headers: utsname.h redux
[linux-2.6.git] / net / sunrpc / svcsock.c
index 2016d9c..ccc5e83 100644 (file)
@@ -38,6 +38,7 @@
 #include <net/checksum.h>
 #include <net/ip.h>
 #include <net/ipv6.h>
+#include <net/tcp.h>
 #include <net/tcp_states.h>
 #include <asm/uaccess.h>
 #include <asm/ioctls.h>
 #include <linux/sunrpc/types.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/sunrpc/xdr.h>
+#include <linux/sunrpc/msg_prot.h>
 #include <linux/sunrpc/svcsock.h>
 #include <linux/sunrpc/stats.h>
-
-/* SMP locking strategy:
- *
- *     svc_pool->sp_lock protects most of the fields of that pool.
- *     svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
- *     when both need to be taken (rare), svc_serv->sv_lock is first.
- *     BKL protects svc_serv->sv_nrthread.
- *     svc_sock->sk_lock protects the svc_sock->sk_deferred list
- *             and the ->sk_info_authunix cache.
- *     svc_sock->sk_xprt.xpt_flags.XPT_BUSY prevents a svc_sock being
- *     enqueued multiply.
- *
- *     Some flags can be set to certain values at any time
- *     providing that certain rules are followed:
- *
- *     XPT_CONN, XPT_DATA, can be set or cleared at any time.
- *             after a set, svc_xprt_enqueue must be called.
- *             after a clear, the socket must be read/accepted
- *              if this succeeds, it must be set again.
- *     XPT_CLOSE can set at any time. It is never cleared.
- *      xpt_ref contains a bias of '1' until XPT_DEAD is set.
- *             so when xprt_ref hits zero, we know the transport is dead
- *             and no-one is using it.
- *      XPT_DEAD can only be set while XPT_BUSY is held which ensures
- *             no other thread will be using the socket or will try to
- *            set XPT_DEAD.
- *
- */
+#include <linux/sunrpc/xprt.h>
 
 #define RPCDBG_FACILITY        RPCDBG_SVCXPRT
 
 
 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
                                         int *errp, int flags);
-static void            svc_delete_xprt(struct svc_xprt *xprt);
 static void            svc_udp_data_ready(struct sock *, int);
 static int             svc_udp_recvfrom(struct svc_rqst *);
 static int             svc_udp_sendto(struct svc_rqst *);
-static void            svc_close_xprt(struct svc_xprt *xprt);
 static void            svc_sock_detach(struct svc_xprt *);
+static void            svc_tcp_sock_detach(struct svc_xprt *);
 static void            svc_sock_free(struct svc_xprt *);
 
-static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
-static int svc_deferred_recv(struct svc_rqst *rqstp);
-static struct cache_deferred_req *svc_defer(struct cache_req *req);
 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
                                          struct sockaddr *, int, int);
-
-/* apparently the "standard" is that clients close
- * idle connections after 5 minutes, servers after
- * 6 minutes
- *   http://www.connectathon.org/talks96/nfstcp.pdf
- */
-static int svc_conn_age_period = 6*60;
-
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 static struct lock_class_key svc_key[2];
 static struct lock_class_key svc_slock_key[2];
 
-static inline void svc_reclassify_socket(struct socket *sock)
+static void svc_reclassify_socket(struct socket *sock)
 {
        struct sock *sk = sock->sk;
        BUG_ON(sock_owned_by_user(sk));
        switch (sk->sk_family) {
        case AF_INET:
                sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
-                   &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]);
+                                             &svc_slock_key[0],
+                                             "sk_xprt.xpt_lock-AF_INET-NFSD",
+                                             &svc_key[0]);
                break;
 
        case AF_INET6:
                sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
-                   &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]);
+                                             &svc_slock_key[1],
+                                             "sk_xprt.xpt_lock-AF_INET6-NFSD",
+                                             &svc_key[1]);
                break;
 
        default:
@@ -126,299 +93,25 @@ static inline void svc_reclassify_socket(struct socket *sock)
        }
 }
 #else
-static inline void svc_reclassify_socket(struct socket *sock)
+static void svc_reclassify_socket(struct socket *sock)
 {
 }
 #endif
 
-static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len)
-{
-       switch (addr->sa_family) {
-       case AF_INET:
-               snprintf(buf, len, "%u.%u.%u.%u, port=%u",
-                       NIPQUAD(((struct sockaddr_in *) addr)->sin_addr),
-                       ntohs(((struct sockaddr_in *) addr)->sin_port));
-               break;
-
-       case AF_INET6:
-               snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u",
-                       NIP6(((struct sockaddr_in6 *) addr)->sin6_addr),
-                       ntohs(((struct sockaddr_in6 *) addr)->sin6_port));
-               break;
-
-       default:
-               snprintf(buf, len, "unknown address type: %d", addr->sa_family);
-               break;
-       }
-       return buf;
-}
-
-/**
- * svc_print_addr - Format rq_addr field for printing
- * @rqstp: svc_rqst struct containing address to print
- * @buf: target buffer for formatted address
- * @len: length of target buffer
- *
- */
-char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
-{
-       return __svc_print_addr(svc_addr(rqstp), buf, len);
-}
-EXPORT_SYMBOL_GPL(svc_print_addr);
-
-/*
- * Queue up an idle server thread.  Must have pool->sp_lock held.
- * Note: this is really a stack rather than a queue, so that we only
- * use as many different threads as we need, and the rest don't pollute
- * the cache.
- */
-static inline void
-svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
-{
-       list_add(&rqstp->rq_list, &pool->sp_threads);
-}
-
-/*
- * Dequeue an nfsd thread.  Must have pool->sp_lock held.
- */
-static inline void
-svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
-{
-       list_del(&rqstp->rq_list);
-}
-
 /*
  * Release an skbuff after use
  */
 static void svc_release_skb(struct svc_rqst *rqstp)
 {
        struct sk_buff *skb = rqstp->rq_xprt_ctxt;
-       struct svc_deferred_req *dr = rqstp->rq_deferred;
 
        if (skb) {
+               struct svc_sock *svsk =
+                       container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
                rqstp->rq_xprt_ctxt = NULL;
 
                dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
-               skb_free_datagram(rqstp->rq_sock->sk_sk, skb);
-       }
-       if (dr) {
-               rqstp->rq_deferred = NULL;
-               kfree(dr);
-       }
-}
-
-/*
- * Queue up a socket with data pending. If there are idle nfsd
- * processes, wake 'em up.
- *
- */
-void svc_xprt_enqueue(struct svc_xprt *xprt)
-{
-       struct svc_serv *serv = xprt->xpt_server;
-       struct svc_pool *pool;
-       struct svc_rqst *rqstp;
-       int cpu;
-
-       if (!(xprt->xpt_flags &
-             ((1<<XPT_CONN)|(1<<XPT_DATA)|(1<<XPT_CLOSE)|(1<<XPT_DEFERRED))))
-               return;
-       if (test_bit(XPT_DEAD, &xprt->xpt_flags))
-               return;
-
-       cpu = get_cpu();
-       pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
-       put_cpu();
-
-       spin_lock_bh(&pool->sp_lock);
-
-       if (!list_empty(&pool->sp_threads) &&
-           !list_empty(&pool->sp_sockets))
-               printk(KERN_ERR
-                      "svc_xprt_enqueue: "
-                      "threads and transports both waiting??\n");
-
-       if (test_bit(XPT_DEAD, &xprt->xpt_flags)) {
-               /* Don't enqueue dead sockets */
-               dprintk("svc: transport %p is dead, not enqueued\n", xprt);
-               goto out_unlock;
-       }
-
-       /* Mark socket as busy. It will remain in this state until the
-        * server has processed all pending data and put the socket back
-        * on the idle list.  We update XPT_BUSY atomically because
-        * it also guards against trying to enqueue the svc_sock twice.
-        */
-       if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
-               /* Don't enqueue socket while already enqueued */
-               dprintk("svc: transport %p busy, not enqueued\n", xprt);
-               goto out_unlock;
-       }
-       BUG_ON(xprt->xpt_pool != NULL);
-       xprt->xpt_pool = pool;
-
-       /* Handle pending connection */
-       if (test_bit(XPT_CONN, &xprt->xpt_flags))
-               goto process;
-
-       /* Handle close in-progress */
-       if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
-               goto process;
-
-       /* Check if we have space to reply to a request */
-       if (!xprt->xpt_ops->xpo_has_wspace(xprt)) {
-               /* Don't enqueue while not enough space for reply */
-               dprintk("svc: no write space, transport %p  not enqueued\n",
-                       xprt);
-               xprt->xpt_pool = NULL;
-               clear_bit(XPT_BUSY, &xprt->xpt_flags);
-               goto out_unlock;
-       }
-
- process:
-       if (!list_empty(&pool->sp_threads)) {
-               rqstp = list_entry(pool->sp_threads.next,
-                                  struct svc_rqst,
-                                  rq_list);
-               dprintk("svc: transport %p served by daemon %p\n",
-                       xprt, rqstp);
-               svc_thread_dequeue(pool, rqstp);
-               if (rqstp->rq_xprt)
-                       printk(KERN_ERR
-                               "svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
-                               rqstp, rqstp->rq_xprt);
-               rqstp->rq_xprt = xprt;
-               svc_xprt_get(xprt);
-               rqstp->rq_reserved = serv->sv_max_mesg;
-               atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
-               BUG_ON(xprt->xpt_pool != pool);
-               wake_up(&rqstp->rq_wait);
-       } else {
-               dprintk("svc: transport %p put into queue\n", xprt);
-               list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
-               BUG_ON(xprt->xpt_pool != pool);
-       }
-
-out_unlock:
-       spin_unlock_bh(&pool->sp_lock);
-}
-EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
-
-/*
- * Dequeue the first socket.  Must be called with the pool->sp_lock held.
- */
-static inline struct svc_sock *
-svc_sock_dequeue(struct svc_pool *pool)
-{
-       struct svc_sock *svsk;
-
-       if (list_empty(&pool->sp_sockets))
-               return NULL;
-
-       svsk = list_entry(pool->sp_sockets.next,
-                         struct svc_sock, sk_xprt.xpt_ready);
-       list_del_init(&svsk->sk_xprt.xpt_ready);
-
-       dprintk("svc: socket %p dequeued, inuse=%d\n",
-               svsk->sk_sk, atomic_read(&svsk->sk_xprt.xpt_ref.refcount));
-
-       return svsk;
-}
-
-/*
- * Having read something from a socket, check whether it
- * needs to be re-enqueued.
- * Note: XPT_DATA only gets cleared when a read-attempt finds
- * no (or insufficient) data.
- */
-static inline void
-svc_sock_received(struct svc_sock *svsk)
-{
-       svsk->sk_xprt.xpt_pool = NULL;
-       clear_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags);
-       svc_xprt_enqueue(&svsk->sk_xprt);
-}
-
-
-/**
- * svc_reserve - change the space reserved for the reply to a request.
- * @rqstp:  The request in question
- * @space: new max space to reserve
- *
- * Each request reserves some space on the output queue of the socket
- * to make sure the reply fits.  This function reduces that reserved
- * space to be the amount of space used already, plus @space.
- *
- */
-void svc_reserve(struct svc_rqst *rqstp, int space)
-{
-       space += rqstp->rq_res.head[0].iov_len;
-
-       if (space < rqstp->rq_reserved) {
-               struct svc_xprt *xprt = rqstp->rq_xprt;
-               atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
-               rqstp->rq_reserved = space;
-
-               svc_xprt_enqueue(xprt);
-       }
-}
-
-static void
-svc_sock_release(struct svc_rqst *rqstp)
-{
-       struct svc_sock *svsk = rqstp->rq_sock;
-
-       rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
-
-       svc_free_res_pages(rqstp);
-       rqstp->rq_res.page_len = 0;
-       rqstp->rq_res.page_base = 0;
-
-
-       /* Reset response buffer and release
-        * the reservation.
-        * But first, check that enough space was reserved
-        * for the reply, otherwise we have a bug!
-        */
-       if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
-               printk(KERN_ERR "RPC request reserved %d but used %d\n",
-                      rqstp->rq_reserved,
-                      rqstp->rq_res.len);
-
-       rqstp->rq_res.head[0].iov_len = 0;
-       svc_reserve(rqstp, 0);
-       rqstp->rq_sock = NULL;
-
-       svc_xprt_put(&svsk->sk_xprt);
-}
-
-/*
- * External function to wake up a server waiting for data
- * This really only makes sense for services like lockd
- * which have exactly one thread anyway.
- */
-void
-svc_wake_up(struct svc_serv *serv)
-{
-       struct svc_rqst *rqstp;
-       unsigned int i;
-       struct svc_pool *pool;
-
-       for (i = 0; i < serv->sv_nrpools; i++) {
-               pool = &serv->sv_pools[i];
-
-               spin_lock_bh(&pool->sp_lock);
-               if (!list_empty(&pool->sp_threads)) {
-                       rqstp = list_entry(pool->sp_threads.next,
-                                          struct svc_rqst,
-                                          rq_list);
-                       dprintk("svc: daemon %p woken up.\n", rqstp);
-                       /*
-                       svc_thread_dequeue(pool, rqstp);
-                       rqstp->rq_sock = NULL;
-                        */
-                       wake_up(&rqstp->rq_wait);
-               }
-               spin_unlock_bh(&pool->sp_lock);
+               skb_free_datagram(svsk->sk_sk, skb);
        }
 }
 
@@ -431,7 +124,9 @@ union svc_pktinfo_u {
 
 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
 {
-       switch (rqstp->rq_sock->sk_sk->sk_family) {
+       struct svc_sock *svsk =
+               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
+       switch (svsk->sk_sk->sk_family) {
        case AF_INET: {
                        struct in_pktinfo *pki = CMSG_DATA(cmh);
 
@@ -459,49 +154,27 @@ static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
 }
 
 /*
- * Generic sendto routine
+ * send routine intended to be shared by the fore- and back-channel
  */
-static int
-svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
+int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
+                   struct page *headpage, unsigned long headoffset,
+                   struct page *tailpage, unsigned long tailoffset)
 {
-       struct svc_sock *svsk = rqstp->rq_sock;
-       struct socket   *sock = svsk->sk_sock;
-       int             slen;
-       union {
-               struct cmsghdr  hdr;
-               long            all[SVC_PKTINFO_SPACE / sizeof(long)];
-       } buffer;
-       struct cmsghdr *cmh = &buffer.hdr;
-       int             len = 0;
        int             result;
        int             size;
        struct page     **ppage = xdr->pages;
        size_t          base = xdr->page_base;
        unsigned int    pglen = xdr->page_len;
        unsigned int    flags = MSG_MORE;
-       char            buf[RPC_MAX_ADDRBUFLEN];
+       int             slen;
+       int             len = 0;
 
        slen = xdr->len;
 
-       if (rqstp->rq_prot == IPPROTO_UDP) {
-               struct msghdr msg = {
-                       .msg_name       = &rqstp->rq_addr,
-                       .msg_namelen    = rqstp->rq_addrlen,
-                       .msg_control    = cmh,
-                       .msg_controllen = sizeof(buffer),
-                       .msg_flags      = MSG_MORE,
-               };
-
-               svc_set_cmsg_data(rqstp, cmh);
-
-               if (sock_sendmsg(sock, &msg, 0) < 0)
-                       goto out;
-       }
-
        /* send head */
        if (slen == xdr->head[0].iov_len)
                flags = 0;
-       len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
+       len = kernel_sendpage(sock, headpage, headoffset,
                                  xdr->head[0].iov_len, flags);
        if (len != xdr->head[0].iov_len)
                goto out;
@@ -525,19 +198,61 @@ svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
                base = 0;
                ppage++;
        }
+
        /* send tail */
        if (xdr->tail[0].iov_len) {
-               result = kernel_sendpage(sock, rqstp->rq_respages[0],
-                                            ((unsigned long)xdr->tail[0].iov_base)
-                                               & (PAGE_SIZE-1),
-                                            xdr->tail[0].iov_len, 0);
-
+               result = kernel_sendpage(sock, tailpage, tailoffset,
+                                  xdr->tail[0].iov_len, 0);
                if (result > 0)
                        len += result;
        }
+
+out:
+       return len;
+}
+
+
+/*
+ * Generic sendto routine
+ */
+static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
+{
+       struct svc_sock *svsk =
+               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
+       struct socket   *sock = svsk->sk_sock;
+       union {
+               struct cmsghdr  hdr;
+               long            all[SVC_PKTINFO_SPACE / sizeof(long)];
+       } buffer;
+       struct cmsghdr *cmh = &buffer.hdr;
+       int             len = 0;
+       unsigned long tailoff;
+       unsigned long headoff;
+       RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
+
+       if (rqstp->rq_prot == IPPROTO_UDP) {
+               struct msghdr msg = {
+                       .msg_name       = &rqstp->rq_addr,
+                       .msg_namelen    = rqstp->rq_addrlen,
+                       .msg_control    = cmh,
+                       .msg_controllen = sizeof(buffer),
+                       .msg_flags      = MSG_MORE,
+               };
+
+               svc_set_cmsg_data(rqstp, cmh);
+
+               if (sock_sendmsg(sock, &msg, 0) < 0)
+                       goto out;
+       }
+
+       tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
+       headoff = 0;
+       len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
+                              rqstp->rq_respages[0], tailoff);
+
 out:
        dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
-               rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len,
+               svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
                xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
 
        return len;
@@ -546,42 +261,76 @@ out:
 /*
  * Report socket names for nfsdfs
  */
-static int one_sock_name(char *buf, struct svc_sock *svsk)
+static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
 {
+       const struct sock *sk = svsk->sk_sk;
+       const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
+                                                       "udp" : "tcp";
        int len;
 
-       switch(svsk->sk_sk->sk_family) {
-       case AF_INET:
-               len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
-                             svsk->sk_sk->sk_protocol==IPPROTO_UDP?
-                             "udp" : "tcp",
-                             NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
-                             inet_sk(svsk->sk_sk)->num);
+       switch (sk->sk_family) {
+       case PF_INET:
+               len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
+                               proto_name,
+                               &inet_sk(sk)->rcv_saddr,
+                               inet_sk(sk)->num);
+               break;
+       case PF_INET6:
+               len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
+                               proto_name,
+                               &inet6_sk(sk)->rcv_saddr,
+                               inet_sk(sk)->num);
                break;
        default:
-               len = sprintf(buf, "*unknown-%d*\n",
-                              svsk->sk_sk->sk_family);
+               len = snprintf(buf, remaining, "*unknown-%d*\n",
+                               sk->sk_family);
+       }
+
+       if (len >= remaining) {
+               *buf = '\0';
+               return -ENAMETOOLONG;
        }
        return len;
 }
 
-int
-svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
+/**
+ * svc_sock_names - construct a list of listener names in a string
+ * @serv: pointer to RPC service
+ * @buf: pointer to a buffer to fill in with socket names
+ * @buflen: size of the buffer to be filled
+ * @toclose: pointer to '\0'-terminated C string containing the name
+ *             of a listener to be closed
+ *
+ * Fills in @buf with a '\n'-separated list of names of listener
+ * sockets.  If @toclose is not NULL, the socket named by @toclose
+ * is closed, and is not included in the output list.
+ *
+ * Returns positive length of the socket name string, or a negative
+ * errno value on error.
+ */
+int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
+                  const char *toclose)
 {
        struct svc_sock *svsk, *closesk = NULL;
        int len = 0;
 
        if (!serv)
                return 0;
+
        spin_lock_bh(&serv->sv_lock);
        list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
-               int onelen = one_sock_name(buf+len, svsk);
-               if (toclose && strcmp(toclose, buf+len) == 0)
+               int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
+               if (onelen < 0) {
+                       len = onelen;
+                       break;
+               }
+               if (toclose && strcmp(toclose, buf + len) == 0)
                        closesk = svsk;
                else
                        len += onelen;
        }
        spin_unlock_bh(&serv->sv_lock);
+
        if (closesk)
                /* Should unregister with portmap, but you cannot
                 * unregister just one protocol...
@@ -591,13 +340,12 @@ svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
                return -ENOENT;
        return len;
 }
-EXPORT_SYMBOL(svc_sock_names);
+EXPORT_SYMBOL_GPL(svc_sock_names);
 
 /*
  * Check input queue length
  */
-static int
-svc_recv_available(struct svc_sock *svsk)
+static int svc_recv_available(struct svc_sock *svsk)
 {
        struct socket   *sock = svsk->sk_sock;
        int             avail, err;
@@ -610,48 +358,31 @@ svc_recv_available(struct svc_sock *svsk)
 /*
  * Generic recvfrom routine.
  */
-static int
-svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen)
+static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
+                       int buflen)
 {
-       struct svc_sock *svsk = rqstp->rq_sock;
+       struct svc_sock *svsk =
+               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
        struct msghdr msg = {
                .msg_flags      = MSG_DONTWAIT,
        };
-       struct sockaddr *sin;
        int len;
 
+       rqstp->rq_xprt_hlen = 0;
+
        len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
                                msg.msg_flags);
 
-       /* sock_recvmsg doesn't fill in the name/namelen, so we must..
-        */
-       memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen);
-       rqstp->rq_addrlen = svsk->sk_remotelen;
-
-       /* Destination address in request is needed for binding the
-        * source address in RPC callbacks later.
-        */
-       sin = (struct sockaddr *)&svsk->sk_local;
-       switch (sin->sa_family) {
-       case AF_INET:
-               rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
-               break;
-       case AF_INET6:
-               rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
-               break;
-       }
-
        dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
                svsk, iov[0].iov_base, iov[0].iov_len, len);
-
        return len;
 }
 
 /*
  * Set socket snd and rcv buffer lengths
  */
-static inline void
-svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
+static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
+                               unsigned int rcv)
 {
 #if 0
        mm_segment_t    oldfs;
@@ -670,14 +401,14 @@ svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
        sock->sk->sk_sndbuf = snd * 2;
        sock->sk->sk_rcvbuf = rcv * 2;
        sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
+       sock->sk->sk_write_space(sock->sk);
        release_sock(sock->sk);
 #endif
 }
 /*
  * INET callback when data has been received on the socket.
  */
-static void
-svc_udp_data_ready(struct sock *sk, int count)
+static void svc_udp_data_ready(struct sock *sk, int count)
 {
        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 
@@ -695,8 +426,7 @@ svc_udp_data_ready(struct sock *sk, int count)
 /*
  * INET callback when space is newly available on the socket.
  */
-static void
-svc_write_space(struct sock *sk)
+static void svc_write_space(struct sock *sk)
 {
        struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
 
@@ -713,30 +443,68 @@ svc_write_space(struct sock *sk)
        }
 }
 
-static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp,
-                                           struct cmsghdr *cmh)
+static void svc_tcp_write_space(struct sock *sk)
 {
-       switch (rqstp->rq_sock->sk_sk->sk_family) {
-       case AF_INET: {
-               struct in_pktinfo *pki = CMSG_DATA(cmh);
-               rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
-               break;
-               }
-       case AF_INET6: {
-               struct in6_pktinfo *pki = CMSG_DATA(cmh);
-               ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
-               break;
-               }
+       struct socket *sock = sk->sk_socket;
+
+       if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
+               clear_bit(SOCK_NOSPACE, &sock->flags);
+       svc_write_space(sk);
+}
+
+/*
+ * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
+ */
+static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
+                                    struct cmsghdr *cmh)
+{
+       struct in_pktinfo *pki = CMSG_DATA(cmh);
+       if (cmh->cmsg_type != IP_PKTINFO)
+               return 0;
+       rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
+       return 1;
+}
+
+/*
+ * See net/ipv6/datagram.c : datagram_recv_ctl
+ */
+static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
+                                    struct cmsghdr *cmh)
+{
+       struct in6_pktinfo *pki = CMSG_DATA(cmh);
+       if (cmh->cmsg_type != IPV6_PKTINFO)
+               return 0;
+       ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
+       return 1;
+}
+
+/*
+ * Copy the UDP datagram's destination address to the rqstp structure.
+ * The 'destination' address in this case is the address to which the
+ * peer sent the datagram, i.e. our local address. For multihomed
+ * hosts, this can change from msg to msg. Note that only the IP
+ * address changes, the port number should remain the same.
+ */
+static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
+                                   struct cmsghdr *cmh)
+{
+       switch (cmh->cmsg_level) {
+       case SOL_IP:
+               return svc_udp_get_dest_address4(rqstp, cmh);
+       case SOL_IPV6:
+               return svc_udp_get_dest_address6(rqstp, cmh);
        }
+
+       return 0;
 }
 
 /*
  * Receive a datagram from a UDP socket.
  */
-static int
-svc_udp_recvfrom(struct svc_rqst *rqstp)
+static int svc_udp_recvfrom(struct svc_rqst *rqstp)
 {
-       struct svc_sock *svsk = rqstp->rq_sock;
+       struct svc_sock *svsk =
+               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
        struct sk_buff  *skb;
        union {
@@ -744,13 +512,14 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
                long            all[SVC_PKTINFO_SPACE / sizeof(long)];
        } buffer;
        struct cmsghdr *cmh = &buffer.hdr;
-       int             err, len;
        struct msghdr msg = {
                .msg_name = svc_addr(rqstp),
                .msg_control = cmh,
                .msg_controllen = sizeof(buffer),
                .msg_flags = MSG_DONTWAIT,
        };
+       size_t len;
+       int err;
 
        if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
            /* udp sockets need large rcvbuf as all pending
@@ -765,11 +534,6 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
                                (serv->sv_nrthreads+3) * serv->sv_max_mesg,
                                (serv->sv_nrthreads+3) * serv->sv_max_mesg);
 
