SIZE);
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->discovery_nqn =
+				!(strcmp(opts->subsysnqn,
+					 NVME_DISC_SUBSYS_NAME));
+			if (opts->discovery_nqn)
+				opts->nr_io_queues = 0;
+			break;
+		case NVMF_OPT_TRADDR:
+			p = match_strdup(args);
+			if (!p) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			opts->traddr = p;
+			break;
+		case NVMF_OPT_TRSVCID:
+			p = match_strdup(args);
+			if (!p) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			opts->trsvcid = p;
+			break;
+		case NVMF_OPT_QUEUE_SIZE:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+			if (token < NVMF_MIN_QUEUE_SIZE ||
+			    token > NVMF_MAX_QUEUE_SIZE) {
+				pr_err("Invalid queue_size %d\n", token);
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->queue_size = token;
+			break;
+		case NVMF_OPT_NR_IO_QUEUES:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+			if (token <= 0) {
+				pr_err("Invalid number of IOQs %d\n", token);
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->nr_io_queues = min_t(unsigned int,
+					num_online_cpus(), token);
+			break;
+		case NVMF_OPT_TL_RETRY_COUNT:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+			if (token < 0) {
+				pr_err("Invalid tl_retry_count %d\n", token);
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->tl_retry_count = token;
+			break;
+		case NVMF_OPT_HOSTNQN:
+			if (opts->host) {
+				pr_err("hostnqn already user-assigned: %s\n",
+				       opts->host->nqn);
+				ret = -EADDRINUSE;
+				goto out;
+			}
+			p = match_strdup(args);
+			if (!p) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			nqnlen = strlen(p);
+			if (nqnlen >= NVMF_NQN_SIZE) {
+				pr_err("%s needs to be < %d bytes\n",
+					p, NVMF_NQN_SIZE);
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->host = nvmf_host_add(p);
+			if (!opts->host) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			break;
+		case NVMF_OPT_RECONNECT_DELAY:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+			if (token <= 0) {
+				pr_err("Invalid reconnect_delay %d\n", token);
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->reconnect_delay = token;
+			break;
+		default:
+			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
+				p);
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (!opts->host) {
+		kref_get(&nvmf_default_host->ref);
+		opts->host = nvmf_default_host;
+	}
+
+out:
+	kfree(options);
+	return ret;
+}
+
+static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
+		unsigned int required_opts)
+{
+	if ((opts->mask & required_opts) != required_opts) {
+		int i;
+
+		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
+			if ((opt_tokens[i].token & required_opts) &&
+			    !(opt_tokens[i].token & opts->mask)) {
+				pr_warn("missing parameter '%s'\n",
+					opt_tokens[i].pattern);
+			}
+		}
+
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
+		unsigned int allowed_opts)
+{
+	if (opts->mask & ~allowed_opts) {
+		int i;
+
+		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
+			if (opt_tokens[i].token & ~allowed_opts) {
+				pr_warn("invalid parameter '%s'\n",
+					opt_tokens[i].pattern);
+			}
+		}
+
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+void nvmf_free_options(struct nvmf_ctrl_options *opts)
+{
+	nvmf_host_put(opts->host);
+	kfree(opts->transport);
+	kfree(opts->traddr);
+	kfree(opts->trsvcid);
+	kfree(opts->subsysnqn);
+	kfree(opts);
+}
+EXPORT_SYMBOL_GPL(nvmf_free_options);
+
+#define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
+#define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
+				 NVMF_OPT_HOSTNQN)
+
+static struct nvme_ctrl *
+nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
+{
+	struct nvmf_ctrl_options *opts;
+	struct nvmf_transport_ops *ops;
+	struct nvme_ctrl *ctrl;
+	int ret;
+
+	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
+	if (!opts)
+		return ERR_PTR(-ENOMEM);
+
+	ret = nvmf_parse_options(opts, buf);
+	if (ret)
+		goto out_free_opts;
+
+	/*
+	 * Check the generic options first as we need a valid transport for
+	 * the lookup below.  Then clear the generic flags so that transport
+	 * drivers don't have to care about them.
+	 */
+	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
+	if (ret)
+		goto out_free_opts;
+	opts->mask &= ~NVMF_REQUIRED_OPTS;
+
+	mutex_lock(&nvmf_transports_mutex);
+	ops = nvmf_lookup_transport(opts);
+	if (!ops) {
+		pr_info("no handler found for transport %s.\n",
+			opts->transport);
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	ret = nvmf_check_required_opts(opts, ops->required_opts);
+	if (ret)
+		goto out_unlock;
+	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
