21 files changed, 970 insertions, 1231 deletions

diff --git a/net/sunrpc/auth.c b/net/sunrpc/auth.c
index 5285ead196c0..247e973544bf 100644
--- a/net/sunrpc/auth.c
+++ b/net/sunrpc/auth.c
@@ -296,7 +296,7 @@ static void
 rpcauth_unhash_cred_locked(struct rpc_cred *cred)
 {
 	hlist_del_rcu(&cred->cr_hash);
-	smp_mb__before_clear_bit();
+	smp_mb__before_atomic();
 	clear_bit(RPCAUTH_CRED_HASHED, &cred->cr_flags);
 }
 
diff --git a/net/sunrpc/auth_gss/auth_gss.c b/net/sunrpc/auth_gss/auth_gss.c
index 36e431ee1c90..b6e440baccc3 100644
--- a/net/sunrpc/auth_gss/auth_gss.c
+++ b/net/sunrpc/auth_gss/auth_gss.c
@@ -143,7 +143,7 @@ gss_cred_set_ctx(struct rpc_cred *cred, struct gss_cl_ctx *ctx)
 	gss_get_ctx(ctx);
 	rcu_assign_pointer(gss_cred->gc_ctx, ctx);
 	set_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
-	smp_mb__before_clear_bit();
+	smp_mb__before_atomic();
 	clear_bit(RPCAUTH_CRED_NEW, &cred->cr_flags);
 }
 
diff --git a/net/sunrpc/auth_gss/gss_mech_switch.c b/net/sunrpc/auth_gss/gss_mech_switch.c
index 27ce26240932..92d5ab99fbf3 100644
--- a/net/sunrpc/auth_gss/gss_mech_switch.c
+++ b/net/sunrpc/auth_gss/gss_mech_switch.c
@@ -218,10 +218,8 @@ static struct gss_api_mech *_gss_mech_get_by_pseudoflavor(u32 pseudoflavor)
 
 	spin_lock(&registered_mechs_lock);
 	list_for_each_entry(pos, &registered_mechs, gm_list) {
-		if (!mech_supports_pseudoflavor(pos, pseudoflavor)) {
-			module_put(pos->gm_owner);
+		if (!mech_supports_pseudoflavor(pos, pseudoflavor))
 			continue;
-		}
 		if (try_module_get(pos->gm_owner))
 			gm = pos;
 		break;
diff --git a/net/sunrpc/auth_gss/svcauth_gss.c b/net/sunrpc/auth_gss/svcauth_gss.c
index 0f73f4507746..4ce5eccec1f6 100644
--- a/net/sunrpc/auth_gss/svcauth_gss.c
+++ b/net/sunrpc/auth_gss/svcauth_gss.c
@@ -1503,6 +1503,7 @@ svcauth_gss_accept(struct svc_rqst *rqstp, __be32 *authp)
 			if (unwrap_integ_data(rqstp, &rqstp->rq_arg,
 					gc->gc_seq, rsci->mechctx))
 				goto garbage_args;
+			rqstp->rq_auth_slack = RPC_MAX_AUTH_SIZE;
 			break;
 		case RPC_GSS_SVC_PRIVACY:
 			/* placeholders for length and seq. number: */
@@ -1511,6 +1512,7 @@ svcauth_gss_accept(struct svc_rqst *rqstp, __be32 *authp)
 			if (unwrap_priv_data(rqstp, &rqstp->rq_arg,
 					gc->gc_seq, rsci->mechctx))
 				goto garbage_args;
+			rqstp->rq_auth_slack = RPC_MAX_AUTH_SIZE * 2;
 			break;
 		default:
 			goto auth_err;
diff --git a/net/sunrpc/backchannel_rqst.c b/net/sunrpc/backchannel_rqst.c
index 3513d559bc45..9761a0da964d 100644
--- a/net/sunrpc/backchannel_rqst.c
+++ b/net/sunrpc/backchannel_rqst.c
@@ -244,10 +244,10 @@ void xprt_free_bc_request(struct rpc_rqst *req)
 	dprintk("RPC:       free backchannel req=%p\n", req);
 
 	req->rq_connect_cookie = xprt->connect_cookie - 1;
-	smp_mb__before_clear_bit();
+	smp_mb__before_atomic();
 	WARN_ON_ONCE(!test_bit(RPC_BC_PA_IN_USE, &req->rq_bc_pa_state));
 	clear_bit(RPC_BC_PA_IN_USE, &req->rq_bc_pa_state);
-	smp_mb__after_clear_bit();
+	smp_mb__after_atomic();
 