-       if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
-               svc_sock_received(svsk);
-               return svc_deferred_recv(rqstp);
-       }
-
        clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
        skb = NULL;
        err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
@@ -783,10 +547,13 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
                        dprintk("svc: recvfrom returned error %d\n", -err);
                        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
                }
-               svc_sock_received(svsk);
+               svc_xprt_received(&svsk->sk_xprt);
                return -EAGAIN;
        }
-       rqstp->rq_addrlen = sizeof(rqstp->rq_addr);
+       len = svc_addr_len(svc_addr(rqstp));
+       if (len == 0)
+               return -EAFNOSUPPORT;
+       rqstp->rq_addrlen = len;
        if (skb->tstamp.tv64 == 0) {
                skb->tstamp = ktime_get_real();
                /* Don't enable netstamp, sunrpc doesn't
@@ -798,23 +565,22 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
        /*
         * Maybe more packets - kick another thread ASAP.
         */
-       svc_sock_received(svsk);
+       svc_xprt_received(&svsk->sk_xprt);
 
        len  = skb->len - sizeof(struct udphdr);
        rqstp->rq_arg.len = len;
 
        rqstp->rq_prot = IPPROTO_UDP;
 
-       if (cmh->cmsg_level != IPPROTO_IP ||
-           cmh->cmsg_type != IP_PKTINFO) {
+       if (!svc_udp_get_dest_address(rqstp, cmh)) {
                if (net_ratelimit())
-                       printk("rpcsvc: received unknown control message:"
-                              "%d/%d\n",
-                              cmh->cmsg_level, cmh->cmsg_type);
+                       printk(KERN_WARNING
+                               "svc: received unknown control message %d/%d; "
+                               "dropping RPC reply datagram\n",
+                                       cmh->cmsg_level, cmh->cmsg_type);
                skb_free_datagram(svsk->sk_sk, skb);
                return 0;
        }
-       svc_udp_get_dest_address(rqstp, cmh);
 
        if (skb_is_nonlinear(skb)) {
                /* we have to copy */
@@ -829,7 +595,8 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
                skb_free_datagram(svsk->sk_sk, skb);
        } else {
                /* we can use it in-place */
-               rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
+               rqstp->rq_arg.head[0].iov_base = skb->data +
+                       sizeof(struct udphdr);
                rqstp->rq_arg.head[0].iov_len = len;
                if (skb_checksum_complete(skb)) {
                        skb_free_datagram(svsk->sk_sk, skb);
@@ -924,10 +691,10 @@ static struct svc_xprt_class svc_udp_class = {
 
 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
 {
-       int one = 1;
-       mm_segment_t oldfs;
+       int err, level, optname, one = 1;
 
        svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
+       clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
        svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
        svsk->sk_sk->sk_write_space = svc_write_space;
 
@@ -939,23 +706,33 @@ static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
                            3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
                            3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
 
-       set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* might have come in before data_ready set up */
+       /* data might have come in before data_ready set up */
+       set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
        set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
 
-       oldfs = get_fs();
-       set_fs(KERNEL_DS);
        /* make sure we get destination address info */
-       svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
-                                      (char __user *)&one, sizeof(one));
-       set_fs(oldfs);
+       switch (svsk->sk_sk->sk_family) {
+       case AF_INET:
+               level = SOL_IP;
+               optname = IP_PKTINFO;
+               break;
+       case AF_INET6:
+               level = SOL_IPV6;
+               optname = IPV6_RECVPKTINFO;
+               break;
+       default:
+               BUG();
+       }
+       err = kernel_setsockopt(svsk->sk_sock, level, optname,
+                                       (char *)&one, sizeof(one));
+       dprintk("svc: kernel_setsockopt returned %d\n", err);
 }
 
 /*
  * A data_ready event on a listening socket means there's a connection
  * pending. Do not use state_change as a substitute for it.
  */
-static void
-svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
+static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
 {
        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 
@@ -987,8 +764,7 @@ svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
 /*
  * A state change on a connected socket means it's dying or dead.
  */
-static void
-svc_tcp_state_change(struct sock *sk)
+static void svc_tcp_state_change(struct sock *sk)
 {
        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 
@@ -1005,8 +781,7 @@ svc_tcp_state_change(struct sock *sk)
                wake_up_interruptible_all(sk->sk_sleep);
 }
 
-static void
-svc_tcp_data_ready(struct sock *sk, int count)
+static void svc_tcp_data_ready(struct sock *sk, int count)
 {
        struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 
@@ -1020,20 +795,6 @@ svc_tcp_data_ready(struct sock *sk, int count)
                wake_up_interruptible(sk->sk_sleep);
 }
 
-static inline int svc_port_is_privileged(struct sockaddr *sin)
-{
-       switch (sin->sa_family) {
-       case AF_INET:
-               return ntohs(((struct sockaddr_in *)sin)->sin_port)
-                       < PROT_SOCK;
-       case AF_INET6:
-               return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
-                       < PROT_SOCK;
-       default:
-               return 0;
-       }
-}
-
 /*
  * Accept a TCP connection
  */
@@ -1047,7 +808,7 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
        struct socket   *newsock;
        struct svc_sock *newsvsk;
        int             err, slen;
-       char            buf[RPC_MAX_ADDRBUFLEN];
+       RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 
        dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
        if (!sock)
@@ -1095,16 +856,13 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
        if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
                                 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
                goto failed;
-       memcpy(&newsvsk->sk_remote, sin, slen);
-       newsvsk->sk_remotelen = slen;
+       svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
        err = kernel_getsockname(newsock, sin, &slen);
        if (unlikely(err < 0)) {
                dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
                slen = offsetof(struct sockaddr, sa_data);
        }
-       memcpy(&newsvsk->sk_local, sin, slen);
-
-       svc_sock_received(newsvsk);
+       svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
 
        if (serv->sv_stats)
                serv->sv_stats->nettcpconn++;
@@ -1117,26 +875,15 @@ failed:
 }
 
 /*
- * Receive data from a TCP socket.
+ * Receive data.
+ * If we haven't gotten the record length yet, get the next four bytes.
+ * Otherwise try to gobble up as much as possible up to the complete
+ * record length.
  */
-static int
-svc_tcp_recvfrom(struct svc_rqst *rqstp)
+static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
 {
-       struct svc_sock *svsk = rqstp->rq_sock;
        struct svc_serv *serv = svsk->sk_xprt.xpt_server;
-       int             len;
-       struct kvec *vec;
-       int pnum, vlen;
-
-       dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
-               svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
-               test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
-               test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
-
-       if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
-               svc_sock_received(svsk);
-               return svc_deferred_recv(rqstp);
-       }
+       int len;
 
        if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
                /* sndbuf needs to have room for one request
@@ -1157,12 +904,8 @@ svc_tcp_recvfrom(struct svc_rqst *rqstp)
 
        clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 
-       /* Receive data. If we haven't got the record length yet, get
-        * the next four bytes. Otherwise try to gobble up as much as
-        * possible up to the complete record length.
-        */
-       if (svsk->sk_tcplen < 4) {
-               unsigned long   want = 4 - svsk->sk_tcplen;
+       if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
+               int             want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
                struct kvec     iov;
 
                iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
@@ -1172,32 +915,32 @@ svc_tcp_recvfrom(struct svc_rqst *rqstp)
                svsk->sk_tcplen += len;
 
                if (len < want) {
-                       dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
-                               len, want);
-                       svc_sock_received(svsk);
-                       return -EAGAIN; /* record header not complete */
+                       dprintk("svc: short recvfrom while reading record "
+                               "length (%d of %d)\n", len, want);
+                       svc_xprt_received(&svsk->sk_xprt);
+                       goto err_again; /* record header not complete */
                }
 