+				ops->allowed_opts | ops->required_opts);
+	if (ret)
+		goto out_unlock;
+
+	ctrl = ops->create_ctrl(dev, opts);
+	if (IS_ERR(ctrl)) {
+		ret = PTR_ERR(ctrl);
+		goto out_unlock;
+	}
+
+	mutex_unlock(&nvmf_transports_mutex);
+	return ctrl;
+
+out_unlock:
+	mutex_unlock(&nvmf_transports_mutex);
+out_free_opts:
+	kfree(opts->host);
+	kfree(opts);
+	return ERR_PTR(ret);
+}
+
+static struct class *nvmf_class;
+static struct device *nvmf_device;
+static DEFINE_MUTEX(nvmf_dev_mutex);
+
+static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
+		size_t count, loff_t *pos)
+{
+	struct seq_file *seq_file = file->private_data;
+	struct nvme_ctrl *ctrl;
+	const char *buf;
+	int ret = 0;
+
+	if (count > PAGE_SIZE)
+		return -ENOMEM;
+
+	buf = memdup_user_nul(ubuf, count);
+	if (IS_ERR(buf))
+		return PTR_ERR(buf);
+
+	mutex_lock(&nvmf_dev_mutex);
+	if (seq_file->private) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
+	if (IS_ERR(ctrl)) {
+		ret = PTR_ERR(ctrl);
+		goto out_unlock;
+	}
+
+	seq_file->private = ctrl;
+
+out_unlock:
+	mutex_unlock(&nvmf_dev_mutex);
+	kfree(buf);
+	return ret ? ret : count;
+}
+
+static int nvmf_dev_show(struct seq_file *seq_file, void *private)
+{
+	struct nvme_ctrl *ctrl;
+	int ret = 0;
+
+	mutex_lock(&nvmf_dev_mutex);
+	ctrl = seq_file->private;
+	if (!ctrl) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
+			ctrl->instance, ctrl->cntlid);
+
+out_unlock:
+	mutex_unlock(&nvmf_dev_mutex);
+	return ret;
+}
+
+static int nvmf_dev_open(struct inode *inode, struct file *file)
+{
+	/*
+	 * The miscdevice code initializes file->private_data, but doesn't
+	 * make use of it later.
+	 */
+	file->private_data = NULL;
+	return single_open(file, nvmf_dev_show, NULL);
+}
+
+static int nvmf_dev_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq_file = file->private_data;
+	struct nvme_ctrl *ctrl = seq_file->private;
+
+	if (ctrl)
+		nvme_put_ctrl(ctrl);
+	return single_release(inode, file);
+}
+
+static const struct file_operations nvmf_dev_fops = {
+	.owner		= THIS_MODULE,
+	.write		= nvmf_dev_write,
+	.read		= seq_read,
+	.open		= nvmf_dev_open,
+	.release	= nvmf_dev_release,
+};
+
+static struct miscdevice nvmf_misc = {
+	.minor		= MISC_DYNAMIC_MINOR,
+	.name           = "nvme-fabrics",
+	.fops		= &nvmf_dev_fops,
+};
+
+static int __init nvmf_init(void)
+{
+	int ret;
+
+	nvmf_default_host = nvmf_host_default();
+	if (!nvmf_default_host)
+		return -ENOMEM;
+
+	nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
+	if (IS_ERR(nvmf_class)) {
+		pr_err("couldn't register class nvme-fabrics\n");
+		ret = PTR_ERR(nvmf_class);
+		goto out_free_host;
+	}
+
+	nvmf_device =
+		device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
+	if (IS_ERR(nvmf_device)) {
+		pr_err("couldn't create nvme-fabris device!\n");
+		ret = PTR_ERR(nvmf_device);
+		goto out_destroy_class;
+	}
+
+	ret = misc_register(&nvmf_misc);
+	if (ret) {
+		pr_err("couldn't register misc device: %d\n", ret);
+		goto out_destroy_device;
+	}
+
+	return 0;
+
+out_destroy_device:
+	device_destroy(nvmf_class, MKDEV(0, 0));
+out_destroy_class:
+	class_destroy(nvmf_class);
+out_free_host:
+	nvmf_host_put(nvmf_default_host);
+	return ret;
+}
+
+static void __exit nvmf_exit(void)
+{
+	misc_deregister(&nvmf_misc);
+	device_destroy(nvmf_class, MKDEV(0, 0));
+	class_destroy(nvmf_class);
+	nvmf_host_put(nvmf_default_host);
+
+	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
+	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
+	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
+	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
+}
+
+MODULE_LICENSE("GPL v2");
+
+module_init(nvmf_init);
+module_exit(nvmf_exit);
diff --git a/drivers/nvme/host/fabrics.h b/drivers/nvme/host/fabrics.h
new file mode 100644
index 0000000..12038308
--- /dev/null
+++ b/drivers/nvme/host/fabrics.h
@@ -0,0 +1,132 @@
+/*
+ * NVMe over Fabrics common host code.
+ * Copyright (c) 2015-2016 HGST, a Western Digital Company.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ */
+#ifndef _NVME_FABRICS_H
+#define _NVME_FABRICS_H 1
+
+#include <linux/in.h>
+#include <linux/inet.h>
+
+#define NVMF_MIN_QUEUE_SIZE	16
+#define NVMF_MAX_QUEUE_SIZE	1024
+#define NVMF_DEF_QUEUE_SIZE	128
+#define NVMF_DEF_RECONNECT_DELAY	10
+
+/*
+ * Define a host as seen by the target.  We allocate one at boot, but also
+ * allow the override it when creating controllers.  This is both to provide
+ * persistence of the Host NQN over multiple boots, and to allow using
+ * multiple ones, for example in a container scenario.  Because we must not