 	if (!xprt_need_to_requeue(xprt)) {
 		/*
diff --git a/net/sunrpc/cache.c b/net/sunrpc/cache.c
index ae333c1845bb..066362141133 100644
--- a/net/sunrpc/cache.c
+++ b/net/sunrpc/cache.c
@@ -374,7 +374,7 @@ void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 	}
 	return;
 out:
-	printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
+	printk(KERN_ERR "RPC: failed to unregister %s cache\n", cd->name);
 }
 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 
diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index 25578afe1548..c0365c14b858 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -832,7 +832,8 @@ static void rpc_async_schedule(struct work_struct *work)
  * @size: requested byte size
  *
  * To prevent rpciod from hanging, this allocator never sleeps,
- * returning NULL if the request cannot be serviced immediately.
+ * returning NULL and suppressing warning if the request cannot be serviced
+ * immediately.
  * The caller can arrange to sleep in a way that is safe for rpciod.
  *
  * Most requests are 'small' (under 2KiB) and can be serviced from a
@@ -845,7 +846,7 @@ static void rpc_async_schedule(struct work_struct *work)
 void *rpc_malloc(struct rpc_task *task, size_t size)
 {
 	struct rpc_buffer *buf;
-	gfp_t gfp = GFP_NOWAIT;
+	gfp_t gfp = GFP_NOWAIT | __GFP_NOWARN;
 
 	if (RPC_IS_SWAPPER(task))
 		gfp |= __GFP_MEMALLOC;
diff --git a/net/sunrpc/sunrpc.h b/net/sunrpc/sunrpc.h
index 14c9f6d1c5ff..f2b7cb540e61 100644
--- a/net/sunrpc/sunrpc.h
+++ b/net/sunrpc/sunrpc.h
@@ -43,6 +43,19 @@ static inline int rpc_reply_expected(struct rpc_task *task)
 		(task->tk_msg.rpc_proc->p_decode != NULL);
 }
 
+static inline int sock_is_loopback(struct sock *sk)
+{
+	struct dst_entry *dst;
+	int loopback = 0;
+	rcu_read_lock();
+	dst = rcu_dereference(sk->sk_dst_cache);
+	if (dst && dst->dev &&
+	    (dst->dev->features & NETIF_F_LOOPBACK))
+		loopback = 1;
+	rcu_read_unlock();
+	return loopback;
+}
+
 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
 		    struct page *headpage, unsigned long headoffset,
 		    struct page *tailpage, unsigned long tailoffset);
diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c
index 06c6ff0cb911..b4737fbdec13 100644
--- a/net/sunrpc/svc_xprt.c
+++ b/net/sunrpc/svc_xprt.c
@@ -597,6 +597,7 @@ static int svc_alloc_arg(struct svc_rqst *rqstp)
 			}
 			rqstp->rq_pages[i] = p;
 		}
+	rqstp->rq_page_end = &rqstp->rq_pages[i];
 	rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
 
 	/* Make arg->head point to first page and arg->pages point to rest */
@@ -730,6 +731,8 @@ static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 		newxpt = xprt->xpt_ops->xpo_accept(xprt);
 		if (newxpt)
 			svc_add_new_temp_xprt(serv, newxpt);
+		else
+			module_put(xprt->xpt_class->xcl_owner);
 	} else if (xprt->xpt_ops->xpo_has_wspace(xprt)) {
 		/* XPT_DATA|XPT_DEFERRED case: */
 		dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
@@ -793,7 +796,7 @@ int svc_recv(struct svc_rqst *rqstp, long timeout)
 
 	clear_bit(XPT_OLD, &xprt->xpt_flags);
 
-	rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
+	rqstp->rq_secure = xprt->xpt_ops->xpo_secure_port(rqstp);
 	rqstp->rq_chandle.defer = svc_defer;
 
 	if (serv->sv_stats)
diff --git a/net/sunrpc/svcauth.c b/net/sunrpc/svcauth.c
index 2af7b0cba43a..79c0f3459b5c 100644
--- a/net/sunrpc/svcauth.c
+++ b/net/sunrpc/svcauth.c
@@ -54,6 +54,8 @@ svc_authenticate(struct svc_rqst *rqstp, __be32 *authp)
 	}
 	spin_unlock(&authtab_lock);
 