                svsk->sk_reclen = ntohl(svsk->sk_reclen);
-               if (!(svsk->sk_reclen & 0x80000000)) {
+               if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
                        /* FIXME: technically, a record can be fragmented,
                         *  and non-terminal fragments will not have the top
                         *  bit set in the fragment length header.
                         *  But apparently no known nfs clients send fragmented
                         *  records. */
                        if (net_ratelimit())
-                               printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
-                                      " (non-terminal)\n",
-                                      (unsigned long) svsk->sk_reclen);
+                               printk(KERN_NOTICE "RPC: multiple fragments "
+                                       "per record not supported\n");
                        goto err_delete;
                }
-               svsk->sk_reclen &= 0x7fffffff;
+
+               svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
                dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
                if (svsk->sk_reclen > serv->sv_max_mesg) {
                        if (net_ratelimit())
-                               printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
-                                      " (large)\n",
-                                      (unsigned long) svsk->sk_reclen);
+                               printk(KERN_NOTICE "RPC: "
+                                       "fragment too large: 0x%08lx\n",
+                                       (unsigned long)svsk->sk_reclen);
                        goto err_delete;
                }
        }
@@ -1210,18 +953,122 @@ svc_tcp_recvfrom(struct svc_rqst *rqstp)
        if (len < svsk->sk_reclen) {
                dprintk("svc: incomplete TCP record (%d of %d)\n",
                        len, svsk->sk_reclen);
-               svc_sock_received(svsk);
-               return -EAGAIN; /* record not complete */
+               svc_xprt_received(&svsk->sk_xprt);
+               goto err_again; /* record not complete */
        }
        len = svsk->sk_reclen;
        set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 
+       return len;
+ error:
+       if (len == -EAGAIN) {
+               dprintk("RPC: TCP recv_record got EAGAIN\n");
+               svc_xprt_received(&svsk->sk_xprt);
+       }
+       return len;
+ err_delete:
+       set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
+ err_again:
+       return -EAGAIN;
+}
+
+static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
+                              struct rpc_rqst **reqpp, struct kvec *vec)
+{
+       struct rpc_rqst *req = NULL;
+       u32 *p;
+       u32 xid;
+       u32 calldir;
+       int len;
+
+       len = svc_recvfrom(rqstp, vec, 1, 8);
+       if (len < 0)
+               goto error;
+
+       p = (u32 *)rqstp->rq_arg.head[0].iov_base;
+       xid = *p++;
+       calldir = *p;
+
+       if (calldir == 0) {
+               /* REQUEST is the most common case */
+               vec[0] = rqstp->rq_arg.head[0];
+       } else {
+               /* REPLY */
+               if (svsk->sk_bc_xprt)
+                       req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid);
+
+               if (!req) {
+                       printk(KERN_NOTICE
+                               "%s: Got unrecognized reply: "
+                               "calldir 0x%x sk_bc_xprt %p xid %08x\n",
+                               __func__, ntohl(calldir),
+                               svsk->sk_bc_xprt, xid);
+                       vec[0] = rqstp->rq_arg.head[0];
+                       goto out;
+               }
+
+               memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
+                      sizeof(struct xdr_buf));
+               /* copy the xid and call direction */
+               memcpy(req->rq_private_buf.head[0].iov_base,
+                      rqstp->rq_arg.head[0].iov_base, 8);
+               vec[0] = req->rq_private_buf.head[0];
+       }
+ out:
+       vec[0].iov_base += 8;
+       vec[0].iov_len -= 8;
+       len = svsk->sk_reclen - 8;
+ error:
+       *reqpp = req;
+       return len;
+}
+
+/*
+ * Receive data from a TCP socket.
+ */
+static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
+{
+       struct svc_sock *svsk =
+               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
+       struct svc_serv *serv = svsk->sk_xprt.xpt_server;
+       int             len;
+       struct kvec *vec;
+       int pnum, vlen;
+       struct rpc_rqst *req = NULL;
+
+       dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
+               svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
+               test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
+               test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
+
+       len = svc_tcp_recv_record(svsk, rqstp);
+       if (len < 0)
+               goto error;
+
        vec = rqstp->rq_vec;
        vec[0] = rqstp->rq_arg.head[0];
        vlen = PAGE_SIZE;
+
+       /*
+        * We have enough data for the whole tcp record. Let's try and read the
+        * first 8 bytes to get the xid and the call direction. We can use this
+        * to figure out if this is a call or a reply to a callback. If
+        * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
+        * In that case, don't bother with the calldir and just read the data.
+        * It will be rejected in svc_process.
+        */
+       if (len >= 8) {
+               len = svc_process_calldir(svsk, rqstp, &req, vec);
+               if (len < 0)
+                       goto err_again;
+               vlen -= 8;
+       }
+
        pnum = 1;
        while (vlen < len) {
-               vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
+               vec[pnum].iov_base = (req) ?
+                       page_address(req->rq_private_buf.pages[pnum - 1]) :
+                       page_address(rqstp->rq_pages[pnum]);
                vec[pnum].iov_len = PAGE_SIZE;
                pnum++;
                vlen += PAGE_SIZE;
@@ -1231,8 +1078,18 @@ svc_tcp_recvfrom(struct svc_rqst *rqstp)
        /* Now receive data */
        len = svc_recvfrom(rqstp, vec, pnum, len);
        if (len < 0)
-               goto error;
+               goto err_again;
 
+       /*
+        * Account for the 8 bytes we read earlier
+        */
+       len += 8;
+
+       if (req) {
+               xprt_complete_rqst(req->rq_task, len);
+               len = 0;
+               goto out;
+       }
        dprintk("svc: TCP complete record (%d bytes)\n", len);
        rqstp->rq_arg.len = len;
        rqstp->rq_arg.page_base = 0;
@@ -1246,38 +1103,37 @@ svc_tcp_recvfrom(struct svc_rqst *rqstp)
        rqstp->rq_xprt_ctxt   = NULL;
        rqstp->rq_prot        = IPPROTO_TCP;
 
+out:
        /* Reset TCP read info */
        svsk->sk_reclen = 0;
        svsk->sk_tcplen = 0;
 
-       svc_sock_received(svsk);
+       svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
+       svc_xprt_received(&svsk->sk_xprt);
        if (serv->sv_stats)
                serv->sv_stats->nettcpcnt++;
 
        return len;
 
- err_delete:
-       set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
-       return -EAGAIN;
-
- error:
+err_again:
        if (len == -EAGAIN) {
                dprintk("RPC: TCP recvfrom got EAGAIN\n");
-               svc_sock_received(svsk);
-       } else {
+               svc_xprt_received(&svsk->sk_xprt);
+               return len;
+       }
+error:
+       if (len != -EAGAIN) {
                printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
                       svsk->sk_xprt.xpt_server->sv_name, -len);
-               goto err_delete;
+               set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
        }
-
-       return len;
+       return -EAGAIN;
 }
 
 /*
  * Send out data on TCP socket.
  */
-static int
-svc_tcp_sendto(struct svc_rqst *rqstp)
+static int svc_tcp_sendto(struct svc_rqst *rqstp)
 {
        struct xdr_buf  *xbufp = &rqstp->rq_res;
        int sent;
@@ -1290,16 +1146,18 @@ svc_tcp_sendto(struct svc_rqst *rqstp)
        reclen = htonl(0x80000000|((xbufp->len ) - 4));
        memcpy(xbufp->head[0].iov_base, &reclen, 4);
 
-       if (test_bit(XPT_DEAD, &rqstp->rq_sock->sk_xprt.xpt_flags))
+       if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
                return -ENOTCONN;
 
        sent = svc_sendto(rqstp, &rqstp->rq_res);
        if (sent != xbufp->len) {
-               printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
-                      rqstp->rq_sock->sk_xprt.xpt_server->sv_name,
+               printk(KERN_NOTICE
+                      "rpc-srv/tcp: %s: %s %d when sending %d bytes "
+                      "- shutting down socket\n",
+                      rqstp->rq_xprt->xpt_server->sv_name,
                       (sent<0)?"got error":"sent only",
                       sent, xbufp->len);
-               set_bit(XPT_CLOSE, &rqstp->rq_sock->sk_xprt.xpt_flags);
+               set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
                svc_xprt_enqueue(rqstp->rq_xprt);
                sent = -EAGAIN;
        }
@@ -1319,26 +1177,17 @@ static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
 
 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
 {
-       struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
-       struct svc_serv *serv = svsk->sk_xprt.xpt_server;
+       struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
+       struct svc_serv *serv = svsk->sk_xprt.xpt_server;
        int required;
-       int wspace;
 
-       /*
-        * Set the SOCK_NOSPACE flag before checking the available
-        * sock space.
-        */
+       if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
+               return 1;
+       required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
+       if (sk_stream_wspace(svsk->sk_sk) >= required)
+               return 1;
        set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
-       required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
-       wspace = sk_stream_wspace(svsk->sk_sk);
-
-       if (wspace < sk_stream_min_wspace(svsk->sk_sk))
-               return 0;
-       if (required * 2 > wspace)
-               return 0;
-
-       clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
-       return 1;
+       return 0;
 }
 
