Skip to contentSkip to conversation

nvme-fabrics: fix un-expected behaviour related to hostnqn and hostid

PATCH Superseded Max Gurtovoy 2 participants 11 messages May 10, 2023 – May 11, 2023 NVMe

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover nvme-fabrics: fix un-expected behaviour related to hostnqn and hostid

Hi Christoph/Sagi/Keith,
I've noticed that we are having some un-expected behaviour in some of the scenarios of connection establisment related to hostnqn and hostid. This is specially around overriding hostid for a host.

For example:
-------------
1. Connect to a target with default hostnqn and hostid (after removing the files from /etc/nvme):
- # nvme connect -t tcp -n testsubsystem_0 -a 1.1.1.1 -s 4420
2. Check hostnqn and hostid

    - #cat /sys/class/nvme/nvme0/host*
       1791bca2-af3c-4e17-a310-6fbdde3b59cb
       nqn.2014-08.org.nvmexpress:uuid:fc4e5383-ce5c-470d-9889-4adbc2dbe969

3. Connect to a target with given hostid from cmdline:
- # nvme connect -t tcp -n testsubsystem_0 -a 1.1.1.1 -s 4420 -I 77720af2-ce59-418b-82ff-7c9df0cea35e Failed to write to /dev/nvme-fabrics: Operation already in progress <--- failed on duplicate connection check
4. Check hostnqn and hostid of first controller

    - #cat /sys/class/nvme/nvme0/host*
       77720af2-ce59-418b-82ff-7c9df0cea35e <--- hostid was changed !!!!
       nqn.2014-08.org.nvmexpress:uuid:fc4e5383-ce5c-470d-9889-4adbc2dbe969
5. Connect to a target with given hostid from cmdline and allow duplicate connections:
    - # nvme connect -t tcp -n testsubsystem_0 -a 1.1.1.1 -s 4420 -I 77720af2-ce59-418b-82ff-7c9df0cea3bb -D
6. Check hostnqn and hostid of first controller again
    - #cat /sys/class/nvme/nvme0/host*
       77720af2-ce59-418b-82ff-7c9df0cea3bb  <--- hostid was changed again !!!!
       nqn.2014-08.org.nvmexpress:uuid:fc4e5383-ce5c-470d-9889-4adbc2dbe969

This seems like a wrong behaviour to me since in case we try to reconnect the target will see that a different values after reconnections and this change the expected behaviour to keep hostid.

According to the spec:

"This feature allows the host to register a Host Identifier with the controller. The Host Identifier is used by the controller to determine whether other controllers in the NVM subsystem are associated with the same host.
"

Also:

"
A host that uses a single Host NQN may employ multiple Host Identifiers to designate elements of the host
that access an NVM subsystem independently of each other (e.g., physical or logical partitions of the host)
"

These patchset fixes the current behaviour and allows one to set different hostIDs for a single hostNQN but not vice versa (as we would like to enable the target also recognize a logical partition of the host).

The first 2 patches are some preparations and cleanups and patch 3/3 is the actual fix.

Max Gurtovoy (3):
nvme-fabrics: unify common code in admin and io queue connect
nvme-fabrics: check hostid using uuid_equal
nvme-fabrics: prevent overriding of existing host

 drivers/nvme/host/fabrics.c | 163 +++++++++++++++++++++++-------------
 drivers/nvme/host/fabrics.h |   3 +-
 2 files changed, 106 insertions(+), 60 deletions(-)
Signature · 1 line
2.18.1
#1 lore raw

PATCH 1/3 nvme-fabrics: unify common code in admin and io queue connect

To simplify code maintenance, it is recommended to avoid duplicating code.

---
 drivers/nvme/host/fabrics.c | 75 ++++++++++++++++++++++---------------
 1 file changed, 44 insertions(+), 31 deletions(-)
drivers/nvme/host/fabrics.c +44 -31
index 0069ebff85df..d11db06d293b 100644
@@ -349,6 +349,46 @@ static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
 	}
 }
 
+static struct nvmf_connect_data *nvmf_connect_data_prepare(struct nvme_ctrl *ctrl,
+		u16 cntlid)
+{
+	struct nvmf_connect_data *data;
+
+	data = kzalloc(sizeof(*data), GFP_KERNEL);
+	if (!data)
+		return NULL;
+
+	uuid_copy(&data->hostid, &ctrl->opts->host->id);
+	data->cntlid = cpu_to_le16(cntlid);
+	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
+	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
+
+	return data;
+}
+
+static void nvmf_connect_cmd_prepare(struct nvme_ctrl *ctrl, u16 qid,
+		struct nvme_command *cmd)
+{
+	cmd->connect.opcode = nvme_fabrics_command;
+	cmd->connect.fctype = nvme_fabrics_type_connect;
+	cmd->connect.qid = cpu_to_le16(qid);
+
+	if (qid) {
+		cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
+	} else {
+		cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
+
+		/*
+		 * For admin queue, set keep-alive timeout in seconds
+		 * granularity (ms * 1000)
+		 */
+		cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
+	}
+
+	if (ctrl->opts->disable_sqflow)
+		cmd->connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
+}
+
 /**
  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
  *				API function.
@@ -377,28 +417,12 @@ int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 	int ret;
 	u32 result;
 