+ * use different Host NQNs with the same Host ID we generate a Host ID and
+ * use this structure to keep track of the relation between the two.
+ */
+struct nvmf_host {
+	struct kref		ref;
+	struct list_head	list;
+	char			nqn[NVMF_NQN_SIZE];
+	uuid_le			id;
+};
+
+/**
+ * enum nvmf_parsing_opts - used to define the sysfs parsing options used.
+ */
+enum {
+	NVMF_OPT_ERR		= 0,
+	NVMF_OPT_TRANSPORT	= 1 << 0,
+	NVMF_OPT_NQN		= 1 << 1,
+	NVMF_OPT_TRADDR		= 1 << 2,
+	NVMF_OPT_TRSVCID	= 1 << 3,
+	NVMF_OPT_QUEUE_SIZE	= 1 << 4,
+	NVMF_OPT_NR_IO_QUEUES	= 1 << 5,
+	NVMF_OPT_TL_RETRY_COUNT	= 1 << 6,
+	NVMF_OPT_HOSTNQN	= 1 << 8,
+	NVMF_OPT_RECONNECT_DELAY = 1 << 9,
+};
+
+/**
+ * struct nvmf_ctrl_options - Used to hold the options specified
+ *			      with the parsing opts enum.
+ * @mask:	Used by the fabrics library to parse through sysfs options
+ *		on adding a NVMe controller.
+ * @transport:	Holds the fabric transport "technology name" (for a lack of
+ *		better description) that will be used by an NVMe controller
+ *		being added.
+ * @subsysnqn:	Hold the fully qualified NQN subystem name (format defined
+ *		in the NVMe specification, "NVMe Qualified Names").
+ * @traddr:	network address that will be used by the host to communicate
+ *		to the added NVMe controller.
+ * @trsvcid:	network port used for host-controller communication.
+ * @queue_size: Number of IO queue elements.
+ * @nr_io_queues: Number of controller IO queues that will be established.
+ * @tl_retry_count: Number of transport layer retries for a fabric queue before
+ *		     kicking upper layer(s) error recovery.
+ * @reconnect_delay: Time between two consecutive reconnect attempts.
+ * @discovery_nqn: indicates if the subsysnqn is the well-known discovery NQN.
+ * @host:	Virtual NVMe host, contains the NQN and Host ID.
+ */
+struct nvmf_ctrl_options {
+	unsigned		mask;
+	char			*transport;
+	char			*subsysnqn;
+	char			*traddr;
+	char			*trsvcid;
+	size_t			queue_size;
+	unsigned int		nr_io_queues;
+	unsigned short		tl_retry_count;
+	unsigned int		reconnect_delay;
+	bool			discovery_nqn;
+	struct nvmf_host	*host;
+};
+
+/*
+ * struct nvmf_transport_ops - used to register a specific
+ *			       fabric implementation of NVMe fabrics.
+ * @entry:		Used by the fabrics library to add the new
+ *			registration entry to its linked-list internal tree.
+ * @name:		Name of the NVMe fabric driver implementation.
+ * @required_opts:	sysfs command-line options that must be specified
+ *			when adding a new NVMe controller.
+ * @allowed_opts:	sysfs command-line options that can be specified
+ *			when adding a new NVMe controller.
+ * @create_ctrl():	function pointer that points to a non-NVMe
+ *			implementation-specific fabric technology
+ *			that would go into starting up that fabric
+ *			for the purpose of conneciton to an NVMe controller
+ *			using that fabric technology.
+ *
+ * Notes:
+ *	1. At minimum, 'required_opts' and 'allowed_opts' should
+ *	   be set to the same enum parsing options defined earlier.
+ *	2. create_ctrl() must be defined (even if it does nothing)
+ */
+struct nvmf_transport_ops {
+	struct list_head	entry;
+	const char		*name;
+	int			required_opts;
+	int			allowed_opts;
+	struct nvme_ctrl	*(*create_ctrl)(struct device *dev,
+					struct nvmf_ctrl_options *opts);
+};
+
+int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val);
+int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val);
+int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val);
+int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl);
+int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid);
+void nvmf_register_transport(struct nvmf_transport_ops *ops);
+void nvmf_unregister_transport(struct nvmf_transport_ops *ops);
+void nvmf_free_options(struct nvmf_ctrl_options *opts);
+const char *nvmf_get_subsysnqn(struct nvme_ctrl *ctrl);
+int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size);
+
+#endif /* _NVME_FABRICS_H */
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index b90e95a..a61c9d1 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -80,6 +80,7 @@ struct nvme_ctrl {
 	spinlock_t lock;
 	const struct nvme_ctrl_ops *ops;
 	struct request_queue *admin_q;
+	struct request_queue *connect_q;
 	struct device *dev;
 	struct kref kref;
 	int instance;
@@ -107,10 +108,19 @@ struct nvme_ctrl {
 	u8 event_limit;
 	u8 vwc;
 	u32 vs;
+	u32 sgls;
 	bool subsystem;
 	unsigned long quirks;
 	struct work_struct scan_work;
 	struct work_struct async_event_work;
+
+	/* Fabrics only */
+	u16 sqsize;
+	u32 ioccsz;
+	u32 iorcsz;
+	u16 icdoff;
+	u16 maxcmd;
+	struct nvmf_ctrl_options *opts;
 };
 
