Linux NFS development
 help / color / mirror / Atom feed
* Re: NLM lock reclaim failure
@ 2006-10-13  5:43 Marc Eshel
  2006-10-13  7:08 ` Neil Brown
  0 siblings, 1 reply; 5+ messages in thread
From: Marc Eshel @ 2006-10-13  5:43 UTC (permalink / raw)
  To: Linux NFS Mailing List

Hi Trond,
Attached is a tested patch for the bug I posted in the following mail
Marc.

nfs-bounces@lists.sourceforge.net wrote on 10/11/2006 04:28:16 PM:

 > I see an NLM lock reclaim failure (using 2.6.17). Let me describe
 > the simple case first.
 >
 > 1. Client gets an NLM lock
 > 2. Server reboots
 > 3. Client sends lock reclaim request
 > 4. On server, nlm_open() gets return code nfserr_dropit; error gets
 >    converted to nlm_lck_denied by nlm_lookup_file().
 > 5. Client gets status=1, the lock reclaim failed
 >
 > This happens _only_ when the first call to nfsd (fh_verify -> exp_find)
 > comes from lockd trying to find the file. Subsequent calls seem to work.
 >
 > nlm_lookup_file() -> nlm_fopen() -> nfsd_open() -> fh_verify() ->
 > exp_find() returns nfserr_dropit
 >
 > This problem may be related to the following commit to lockd (Trond's
 > tree):
 > commit: 26bcbf965f857c710adafd16cf424f043006b5dd
 > lockd: stop abusing file_lock_list
 >
 > The client earlier retained the list of outstanding locks regardless of
 > reclaim return code. The change above removes all locks from the list
 > and only adds them back after a successful reclaim. Since the reclaim
 > fails, it is removed from the list; so the client will not try to
 > reclaim the lock on a second failure.
 >
 > Note that the problem is not on the client side - it seems the server
 > that should drop the rpc when it gets nfserr_dropit error and not
 > convert it to an nlm_lck_denied.
 >
 > It get even more complicated - when the server doesn't respond
 > immediately to the reclaim, client will retry the rpc, on the second
 > reclaim call, the server will grant the lock to the client but since the
 > client already got the nlm_lck_denied from the first call, it has
 > dropped it from it list. So now we have a lock on the server for a
 > client that doesn't know that it holds the lock.
 >
 > Marc.
 >

Index: lockd/svc.c
===================================================================
RCS file: /cvs/nfsv4/cvs/pnfs/fs/lockd/svc.c,v
retrieving revision 1.1.1.4
diff -u -r1.1.1.4 svc.c
--- lockd/svc.c	12 Jul 2006 19:53:39 -0000	1.1.1.4
+++ lockd/svc.c	13 Oct 2006 04:03:13 -0000
@@ -34,6 +34,7 @@
 #include <linux/sunrpc/svcsock.h>
 #include <linux/lockd/lockd.h>
 #include <linux/nfs.h>
+#include <linux/nfsd/nfsd.h>
 
 #define NLMDBG_FACILITY		NLMDBG_SVC
 #define LOCKD_BUFSIZE		(1024 + NLMSVC_XDRSIZE)
@@ -320,6 +321,43 @@
 }
 EXPORT_SYMBOL(lockd_down);
 
+int
+nlmsvc_dispatch(struct svc_rqst *rqstp, u32 *statp)
+{
+	struct svc_procedure	*procp;
+	kxdrproc_t		xdr;
+	struct kvec *		argv = &rqstp->rq_arg.head[0];
+	struct kvec *		resv = &rqstp->rq_res.head[0];
+
+	dprintk("lockd: nlmsvc_dispatch vers %d proc %d\n",
+				rqstp->rq_vers, rqstp->rq_proc);
+
+	procp = rqstp->rq_procinfo;
+
+	/* Decode arguments */
+	xdr = procp->pc_decode;
+
+	if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) {
+		dprintk("lockd: failed to decode arguments!\n");
+		*statp = rpc_garbage_args;
+		return 1;
+	}
+	/* Now call the procedure handler, and encode status. */
+	*statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
+	if (((struct nlm_res  *)(rqstp->rq_resp))->status == nfserr_dropit) {
+		dprintk("lockd: dropping request!\n");
+		return 0;
+	}
+	/* Encode reply */
+	if (*statp == rpc_success && (xdr = procp->pc_encode)
+	    && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
+			dprintk("lockd: failed to encode reply\n");
+			/* serv->sv_stats->rpcsystemerr++; */
+			*statp = rpc_system_err;
+	}
+	return 1;
+}
+
 /*
  * Sysctl parameters (same as module parameters, different interface).
  */
@@ -484,12 +522,14 @@
 		.vs_vers	= 1,
 		.vs_nproc	= 17,
 		.vs_proc	= nlmsvc_procedures,
+		.vs_dispatch	= nlmsvc_dispatch,
 		.vs_xdrsize	= NLMSVC_XDRSIZE,
 };
 static struct svc_version	nlmsvc_version3 = {
 		.vs_vers	= 3,
 		.vs_nproc	= 24,
 		.vs_proc	= nlmsvc_procedures,
+		.vs_dispatch	= nlmsvc_dispatch,
 		.vs_xdrsize	= NLMSVC_XDRSIZE,
 };
 #ifdef CONFIG_LOCKD_V4
@@ -497,6 +537,7 @@
 		.vs_vers	= 4,
 		.vs_nproc	= 24,
 		.vs_proc	= nlmsvc_procedures4,
+		.vs_dispatch	= nlmsvc_dispatch,
 		.vs_xdrsize	= NLMSVC_XDRSIZE,
 };
 #endif
Index: lockd/svcsubs.c
===================================================================
RCS file: /cvs/nfsv4/cvs/pnfs/fs/lockd/svcsubs.c,v
retrieving revision 1.1.1.4
diff -u -r1.1.1.4 svcsubs.c
--- lockd/svcsubs.c	12 Jul 2006 19:53:39 -0000	1.1.1.4
+++ lockd/svcsubs.c	13 Oct 2006 04:03:13 -0000
@@ -16,6 +16,7 @@
 #include <linux/sunrpc/clnt.h>
 #include <linux/nfsd/nfsfh.h>
 #include <linux/nfsd/export.h>
+#include <linux/nfsd/nfsd.h>
 #include <linux/lockd/lockd.h>
 #include <linux/lockd/share.h>
 #include <linux/lockd/sm_inter.h>
@@ -136,12 +137,14 @@
 
