[Devel] Re: [PATCH] SUNRPC: set desired file system root before connecting local transports

[Myklebust, Trond]

On Wed, 2012-03-07 at 12:34 +0400, Stanislav Kinsbursky wrote:
> 07.03.2012 04:19, Myklebust, Trond пишет:
> > On Wed, 2012-02-29 at 18:59 +0400, Stanislav Kinsbursky wrote:
> >> Today, there is a problem in connecting of local SUNRPC thansports. These
> >> transports uses UNIX sockets and connection itself is done by rpciod workqueue.
> >> But UNIX sockets lookup is done in context of process file system root. I.e.
> >> all local thunsports are connecting in rpciod context.
> >> This works nice until we will try to mount NFS from process with other root -
> >> for example in container. This container can have it's own (nested) root and
> >> rcpbind process, listening on it's own unix sockets. But NFS mount attempt in
> >> this container will register new service (Lockd for example) in global rpcbind
> >> - not containers's one.
> >> This patch solves the problem by switching rpciod kernel thread's file system
> >> root to right one (stored on transport) while connecting of local transports.
> >>
> >> Signed-off-by: Stanislav Kinsbursky<skinsbursky at parallels.com>
> >>
> >> ---
> >>   fs/fs_struct.c        |    1 +
> >>   net/sunrpc/xprtsock.c |   32 ++++++++++++++++++++++++++++++--
> >>   2 files changed, 31 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/fs/fs_struct.c b/fs/fs_struct.c
> >> index 78b519c..0f984c3 100644
> >> --- a/fs/fs_struct.c
> >> +++ b/fs/fs_struct.c
> >> @@ -36,6 +36,7 @@ void set_fs_root(struct fs_struct *fs, struct path *path)
> >>   	if (old_root.dentry)
> >>   		path_put_longterm(&old_root);
> >>   }
> >> +EXPORT_SYMBOL_GPL(set_fs_root);
> >>
> >>   /*
> >>    * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
> >> diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
> >> index 4c8281d..c94c181 100644
> >> --- a/net/sunrpc/xprtsock.c
> >> +++ b/net/sunrpc/xprtsock.c
> >> @@ -37,6 +37,7 @@
> >>   #include<linux/sunrpc/svcsock.h>
> >>   #include<linux/sunrpc/xprtsock.h>
> >>   #include<linux/file.h>
> >> +#include<linux/fs_struct.h>
> >>   #ifdef CONFIG_SUNRPC_BACKCHANNEL
> >>   #include<linux/sunrpc/bc_xprt.h>
> >>   #endif
> >> @@ -255,6 +256,11 @@ struct sock_xprt {
> >>   	void			(*old_state_change)(struct sock *);
> >>   	void			(*old_write_space)(struct sock *);
> >>   	void			(*old_error_report)(struct sock *);
> >> +
> >> +	/*
> >> +	 * Saved transport creator root. Required for local transports only.
> >> +	 */
> >> +	struct path		root;
> >>   };
> >>
> >>   /*
> >> @@ -1891,6 +1897,7 @@ static void xs_local_setup_socket(struct work_struct *work)
> >>   	struct rpc_xprt *xprt =&transport->xprt;
> >>   	struct socket *sock;
> >>   	int status = -EIO;
> >> +	struct path root;
> >>
> >>   	if (xprt->shutdown)
> >>   		goto out;
> >> @@ -1908,7 +1915,14 @@ static void xs_local_setup_socket(struct work_struct *work)
> >>   	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
> >>   			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
> >>
> >> +	get_fs_root(current->fs,&root);
> >> +	set_fs_root(current->fs,&transport->root);
> >> +
> >>   	status = xs_local_finish_connecting(xprt, sock);
> >> +
> >> +	set_fs_root(current->fs,&root);
> >> +	path_put(&root);
> >> +
> >>   	switch (status) {
> >>   	case 0:
> >
> > Hi Stanislav,
> >
> > What happens here if the mount namespace of the process that originally
> > created the sock_xprt no longer exists? Should we care about that case?
> >
> 
> Hi, Trond.
> Looks like this is not a problem, because process fs->root->mnt usage counter 
> was increased on transport creation.
> IOW, transport holds current root and thus mount namespace can't disappear.
> 

That pins the root struct vfsmount, but it doesn't pin the actual
process mount namespace. If the process is dead, then it is quite
possible that the struct mnt_namespace is gone, in which case while you
are pinning the root (i.e. '/'), submounts such as '/var' may be gone.

OTOH, I suppose that you can argue that if the mnt_namespace is gone,
then rpcbind can't be listening on /var/run/rpcbind.sock and so you are
screwed anyway...

OK.... Please just resend the patch, Ccing Al Viro and Christoph so that
we can get their opinion.

-- 
Trond Myklebust
Linux NFS client maintainer

NetApp
Trond.Myklebust at netapp.com
www.netapp.com