Messages in this thread Patch in this message | | | Subject | Re: lockd: couldn't create RPC handle for (host) | From | Trond Myklebust <> | Date | Sun, 18 Dec 2005 17:24:24 -0500 |
| |
On Sun, 2005-12-18 at 15:00 -0500, Ryan Richter wrote: > On Sun, Dec 18, 2005 at 02:31:07PM -0500, Trond Myklebust wrote: > > On Sun, 2005-12-18 at 13:01 -0500, Ryan Richter wrote: > > > Code: 48 39 78 18 75 1c 8b 86 8c 00 00 00 a8 01 74 12 83 c8 02 89 > > > RIP <ffffffff801dbd9e>{nlmclnt_mark_reclaim+62} RSP <ffff81007dfade70> > > > CR2: 0000000000000018 > > > > Looks like the global lock list is corrupted. Could you cat the contents > > of /proc/locks? > > $ cat /proc/locks > 1: POSIX ADVISORY WRITE 1657 00:0e:1771273 0 EOF > 2: FLOCK ADVISORY WRITE 1486 00:0e:1770759 0 EOF > 3: FLOCK ADVISORY WRITE 1478 00:0e:1770399 0 EOF
OK. I think this client patch ought to fix the Oopses. It should apply to a 2.6.14 kernel as well as 2.6.15-rc5.
Cheers, Trond
NLM: Fix Oops in nlmclnt_mark_reclaim()
When mixing -olock and -onolock mounts on the same client, we have to check that fl->fl_u.nfs_fl.owner is set before dereferencing it.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com> ---
fs/lockd/clntlock.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/fs/lockd/clntlock.c b/fs/lockd/clntlock.c index 006bb9e..3eaf6e7 100644 --- a/fs/lockd/clntlock.c +++ b/fs/lockd/clntlock.c @@ -157,6 +157,8 @@ void nlmclnt_mark_reclaim(struct nlm_hos inode = fl->fl_file->f_dentry->d_inode; if (inode->i_sb->s_magic != NFS_SUPER_MAGIC) continue; + if (fl->fl_u.nfs_fl.owner == NULL) + continue; if (fl->fl_u.nfs_fl.owner->host != host) continue; if (!(fl->fl_u.nfs_fl.flags & NFS_LCK_GRANTED)) @@ -226,6 +228,8 @@ restart: inode = fl->fl_file->f_dentry->d_inode; if (inode->i_sb->s_magic != NFS_SUPER_MAGIC) continue; + if (fl->fl_u.nfs_fl.owner == NULL) + continue; if (fl->fl_u.nfs_fl.owner->host != host) continue; if (!(fl->fl_u.nfs_fl.flags & NFS_LCK_RECLAIM)) | |