+	rqstp->rq_auth_slack = 0;
+
 	rqstp->rq_authop = aops;
 	return aops->accept(rqstp, authp);
 }
diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 43bcb4699d69..b507cd327d9b 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -400,6 +400,12 @@ static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
 	release_sock(sock->sk);
 #endif
 }
+
+static int svc_sock_secure_port(struct svc_rqst *rqstp)
+{
+	return svc_port_is_privileged(svc_addr(rqstp));
+}
+
 /*
  * INET callback when data has been received on the socket.
  */
@@ -678,6 +684,7 @@ static struct svc_xprt_ops svc_udp_ops = {
 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
 	.xpo_has_wspace = svc_udp_has_wspace,
 	.xpo_accept = svc_udp_accept,
+	.xpo_secure_port = svc_sock_secure_port,
 };
 
 static struct svc_xprt_class svc_udp_class = {
@@ -842,8 +849,7 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
 	 * tell us anything.  For now just warn about unpriv connections.
 	 */
 	if (!svc_port_is_privileged(sin)) {
-		dprintk(KERN_WARNING
-			"%s: connect from unprivileged port: %s\n",
+		dprintk("%s: connect from unprivileged port: %s\n",
 			serv->sv_name,
 			__svc_print_addr(sin, buf, sizeof(buf)));
 	}
@@ -867,6 +873,10 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
 	}
 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
 
+	if (sock_is_loopback(newsock->sk))
+		set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
+	else
+		clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
 	if (serv->sv_stats)
 		serv->sv_stats->nettcpconn++;
 
@@ -1112,6 +1122,7 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
 
 	rqstp->rq_xprt_ctxt   = NULL;
 	rqstp->rq_prot	      = IPPROTO_TCP;
+	rqstp->rq_local	      = !!test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags);
 
 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
 	calldir = p[1];
@@ -1234,6 +1245,7 @@ static struct svc_xprt_ops svc_tcp_bc_ops = {
 	.xpo_detach = svc_bc_tcp_sock_detach,
 	.xpo_free = svc_bc_sock_free,
 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
+	.xpo_secure_port = svc_sock_secure_port,
 };
 
 static struct svc_xprt_class svc_tcp_bc_class = {
@@ -1272,6 +1284,7 @@ static struct svc_xprt_ops svc_tcp_ops = {
 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
 	.xpo_has_wspace = svc_tcp_has_wspace,
 	.xpo_accept = svc_tcp_accept,
+	.xpo_secure_port = svc_sock_secure_port,
 };
 
 static struct svc_xprt_class svc_tcp_class = {
diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
index dd97ba3c4456..23fb4e75e245 100644
--- a/net/sunrpc/xdr.c
+++ b/net/sunrpc/xdr.c
@@ -462,6 +462,7 @@ void xdr_init_encode(struct xdr_stream *xdr, struct xdr_buf *buf, __be32 *p)
 	struct kvec *iov = buf->head;
 	int scratch_len = buf->buflen - buf->page_len - buf->tail[0].iov_len;
 
+	xdr_set_scratch_buffer(xdr, NULL, 0);
 	BUG_ON(scratch_len < 0);
 	xdr->buf = buf;
 	xdr->iov = iov;
@@ -482,6 +483,73 @@ void xdr_init_encode(struct xdr_stream *xdr, struct xdr_buf *buf, __be32 *p)
 EXPORT_SYMBOL_GPL(xdr_init_encode);
 