 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
@@ -1353,7 +1202,7 @@ static struct svc_xprt_ops svc_tcp_ops = {
        .xpo_recvfrom = svc_tcp_recvfrom,
        .xpo_sendto = svc_tcp_sendto,
        .xpo_release_rqst = svc_release_skb,
-       .xpo_detach = svc_sock_detach,
+       .xpo_detach = svc_tcp_sock_detach,
        .xpo_free = svc_sock_free,
        .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
        .xpo_has_wspace = svc_tcp_has_wspace,
@@ -1382,10 +1231,9 @@ void svc_cleanup_xprt_sock(void)
 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
 {
        struct sock     *sk = svsk->sk_sk;
-       struct tcp_sock *tp = tcp_sk(sk);
 
        svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
-
+       set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
        if (sk->sk_state == TCP_LISTEN) {
                dprintk("setting up TCP socket for listening\n");
                set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
@@ -1395,12 +1243,12 @@ static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
                dprintk("setting up TCP socket for reading\n");
                sk->sk_state_change = svc_tcp_state_change;
                sk->sk_data_ready = svc_tcp_data_ready;
-               sk->sk_write_space = svc_write_space;
+               sk->sk_write_space = svc_tcp_write_space;
 
                svsk->sk_reclen = 0;
                svsk->sk_tcplen = 0;
 
-               tp->nonagle = 1;        /* disable Nagle's algorithm */
+               tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
 
                /* initialise setting must have enough space to
                 * receive and respond to one request.
@@ -1417,8 +1265,7 @@ static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
        }
 }
 
-void
-svc_sock_update_bufs(struct svc_serv *serv)
+void svc_sock_update_bufs(struct svc_serv *serv)
 {
        /*
         * The number of server threads has changed. Update
@@ -1439,282 +1286,7 @@ svc_sock_update_bufs(struct svc_serv *serv)
        }
        spin_unlock_bh(&serv->sv_lock);
 }
-
-/*
- * Make sure that we don't have too many active connections.  If we
- * have, something must be dropped.
- *
- * There's no point in trying to do random drop here for DoS
- * prevention. The NFS clients does 1 reconnect in 15 seconds. An
- * attacker can easily beat that.
- *
- * The only somewhat efficient mechanism would be if drop old
- * connections from the same IP first. But right now we don't even
- * record the client IP in svc_sock.
- */
-static void svc_check_conn_limits(struct svc_serv *serv)
-{
-       if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
-               struct svc_sock *svsk = NULL;
-               spin_lock_bh(&serv->sv_lock);
-               if (!list_empty(&serv->sv_tempsocks)) {
-                       if (net_ratelimit()) {
-                               /* Try to help the admin */
-                               printk(KERN_NOTICE "%s: too many open TCP "
-                                      "sockets, consider increasing the "
-                                      "number of nfsd threads\n",
-                                      serv->sv_name);
-                       }
-                       /*
-                        * Always select the oldest socket. It's not fair,
-                        * but so is life
-                        */
-                       svsk = list_entry(serv->sv_tempsocks.prev,
-                                         struct svc_sock,
-                                         sk_xprt.xpt_list);
-                       set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
-                       svc_xprt_get(&svsk->sk_xprt);
-               }
-               spin_unlock_bh(&serv->sv_lock);
-
-               if (svsk) {
-                       svc_xprt_enqueue(&svsk->sk_xprt);
-                       svc_xprt_put(&svsk->sk_xprt);
-               }
-       }
-}
-
-/*
- * Receive the next request on any socket.  This code is carefully
- * organised not to touch any cachelines in the shared svc_serv
- * structure, only cachelines in the local svc_pool.
- */
-int
-svc_recv(struct svc_rqst *rqstp, long timeout)
-{
-       struct svc_sock         *svsk = NULL;
-       struct svc_serv         *serv = rqstp->rq_server;
-       struct svc_pool         *pool = rqstp->rq_pool;
-       int                     len, i;
-       int                     pages;
-       struct xdr_buf          *arg;
-       DECLARE_WAITQUEUE(wait, current);
-
-       dprintk("svc: server %p waiting for data (to = %ld)\n",
-               rqstp, timeout);
-
-       if (rqstp->rq_sock)
-               printk(KERN_ERR
-                       "svc_recv: service %p, socket not NULL!\n",
-                        rqstp);
-       if (waitqueue_active(&rqstp->rq_wait))
-               printk(KERN_ERR
-                       "svc_recv: service %p, wait queue active!\n",
-                        rqstp);
-
-
-       /* now allocate needed pages.  If we get a failure, sleep briefly */
-       pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
-       for (i=0; i < pages ; i++)
-               while (rqstp->rq_pages[i] == NULL) {
-                       struct page *p = alloc_page(GFP_KERNEL);
-                       if (!p)
-                               schedule_timeout_uninterruptible(msecs_to_jiffies(500));
-                       rqstp->rq_pages[i] = p;
-               }
-       rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
-       BUG_ON(pages >= RPCSVC_MAXPAGES);
-
-       /* Make arg->head point to first page and arg->pages point to rest */
-       arg = &rqstp->rq_arg;
-       arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
-       arg->head[0].iov_len = PAGE_SIZE;
-       arg->pages = rqstp->rq_pages + 1;
-       arg->page_base = 0;
-       /* save at least one page for response */
-       arg->page_len = (pages-2)*PAGE_SIZE;
-       arg->len = (pages-1)*PAGE_SIZE;
-       arg->tail[0].iov_len = 0;
-
-       try_to_freeze();
-       cond_resched();
-       if (signalled())
-               return -EINTR;
-
-       spin_lock_bh(&pool->sp_lock);
-       if ((svsk = svc_sock_dequeue(pool)) != NULL) {
-               rqstp->rq_sock = svsk;
-               svc_xprt_get(&svsk->sk_xprt);
-               rqstp->rq_reserved = serv->sv_max_mesg;
-               atomic_add(rqstp->rq_reserved, &svsk->sk_xprt.xpt_reserved);
-       } else {
-               /* No data pending. Go to sleep */
-               svc_thread_enqueue(pool, rqstp);
-
-               /*
-                * We have to be able to interrupt this wait
-                * to bring down the daemons ...
-                */
-               set_current_state(TASK_INTERRUPTIBLE);
-               add_wait_queue(&rqstp->rq_wait, &wait);
-               spin_unlock_bh(&pool->sp_lock);
-
-               schedule_timeout(timeout);
-
-               try_to_freeze();
-
-               spin_lock_bh(&pool->sp_lock);
-               remove_wait_queue(&rqstp->rq_wait, &wait);
-
-               if (!(svsk = rqstp->rq_sock)) {
-                       svc_thread_dequeue(pool, rqstp);
-                       spin_unlock_bh(&pool->sp_lock);
-                       dprintk("svc: server %p, no data yet\n", rqstp);
-                       return signalled()? -EINTR : -EAGAIN;
-               }
-       }
-       spin_unlock_bh(&pool->sp_lock);
-
-       len = 0;
-       if (test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)) {
-               dprintk("svc_recv: found XPT_CLOSE\n");
-               svc_delete_xprt(&svsk->sk_xprt);
-       } else if (test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags)) {
-               struct svc_xprt *newxpt;
-               newxpt = svsk->sk_xprt.xpt_ops->xpo_accept(&svsk->sk_xprt);
-               if (newxpt) {
-                       /*
-                        * We know this module_get will succeed because the
-                        * listener holds a reference too
-                        */
-                       __module_get(newxpt->xpt_class->xcl_owner);
-                       svc_check_conn_limits(svsk->sk_xprt.xpt_server);
-               }
-               svc_sock_received(svsk);
-       } else {
-               dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
-                       rqstp, pool->sp_id, svsk,
-                       atomic_read(&svsk->sk_xprt.xpt_ref.refcount));
-               len = svsk->sk_xprt.xpt_ops->xpo_recvfrom(rqstp);
-               dprintk("svc: got len=%d\n", len);
-       }
-
-       /* No data, incomplete (TCP) read, or accept() */
-       if (len == 0 || len == -EAGAIN) {
-               rqstp->rq_res.len = 0;
-               svc_sock_release(rqstp);
-               return -EAGAIN;
-       }
-       svsk->sk_lastrecv = get_seconds();
-       clear_bit(XPT_OLD, &svsk->sk_xprt.xpt_flags);
-
-       rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
-       rqstp->rq_chandle.defer = svc_defer;
-
-       if (serv->sv_stats)
-               serv->sv_stats->netcnt++;
-       return len;
-}
-
-/*
- * Drop request
- */
-void
-svc_drop(struct svc_rqst *rqstp)
-{
-       dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
-       svc_sock_release(rqstp);
-}
-
-/*
- * Return reply to client.
- */
-int
-svc_send(struct svc_rqst *rqstp)
-{
-       struct svc_xprt *xprt;
-       int             len;
-       struct xdr_buf  *xb;
-
-       xprt = rqstp->rq_xprt;
-       if (!xprt)
-               return -EFAULT;
-
-       /* release the receive skb before sending the reply */
-       rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
-
-       /* calculate over-all length */
-       xb = & rqstp->rq_res;
-       xb->len = xb->head[0].iov_len +
-               xb->page_len +
-               xb->tail[0].iov_len;
-
-       /* Grab mutex to serialize outgoing data. */
-       mutex_lock(&xprt->xpt_mutex);
-       if (test_bit(XPT_DEAD, &xprt->xpt_flags))
-               len = -ENOTCONN;
-       else
-               len = xprt->xpt_ops->xpo_sendto(rqstp);
-       mutex_unlock(&xprt->xpt_mutex);
-       svc_sock_release(rqstp);
-
-       if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
-               return 0;
-       return len;
-}
-
-/*
- * Timer function to close old temporary sockets, using
- * a mark-and-sweep algorithm.
- */
-static void
-svc_age_temp_sockets(unsigned long closure)
-{
-       struct svc_serv *serv = (struct svc_serv *)closure;
-       struct svc_sock *svsk;
-       struct list_head *le, *next;
-       LIST_HEAD(to_be_aged);
-
-       dprintk("svc_age_temp_sockets\n");
-
-       if (!spin_trylock_bh(&serv->sv_lock)) {
-               /* busy, try again 1 sec later */
-               dprintk("svc_age_temp_sockets: busy\n");
-               mod_timer(&serv->sv_temptimer, jiffies + HZ);
-               return;
-       }
-
-       list_for_each_safe(le, next, &serv->sv_tempsocks) {
-               svsk = list_entry(le, struct svc_sock, sk_xprt.xpt_list);
-
-               if (!test_and_set_bit(XPT_OLD, &svsk->sk_xprt.xpt_flags))
-                       continue;
-               if (atomic_read(&svsk->sk_xprt.xpt_ref.refcount) > 1
-                   || test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags))
-                       continue;
-               svc_xprt_get(&svsk->sk_xprt);
-               list_move(le, &to_be_aged);
-               set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
-               set_bit(XPT_DETACHED, &svsk->sk_xprt.xpt_flags);
-       }
-       spin_unlock_bh(&serv->sv_lock);
-
-       while (!list_empty(&to_be_aged)) {
-               le = to_be_aged.next;
-               /* fiddling the sk_xprt.xpt_list node is safe 'cos we're XPT_DETACHED */
-               list_del_init(le);
-               svsk = list_entry(le, struct svc_sock, sk_xprt.xpt_list);
-
-               dprintk("queuing svsk %p for closing, %lu seconds old\n",
-                       svsk, get_seconds() - svsk->sk_lastrecv);
-
-               /* a thread will dequeue and close it soon */
-               svc_xprt_enqueue(&svsk->sk_xprt);
-               svc_xprt_put(&svsk->sk_xprt);
-       }
-
-       mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
-}
+EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
 