 out_free:
 	kfree(file);
+	if (nfserr != nfserr_dropit) {
 #ifdef CONFIG_LOCKD_V4
-	if (nfserr == 1)
-		nfserr = nlm4_stale_fh;
-	else
+		if (nfserr == 1)
+			nfserr = nlm4_stale_fh;
+		else
 #endif
-	nfserr = nlm_lck_denied;
+		nfserr = nlm_lck_denied;
+	}
 	goto out_unlock;
 }
 
Index: nfsd/lockd.c
===================================================================
RCS file: /cvs/nfsv4/cvs/pnfs/fs/nfsd/lockd.c,v
retrieving revision 1.1.1.3
diff -u -r1.1.1.3 lockd.c
--- nfsd/lockd.c	4 Apr 2006 17:55:25 -0000	1.1.1.3
+++ nfsd/lockd.c	13 Oct 2006 04:03:14 -0000
@@ -49,6 +49,8 @@
 		return 0;
 	case nfserr_stale:
 		return 1;
+	case nfserr_dropit:
+		return nfserr;
 	default:
 		return 2;
 	}



-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

^ permalink raw reply	[flat|nested] 5+ messages in thread
* NLM lock reclaim failure
@ 2006-10-11 23:28 Marc Eshel
  0 siblings, 0 replies; 5+ messages in thread
From: Marc Eshel @ 2006-10-11 23:28 UTC (permalink / raw)
  To: Linux NFS Mailing List

I see an NLM lock reclaim failure (using 2.6.17). Let me describe 
the simple case first.

1. Client gets an NLM lock
2. Server reboots
3. Client sends lock reclaim request
4. On server, nlm_open() gets return code nfserr_dropit; error gets 
   converted to nlm_lck_denied by nlm_lookup_file().
5. Client gets status=1, the lock reclaim failed

This happens _only_ when the first call to nfsd (fh_verify -> exp_find) 
comes from lockd trying to find the file. Subsequent calls seem to work.

nlm_lookup_file() -> nlm_fopen() -> nfsd_open() -> fh_verify() -> 
exp_find() returns nfserr_dropit

This problem may be related to the following commit to lockd (Trond's 
tree):
commit: 26bcbf965f857c710adafd16cf424f043006b5dd
lockd: stop abusing file_lock_list

The client earlier retained the list of outstanding locks regardless of 
reclaim return code. The change above removes all locks from the list 
and only adds them back after a successful reclaim. Since the reclaim 
fails, it is removed from the list; so the client will not try to 
reclaim the lock on a second failure.

Note that the problem is not on the client side - it seems the server 
that should drop the rpc when it gets nfserr_dropit error and not 
convert it to an nlm_lck_denied.

It get even more complicated - when the server doesn't respond 
immediately to the reclaim, client will retry the rpc, on the second 
reclaim call, the server will grant the lock to the client but since the 
client already got the nlm_lck_denied from the first call, it has 
dropped it from it list. So now we have a lock on the server for a 
client that doesn't know that it holds the lock.

Marc.



-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

^ permalink raw reply	[flat|nested] 5+ messages in thread
end of thread, other threads:[~2006-10-16 23:17 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2006-10-13  5:43 NLM lock reclaim failure Marc Eshel
2006-10-13  7:08 ` Neil Brown
2006-10-13 18:22   ` Marc Eshel
2006-10-16 23:17     ` Neil Brown
  -- strict thread matches above, loose matches on Subject: below --
2006-10-11 23:28 Marc Eshel

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox