Skip to contentSkip to conversation

nfs4v stalls transfering files, stalls

DISCUSSION Konstantin Kletschke 1 participant 2 messages Aug 25, 2008

lore
Share & tools
lore
#0 lore raw

Recently I experience oddities with my home nfsv4 client/server connection which worked long very well. I don't know exactly with which version it begun. I experienced once in a while, that I delete files but after rebooting the computers the next day or after doing an ls on the directory, the files are still there. While I did not take this sriously first I now have the problem copying an ~600MB avi onto the nfs share. After 25M or 30M (each time after this amount already transferred) the connection stalls:

nfs: server 10.10.0.1 not responding, timed out

On my client I have 2.6.17_rc3 and this fstab entry:

10.10.0.1:/pub/ /pub nfs4 tcp,async,hard,intr,noauto,user,sec=sys 0 0

this is only an example, I tried udp, soft and timeo=8 instead of hard and intr.

On server I had 2.6.25 with this exports:

/exports 10.10.0.0/255.255.0.0(rw,fsid=0,insecure,no_subtree_check,sync) /exports/pub 10.10.0.0/255.255.0.0(rw,nohide,insecure,no_subtree_check,async)

When this happened, the server had this in dmesg:

RPC: bad TCP reclen 0x0010011a (non-terminal)

or:

RPC: bad TCP reclen 0x3b468abe (large)

In 2.6.27_rc (runnning now), the part in svchost.c was spelled other, it says now:

RPC: fragment too large: 0x6be367a9
or
RPC: fragment too large: 0x1287087c

Starting the nfs stuff yields into

RPC: multiple fragments per record not supported

with no other regression though.

What can I do to narrow this down?

Kind Regards, Konsti

Signature · 2 lines
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
#1 lore raw

Hm, may be the ethernet driver atl1 is the culprit.
I stumpled upon "sendfile() broken with 2.6.26 + Apache 2 ?" and especially http://marc.info/?l=linux-kernel&m=121619345703525&w=2

So I did "ethtool -K eth0 tso off" and I was able to transfer the file immediately twice, all three (origin, copy 1 and 2) have identical md5sum and the transferrate was reasonable at ca. 12MB/s.

Regards, Konsti

Signature · 2 lines
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF