Gentoo Archives: gentoo-user

From: Matthias Bethke <matthias@×××××××.de>
To: gentoo-user@l.g.o
Subject: [gentoo-user] NFS vs. jumbo frames
Date: Mon, 23 Apr 2007 14:58:22
Message-Id: 20070423144227.GE14406@huxley
1 I've been fiddling with this for some days and can't but assume it's a
2 bug in one of the Gentoo patches to either the kernel or NFS tools:
3 Basically, NFS locking breaks as soon as I enable jumbo frames on both
4 server and client.
5 touch foobar
6 flock foobar ls
7 works fine in my NFS-mounted home with an MTU of 1500. An MTU of 9000 is
8 great for general net throughput so I wanted to use it on both the
9 server and the clients, but the above sequence hangs indefinitely when I
10 try. I'm aware flock() isn't supposed to work correctly with NFS anyway,
11 but all kinds of stuff depends on it at least pretending to.
12 The strange thing is, SuSE 10.1 as a client works fine with jumbo
13 frames, just my Gentoo box doesn't. I tried enabling nfs_debug with
14 sysctl and sniffing the wire with tcpdump and wireshark but with my
15 pretty basic knowledge of NFS workings I didn't spot anything
16 conspicuous other than that
17 lookup(msbethke/foobar)
18 nfs_update_inode(0:18/3424742 ct=1 info=0x6)
19 nfs_fhget(0:18/1081970 ct=1)
20 permission(0:18/1081970), mask=0x4, res=0
21 seems to be the exchange after which the hang occurs.
22 Our server is running 2.6.18-hardened-r6 and nfs-utils-1.0.12. The
23 clients are mostly SuSE 10.1 boxes with kernel 2.6.16.21-0.21-smp and
24 nfs-utils-1.0.7-36 while my workstation has 2.6.20-gentoo-r6 (was
25 linux-2.6.19-gentoo-r5 before) and the same ns-utils as the server.
26
27 --
28 I prefer encrypted and signed messages. KeyID: FAC37665
29 Fingerprint: 8C16 3F0m A6FC DF0D 19B0 8DEF 48D9 1700 FAC3 7665

Replies

Subject Author
Re: [gentoo-user] NFS vs. jumbo frames Uwe Thiem <uwix@××××.na>