 /*
  * Initialize socket for RPC use and create svc_sock struct
@@ -1727,7 +1299,6 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
        struct svc_sock *svsk;
        struct sock     *inet;
        int             pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
-       int             is_temporary = flags & SVC_SOCK_TEMPORARY;
 
        dprintk("svc: svc_setup_socket %p\n", sock);
        if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
@@ -1739,7 +1310,7 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
 
        /* Register socket with portmapper */
        if (*errp >= 0 && pmap_register)
-               *errp = svc_register(serv, inet->sk_protocol,
+               *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
                                     ntohs(inet_sk(inet)->sport));
 
        if (*errp < 0) {
@@ -1747,16 +1318,12 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
                return NULL;
        }
 
-       set_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags);
        inet->sk_user_data = svsk;
        svsk->sk_sock = sock;
        svsk->sk_sk = inet;
        svsk->sk_ostate = inet->sk_state_change;
        svsk->sk_odata = inet->sk_data_ready;
        svsk->sk_owspace = inet->sk_write_space;
-       svsk->sk_lastrecv = get_seconds();
-       spin_lock_init(&svsk->sk_lock);
-       INIT_LIST_HEAD(&svsk->sk_deferred);
 
        /* Initialize the socket */
        if (sock->type == SOCK_DGRAM)
@@ -1764,34 +1331,25 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
        else
                svc_tcp_init(svsk, serv);
 
-       spin_lock_bh(&serv->sv_lock);
-       if (is_temporary) {
-               set_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
-               list_add(&svsk->sk_xprt.xpt_list, &serv->sv_tempsocks);
-               serv->sv_tmpcnt++;
-               if (serv->sv_temptimer.function == NULL) {
-                       /* setup timer to age temp sockets */
-                       setup_timer(&serv->sv_temptimer, svc_age_temp_sockets,
-                                       (unsigned long)serv);
-                       mod_timer(&serv->sv_temptimer,
-                                       jiffies + svc_conn_age_period * HZ);
-               }
-       } else {
-               clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
-               list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
-       }
-       spin_unlock_bh(&serv->sv_lock);
-
        dprintk("svc: svc_setup_socket created %p (inet %p)\n",
                                svsk, svsk->sk_sk);
 
        return svsk;
 }
 
-int svc_addsock(struct svc_serv *serv,
-               int fd,
-               char *name_return,
-               int *proto)
+/**
+ * svc_addsock - add a listener socket to an RPC service
+ * @serv: pointer to RPC service to which to add a new listener
+ * @fd: file descriptor of the new listener
+ * @name_return: pointer to buffer to fill in with name of listener
+ * @len: size of the buffer
+ *
+ * Fills in socket name and returns positive length of name if successful.
+ * Name is terminated with '\n'.  On error, returns a negative errno
+ * value.
+ */
+int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
+               const size_t len)
 {
        int err = 0;
        struct socket *so = sockfd_lookup(fd, &err);
@@ -1807,18 +1365,31 @@ int svc_addsock(struct svc_serv *serv,
        else if (so->state > SS_UNCONNECTED)
                err = -EISCONN;
        else {
-               svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
+               if (!try_module_get(THIS_MODULE))
+                       err = -ENOENT;
+               else
+                       svsk = svc_setup_socket(serv, so, &err,
+                                               SVC_SOCK_DEFAULTS);
                if (svsk) {
-                       svc_sock_received(svsk);
+                       struct sockaddr_storage addr;
+                       struct sockaddr *sin = (struct sockaddr *)&addr;
+                       int salen;
+                       if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
+                               svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
+                       clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
+                       spin_lock_bh(&serv->sv_lock);
+                       list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
+                       spin_unlock_bh(&serv->sv_lock);
+                       svc_xprt_received(&svsk->sk_xprt);
                        err = 0;
-               }
+               } else
+                       module_put(THIS_MODULE);
        }
        if (err) {
                sockfd_put(so);
                return err;
        }
-       if (proto) *proto = so->sk->sk_protocol;
-       return one_sock_name(name_return, svsk);
+       return svc_one_sock_name(svsk, name_return, len);
 }
 EXPORT_SYMBOL_GPL(svc_addsock);
 
@@ -1834,7 +1405,12 @@ static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
        struct socket   *sock;
        int             error;
        int             type;
-       char            buf[RPC_MAX_ADDRBUFLEN];
+       struct sockaddr_storage addr;
+       struct sockaddr *newsin = (struct sockaddr *)&addr;
+       int             newlen;
+       int             family;
+       int             val;
+       RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 
        dprintk("svc: svc_create_socket(%s, %d, %s)\n",
                        serv->sv_program->pg_name, protocol,
@@ -1845,27 +1421,53 @@ static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
                                "sockets supported\n");
                return ERR_PTR(-EINVAL);
        }
+
        type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
+       switch (sin->sa_family) {
+       case AF_INET6:
+               family = PF_INET6;
+               break;
+       case AF_INET:
+               family = PF_INET;
+               break;
+       default:
+               return ERR_PTR(-EINVAL);
+       }
 
-       error = sock_create_kern(sin->sa_family, type, protocol, &sock);
+       error = sock_create_kern(family, type, protocol, &sock);
        if (error < 0)
                return ERR_PTR(error);
 
        svc_reclassify_socket(sock);
 
