On Mon, 20 Jun 2016, Christos Zoulas wrote:
On Jun 20, 10:24pm, [email protected]
([email protected]) wrote:
-- Subject: Re: nfs client kernel crash
| On Fri, 17 Jun 2016, Christos Zoulas wrote:
|
| > | I am mounting with: nosuid,userquota,rw,tcp,soft,intr
| > |
| > | Is this a problem?
| >
| > Yes :-), I'd get rid of "soft" first.
| >
| > christos
| >
| I changed the nfs options to: userquota,nosuid,rw,tcp,bg
|
| The message is still there.
The interrupted message? If yes, this is wrong. Do you want to instrument
the code or should I and send you a patch?
Since I have removed the nfs mount options soft and intr, I have not seen
any interrupted message.
Since I have applied the patch there was no more crash. The PR/51215 can
be closed.
In my opinion, the NFS client implementation is still faulty. In the last
days I have created and deleted a mirror of Scientific Linux for several
times. This is a very large tree. Some directories have up to 10,000
entries. In some experiments I tried to clear the tree recursively and got
some error messages like:
rm -rf scientific/
rm: scientific/50/i386/updates/security: Directory not empty
rm: scientific/50/i386/updates: Directory not empty
rm: scientific/50/i386: Directory not empty
rm: scientific/50/x86_64/sites/Fermi/updates/security: Directory not empty
rm: scientific/50/x86_64/sites/Fermi/updates: Directory not empty
rm: scientific/50/x86_64/sites/Fermi: Directory not empty
rm: scientific/50/x86_64/sites: Directory not empty
rm: scientific/50/x86_64: Directory not empty
rm: scientific/50: Directory not empty
rm: scientific: Directory not empty
A new 'rm -rf' call deletes the remaining data. This behavior occurs
randomly. Often in a directory with many files. I observed the behavior
with different nfs servers.
I have not found any way to reproduce the behavior safely and with little
effort :-(
Regards
Uwe