 /**
+ * xdr_commit_encode - Ensure all data is written to buffer
+ * @xdr: pointer to xdr_stream
+ *
+ * We handle encoding across page boundaries by giving the caller a
+ * temporary location to write to, then later copying the data into
+ * place; xdr_commit_encode does that copying.
+ *
+ * Normally the caller doesn't need to call this directly, as the
+ * following xdr_reserve_space will do it.  But an explicit call may be
+ * required at the end of encoding, or any other time when the xdr_buf
+ * data might be read.
+ */
+void xdr_commit_encode(struct xdr_stream *xdr)
+{
+	int shift = xdr->scratch.iov_len;
+	void *page;
+
+	if (shift == 0)
+		return;
+	page = page_address(*xdr->page_ptr);
+	memcpy(xdr->scratch.iov_base, page, shift);
+	memmove(page, page + shift, (void *)xdr->p - page);
+	xdr->scratch.iov_len = 0;
+}
+EXPORT_SYMBOL_GPL(xdr_commit_encode);
+
+__be32 *xdr_get_next_encode_buffer(struct xdr_stream *xdr, size_t nbytes)
+{
+	static __be32 *p;
+	int space_left;
+	int frag1bytes, frag2bytes;
+
+	if (nbytes > PAGE_SIZE)
+		return NULL; /* Bigger buffers require special handling */
+	if (xdr->buf->len + nbytes > xdr->buf->buflen)
+		return NULL; /* Sorry, we're totally out of space */
+	frag1bytes = (xdr->end - xdr->p) << 2;
+	frag2bytes = nbytes - frag1bytes;
+	if (xdr->iov)
+		xdr->iov->iov_len += frag1bytes;
+	else
+		xdr->buf->page_len += frag1bytes;
+	xdr->page_ptr++;
+	xdr->iov = NULL;
+	/*
+	 * If the last encode didn't end exactly on a page boundary, the
+	 * next one will straddle boundaries.  Encode into the next
+	 * page, then copy it back later in xdr_commit_encode.  We use
+	 * the "scratch" iov to track any temporarily unused fragment of
+	 * space at the end of the previous buffer:
+	 */
+	xdr->scratch.iov_base = xdr->p;
+	xdr->scratch.iov_len = frag1bytes;
+	p = page_address(*xdr->page_ptr);
+	/*
+	 * Note this is where the next encode will start after we've
+	 * shifted this one back:
+	 */
+	xdr->p = (void *)p + frag2bytes;
+	space_left = xdr->buf->buflen - xdr->buf->len;
+	xdr->end = (void *)p + min_t(int, space_left, PAGE_SIZE);
+	xdr->buf->page_len += frag2bytes;
+	xdr->buf->len += nbytes;
+	return p;
+}
+
+/**
  * xdr_reserve_space - Reserve buffer space for sending
  * @xdr: pointer to xdr_stream
  * @nbytes: number of bytes to reserve
@@ -495,20 +563,122 @@ __be32 * xdr_reserve_space(struct xdr_stream *xdr, size_t nbytes)
 	__be32 *p = xdr->p;
 	__be32 *q;
 
+	xdr_commit_encode(xdr);
 	/* align nbytes on the next 32-bit boundary */
 	nbytes += 3;
 	nbytes &= ~3;
 	q = p + (nbytes >> 2);
 	if (unlikely(q > xdr->end || q < p))
-		return NULL;
+		return xdr_get_next_encode_buffer(xdr, nbytes);
 	xdr->p = q;
-	xdr->iov->iov_len += nbytes;
+	if (xdr->iov)
+		xdr->iov->iov_len += nbytes;
+	else
+		xdr->buf->page_len += nbytes;
 	xdr->buf->len += nbytes;
 	return p;
 }
 EXPORT_SYMBOL_GPL(xdr_reserve_space);
 