+       /*
+        * If this is an PF_INET6 listener, we want to avoid
+        * getting requests from IPv4 remotes.  Those should
+        * be shunted to a PF_INET listener via rpcbind.
+        */
+       val = 1;
+       if (family == PF_INET6)
+               kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
+                                       (char *)&val, sizeof(val));
+
        if (type == SOCK_STREAM)
                sock->sk->sk_reuse = 1;         /* allow address reuse */
        error = kernel_bind(sock, sin, len);
        if (error < 0)
                goto bummer;
 
+       newlen = len;
+       error = kernel_getsockname(sock, newsin, &newlen);
+       if (error < 0)
+               goto bummer;
+
        if (protocol == IPPROTO_TCP) {
                if ((error = kernel_listen(sock, 64)) < 0)
                        goto bummer;
        }
 
        if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
-               svc_sock_received(svsk);
+               svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
                return (struct svc_xprt *)svsk;
        }
 
@@ -1890,6 +1492,24 @@ static void svc_sock_detach(struct svc_xprt *xprt)
        sk->sk_state_change = svsk->sk_ostate;
        sk->sk_data_ready = svsk->sk_odata;
        sk->sk_write_space = svsk->sk_owspace;
+
+       if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
+               wake_up_interruptible(sk->sk_sleep);
+}
+
+/*
+ * Disconnect the socket, and reset the callbacks
+ */
+static void svc_tcp_sock_detach(struct svc_xprt *xprt)
+{
+       struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
+
+       dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
+
+       svc_sock_detach(xprt);
+
+       if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
+               kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
 }
 
 /*
@@ -1900,8 +1520,6 @@ static void svc_sock_free(struct svc_xprt *xprt)
        struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
        dprintk("svc: svc_sock_free(%p)\n", svsk);
 
-       if (svsk->sk_info_authunix != NULL)
-               svcauth_unix_info_release(svsk->sk_info_authunix);
        if (svsk->sk_sock->file)
                sockfd_put(svsk->sk_sock);
        else
@@ -1910,159 +1528,40 @@ static void svc_sock_free(struct svc_xprt *xprt)
 }
 
 /*
- * Remove a dead transport
- */
-static void svc_delete_xprt(struct svc_xprt *xprt)
-{
-       struct svc_serv *serv = xprt->xpt_server;
-
-       dprintk("svc: svc_delete_xprt(%p)\n", xprt);
-       xprt->xpt_ops->xpo_detach(xprt);
-
-       spin_lock_bh(&serv->sv_lock);
-       if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags))
-               list_del_init(&xprt->xpt_list);
-       /*
-        * We used to delete the transport from whichever list
-        * it's sk_xprt.xpt_ready node was on, but we don't actually
-        * need to.  This is because the only time we're called
-        * while still attached to a queue, the queue itself
-        * is about to be destroyed (in svc_destroy).
-        */
-       if (!test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) {
-               BUG_ON(atomic_read(&xprt->xpt_ref.refcount) < 2);
-               if (test_bit(XPT_TEMP, &xprt->xpt_flags))
-                       serv->sv_tmpcnt--;
-               svc_xprt_put(xprt);
-       }
-       spin_unlock_bh(&serv->sv_lock);
-}
-
-static void svc_close_xprt(struct svc_xprt *xprt)
-{
-       set_bit(XPT_CLOSE, &xprt->xpt_flags);
-       if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
-               /* someone else will have to effect the close */
-               return;
-
-       svc_xprt_get(xprt);
-       svc_delete_xprt(xprt);
-       clear_bit(XPT_BUSY, &xprt->xpt_flags);
-       svc_xprt_put(xprt);
-}
-
-void svc_close_all(struct list_head *xprt_list)
-{
-       struct svc_xprt *xprt;
-       struct svc_xprt *tmp;
-
-       list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) {
-               set_bit(XPT_CLOSE, &xprt->xpt_flags);
-               if (test_bit(XPT_BUSY, &xprt->xpt_flags)) {
-                       /* Waiting to be processed, but no threads left,
-                        * So just remove it from the waiting list
-                        */
-                       list_del_init(&xprt->xpt_ready);
-                       clear_bit(XPT_BUSY, &xprt->xpt_flags);
-               }
-               svc_close_xprt(xprt);
-       }
-}
-
-/*
- * Handle defer and revisit of requests
+ * Create a svc_xprt.
+ *
+ * For internal use only (e.g. nfsv4.1 backchannel).
+ * Callers should typically use the xpo_create() method.
  */
-
-static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
+struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot)
 {
-       struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
        struct svc_sock *svsk;
+       struct svc_xprt *xprt = NULL;
 
-       if (too_many) {
-               svc_xprt_put(&dr->svsk->sk_xprt);
-               kfree(dr);
-               return;
-       }
-       dprintk("revisit queued\n");
-       svsk = dr->svsk;
-       dr->svsk = NULL;
-       spin_lock(&svsk->sk_lock);
-       list_add(&dr->handle.recent, &svsk->sk_deferred);
-       spin_unlock(&svsk->sk_lock);
-       set_bit(XPT_DEFERRED, &svsk->sk_xprt.xpt_flags);
-       svc_xprt_enqueue(&svsk->sk_xprt);
-       svc_xprt_put(&svsk->sk_xprt);
-}
-
-static struct cache_deferred_req *
-svc_defer(struct cache_req *req)
-{
-       struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
-       int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
-       struct svc_deferred_req *dr;
-
-       if (rqstp->rq_arg.page_len)
-               return NULL; /* if more than a page, give up FIXME */
-       if (rqstp->rq_deferred) {
-               dr = rqstp->rq_deferred;
-               rqstp->rq_deferred = NULL;
-       } else {
-               int skip  = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
-               /* FIXME maybe discard if size too large */
-               dr = kmalloc(size, GFP_KERNEL);
-               if (dr == NULL)
-                       return NULL;
-
-               dr->handle.owner = rqstp->rq_server;
-               dr->prot = rqstp->rq_prot;
-               memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
-               dr->addrlen = rqstp->rq_addrlen;
-               dr->daddr = rqstp->rq_daddr;
-               dr->argslen = rqstp->rq_arg.len >> 2;
-               memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
-       }
-       svc_xprt_get(rqstp->rq_xprt);
-       dr->svsk = rqstp->rq_sock;
+       dprintk("svc: %s\n", __func__);
+       svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
+       if (!svsk)
+               goto out;
 
-       dr->handle.revisit = svc_revisit;
-       return &dr->handle;
+       xprt = &svsk->sk_xprt;
+       if (prot == IPPROTO_TCP)
+               svc_xprt_init(&svc_tcp_class, xprt, serv);
+       else if (prot == IPPROTO_UDP)
+               svc_xprt_init(&svc_udp_class, xprt, serv);
+       else
+               BUG();
+out:
+       dprintk("svc: %s return %p\n", __func__, xprt);
+       return xprt;
 }
+EXPORT_SYMBOL_GPL(svc_sock_create);
 
 /*
- * recv data from a deferred request into an active one
+ * Destroy a svc_sock.
  */
-static int svc_deferred_recv(struct svc_rqst *rqstp)
+void svc_sock_destroy(struct svc_xprt *xprt)
 {
-       struct svc_deferred_req *dr = rqstp->rq_deferred;
-
-       rqstp->rq_arg.head[0].iov_base = dr->args;
-       rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
-       rqstp->rq_arg.page_len = 0;
-       rqstp->rq_arg.len = dr->argslen<<2;
-       rqstp->rq_prot        = dr->prot;
-       memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
-       rqstp->rq_addrlen     = dr->addrlen;
-       rqstp->rq_daddr       = dr->daddr;
-       rqstp->rq_respages    = rqstp->rq_pages;
-       return dr->argslen<<2;
-}
-
-
-static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
-{
-       struct svc_deferred_req *dr = NULL;
-
-       if (!test_bit(XPT_DEFERRED, &svsk->sk_xprt.xpt_flags))
-               return NULL;
-       spin_lock(&svsk->sk_lock);
-       clear_bit(XPT_DEFERRED, &svsk->sk_xprt.xpt_flags);
-       if (!list_empty(&svsk->sk_deferred)) {
-               dr = list_entry(svsk->sk_deferred.next,
-                               struct svc_deferred_req,
-                               handle.recent);
-               list_del_init(&dr->handle.recent);
-               set_bit(XPT_DEFERRED, &svsk->sk_xprt.xpt_flags);
-       }
-       spin_unlock(&svsk->sk_lock);
-       return dr;
+       if (xprt)
+               kfree(container_of(xprt, struct svc_sock, sk_xprt));
 }
+EXPORT_SYMBOL_GPL(svc_sock_destroy);