 /*
@@ -146,6 +156,7 @@ struct nvme_ns {
 struct nvme_ctrl_ops {
 	const char *name;
 	struct module *module;
+	bool is_fabrics;
 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
-- 
2.1.4

[PATCH 7/8] nvme.h: Add keep-alive opcode and identify controller attribute

[Christoph Hellwig]

From: Sagi Grimberg <sagi@grimberg.me>

KAS: keep-alive support and granularity of kato in units of 100 ms
nvme_admin_keep_alive opcode: 0x18

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 include/linux/nvme.h | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index 7525030..d8b37ba 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -202,7 +202,9 @@ struct nvme_id_ctrl {
 	__u8			apsta;
 	__le16			wctemp;
 	__le16			cctemp;
-	__u8			rsvd270[242];
+	__u8			rsvd270[50];
+	__le16			kas;
+	__u8			rsvd322[190];
 	__u8			sqes;
 	__u8			cqes;
 	__le16			maxcmd;
@@ -556,6 +558,7 @@ enum nvme_admin_opcode {
 	nvme_admin_async_event		= 0x0c,
 	nvme_admin_activate_fw		= 0x10,
 	nvme_admin_download_fw		= 0x11,
+	nvme_admin_keep_alive		= 0x18,
 	nvme_admin_format_nvm		= 0x80,
 	nvme_admin_security_send	= 0x81,
 	nvme_admin_security_recv	= 0x82,
@@ -580,6 +583,7 @@ enum {
 	NVME_FEAT_WRITE_ATOMIC	= 0x0a,
 	NVME_FEAT_ASYNC_EVENT	= 0x0b,
 	NVME_FEAT_AUTO_PST	= 0x0c,
+	NVME_FEAT_KATO		= 0x0f,
 	NVME_FEAT_SW_PROGRESS	= 0x80,
 	NVME_FEAT_HOST_ID	= 0x81,
 	NVME_FEAT_RESV_MASK	= 0x82,
-- 
2.1.4

[PATCH 8/8] nvme: add keep-alive support

[Christoph Hellwig]

Periodic keep-alive is a mandatory feature in NVMe over Fabrics, and
optional in NVMe 1.2.1 for PCIe.  This patch adds periodic keep-alive
sent from the host to verify that the controller is still responsive
and vice-versa.  The keep-alive timeout is user-defined (with
keep_alive_tmo connection parameter) and defaults to 5 seconds.

In order to avoid a race condition where the host sends a keep-alive
competing with the target side keep-alive timeout expiration, the host
adds a grace period of 10 seconds when publishing the keep-alive timeout
to the target.

In case a keep-alive failed (or timed out), a transport specific error
recovery kicks in.

For now only NVMe over Fabrics is wired up to support keep alive, but
we can add PCIe support easily once controllers actually supporting it
become available.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Steve Wise <swise@chelsio.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/core.c    | 75 +++++++++++++++++++++++++++++++++++++++++++++
 drivers/nvme/host/fabrics.c | 33 +++++++++++++++++++-
 drivers/nvme/host/fabrics.h |  3 ++
 drivers/nvme/host/nvme.h    |  8 +++++
 4 files changed, 118 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 1f70393..6f4361b 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -30,6 +30,7 @@
 #include <asm/unaligned.h>
 
 #include "nvme.h"
+#include "fabrics.h"
 
 #define NVME_MINORS		(1U << MINORBITS)
 
@@ -463,6 +464,73 @@ int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 			result, timeout);
 }
 