 /**
+ * xdr_truncate_encode - truncate an encode buffer
+ * @xdr: pointer to xdr_stream
+ * @len: new length of buffer
+ *
+ * Truncates the xdr stream, so that xdr->buf->len == len,
+ * and xdr->p points at offset len from the start of the buffer, and
+ * head, tail, and page lengths are adjusted to correspond.
+ *
+ * If this means moving xdr->p to a different buffer, we assume that
+ * that the end pointer should be set to the end of the current page,
+ * except in the case of the head buffer when we assume the head
+ * buffer's current length represents the end of the available buffer.
+ *
+ * This is *not* safe to use on a buffer that already has inlined page
+ * cache pages (as in a zero-copy server read reply), except for the
+ * simple case of truncating from one position in the tail to another.
+ *
+ */
+void xdr_truncate_encode(struct xdr_stream *xdr, size_t len)
+{
+	struct xdr_buf *buf = xdr->buf;
+	struct kvec *head = buf->head;
+	struct kvec *tail = buf->tail;
+	int fraglen;
+	int new, old;
+
+	if (len > buf->len) {
+		WARN_ON_ONCE(1);
+		return;
+	}
+	xdr_commit_encode(xdr);
+
+	fraglen = min_t(int, buf->len - len, tail->iov_len);
+	tail->iov_len -= fraglen;
+	buf->len -= fraglen;
+	if (tail->iov_len && buf->len == len) {
+		xdr->p = tail->iov_base + tail->iov_len;
+		/* xdr->end, xdr->iov should be set already */
+		return;
+	}
+	WARN_ON_ONCE(fraglen);
+	fraglen = min_t(int, buf->len - len, buf->page_len);
+	buf->page_len -= fraglen;
+	buf->len -= fraglen;
+
+	new = buf->page_base + buf->page_len;
+	old = new + fraglen;
+	xdr->page_ptr -= (old >> PAGE_SHIFT) - (new >> PAGE_SHIFT);
+
+	if (buf->page_len && buf->len == len) {
+		xdr->p = page_address(*xdr->page_ptr);
+		xdr->end = (void *)xdr->p + PAGE_SIZE;
+		xdr->p = (void *)xdr->p + (new % PAGE_SIZE);
+		/* xdr->iov should already be NULL */
+		return;
+	}
+	if (fraglen) {
+		xdr->end = head->iov_base + head->iov_len;
+		xdr->page_ptr--;
+	}
+	/* (otherwise assume xdr->end is already set) */
+	head->iov_len = len;
+	buf->len = len;
+	xdr->p = head->iov_base + head->iov_len;
+	xdr->iov = buf->head;
+}
+EXPORT_SYMBOL(xdr_truncate_encode);
+
+/**
+ * xdr_restrict_buflen - decrease available buffer space
+ * @xdr: pointer to xdr_stream
+ * @newbuflen: new maximum number of bytes available
+ *
+ * Adjust our idea of how much space is available in the buffer.
+ * If we've already used too much space in the buffer, returns -1.
+ * If the available space is already smaller than newbuflen, returns 0
+ * and does nothing.  Otherwise, adjusts xdr->buf->buflen to newbuflen
+ * and ensures xdr->end is set at most offset newbuflen from the start
+ * of the buffer.
+ */
+int xdr_restrict_buflen(struct xdr_stream *xdr, int newbuflen)
+{
+	struct xdr_buf *buf = xdr->buf;
+	int left_in_this_buf = (void *)xdr->end - (void *)xdr->p;
+	int end_offset = buf->len + left_in_this_buf;
+
+	if (newbuflen < 0 || newbuflen < buf->len)
+		return -1;
+	if (newbuflen > buf->buflen)
+		return 0;
+	if (newbuflen < end_offset)
+		xdr->end = (void *)xdr->end + newbuflen - end_offset;
+	buf->buflen = newbuflen;
+	return 0;
+}
+EXPORT_SYMBOL(xdr_restrict_buflen);
+
+/**
  * xdr_write_pages - Insert a list of pages into an XDR buffer for sending
  * @xdr: pointer to xdr_stream
  * @pages: list of pages
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index d173f79947c6..c3b2b3369e52 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -71,24 +71,6 @@ static void	 xprt_destroy(struct rpc_xprt *xprt);
 static DEFINE_SPINLOCK(xprt_list_lock);
 static LIST_HEAD(xprt_list);
 
-/*
- * The transport code maintains an estimate on the maximum number of out-
- * standing RPC requests, using a smoothed version of the congestion
- * avoidance implemented in 44BSD. This is basically the Van Jacobson
- * congestion algorithm: If a retransmit occurs, the congestion window is
- * halved; otherwise, it is incremented by 1/cwnd when
- *
- *	-	a reply is received and
- *	-	a full number of requests are outstanding and
- *	-	the congestion window hasn't been updated recently.
- */
-#define RPC_CWNDSHIFT		(8U)
-#define RPC_CWNDSCALE		(1U << RPC_CWNDSHIFT)
-#define RPC_INITCWND		RPC_CWNDSCALE
-#define RPC_MAXCWND(xprt)	((xprt)->max_reqs << RPC_CWNDSHIFT)
-
-#define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
-
 /**
  * xprt_register_transport - register a transport implementation
  * @transport: transport to register
@@ -230,9 +212,9 @@ static void xprt_clear_locked(struct rpc_xprt *xprt)
 {
 	xprt->snd_task = NULL;
 	if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
-		smp_mb__before_clear_bit();
+		smp_mb__before_atomic();
 		clear_bit(XPRT_LOCKED, &xprt->state);
-		smp_mb__after_clear_bit();
+		smp_mb__after_atomic();
 	} else
 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
 }
@@ -446,7 +428,15 @@ EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
  * @task: recently completed RPC request used to adjust window
  * @result: result code of completed RPC request
  *
- * We use a time-smoothed congestion estimator to avoid heavy oscillation.
+ * The transport code maintains an estimate on the maximum number of out-
+ * standing RPC requests, using a smoothed version of the congestion
+ * avoidance implemented in 44BSD. This is basically the Van Jacobson
+ * congestion algorithm: If a retransmit occurs, the congestion window is
+ * halved; otherwise, it is incremented by 1/cwnd when
+ *
+ *	-	a reply is received and
+ *	-	a full number of requests are outstanding and
+ *	-	the congestion window hasn't been updated recently.
  */
 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 {
diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
index 96ead526b125..693966d3f33b 100644
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -78,8 +78,7 @@ static const char transfertypes[][12] = {
  * elements. Segments are then coalesced when registered, if possible
  * within the selected memreg mode.
  *
- * Note, this routine is never called if the connection's memory
- * registration strategy is 0 (bounce buffers).
+ * Returns positive number of segments converted, or a negative errno.
  */
 
