Skip to contentSkip to conversation

NFS Locking violates protocol spec (incompatible with FreeBSD)

DISCUSSION Kris Kennaway 3 participants 5 messages Nov 7, 2003

lore
Share & tools
lore
#0 lore raw

Hello,

In http://lxr.linux.no/source/include/linux/lockd/xdr.h?v=2.6.0-test7
can be found the following comment:

35 /*

36  *      NLM cookies. Technically they can be 1K, Nobody uses over 8 bytes
37  *      however.
38  */

39
40 struct nlm_cookie
41 {
42 unsigned char data[8];
43 unsigned int len;
44 };

Unfortunately, this is incorrect: FreeBSD 5.x's rpc.lockd uses a 16 byte cookie, and therefore FreeBSD 5.x NFS clients cannot interoperate with Linux when NFS locking is enabled.

http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/56461

contains more details about this problem, including a workaround for FreeBSD to limit the cookie size to 8 bytes. Obviously, it would be better for this bug to be fixed in Linux, since Linux is non-conformant to the protocol.

Kris

#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == Kris Kennaway <[email protected]> writes:

> contains more details about this problem, including a > workaround for FreeBSD to limit the cookie size to 8 bytes. > Obviously, it would be better for this bug to be fixed in > Linux, since Linux is non-conformant to the protocol.

Yes. I saw a mail with a justification for why you want to be able to wait on > 2^64 outstanding lock requests to a single lockd server, and was highly amused.
I'm still hoping the person who decided that he needed 1024 byte long cookies will own up some day. OTOH, he might still be busy testing his locking code for cookie wraparound...

Anyhow, a patch exists (written by Greg Banks), and can be found as

http://www.fys.uio.no/~trondmy/src/Linux-2.4.x/2.4.23-pre9/linux-2.4.23-01-fix_osx.dif

No. It does not extend the cookie size to 1k...

Cheers,
Trond

#2 lore raw
Quoted 16 lines · Trond Myklebust
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == Kris Kennaway <[email protected]> writes:

> contains more details about this problem, including a > workaround for FreeBSD to limit the cookie size to 8 bytes. > Obviously, it would be better for this bug to be fixed in > Linux, since Linux is non-conformant to the protocol.

Yes. I saw a mail with a justification for why you want to be able to wait on > 2^64 outstanding lock requests to a single lockd server, and was highly amused.
I'm still hoping the person who decided that he needed 1024 byte long cookies will own up some day. OTOH, he might still be busy testing his locking code for cookie wraparound...

Quoted 6 lines

Anyhow, a patch exists (written by Greg Banks), and can be found as

http://www.fys.uio.no/~trondmy/src/Linux-2.4.x/2.4.23-pre9/linux-2.4.23-01-fix_osx.dif

No. It does not extend the cookie size to 1k...

Thanks..obviously we'd like a fix to be committed to Linux so that it interoperates out of the box with FreeBSD. What are the chances of this?

Kris

#3 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == Kris Kennaway <[email protected]> writes:

> Thanks..obviously we'd like a fix to be committed to Linux so > that it interoperates out of the box with FreeBSD. What are > the chances of this?

For the moment I'm still waiting for confirmation that this patch does actually fix the problem on both FreeBSD and OSX/Panther. Once that is done, it's up to Marcelo to tell me when he wants it into 2.4.x. He has already announce that he only wants critical bugfixes in 2.4.23, though, so I would guess that we will have to wait for 2.4.24.

As for 2.6.x, I expect we will have to wait until the code freeze lifts. When it does, I already have a backlog of lockd bugfixes to port forward from 2.4.x, so it will make sense to merge it together with that.

Then we just have to wait for the distributions to catch up ;-)

Cheers,
Trond

#4 lore raw
Quoted 12 lines · Trond Myklebust
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == Kris Kennaway <[email protected]> writes:

> Thanks..obviously we'd like a fix to be committed to Linux so > that it interoperates out of the box with FreeBSD. What are > the chances of this?

For the moment I'm still waiting for confirmation that this patch does actually fix the problem on both FreeBSD and OSX/Panther. Once that is done, it's up to Marcelo to tell me when he wants it into 2.4.x. He has already announce that he only wants critical bugfixes in 2.4.23, though, so I would guess that we will have to wait for 2.4.24.

Quoted 6 lines

As for 2.6.x, I expect we will have to wait until the code freeze lifts. When it does, I already have a backlog of lockd bugfixes to port forward from 2.4.x, so it will make sense to merge it together with that.

Then we just have to wait for the distributions to catch up ;-)

Thanks. Unfortunately the Linux server I experience this problem with is not something I can play with, but I've CC'ed two people who experience this problem with FreeBSD: perhaps they can test the patch
(http://www.fys.uio.no/~trondmy/src/Linux-2.4.x/2.4.23-pre9/linux-2.4.23-01-fix_osx.dif)

Kris