+static void nvme_keep_alive_end_io(struct request *rq, int error)
+{
+	struct nvme_ctrl *ctrl = rq->end_io_data;
+
+	blk_mq_free_request(rq);
+
+	if (error) {
+		dev_err(ctrl->device,
+			"failed nvme_keep_alive_end_io error=%d\n", error);
+		return;
+	}
+
+	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
+}
+
+static int nvme_keep_alive(struct nvme_ctrl *ctrl)
+{
+	struct nvme_command c;
+	struct request *rq;
+
+	memset(&c, 0, sizeof(c));
+	c.common.opcode = nvme_admin_keep_alive;
+
+	rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED, 0);
+	if (IS_ERR(rq))
+		return PTR_ERR(rq);
+
+	rq->timeout = ctrl->kato * HZ;
+	rq->end_io_data = ctrl;
+
+	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
+
+	return 0;
+}
+
+static void nvme_keep_alive_work(struct work_struct *work)
+{
+	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
+			struct nvme_ctrl, ka_work);
+
+	if (nvme_keep_alive(ctrl)) {
+		/* allocation failure, reset the controller */
+		dev_err(ctrl->device, "keep-alive failed\n");
+		ctrl->ops->reset_ctrl(ctrl);
+		return;
+	}
+}
+
+void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
+{
+	if (unlikely(ctrl->kato == 0))
+		return;
+
+	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
+	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
+}
+EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
+
+void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
+{
+	if (unlikely(ctrl->kato == 0))
+		return;
+
+	cancel_delayed_work_sync(&ctrl->ka_work);
+}
+EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
+
 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 {
 	struct nvme_command c = { };
@@ -1179,6 +1247,7 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
 
 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
 	ctrl->sgls = le32_to_cpu(id->sgls);
+	ctrl->kas = le16_to_cpu(id->kas);
 
 	if (ctrl->ops->is_fabrics) {
 		ctrl->icdoff = le16_to_cpu(id->icdoff);
@@ -1192,6 +1261,12 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
 		 */
 		if (ctrl->cntlid != le16_to_cpu(id->cntlid))
 			ret = -EINVAL;
+
+		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
+			dev_err(ctrl->dev,
+				"keep-alive support is mandatory for fabrics\n");
+			ret = -EINVAL;
+		}
 	} else {
 		ctrl->cntlid = le16_to_cpu(id->cntlid);
 	}
diff --git a/drivers/nvme/host/fabrics.c b/drivers/nvme/host/fabrics.c
index 8372eb6..ee4b7f1 100644
--- a/drivers/nvme/host/fabrics.c
+++ b/drivers/nvme/host/fabrics.c
@@ -360,6 +360,12 @@ int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 	cmd.connect.fctype = nvme_fabrics_type_connect;
 	cmd.connect.qid = 0;
 	cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
+	/*
+	 * Set keep-alive timeout in seconds granularity (ms * 1000)
+	 * and add a grace period for controller kato enforcement
+	 */
+	cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
+		cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
 
 	data = kzalloc(sizeof(*data), GFP_KERNEL);
 	if (!data)
@@ -500,6 +506,7 @@ static const match_table_t opt_tokens = {
 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
 	{ NVMF_OPT_TL_RETRY_COUNT,	"tl_retry_count=%d"	},
 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
+	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
 	{ NVMF_OPT_ERR,			NULL			}
 };
@@ -611,6 +618,28 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 			}
 			opts->tl_retry_count = token;
 			break;
+		case NVMF_OPT_KATO:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+
+			if (opts->discovery_nqn) {
+				pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
+				ret = -EINVAL;
+				goto out;
+			}
+
+			if (token < 0) {
+				pr_err("Invalid keep_alive_tmo %d\n", token);
+				ret = -EINVAL;
+				goto out;
+			} else if (token == 0) {
+				/* Allowed for debug */
+				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
+			}
+			opts->kato = token;
+			break;
 		case NVMF_OPT_HOSTNQN:
 			if (opts->host) {
 				pr_err("hostnqn already user-assigned: %s\n",
@@ -662,6 +691,8 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 	}
 
 out:
+	if (!opts->discovery_nqn && !opts->kato)
+		opts->kato = NVME_DEFAULT_KATO;
 	kfree(options);
 	return ret;
 }
@@ -718,7 +749,7 @@ EXPORT_SYMBOL_GPL(nvmf_free_options);
 
 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
-				 NVMF_OPT_HOSTNQN)
+				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN)
 
 static struct nvme_ctrl *
 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
diff --git a/drivers/nvme/host/fabrics.h b/drivers/nvme/host/fabrics.h
index 12038308..b540674 100644
--- a/drivers/nvme/host/fabrics.h
+++ b/drivers/nvme/host/fabrics.h
@@ -49,6 +49,7 @@ enum {
 	NVMF_OPT_QUEUE_SIZE	= 1 << 4,
 	NVMF_OPT_NR_IO_QUEUES	= 1 << 5,
 	NVMF_OPT_TL_RETRY_COUNT	= 1 << 6,
+	NVMF_OPT_KATO		= 1 << 7,
 	NVMF_OPT_HOSTNQN	= 1 << 8,
 	NVMF_OPT_RECONNECT_DELAY = 1 << 9,
 };
@@ -72,6 +73,7 @@ enum {
  *		     kicking upper layer(s) error recovery.
  * @reconnect_delay: Time between two consecutive reconnect attempts.
  * @discovery_nqn: indicates if the subsysnqn is the well-known discovery NQN.
+ * @kato:	Keep-alive timeout.
  * @host:	Virtual NVMe host, contains the NQN and Host ID.
  */
 struct nvmf_ctrl_options {
@@ -85,6 +87,7 @@ struct nvmf_ctrl_options {
 	unsigned short		tl_retry_count;
 	unsigned int		reconnect_delay;
 	bool			discovery_nqn;
+	unsigned int		kato;
 	struct nvmf_host	*host;
 };
 
