Re: [BUG] 2.6.24-rc2-mm1 - kernel bug on nfs v4

From: Trond Myklebust
Date: Sat Nov 17 2007 - 13:59:10 EST



On Sat, 2007-11-17 at 18:53 +0100, Torsten Kaiser wrote:
> On Nov 16, 2007 3:15 PM, Kamalesh Babulal <kamalesh@xxxxxxxxxxxxxxxxxx> wrote:
> > Hi Andrew,
> >
> > The kernel enters the xmon state while running the file system
> > stress on nfs v4 mounted partition.
> [snip]
> > 0:mon> t
> > [c0000000dbd4fb50] c000000000069768 .__wake_up+0x54/0x88
> > [c0000000dbd4fc00] d00000000086b890 .nfs_sb_deactive+0x44/0x58 [nfs]
> > [c0000000dbd4fc80] d000000000872658 .nfs_free_unlinkdata+0x2c/0x74 [nfs]
> > [c0000000dbd4fd10] d000000000598510 .rpc_release_calldata+0x50/0x74 [sunrpc]
> > [c0000000dbd4fda0] c00000000008d960 .run_workqueue+0x10c/0x1f4
> > [c0000000dbd4fe50] c00000000008ec70 .worker_thread+0x118/0x138
> > [c0000000dbd4ff00] c0000000000939f4 .kthread+0x78/0xc4
> > [c0000000dbd4ff90] c00000000002b060 .kernel_thread+0x4c/0x68

Could you try with the attached patch.

Cheers
Trond
--- Begin Message --- We should really only be calling nfs_sb_deactive() at the end of an RPC
call, to balance the nfs_sb_active() call in nfs_do_call_unlink(). OTOH,
nfs_free_unlinkdata() can be called from a variety of other situations.

Fix is to move the call to nfs_sb_deactive() into
nfs_async_unlink_release().

Signed-off-by: Trond Myklebust <Trond.Myklebust@xxxxxxxxxx>
---

fs/nfs/unlink.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/fs/nfs/unlink.c b/fs/nfs/unlink.c
index b97d3bb..c90862a 100644
--- a/fs/nfs/unlink.c
+++ b/fs/nfs/unlink.c
@@ -31,7 +31,6 @@ struct nfs_unlinkdata {
static void
nfs_free_unlinkdata(struct nfs_unlinkdata *data)
{
- nfs_sb_deactive(NFS_SERVER(data->dir));
iput(data->dir);
put_rpccred(data->cred);
kfree(data->args.name.name);
@@ -116,6 +115,7 @@ static void nfs_async_unlink_release(void *calldata)
struct nfs_unlinkdata *data = calldata;

nfs_dec_sillycount(data->dir);
+ nfs_sb_deactive(NFS_SERVER(data->dir));
nfs_free_unlinkdata(data);
}


--- End Message ---