Overwriting a file does not update mtime, only ctime? Is that normal?

[PMc]

root@vbuild3:/usr/ports # mkdir XX
root@vbuild3:/usr/ports # cd XX
root@vbuild3:/usr/ports/XX # cat > somecrap
crap1
crap2
root@vbuild3:/usr/ports/XX # ls -l ; ls -lc
total 6
-rw-r--r--  1 root wheel 12 Oct  3 06:16 somecrap
total 6
-rw-r--r--  1 root wheel 12 Oct  3 06:16 somecrap
root@vbuild3:/usr/ports/XX # sleep 120
root@vbuild3:/usr/ports/XX # cat > somecrap
crap1
crap2
root@vbuild3:/usr/ports/XX # ls -l ; ls -lc
total 6
-rw-r--r--  1 root wheel 12 Oct  3 06:16 somecrap
total 6
-rw-r--r--  1 root wheel 12 Oct  3 06:19 somecrap

 

[covacat]

i can't reproduce it on 14.4 ufs or zfs

 

[gpw928]

The file contents changed, so mtime should have changed. We need to know more information about the file system...

 

[PMc]

NFS4. Sorry, I was in a hurry earlier.
Release is 14.5.

This exists for a while (I seem to see it in logs from April), and it should have done real havoc (masses of files getting deleted because apparently unchanged), but for some other reason it hasn't. I'm trying to get the issue contained and figure out what it might have happened, before searching which options might cause it.

For sure I have seen it persist for half an hour at least, which is longer than any acceptable caching. And I am not sure if caching should happen at all on the node that did the file write.

 

[covacat]

does the problem show on both client and server ?

 

[PMc]

Only on client.
And it seems to be (another) bug in nfscbd(8), doesnt appear when that is not activated,

 

[mer]

Is the overwriting being done from the client?
If you touch the file from the client does mtime update?

 

[cracauer@]

What is the server, anyway (OS, version etc.)?

 

[PMc]

Is the overwriting being done from the client?

Yes.

If you touch the file from the client does mtime update?

Yes.

What is the server, anyway (OS, version etc.)?

All the same, Rel. 14.5-p1

 

[cracauer@]

All the same, Rel. 14.5-p1

What is sysctl vfs.nfsd.issue_delegations set to?

 

[PMc]

What is sysctl vfs.nfsd.issue_delegations set to?

Enabled, i.e. "1" on the server. "0" on the client.
Then on the client there is nfscbd_enable="YES" in rc.conf. And when I remove that, the problem goes away.

Anyway, this is not the first issue there. Last year I had network stalls, where the NFS would just freeze for 2-3 minutes. I was persistent, and was finally able to track that down sufficiently to pinpoint the cause - it was also NFSCBD. Rick Macklem was then able to fix it, per PR 289711, which finally arrived in Rel. 14.5

So now there is apparently another issue, and I start to get a bad feeling bothering Rick again. He is certainly one of the good ones, always quick in tackling reported PRs.
And anyway I have now already created two PRs in one upgrade session.

 

[cracauer@]

And anyway I have now already created two PRs in one upgrade session.

Nah. It's good to fine these bugs.