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index a61c9d1..d25eaab 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -38,6 +38,9 @@ extern unsigned char admin_timeout;
 extern unsigned char shutdown_timeout;
 #define SHUTDOWN_TIMEOUT	(shutdown_timeout * HZ)
 
+#define NVME_DEFAULT_KATO	5
+#define NVME_KATO_GRACE		10
+
 enum {
 	NVME_NS_LBA		= 0,
 	NVME_NS_LIGHTNVM	= 1,
@@ -109,10 +112,13 @@ struct nvme_ctrl {
 	u8 vwc;
 	u32 vs;
 	u32 sgls;
+	u16 kas;
+	unsigned int kato;
 	bool subsystem;
 	unsigned long quirks;
 	struct work_struct scan_work;
 	struct work_struct async_event_work;
+	struct delayed_work ka_work;
 
 	/* Fabrics only */
 	u16 sqsize;
@@ -273,6 +279,8 @@ int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 			dma_addr_t dma_addr, u32 *result);
 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
+void nvme_start_keep_alive(struct nvme_ctrl *ctrl);
+void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
 
 struct sg_io_hdr;
 
-- 
2.1.4

Re: [PATCH 4/8] nvme: add fabrics sysfs attributes

[Johannes Thumshirn]

On Mon, Jun 06, 2016 at 11:21:55PM +0200, Christoph Hellwig wrote:
> - delete_controller: This attribute allows to delete a controller.
>   A driver is not obligated to support it (pci doesn't) so it is
>   created only if the driver supports it. The new fabrics drivers
>   will support it (essentialy a disconnect operation).
> 
>   Usage:
>   echo > /sys/class/nvme/nvme0/delete_controller
> 
> - subsysnqn: This attribute shows the subsystem nqn of the configured
>   device. If a driver does not implement the get_subsysnqn method, the
>   file will not appear in sysfs.
> 
> - transport: This attribute shows the transport name. Added a "name"
>   field to struct nvme_ctrl_ops.
> 
>   For loop,
>   cat /sys/class/nvme/nvme0/transport
>   loop
> 
>   For RDMA,
>   cat /sys/class/nvme/nvme0/transport
>   rdma
> 
>   For PCIe,
>   cat /sys/class/nvme/nvme0/transport
>   pcie
> 
> - address: This attributes shows the controller address. The fabrics
>   drivers that will implement get_address can show the address of the
>   connected controller.
> 
>   example:
>   cat /sys/class/nvme/nvme0/address
>   traddr=192.168.2.2,trsvcid=1023

Can you please add this to Documentation/ABI/ as well?

-- 
Johannes Thumshirn                                          Storage
jthumshirn@suse.de                                +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850

Re: [PATCH 4/8] nvme: add fabrics sysfs attributes

[Christoph Hellwig]

On Tue, Jun 07, 2016 at 10:52:17AM +0200, Johannes Thumshirn wrote:
> Can you please add this to Documentation/ABI/ as well?

None of the NVMe files are documented there..  But maybe it's worth
a separate little project to document all of them.

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Keith Busch]

On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:
> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
> +		unsigned int flags, unsigned int hctx_idx)
> +{
> +	struct blk_mq_hw_ctx *hctx;
> +	struct blk_mq_ctx *ctx;
> +	struct request *rq;
> +	struct blk_mq_alloc_data alloc_data;
> +	int ret;
> +
> +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> +	if (ret)
> +		return ERR_PTR(ret);
> +
> +	hctx = q->queue_hw_ctx[hctx_idx];

We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
getting the hctx. Even if hctx_idx was origially valid, it's possible
(though unlikely) blk_queue_enter waits on reallocating h/w contexts,
which can make hctx_idx invalid.

Re: [PATCH 4/8] nvme: add fabrics sysfs attributes

[Christoph Hellwig]

FYI, this patch should be

From: Ming Lin <ming.l@ssi.samsung.com>

unfortunately git is really eager to lose these when having the slightest
conflicts in a rebase..

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Ming Lin]

On Tue, Jun 7, 2016 at 7:57 AM, Keith Busch <keith.busch@intel.com> wrote:
> On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:
>> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
>> +             unsigned int flags, unsigned int hctx_idx)
>> +{
>> +     struct blk_mq_hw_ctx *hctx;
>> +     struct blk_mq_ctx *ctx;
>> +     struct request *rq;
>> +     struct blk_mq_alloc_data alloc_data;
>> +     int ret;
>> +
>> +     ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
>> +     if (ret)
>> +             return ERR_PTR(ret);
>> +
>> +     hctx = q->queue_hw_ctx[hctx_idx];
>
> We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
> getting the hctx. Even if hctx_idx was origially valid, it's possible
> (though unlikely) blk_queue_enter waits on reallocating h/w contexts,
> which can make hctx_idx invalid.