-	cmd.connect.opcode = nvme_fabrics_command;
-	cmd.connect.fctype = nvme_fabrics_type_connect;
-	cmd.connect.qid = 0;
-	cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
-
-	/*
-	 * Set keep-alive timeout in seconds granularity (ms * 1000)
-	 */
-	cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000);
-
-	if (ctrl->opts->disable_sqflow)
-		cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
+	nvmf_connect_cmd_prepare(ctrl, 0, &cmd);
 
-	data = kzalloc(sizeof(*data), GFP_KERNEL);
+	data = nvmf_connect_data_prepare(ctrl, 0xffff);
 	if (!data)
 		return -ENOMEM;
 
-	uuid_copy(&data->hostid, &ctrl->opts->host->id);
-	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->fabrics_q, &cmd, &res,
 			data, sizeof(*data), NVME_QID_ANY, 1,
 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
@@ -468,23 +492,12 @@ int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
 	int ret;
 	u32 result;
 
-	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);
+	nvmf_connect_cmd_prepare(ctrl, qid, &cmd);
 
-	if (ctrl->opts->disable_sqflow)
-		cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
-
-	data = kzalloc(sizeof(*data), GFP_KERNEL);
+	data = nvmf_connect_data_prepare(ctrl, ctrl->cntlid);
 	if (!data)
 		return -ENOMEM;
 
-	uuid_copy(&data->hostid, &ctrl->opts->host->id);
-	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, &res,
 			data, sizeof(*data), qid, 1,
 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
Signature · 1 line
2.18.1
#2 lore raw

Re: [PATCH 1/3] nvme-fabrics: unify common code in admin and io queue connect

Quoted 1 line · Max Gurtovoy

+static struct nvmf_connect_data *nvmf_connect_data_prepare(struct nvme_ctrl *ctrl,

Overly long line. Maybe shorten the function name a little?

drivers/nvme/host/fabrics.c static void nvmf_log_connect_error(struct nvme_ctrl *ctrl, · line 386

Quoted 11 lines
drivers/nvme/host/fabrics.c +11 -0
+	if (qid) {
+		cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
+	} else {
+		cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
+
+		/*
+		 * For admin queue, set keep-alive timeout in seconds
+		 * granularity (ms * 1000)
+		 */
+		cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
+	}

Not sure the "For admin queue, " part here adds any value. That's pretty obvious from the code.

#3 lore raw

PATCH 2/3 nvme-fabrics: check hostid using uuid_equal

In reply to Max Gurtovoy

Use a dedicated function to match uuids instead of duplicating it.

---
 drivers/nvme/host/fabrics.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/nvme/host/fabrics.h +2 -1
index dcac3df8a5f7..0f3763283222 100644
@@ -181,8 +181,9 @@ nvmf_ctlr_matches_baseopts(struct nvme_ctrl *ctrl,
 	    ctrl->state == NVME_CTRL_DEAD ||
 	    strcmp(opts->subsysnqn, ctrl->opts->subsysnqn) ||
 	    strcmp(opts->host->nqn, ctrl->opts->host->nqn) ||
-	    memcmp(&opts->host->id, &ctrl->opts->host->id, sizeof(uuid_t)))
+	    !uuid_equal(&opts->host->id, &ctrl->opts->host->id)) {
 		return false;
+	}
 
 	return true;
 }
Signature · 1 line
2.18.1
#4 lore raw

Re: [PATCH 2/3] nvme-fabrics: check hostid using uuid_equal

drivers/nvme/host/fabrics.h nvmf_ctlr_matches_baseopts(struct nvme_ctrl *ctrl, · line 186

Quoted 12 lines · Max Gurtovoy
drivers/nvme/host/fabrics.h +2 -1
index dcac3df8a5f7..0f3763283222 100644
@@ -181,8 +181,9 @@ nvmf_ctlr_matches_baseopts(struct nvme_ctrl *ctrl,
 	    ctrl->state == NVME_CTRL_DEAD ||
 	    strcmp(opts->subsysnqn, ctrl->opts->subsysnqn) ||
 	    strcmp(opts->host->nqn, ctrl->opts->host->nqn) ||
-	    memcmp(&opts->host->id, &ctrl->opts->host->id, sizeof(uuid_t)))
+	    !uuid_equal(&opts->host->id, &ctrl->opts->host->id)) {
 		return false;
+	}

Please don't add the pointless braces. The actual change looks fine to me, though.

