Skip to contentSkip to conversation

nvme-cli: attach ns to local controller if none specified

PATCH Superseded Keith Busch 6 participants 8 messages Mar 31, 2025 – Apr 3, 2025 NVMe

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme-cli: attach ns to local controller if none specified

From: Keith Busch <[email protected]>

Assume the user meant to attach the namespace to the controller the command was sent to if the user didn't provide a controller id list.

---
 nvme.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
nvme.c +12 -4
index bb412843..7a2c2e9f 100644
@@ -2953,9 +2953,6 @@ static int nvme_attach_ns(int argc, char **argv, int attach, const char *desc, s
 
 	num = argconfig_parse_comma_sep_array_u16(cfg.cntlist,
 						  list, ARRAY_SIZE(list));
-	if (!num)
-		fprintf(stderr, "warning: empty controller-id list will result in no actual change in namespace attachment\n");
-
 	if (num == -1) {
 		nvme_show_error("%s: controller id list is malformed", cmd->name);
 		return -EINVAL;
@@ -2965,7 +2962,18 @@ static int nvme_attach_ns(int argc, char **argv, int attach, const char *desc, s
 	if (!cntlist)
 		return -ENOMEM;
 
-	nvme_init_ctrl_list(cntlist, num, list);
+	if (argconfig_parse_seen(opts, "controllers")) {
+		nvme_init_ctrl_list(cntlist, num, list);
+	} else {
+		struct nvme_id_ctrl ctrl = { 0 };
+
+		if (nvme_cli_identify_ctrl(dev, &ctrl)) {
+			perror("identify-ctrl");
+			return -errno;
+		}
+		cntlist->identifier[0] = ctrl.cntlid;
+	}
+
 
 	if (attach)
 		err = nvme_cli_ns_attach_ctrls(dev, cfg.namespace_id,
Signature · 1 line
2.47.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 7 lines · Keith Busch

From: Keith Busch<[email protected]>

Assume the user meant to attach the namespace to the controller the command was sent to if the user didn't provide a controller id list.

this will not be backward compatible hence all the existing deployments that are resulting in error [1] will now attach the default controller, I believe that is what is expected ?

if not then do we need add a version check so it will stay as backward compatible ?

Looks good.

-ck

[1]
- fprintf(stderr, "warning: empty controller-id list will result in no actual change in namespace attachment\n");

maintainer
#2 lore raw
Quoted 6 lines · Chaitanya Kulkarni
Quoted 5 lines · Keith Busch

From: Keith Busch<[email protected]>

Assume the user meant to attach the namespace to the controller the command was sent to if the user didn't provide a controller id list.

Quoted 6 lines
Quoted 2 lines

this will not be backward compatible hence all the existing deployments that are resulting in error [1] will now attach the default controller, I believe that is what is expected ?

Yes, this changes behavior. I don't think a user would purposefully request to attach a namespace to nothing. If they explicitly request that via '-c ""', then that request will still be honored. But if they omit the list entirely, I think it's reasonable to assume you want the controller servicing the command to attach the namespace to itself.

It's really a nice workflow optimization since you can skip finding the controller id when you just want local attachment.

Quoted 2 lines

if not then do we need add a version check so it will stay as backward compatible ?

I don't follow what you mean here. What version of what component could be checked to guide this behavior?

#3 lore raw
Quoted 12 lines · Keith Busch
Quoted 11 lines · Chaitanya Kulkarni
Quoted 7 lines · Keith Busch

From: Keith Busch<[email protected]>

Assume the user meant to attach the namespace to the controller the command was sent to if the user didn't provide a controller id list.

this will not be backward compatible hence all the existing deployments that are resulting in error [1] will now attach the default controller, I believe that is what is expected ?

Quoted 9 lines

Yes, this changes behavior. I don't think a user would purposefully request to attach a namespace to nothing. If they explicitly request that via '-c ""', then that request will still be honored. But if they omit the list entirely, I think it's reasonable to assume you want the controller servicing the command to attach the namespace to itself.

It's really a nice workflow optimization since you can skip finding the controller id when you just want local attachment.