 static int
@@ -102,10 +101,17 @@ rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos,
 	page_base = xdrbuf->page_base & ~PAGE_MASK;
 	p = 0;
 	while (len && n < nsegs) {
+		if (!ppages[p]) {
+			/* alloc the pagelist for receiving buffer */
+			ppages[p] = alloc_page(GFP_ATOMIC);
+			if (!ppages[p])
+				return -ENOMEM;
+		}
 		seg[n].mr_page = ppages[p];
 		seg[n].mr_offset = (void *)(unsigned long) page_base;
 		seg[n].mr_len = min_t(u32, PAGE_SIZE - page_base, len);
-		BUG_ON(seg[n].mr_len > PAGE_SIZE);
+		if (seg[n].mr_len > PAGE_SIZE)
+			return -EIO;
 		len -= seg[n].mr_len;
 		++n;
 		++p;
@@ -114,7 +120,7 @@ rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos,
 
 	/* Message overflows the seg array */
 	if (len && n == nsegs)
-		return 0;
+		return -EIO;
 
 	if (xdrbuf->tail[0].iov_len) {
 		/* the rpcrdma protocol allows us to omit any trailing
@@ -123,7 +129,7 @@ rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos,
 			return n;
 		if (n == nsegs)
 			/* Tail remains, but we're out of segments */
-			return 0;
+			return -EIO;
 		seg[n].mr_page = NULL;
 		seg[n].mr_offset = xdrbuf->tail[0].iov_base;
 		seg[n].mr_len = xdrbuf->tail[0].iov_len;
@@ -164,15 +170,17 @@ rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos,
  *  Reply chunk (a counted array):
  *   N elements:
  *    1 - N - HLOO - HLOO - ... - HLOO
+ *
+ * Returns positive RPC/RDMA header size, or negative errno.
  */
 
-static unsigned int
+static ssize_t
 rpcrdma_create_chunks(struct rpc_rqst *rqst, struct xdr_buf *target,
 		struct rpcrdma_msg *headerp, enum rpcrdma_chunktype type)
 {
 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
 	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
-	int nsegs, nchunks = 0;
+	int n, nsegs, nchunks = 0;
 	unsigned int pos;
 	struct rpcrdma_mr_seg *seg = req->rl_segments;
 	struct rpcrdma_read_chunk *cur_rchunk = NULL;
@@ -198,12 +206,11 @@ rpcrdma_create_chunks(struct rpc_rqst *rqst, struct xdr_buf *target,
 		pos = target->head[0].iov_len;
 
 	nsegs = rpcrdma_convert_iovs(target, pos, type, seg, RPCRDMA_MAX_SEGS);
-	if (nsegs == 0)
-		return 0;
+	if (nsegs < 0)
+		return nsegs;
 
 	do {
-		/* bind/register the memory, then build chunk from result. */
-		int n = rpcrdma_register_external(seg, nsegs,
+		n = rpcrdma_register_external(seg, nsegs,
 						cur_wchunk != NULL, r_xprt);
 		if (n <= 0)
 			goto out;
@@ -248,10 +255,6 @@ rpcrdma_create_chunks(struct rpc_rqst *rqst, struct xdr_buf *target,
 	/* success. all failures return above */
 	req->rl_nchunks = nchunks;
 
-	BUG_ON(nchunks == 0);
-	BUG_ON((r_xprt->rx_ia.ri_memreg_strategy == RPCRDMA_FRMR)
-	       && (nchunks > 3));
-
 	/*
 	 * finish off header. If write, marshal discrim and nchunks.
 	 */
@@ -278,8 +281,8 @@ rpcrdma_create_chunks(struct rpc_rqst *rqst, struct xdr_buf *target,
 out:
 	for (pos = 0; nchunks--;)
 		pos += rpcrdma_deregister_external(
-				&req->rl_segments[pos], r_xprt, NULL);
-	return 0;
+				&req->rl_segments[pos], r_xprt);
+	return n;
 }
 