#5 lore raw

Re: [PATCH 2/3] nvme-fabrics: check hostid using uuid_equal

Quoted 16 lines · Christoph Hellwig

drivers/nvme/host/fabrics.h nvmf_ctlr_matches_baseopts(struct nvme_ctrl *ctrl, · line 186

Quoted 12 lines · Max Gurtovoy
drivers/nvme/host/fabrics.h +2 -1
index dcac3df8a5f7..0f3763283222 100644
@@ -181,8 +181,9 @@ nvmf_ctlr_matches_baseopts(struct nvme_ctrl *ctrl,
  	    ctrl->state == NVME_CTRL_DEAD ||
  	    strcmp(opts->subsysnqn, ctrl->opts->subsysnqn) ||
  	    strcmp(opts->host->nqn, ctrl->opts->host->nqn) ||
-	    memcmp(&opts->host->id, &ctrl->opts->host->id, sizeof(uuid_t)))
+	    !uuid_equal(&opts->host->id, &ctrl->opts->host->id)) {
  		return false;
+	}

Please don't add the pointless braces. The actual change looks fine to me, though.

sorry. probably I forgot to remove after finished debugging..

#6 lore raw

PATCH 3/3 nvme-fabrics: prevent overriding of existing host

In reply to Max Gurtovoy

When first connecting a target using the "default" host parameters, setting the hostid from the command line during a subsequent connection establishment would override the "default" hostid parameter. This would cause an existing connection that is already using the host definitions to lose its hostid.

To address this issue, the code has been modified to assign a new host structure if hostid is given as input during the connect command.

---
 drivers/nvme/host/fabrics.c | 88 +++++++++++++++++++++++++------------
 1 file changed, 60 insertions(+), 28 deletions(-)
drivers/nvme/host/fabrics.c +60 -28
index d11db06d293b..7aca456e94ec 100644
@@ -21,35 +21,67 @@ static DEFINE_MUTEX(nvmf_hosts_mutex);
 
 static struct nvmf_host *nvmf_default_host;
 
-static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
+/**
+ * __nvmf_host_find() - Find a matching to a previously created host
+ * @hostnqn: Host NQN to match
+ * @id: Host ID to match
+ *
+ * We defined a host as seen by the target, thus we must not use different
+ * Host NQNs with the same Host ID. We do allow having same Host NQN with
+ * different Host IDs.
+ *
+ * Return: Returns host pointer on success, NULL in case of no match or
+ *         ERR_PTR(-EINVAL) in case of error match.
+ */
+static struct nvmf_host *__nvmf_host_find(const char *hostnqn, uuid_t *id)
 {
 	struct nvmf_host *host;
 
+	lockdep_assert_held(&nvmf_hosts_mutex);
+
 	list_for_each_entry(host, &nvmf_hosts, list) {
-		if (!strcmp(host->nqn, hostnqn))
-			return host;
+		if (uuid_equal(&host->id, id)) {
+			if (!strcmp(host->nqn, hostnqn))
+				return host;
+			else
+				return ERR_PTR(-EINVAL);
+		}
 	}
 
 	return NULL;
 }
 
-static struct nvmf_host *nvmf_host_add(const char *hostnqn)
+static struct nvmf_host *nvmf_host_alloc(const char *hostnqn, uuid_t *id)
+{
+	struct nvmf_host *host;
+
+	host = kmalloc(sizeof(*host), GFP_KERNEL);
+	if (!host)
+		return NULL;
+
+	kref_init(&host->ref);
+	uuid_copy(&host->id, id);
+	strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
+
+	return host;
+}
+
+static struct nvmf_host *nvmf_host_add(const char *hostnqn, uuid_t *id)
 {
 	struct nvmf_host *host;
 
 	mutex_lock(&nvmf_hosts_mutex);
-	host = __nvmf_host_find(hostnqn);
-	if (host) {
+	host = __nvmf_host_find(hostnqn, id);
+	if (IS_ERR(host)) {
+		goto out_unlock;
+	} else if (host) {
 		kref_get(&host->ref);
 		goto out_unlock;
 	}
 
-	host = kmalloc(sizeof(*host), GFP_KERNEL);
+	host = nvmf_host_alloc(hostnqn, id);
 	if (!host)
-		goto out_unlock;
-
-	kref_init(&host->ref);
-	strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
+		return ERR_PTR(-ENOMEM);
 
 	list_add_tail(&host->list, &nvmf_hosts);
 out_unlock:
@@ -60,16 +92,17 @@ static struct nvmf_host *nvmf_host_add(const char *hostnqn)
 static struct nvmf_host *nvmf_host_default(void)
 {
 	struct nvmf_host *host;
+	char nqn[NVMF_NQN_SIZE];
+	uuid_t id;
 
