Skip to contentSkip to conversation

knfsd: Fix a race in closing NFSd connections.

PATCH In review NeilBrown 2 participants 3 messages Feb 7, 2007 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 20070207110636.9166.patches@notabene

NeilBrown

author
#1 lore raw

PATCH 1/1 knfsd: Fix a race in closing NFSd connections.

There is a bug introduced in 2.6.19 that can cause badness in the NFS server, particularly if you get clients closing connections while the server is trying to write a reply.

This patch fixed the bug and improves the handling of socket closing.

The patch was made against -mm, and applies correctly with a little bit of fuzz to 2.6.19 and 2.6.20. It should be considered at least for 2.6.20.1.

Thanks,
NeilBrown

### Comments for Changeset

If you lose this race, it can iput a socket inode twice and you get a BUG in fs/inode.c

When I added the option for user-space to close a socket, I added some cruft to svc_delete_socket so that I could call that function when closing a socket per user-space request.

This was the wrong thing to do. I should have just set SK_CLOSE and let normal mechanisms do the work.

Not only wrong, but buggy. The locking is all wrong and it openned up a race where-by a socket could be closed twice.

So this patch:
Introduces svc_close_socket which sets SK_CLOSE then either leave the close up to a thread, or calls svc_delete_socket if it can get SK_BUSY.

Adds a bias to sk_busy which is removed when SK_DEAD is set, This avoid races around shutting down the socket.

Changes several 'spin_lock' to 'spin_lock_bh' where the _bh was missing.

Bugzilla-url: http://bugzilla.kernel.org/show_bug.cgi?id=7916

### Diffstat output

 ./include/linux/sunrpc/svcsock.h |    2 -
 ./net/sunrpc/svc.c               |    4 +--
 ./net/sunrpc/svcsock.c           |   52 ++++++++++++++++++++++++++++-----------
 3 files changed, 41 insertions(+), 17 deletions(-)
include/linux/sunrpc/svcsock.h +1 -1
diff .prev/include/linux/sunrpc/svcsock.h ./include/linux/sunrpc/svcsock.h
@@ -66,7 +66,7 @@ struct svc_sock {
  * Function prototypes.
  */
 int		svc_makesock(struct svc_serv *, int, unsigned short, int flags);
-void		svc_delete_socket(struct svc_sock *);
+void		svc_close_socket(struct svc_sock *);
 int		svc_recv(struct svc_rqst *, long);
 int		svc_send(struct svc_rqst *);
 void		svc_drop(struct svc_rqst *);
net/sunrpc/svc.c +2 -2
diff .prev/net/sunrpc/svc.c ./net/sunrpc/svc.c
@@ -386,7 +386,7 @@ svc_destroy(struct svc_serv *serv)
 		svsk = list_entry(serv->sv_tempsocks.next,
 				  struct svc_sock,
 				  sk_list);
-		svc_delete_socket(svsk);
+		svc_close_socket(svsk);
 	}
 	if (serv->sv_shutdown)
 		serv->sv_shutdown(serv);
@@ -395,7 +395,7 @@ svc_destroy(struct svc_serv *serv)
 		svsk = list_entry(serv->sv_permsocks.next,
 				  struct svc_sock,
 				  sk_list);
-		svc_delete_socket(svsk);
+		svc_close_socket(svsk);
 	}
 	
 	cache_clean_deferred(serv);
net/sunrpc/svcsock.c +38 -14

120 lines not on this page · raw message

#2 lore raw
Quoted 18 lines · On Wed, 7 Feb 2007 11:10:26 +1100 NeilBrown <[email protected]> wrote:

If you lose this race, it can iput a socket inode twice and you get a BUG in fs/inode.c

When I added the option for user-space to close a socket, I added some cruft to svc_delete_socket so that I could call that function when closing a socket per user-space request.

This was the wrong thing to do. I should have just set SK_CLOSE and let normal mechanisms do the work.

Not only wrong, but buggy. The locking is all wrong and it openned up a race where-by a socket could be closed twice.

So this patch:
Introduces svc_close_socket which sets SK_CLOSE then either leave the close up to a thread, or calls svc_delete_socket if it can get SK_BUSY.

Quoted 6 lines

Adds a bias to sk_busy which is removed when SK_DEAD is set, This avoid races around shutting down the socket.

Changes several 'spin_lock' to 'spin_lock_bh' where the _bh was missing.

This patch assumes the presence of knfsd-sunrpc-allow-creating-an-rpc-service-without-registering-with-portmapper.patch which obviously isn't appropriate for -stable.

Please confirm that the patch which I merged is still OK, thanks.

-------------------------------------------------------------------------
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 - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

author
#3 lore raw
Quoted 4 lines · On Tuesday February 6, [email protected] wrote:

This patch assumes the presence of knfsd-sunrpc-allow-creating-an-rpc-service-without-registering-with-portmapper.patch which obviously isn't appropriate for -stable.

Only because they change adjacent lines. There is no semantic dependency.
The patch I sent will apply *correctly* against -stable with "--fuzz=3".
I cannot send a fuzz-free patch if that would make Mr Stable happy.

Quoted 2 lines

Please confirm that the patch which I merged is still OK, thanks.

Yes, the patch which you merged is OK.

Thanks,
NeilBrown