Yes, I'll update it.

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Ming Lin]

On Tue, Jun 7, 2016 at 8:27 AM, Ming Lin <mlin@kernel.org> wrote:
> On Tue, Jun 7, 2016 at 7:57 AM, Keith Busch <keith.busch@intel.com> wrote:
>> On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:
>>> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
>>> +             unsigned int flags, unsigned int hctx_idx)
>>> +{
>>> +     struct blk_mq_hw_ctx *hctx;
>>> +     struct blk_mq_ctx *ctx;
>>> +     struct request *rq;
>>> +     struct blk_mq_alloc_data alloc_data;
>>> +     int ret;
>>> +
>>> +     ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
>>> +     if (ret)
>>> +             return ERR_PTR(ret);
>>> +
>>> +     hctx = q->queue_hw_ctx[hctx_idx];
>>
>> We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
>> getting the hctx. Even if hctx_idx was origially valid, it's possible
>> (though unlikely) blk_queue_enter waits on reallocating h/w contexts,
>> which can make hctx_idx invalid.
>
> Yes, I'll update it.

How about this?

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 7bb45ed..b59d2ef 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -279,6 +279,11 @@ struct request *blk_mq_alloc_request_hctx(struct
request_queue *q, int rw,
        if (ret)
                return ERR_PTR(ret);

+       if (hctx_idx >= q->nr_hw_queues) {
+               blk_queue_exit(q);
+               return ERR_PTR(-EINVAL);
+       }
+
        hctx = q->queue_hw_ctx[hctx_idx];
        ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Jens Axboe]

On 06/06/2016 03:21 PM, Christoph Hellwig wrote:
> From: Ming Lin <ming.l@ssi.samsung.com>
>
> For some protocols like NVMe over Fabrics we need to be able to send
> initialization commands to a specific queue.
>
> Based on an earlier patch from Christoph Hellwig <hch@lst.de>.
>
> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>   block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
>   include/linux/blk-mq.h |  2 ++
>   2 files changed, 35 insertions(+)
>
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 29cbc1b..7bb45ed 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
>   }
>   EXPORT_SYMBOL(blk_mq_alloc_request);
>
> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
> +		unsigned int flags, unsigned int hctx_idx)
> +{
> +	struct blk_mq_hw_ctx *hctx;
> +	struct blk_mq_ctx *ctx;
> +	struct request *rq;
> +	struct blk_mq_alloc_data alloc_data;
> +	int ret;
> +
> +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> +	if (ret)
> +		return ERR_PTR(ret);
> +
> +	hctx = q->queue_hw_ctx[hctx_idx];
> +	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
> +
> +	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
> +
> +	rq = __blk_mq_alloc_request(&alloc_data, rw);
> +	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
> +		__blk_mq_run_hw_queue(hctx);
> +
> +		rq =  __blk_mq_alloc_request(&alloc_data, rw);
> +	}

Why are we duplicating this code here? If NOWAIT isn't set, then we'll
always return a request. bt_get() will run the queue for us, if it needs
to. blk_mq_alloc_request() does this too, and I'm guessing that code was
just copied. I'll fix that up. Looks like this should just be:

	rq = __blk_mq_alloc_request(&alloc_data, rw);
	if (rq)
		return rq;

	blk_queue_exit(q);
	return ERR_PTR(-EWOULDBLOCK);

for this case.

-- 
Jens Axboe

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Ming Lin]

On Tue, 2016-06-07 at 22:49 -0600, Jens Axboe wrote:
> On 06/06/2016 03:21 PM, Christoph Hellwig wrote:
> > From: Ming Lin <ming.l@ssi.samsung.com>
> > 
> > For some protocols like NVMe over Fabrics we need to be able to
> > send
> > initialization commands to a specific queue.
> > 
> > Based on an earlier patch from Christoph Hellwig <hch@lst.de>.
> > 
> > Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > ---
> >   block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
> >   include/linux/blk-mq.h |  2 ++
> >   2 files changed, 35 insertions(+)
> > 
> > diff --git a/block/blk-mq.c b/block/blk-mq.c
> > index 29cbc1b..7bb45ed 100644
> > --- a/block/blk-mq.c
> > +++ b/block/blk-mq.c
> > @@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct
> > request_queue *q, int rw,
> >   }
> >   EXPORT_SYMBOL(blk_mq_alloc_request);
> > 
> > +struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
> > int rw,
> > +		unsigned int flags, unsigned int hctx_idx)
> > +{
> > +	struct blk_mq_hw_ctx *hctx;
> > +	struct blk_mq_ctx *ctx;
> > +	struct request *rq;
> > +	struct blk_mq_alloc_data alloc_data;
> > +	int ret;
> > +
> > +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> > +	if (ret)
> > +		return ERR_PTR(ret);
> > +
> > +	hctx = q->queue_hw_ctx[hctx_idx];
> > +	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
> > +
> > +	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
> > +
> > +	rq = __blk_mq_alloc_request(&alloc_data, rw);
> > +	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
> > +		__blk_mq_run_hw_queue(hctx);
> > +
> > +		rq =  __blk_mq_alloc_request(&alloc_data, rw);
> > +	}
> 
> Why are we duplicating this code here? If NOWAIT isn't set, then
> we'll
> always return a request. bt_get() will run the queue for us, if it
> needs
> to. blk_mq_alloc_request() does this too, and I'm guessing that code
> was
> just copied. I'll fix that up. Looks like this should just be:
> 
> 	rq = __blk_mq_alloc_request(&alloc_data, rw);
> 	if (rq)
> 		return rq;
> 
> 	blk_queue_exit(q);
> 	return ERR_PTR(-EWOULDBLOCK);
> 
> for this case.