 /*
@@ -361,6 +364,8 @@ rpcrdma_inline_pullup(struct rpc_rqst *rqst, int pad)
  *  [1] -- the RPC header/data, marshaled by RPC and the NFS protocol.
  *  [2] -- optional padding.
  *  [3] -- if padded, header only in [1] and data here.
+ *
+ * Returns zero on success, otherwise a negative errno.
  */
 
 int
@@ -370,7 +375,8 @@ rpcrdma_marshal_req(struct rpc_rqst *rqst)
 	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
 	char *base;
-	size_t hdrlen, rpclen, padlen;
+	size_t rpclen, padlen;
+	ssize_t hdrlen;
 	enum rpcrdma_chunktype rtype, wtype;
 	struct rpcrdma_msg *headerp;
 
@@ -441,14 +447,10 @@ rpcrdma_marshal_req(struct rpc_rqst *rqst)
 	/* The following simplification is not true forever */
 	if (rtype != rpcrdma_noch && wtype == rpcrdma_replych)
 		wtype = rpcrdma_noch;
-	BUG_ON(rtype != rpcrdma_noch && wtype != rpcrdma_noch);
-
-	if (r_xprt->rx_ia.ri_memreg_strategy == RPCRDMA_BOUNCEBUFFERS &&
-	    (rtype != rpcrdma_noch || wtype != rpcrdma_noch)) {
-		/* forced to "pure inline"? */
-		dprintk("RPC:       %s: too much data (%d/%d) for inline\n",
-			__func__, rqst->rq_rcv_buf.len, rqst->rq_snd_buf.len);
-		return -1;
+	if (rtype != rpcrdma_noch && wtype != rpcrdma_noch) {
+		dprintk("RPC:       %s: cannot marshal multiple chunk lists\n",
+			__func__);
+		return -EIO;
 	}
 
 	hdrlen = 28; /*sizeof *headerp;*/
@@ -474,8 +476,11 @@ rpcrdma_marshal_req(struct rpc_rqst *rqst)
 			headerp->rm_body.rm_padded.rm_pempty[1] = xdr_zero;
 			headerp->rm_body.rm_padded.rm_pempty[2] = xdr_zero;
 			hdrlen += 2 * sizeof(u32); /* extra words in padhdr */
-			BUG_ON(wtype != rpcrdma_noch);
-
+			if (wtype != rpcrdma_noch) {
+				dprintk("RPC:       %s: invalid chunk list\n",
+					__func__);
+				return -EIO;
+			}
 		} else {
 			headerp->rm_body.rm_nochunks.rm_empty[0] = xdr_zero;
 			headerp->rm_body.rm_nochunks.rm_empty[1] = xdr_zero;
@@ -492,8 +497,7 @@ rpcrdma_marshal_req(struct rpc_rqst *rqst)
 			 * on receive. Therefore, we request a reply chunk
 			 * for non-writes wherever feasible and efficient.
 			 */
-			if (wtype == rpcrdma_noch &&
-			    r_xprt->rx_ia.ri_memreg_strategy > RPCRDMA_REGISTER)
+			if (wtype == rpcrdma_noch)
 				wtype = rpcrdma_replych;
 		}
 	}
@@ -511,9 +515,8 @@ rpcrdma_marshal_req(struct rpc_rqst *rqst)
 		hdrlen = rpcrdma_create_chunks(rqst,
 					&rqst->rq_rcv_buf, headerp, wtype);
 	}
-
-	if (hdrlen == 0)
-		return -1;
+	if (hdrlen < 0)
+		return hdrlen;
 
 	dprintk("RPC:       %s: %s: hdrlen %zd rpclen %zd padlen %zd"
 		" headerp 0x%p base 0x%p lkey 0x%x\n",
@@ -680,15 +683,11 @@ rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
 	rqst->rq_private_buf = rqst->rq_rcv_buf;
 }
 
-/*
- * This function is called when an async event is posted to
- * the connection which changes the connection state. All it
- * does at this point is mark the connection up/down, the rpc
- * timers do the rest.
- */
 void
-rpcrdma_conn_func(struct rpcrdma_ep *ep)
+rpcrdma_connect_worker(struct work_struct *work)
 {
+	struct rpcrdma_ep *ep =
+		container_of(work, struct rpcrdma_ep, rep_connect_worker.work);
 	struct rpc_xprt *xprt = ep->rep_xprt;
 
 	spin_lock_bh(&xprt->transport_lock);
@@ -705,13 +704,15 @@ rpcrdma_conn_func(struct rpcrdma_ep *ep)
 }
 
