ksmbd stuck on write event after attempt to shutdown

[consp]

After I tried shutting down ksmbd gave these errors and could not be stopped:

[82661.298655] ksmbd: sock_read failed: -108
[82661.559100] ksmbd: smb_direct: Unexpected RDMA CM event. cm_id=00000000ea775640, event=timewait exit (15)
[82667.180135] ksmbd: Unable to close RPC pipe 2
[82667.180351] ksmbd: Unable to close RPC pipe 1
[82667.180572] ksmbd: Unable to close RPC pipe 0

Console output of the stuck part on reboot, I do not have the logs:

Have not been able to reproduce it but it might have had to do with a change in the filesystem since a disk was remounted.

Hopefully the information is of use to you!

[namjaejeon]

ksmbd: smb_direct: Unexpected RDMA CM event. cm_id=00000000ea775640, event=timewait exit (15)

@hclee any opinion?

[namjaejeon]

@consp Let me know your NIC model name and kernel version.

[hclee]

ksmbd: smb_direct: Unexpected RDMA CM event. cm_id=00000000ea775640, event=timewait exit (15)

@hclee any opinion?

We have to terminate the loop for handling connection if RDMA_CM_EVENT_TIMEWAIT_EXIT is received like RDMA_CM_EVENT_DISCONNECTED.

[consp]

@consp Let me know your NIC model name and kernel version.

5.17 xanmod (also had it with the mainline version now) and a change to allow the ConnectX 4 card to be detected.
Card: ConnectX 4 MCX455A

[namjaejeon]

We have to terminate the loop for handling connection if RDMA_CM_EVENT_TIMEWAIT_EXIT is received like RDMA_CM_EVENT_DISCONNECTED.

@hclee So, what are you going to do ?

[hclee]

We have to terminate the loop for handling connection if RDMA_CM_EVENT_TIMEWAIT_EXIT is received like RDMA_CM_EVENT_DISCONNECTED.

@hclee So, what are you going to do ?

I will send a patch for this issue.

[namjaejeon]

@hclee Thanks!
@consp I have applied the patch(bb974ae) for this issue. Can you check it ?

[consp]

@hclee Thanks! @consp I have applied the patch(bb974ae) for this issue. Can you check it ?

Will install it this week, might take a while to show up/not show up as it doesn't happen very often.

[consp]

So far so good, the things I got it to sometimes break don't do anything bad anymore and so far no issues with ksmbd.control -s. Stability seems ok, but I will have to leave it running for longer to be sure.

[namjaejeon]

@consp I guess this issue may be fixed with this patch(5039a45)

[consp]

@consp I guess this issue may be fixed with this patch(5039a45)

Looks like it might be the case, "racy" sounds like what I experienced. I'll have a look at it when I get back from holiday.

[consp]

Fixed is correct, I do not get any issues anymore. @namjaejeon tnx for the fix!
ksmbd restarted several times as that is part of the server hibernation procedure, power is expensive these days and I've seen no messages so far.

[namjaejeon]

@consp Thanks for your checking it:) Let me know if you have any issue while using ksmbd.