Yes,

But the bt_get() reminds me that this patch actually has a problem.

blk_mq_alloc_request_hctx() ->
  __blk_mq_alloc_request() ->
    blk_mq_get_tag() -> 
      __blk_mq_get_tag() ->
        bt_get() ->
          blk_mq_put_ctx(data->ctx);

Here are blk_mq_get_ctx() and blk_mq_put_ctx().

static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
{       
        return __blk_mq_get_ctx(q, get_cpu());
} 

static inline void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
{
        put_cpu();
}

blk_mq_alloc_request_hctx() calls __blk_mq_get_ctx() instead
of blk_mq_get_ctx(). Then reason is the "hctx" could belong to other
cpu. So blk_mq_get_ctx() doesn't work.

But then above put_cpu() in blk_mq_put_ctx() will trigger a WARNING
because we didn't do get_cpu() in blk_mq_alloc_request_hctx()

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Christoph Hellwig]

On Tue, Jun 07, 2016 at 10:49:22PM -0600, Jens Axboe wrote:
> Why are we duplicating this code here? If NOWAIT isn't set, then we'll
> always return a request. bt_get() will run the queue for us, if it needs
> to. blk_mq_alloc_request() does this too, and I'm guessing that code was
> just copied. I'll fix that up. Looks like this should just be:

The generic code is a bit of a mess as it does the nowait optimization
in two different cases, I'll send you a separate cleanup patch for that.

Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Christoph Hellwig]

What we really nee to do is to always set the NOWAIT flag (we have a
reserved tag for connect anyway), and thus never trigger the code
deep down in bt_get that might switch to a different hctx.

Below is the version that I've tested together with the NVMe change
to use the NOWAIT flag:

---
>From 2346a4fdcc57c70a671e1329f7550d91d4e9d8a8 Mon Sep 17 00:00:00 2001
From: Ming Lin <ming.l@ssi.samsung.com>
Date: Mon, 30 Nov 2015 19:45:48 +0100
Subject: blk-mq: add blk_mq_alloc_request_hctx

For some protocols like NVMe over Fabrics we need to be able to send
initialization commands to a specific queue.

Based on an earlier patch from Christoph Hellwig <hch@lst.de>.

Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
[hch: disallow sleeping allocation, req_op fixes]
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/blk-mq.c         | 39 +++++++++++++++++++++++++++++++++++++++
 include/linux/blk-mq.h |  2 ++
 2 files changed, 41 insertions(+)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 13f4603..7aa60c4 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -267,6 +267,45 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 }
 EXPORT_SYMBOL(blk_mq_alloc_request);
 
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+		unsigned int flags, unsigned int hctx_idx)
+{
+	struct blk_mq_hw_ctx *hctx;
+	struct blk_mq_ctx *ctx;
+	struct request *rq;
+	struct blk_mq_alloc_data alloc_data;
+	int ret;
+
+	/*
+	 * If the tag allocator sleeps we could get an allocation for a
+	 * different hardware context.  No need to complicate the low level
+	 * allocator for this for the rare use case of a command tied to
+	 * a specific queue.
+	 */
+	if (WARN_ON_ONCE(!(flags & BLK_MQ_REQ_NOWAIT)))
+		return ERR_PTR(-EINVAL);
+
+	if (hctx_idx >= q->nr_hw_queues)
+		return ERR_PTR(-EIO);
+
+	ret = blk_queue_enter(q, true);
+	if (ret)
+		return ERR_PTR(ret);
+
+	hctx = q->queue_hw_ctx[hctx_idx];
+	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
+
+	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
+	rq = __blk_mq_alloc_request(&alloc_data, rw, 0);
+	if (!rq) {
+		blk_queue_exit(q);
+		return ERR_PTR(-EWOULDBLOCK);
+	}
+
+	return rq;
+}
+EXPORT_SYMBOL_GPL(blk_mq_alloc_request_hctx);
+
 static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx, struct request *rq)
 {
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index faa7d5c2..e43bbff 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -198,6 +198,8 @@ enum {
 
 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 		unsigned int flags);
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
+		unsigned int flags, unsigned int hctx_idx);
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
 
-- 
2.1.4