[PATCH] knfsd: use new lock for svc_sock deferred list
Greg Banks [Mon, 2 Oct 2006 09:17:55 +0000 (02:17 -0700)]
Protect the svc_sock->sk_deferred list with a new lock svc_sock->sk_defer_lock
instead of svc_serv->sv_lock.  Using the more fine-grained lock reduces the
number of places we need to take the svc_serv lock.

Signed-off-by: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

include/linux/sunrpc/svcsock.h
net/sunrpc/svcsock.c

index 17cb834..7766a10 100644 (file)
@@ -36,6 +36,7 @@ struct svc_sock {
 
        int                     sk_reserved;    /* space on outq that is reserved */
 
+       spinlock_t              sk_defer_lock;  /* protects sk_deferred */
        struct list_head        sk_deferred;    /* deferred requests that need to
                                                 * be revisted */
        struct mutex            sk_mutex;       /* to serialize sending data */
index d836031..bdb5c28 100644 (file)
@@ -47,6 +47,7 @@
 /* SMP locking strategy:
  *
  *     svc_serv->sv_lock protects most stuff for that service.
+ *     svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list
  *
  *     Some flags can be set to certain values at any time
  *     providing that certain rules are followed:
@@ -1416,6 +1417,7 @@ svc_setup_socket(struct svc_serv *serv, struct socket *sock,
        svsk->sk_server = serv;
        atomic_set(&svsk->sk_inuse, 0);
        svsk->sk_lastrecv = get_seconds();
+       spin_lock_init(&svsk->sk_defer_lock);
        INIT_LIST_HEAD(&svsk->sk_deferred);
        INIT_LIST_HEAD(&svsk->sk_ready);
        mutex_init(&svsk->sk_mutex);
@@ -1594,7 +1596,6 @@ svc_makesock(struct svc_serv *serv, int protocol, unsigned short port)
 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
 {
        struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
-       struct svc_serv *serv = dreq->owner;
        struct svc_sock *svsk;
 
        if (too_many) {
@@ -1605,9 +1606,9 @@ static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
        dprintk("revisit queued\n");
        svsk = dr->svsk;
        dr->svsk = NULL;
-       spin_lock_bh(&serv->sv_lock);
+       spin_lock_bh(&svsk->sk_defer_lock);
        list_add(&dr->handle.recent, &svsk->sk_deferred);
-       spin_unlock_bh(&serv->sv_lock);
+       spin_unlock_bh(&svsk->sk_defer_lock);
        set_bit(SK_DEFERRED, &svsk->sk_flags);
        svc_sock_enqueue(svsk);
        svc_sock_put(svsk);
@@ -1667,11 +1668,10 @@ static int svc_deferred_recv(struct svc_rqst *rqstp)
 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
 {
        struct svc_deferred_req *dr = NULL;
-       struct svc_serv *serv = svsk->sk_server;
        
        if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
                return NULL;
-       spin_lock_bh(&serv->sv_lock);
+       spin_lock_bh(&svsk->sk_defer_lock);
        clear_bit(SK_DEFERRED, &svsk->sk_flags);
        if (!list_empty(&svsk->sk_deferred)) {
                dr = list_entry(svsk->sk_deferred.next,
@@ -1680,6 +1680,6 @@ static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
                list_del_init(&dr->handle.recent);
                set_bit(SK_DEFERRED, &svsk->sk_flags);
        }
-       spin_unlock_bh(&serv->sv_lock);
+       spin_unlock_bh(&svsk->sk_defer_lock);
        return dr;
 }