-	host = kmalloc(sizeof(*host), GFP_KERNEL);
+	uuid_gen(&id);
+	snprintf(nqn, NVMF_NQN_SIZE,
+		"nqn.2014-08.org.nvmexpress:uuid:%pUb", &id);
+
+	host = nvmf_host_alloc(nqn, &id);
 	if (!host)
 		return NULL;
 
-	kref_init(&host->ref);
-	uuid_gen(&host->id);
-	snprintf(host->nqn, NVMF_NQN_SIZE,
-		"nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
-
 	mutex_lock(&nvmf_hosts_mutex);
 	list_add_tail(&host->list, &nvmf_hosts);
 	mutex_unlock(&nvmf_hosts_mutex);
@@ -634,6 +667,7 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 	size_t nqnlen  = 0;
 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
 	uuid_t hostid;
+	char hostnqn[NVMF_NQN_SIZE];
 
 	/* Set defaults */
 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
@@ -650,7 +684,9 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 	if (!options)
 		return -ENOMEM;
 
-	uuid_gen(&hostid);
+	/* use default host if not given by user space */
+	uuid_copy(&hostid, &nvmf_default_host->id);
+	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
 
 	while ((p = strsep(&o, ",\n")) != NULL) {
 		if (!*p)
@@ -796,12 +832,8 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 				ret = -EINVAL;
 				goto out;
 			}
-			opts->host = nvmf_host_add(p);
+			strscpy(hostnqn, p, NVMF_NQN_SIZE);
 			kfree(p);
-			if (!opts->host) {
-				ret = -ENOMEM;
-				goto out;
-			}
 			break;
 		case NVMF_OPT_RECONNECT_DELAY:
 			if (match_int(args, &token)) {
@@ -958,13 +990,13 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
 	}
 
-	if (!opts->host) {
-		kref_get(&nvmf_default_host->ref);
-		opts->host = nvmf_default_host;
+	opts->host = nvmf_host_add(hostnqn, &hostid);
+	if (IS_ERR(opts->host)) {
+		ret = PTR_ERR(opts->host);
+		opts->host = NULL;
+		goto out;
 	}
 
-	uuid_copy(&opts->host->id, &hostid);
-
 out:
 	kfree(options);
 	return ret;
Signature · 1 line
2.18.1
#7 lore raw

Re: [PATCH 3/3] nvme-fabrics: prevent overriding of existing host

Quoted 5 lines · Max Gurtovoy

When first connecting a target using the "default" host parameters, setting the hostid from the command line during a subsequent connection establishment would override the "default" hostid parameter. This would cause an existing connection that is already using the host definitions to lose its hostid.

We need to reject that second connect - hostids and hostnqns should have a 1:1 relationship.

#8 lore raw

Re: [PATCH 3/3] nvme-fabrics: prevent overriding of existing host

Quoted 9 lines · Christoph Hellwig
Quoted 5 lines · Max Gurtovoy

When first connecting a target using the "default" host parameters, setting the hostid from the command line during a subsequent connection establishment would override the "default" hostid parameter. This would cause an existing connection that is already using the host definitions to lose its hostid.

We need to reject that second connect - hostids and hostnqns should have a 1:1 relationship.

I first though this way too but the specification allows an NQN to have logical partitions marked as hostIds:

NVMe over fabrics spec 1.1a section 3.3 (p 27):

"A host that uses a single Host NQN may employ multiple Host Identifiers to designate elements of the host that access an NVM subsystem independently of each other (e.g., physical or logical partitions of the host). Alternatively, a host may employ multiple Host NQN values to cause each element to be treated as a separate host by an NVM subsystem."

And actually this was probably the idea of the initial implementation in the driver.

I've just fixed the bug in the implementation.

#9 lore raw

Re: [PATCH 3/3] nvme-fabrics: prevent overriding of existing host

Quoted 4 lines · Max Gurtovoy

I first though this way too but the specification allows an NQN to have logical partitions marked as hostIds:

NVMe over fabrics spec 1.1a section 3.3 (p 27):

This is simply broken and we must no support it.

#10 lore raw

Re: [PATCH 3/3] nvme-fabrics: prevent overriding of existing host

Quoted 7 lines · Christoph Hellwig
Quoted 4 lines · Max Gurtovoy

I first though this way too but the specification allows an NQN to have logical partitions marked as hostIds:

NVMe over fabrics spec 1.1a section 3.3 (p 27):

This is simply broken and we must no support it.

I see.
So I'll re-spin it with 1:1 mapping between hostnqn and hostid. This means we won't allow any nqn to have multiple ids and vice versa.