 /*
- * This function is called when memory window unbind which we are waiting
- * for completes. Just use rr_func (zeroed by upcall) to signal completion.
+ * This function is called when an async event is posted to
+ * the connection which changes the connection state. All it
+ * does at this point is mark the connection up/down, the rpc
+ * timers do the rest.
  */
-static void
-rpcrdma_unbind_func(struct rpcrdma_rep *rep)
+void
+rpcrdma_conn_func(struct rpcrdma_ep *ep)
 {
-	wake_up(&rep->rr_unbind);
+	schedule_delayed_work(&ep->rep_connect_worker, 0);
 }
 
 /*
@@ -728,7 +729,8 @@ rpcrdma_reply_handler(struct rpcrdma_rep *rep)
 	struct rpc_xprt *xprt = rep->rr_xprt;
 	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 	__be32 *iptr;
-	int i, rdmalen, status;
+	int rdmalen, status;
+	unsigned long cwnd;
 
 	/* Check status. If bad, signal disconnect and return rep to pool */
 	if (rep->rr_len == ~0U) {
@@ -783,6 +785,7 @@ repost:
 
 	/* from here on, the reply is no longer an orphan */
 	req->rl_reply = rep;
+	xprt->reestablish_timeout = 0;
 
 	/* check for expected message types */
 	/* The order of some of these tests is important. */
@@ -857,26 +860,10 @@ badheader:
 		break;
 	}
 
-	/* If using mw bind, start the deregister process now. */
-	/* (Note: if mr_free(), cannot perform it here, in tasklet context) */
-	if (req->rl_nchunks) switch (r_xprt->rx_ia.ri_memreg_strategy) {
-	case RPCRDMA_MEMWINDOWS:
-		for (i = 0; req->rl_nchunks-- > 1;)
-			i += rpcrdma_deregister_external(
-				&req->rl_segments[i], r_xprt, NULL);
-		/* Optionally wait (not here) for unbinds to complete */
-		rep->rr_func = rpcrdma_unbind_func;
-		(void) rpcrdma_deregister_external(&req->rl_segments[i],
-						   r_xprt, rep);
-		break;
-	case RPCRDMA_MEMWINDOWS_ASYNC:
-		for (i = 0; req->rl_nchunks--;)
-			i += rpcrdma_deregister_external(&req->rl_segments[i],
-							 r_xprt, NULL);
-		break;
-	default:
-		break;
-	}
+	cwnd = xprt->cwnd;
+	xprt->cwnd = atomic_read(&r_xprt->rx_buf.rb_credits) << RPC_CWNDSHIFT;
+	if (xprt->cwnd > cwnd)
+		xprt_release_rqst_cong(rqst->rq_task);
 
 	dprintk("RPC:       %s: xprt_complete_rqst(0x%p, 0x%p, %d)\n",
 			__func__, xprt, rqst, status);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index 8d904e4eef15..8f92a61ee2df 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
  * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -69,7 +70,8 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
 
 	/* Set up the XDR head */
 	rqstp->rq_arg.head[0].iov_base = page_address(page);
-	rqstp->rq_arg.head[0].iov_len = min(byte_count, ctxt->sge[0].length);
+	rqstp->rq_arg.head[0].iov_len =
+		min_t(size_t, byte_count, ctxt->sge[0].length);
 	rqstp->rq_arg.len = byte_count;
 	rqstp->rq_arg.buflen = byte_count;
 
@@ -85,7 +87,7 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
 		page = ctxt->pages[sge_no];
 		put_page(rqstp->rq_pages[sge_no]);
 		rqstp->rq_pages[sge_no] = page;
-		bc -= min(bc, ctxt->sge[sge_no].length);
+		bc -= min_t(u32, bc, ctxt->sge[sge_no].length);
 		rqstp->rq_arg.buflen += ctxt->sge[sge_no].length;
 		sge_no++;
 	}
@@ -113,291 +115,265 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
 	rqstp->rq_arg.tail[0].iov_len = 0;
 }
 
-/* Encode a read-chunk-list as an array of IB SGE
- *
- * Assumptions:
- * - chunk[0]->position points to pages[0] at an offset of 0
- * - pages[] is not physically or virtually contiguous and consists of
- *   PAGE_SIZE elements.
- *
- * Output:
- * - sge array pointing into pages[] array.
- * - chunk_sge array specifying sge index and count for each
- *   chunk in the read list
- *
- */
-static int map_read_chunks(struct svcxprt_rdma *xprt,
-			   struct svc_rqst *rqstp,
-			   struct svc_rdma_op_ctxt *head,
-			   struct rpcrdma_msg *rmsgp,
-			   struct svc_rdma_req_map *rpl_map,
-			   s