Yes, someone who's using nvme-cli from very start I totally agree :).

Quoted 4 lines
Quoted 2 lines

if not then do we need add a version check so it will stay as backward compatible ?

I don't follow what you mean here. What version of what component could be checked to guide this behavior?

if we can only enable this functionality for nvme-cli/libnvme for the current
version and any other versions after current version.

In case nvme-cli or libnvme has a version below the current version, then it can follow the original behavior, will this work or not ?

-ck

#4 lore raw
Quoted 9 lines · Chaitanya Kulkarni
Quoted 5 lines · Keith Busch
Quoted 2 lines · Chaitanya Kulkarni

if not then do we need add a version check so it will stay as backward compatible ?

I don't follow what you mean here. What version of what component could be checked to guide this behavior?

if we can only enable this functionality for nvme-cli/libnvme for the current version and any other versions after current version.

This is a new feature which is backwards compatible as I understand it. If a user provides the -c option it still does exactly what it supposed to do.

Quoted 2 lines

In case nvme-cli or libnvme has a version below the current version, then it can follow the original behavior, will this work or not ?

Older version will complain, that you need to provide the controller id, newer ones will just assume you wanted the first controller.

For extra points, this could be documented in the docs :)

maintainer
#5 lore raw
In reply to Keith Busch

nvme.c static int nvme_attach_ns(int argc, char **argv, int attach, const char *desc, s · line 2974

Quoted 8 lines · Keith Busch
nvme.c +8 -0
+	} else {
+		struct nvme_id_ctrl ctrl = { 0 };
+
+		if (nvme_cli_identify_ctrl(dev, &ctrl)) {
+			perror("identify-ctrl");
+			return -errno;
+		}
+		cntlist->identifier[0] = ctrl.cntlid;

This is not a zero-based-value, so this will need a

		cntlist->num = cpu_to_le16(1);

in this section.

#6 lore raw
In reply to Keith Busch
Quoted 30 lines · Keith Busch

From: Keith Busch <[email protected]>

Assume the user meant to attach the namespace to the controller the command was sent to if the user didn't provide a controller id list.

---
 nvme.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
nvme.c +0 -4
index bb412843..7a2c2e9f 100644
@@ -2953,9 +2953,6 @@ static int nvme_attach_ns(int argc, char **argv, int attach, const char *desc, s
 
 	num = argconfig_parse_comma_sep_array_u16(cfg.cntlist,
 						  list, ARRAY_SIZE(list));
-	if (!num)
-		fprintf(stderr, "warning: empty controller-id list will result in no actual change in namespace attachment\n");
-
 	if (num == -1) {
 		nvme_show_error("%s: controller id list is malformed", cmd->name);
 		return -EINVAL;
@@ -2965,7 +2962,18 @@ static int nvme_attach_ns(int argc, char **argv, int attach, const char *desc, s
 	if (!cntlist)
 		return -ENOMEM;
 
-	nvme_init_ctrl_list(cntlist, num, list);

nvme.c static int nvme_attach_ns(int argc, char **argv, int attach, const char *desc, s · line 2974

Quoted 10 lines
nvme.c +10 -0
+	if (argconfig_parse_seen(opts, "controllers")) {
+		nvme_init_ctrl_list(cntlist, num, list);
+	} else {
+		struct nvme_id_ctrl ctrl = { 0 };
+
+		if (nvme_cli_identify_ctrl(dev, &ctrl)) {
+			perror("identify-ctrl");
+			return -errno;
+		}
+		cntlist->identifier[0] = ctrl.cntlid;

Thanks Keith for working on this!

I think we also need to set the num of cntrls in the list here. In this particular case, we need to add the following line after we initialize the cntrl identifier:

cntlist->num = cpu_to_le16(1);

Otherwise changes looks good to me.

Thanks,
--Nilay

#7 lore raw
In reply to Keith Busch
Quoted 4 lines · Keith Busch

From: Keith Busch <[email protected]>

Assume the user meant to attach the namespace to the controller the command was sent to if the user didn't provide a controller id list.

This looks good, but I think it also needs a documentation update.