Skip to contentSkip to conversation

nvme-fabrics: add ENOKEY to no retry criteria for authentication failures

PATCH Applied Justin Tee 5 participants 8 messages Nov 17, 2025 – Nov 30, 2025 NVMe

lore
Share & tools
lore

Justin Tee

author
#0 lore raw

PATCH 1/1 nvme-fabrics: add ENOKEY to no retry criteria for authentication failures

With authentication, in addition to EKEYREJECTED there is also no point in retrying reconnects when status is ENOKEY. Thus, add -ENOKEY as another criteria to determine when to stop retries.

---
 drivers/nvme/host/fabrics.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/fabrics.c +1 -1
index 3d4d6d8e88c4..fb7765bf0cdf 100644
@@ -594,7 +594,7 @@ bool nvmf_should_reconnect(struct nvme_ctrl *ctrl, int status)
 	if (status > 0 && (status & NVME_STATUS_DNR))
 		return false;
 
-	if (status == -EKEYREJECTED)
+	if (status == -EKEYREJECTED || status == -ENOKEY)
 		return false;
 
 	if (ctrl->opts->max_reconnects == -1 ||
Signature · 1 line
2.38.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 17 lines · Justin Tee

With authentication, in addition to EKEYREJECTED there is also no point in retrying reconnects when status is ENOKEY. Thus, add -ENOKEY as another criteria to determine when to stop retries.

---
  drivers/nvme/host/fabrics.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/fabrics.c +0 -0
index 3d4d6d8e88c4..fb7765bf0cdf 100644

drivers/nvme/host/fabrics.c bool nvmf_should_reconnect(struct nvme_ctrl *ctrl, int status) · line 600

Quoted 8 lines
drivers/nvme/host/fabrics.c +1 -1
@@ -594,7 +594,7 @@ bool nvmf_should_reconnect(struct nvme_ctrl *ctrl, int status)
  	if (status > 0 && (status & NVME_STATUS_DNR))
  		return false;
 
-	if (status == -EKEYREJECTED)
+	if (status == -EKEYREJECTED || status == -ENOKEY)
  		return false;
 
  	if (ctrl->opts->max_reconnects == -1 ||

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                  Kernel Storage Architect
[email protected]                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
Tested-by Reviewed-by
#2 lore raw

Hi Justin,

Sorry for the late response. I was on vacation.

Quoted 9 lines · Hannes Reinecke
Quoted 8 lines · Justin Tee

With authentication, in addition to EKEYREJECTED there is also no point in retrying reconnects when status is ENOKEY. Thus, add -ENOKEY as another criteria to determine when to stop retries.

The patch itself is okay, nothing against it. But it wont close the bug report. blktests nvme/041 just passes because we don't check the return code from 'nvme connect' yet. I wanted to add this for a while, time to do it.

Furthermore, the test now races with the kernel. If the worker thread in the kernel is delayed, the 'nvme disconnect' might still see a controller:

Test authenticated connection
-disconnected 1 controller(s)
+disconnected 0 controller(s)

Anyway, I am fine with this change just without the 'Closes' tag.

author
#3 lore raw

Hi Daniel,

Quoted 8 lines

The patch itself is okay, nothing against it. But it wont close the bug report. blktests nvme/041 just passes because we don't check the return code from 'nvme connect' yet. I wanted to add this for a while, time to do it.

Furthermore, the test now races with the kernel. If the worker thread in the kernel is delayed, the 'nvme disconnect' might still see a controller:

Quoted 8 lines

Test authenticated connection
-disconnected 1 controller(s)
+disconnected 0 controller(s)

Anyway, I am fine with this change just without the 'Closes' tag.

Was under the impression that the issue with 041 is that there is a lingering reconnect attempt leftover from the “Test unauthenticated connection (should fail)” portion that interferes with the next “Test authenticated connection” portion of the blktest.

Through correcting the nvmf_should_reconnect routine, which stops retrying reconnect when -ENOKEY, there would be no interference in the second part of the test when nvme connect is with an actual key. So, there will be a controller to nvme disconnect, as expected, for the “Test authenticated connection” portion of the test.

Regards,
Justin

#4 lore raw

Hi Justin,

Quoted 4 lines · Justin Tee

Was under the impression that the issue with 041 is that there is a lingering reconnect attempt leftover from the “Test unauthenticated connection (should fail)” portion that interferes with the next “Test authenticated connection” portion of the blktest.

Yes, that was one part of the problem. FC keept continuing reconnecting with an invalid key. The other transport would also try to reconnect with an invalid key after the first initial connection. This is a long standing problem and your patch fixes it for all transport.

As I said your fix is good and addresses the reconnect with an invalid key problem. Though it doesn't address the issue that 'nvme connect' will still return success.

Quoted 5 lines

Through correcting the nvmf_should_reconnect routine, which stops retrying reconnect when -ENOKEY, there would be no interference in the second part of the test when nvme connect is with an actual key. So, there will be a controller to nvme disconnect, as expected, for the “Test authenticated connection” portion of the test.

All the nvme tests which use 'nvme connect' do not check the return code yet. Something I am going to add now. This should ensure that all the transport behave from userpoint of view the same. There is no exception for a transport to behave differently IMO. On the other hand we can't break existing APIs, so we have to come up with an opt-in solution here.

Hope this makes it a bit more clearer.

Thanks a lot,
Daniel

#5 lore raw
Quoted 3 lines · Daniel Wagner

As I said your fix is good and addresses the reconnect with an invalid key problem. Though it doesn't address the issue that 'nvme connect' will still return success.

To my surprise this is not happening. nvme/041 is behaving for all transport the same. I've extended blktests to check the return code for 'nvme connect'

https://github.com/linux-blktests/blktests/pull/216

So your patch fixes this problem completely. Nice!

Thanks,
Daniel

maintainer Applied
#6 lore raw
In reply to Justin Tee
Quoted 3 lines · Justin Tee

With authentication, in addition to EKEYREJECTED there is also no point in retrying reconnects when status is ENOKEY. Thus, add -ENOKEY as another criteria to determine when to stop retries.

Thanks, applied to nvme-6.19.