Merge branch 'hotfixes' into devel

20 files changed, 187 insertions, 116 deletions

diff --git a/net/bluetooth/bnep/bnep.h b/net/bluetooth/bnep/bnep.h
index a299228..e69244dd 100644
--- a/net/bluetooth/bnep/bnep.h
+++ b/net/bluetooth/bnep/bnep.h
@@ -174,7 +174,7 @@ struct bnep_session {
 
 void bnep_net_setup(struct net_device *dev);
 int bnep_sock_init(void);
-int bnep_sock_cleanup(void);
+void bnep_sock_cleanup(void);
 
 static inline int bnep_mc_hash(__u8 *addr)
 {
diff --git a/net/bluetooth/bnep/sock.c b/net/bluetooth/bnep/sock.c
index 81065e5..201e5b1 100644
--- a/net/bluetooth/bnep/sock.c
+++ b/net/bluetooth/bnep/sock.c
@@ -257,12 +257,10 @@ error:
 	return err;
 }
 
-int __exit bnep_sock_cleanup(void)
+void __exit bnep_sock_cleanup(void)
 {
 	if (bt_sock_unregister(BTPROTO_BNEP) < 0)
 		BT_ERR("Can't unregister BNEP socket");
 
 	proto_unregister(&bnep_proto);
-
-	return 0;
 }
diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 930b58e..aec6929 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -902,8 +902,6 @@ int hci_unregister_dev(struct hci_dev *hdev)
 
 	BT_DBG("%p name %s type %d", hdev, hdev->name, hdev->type);
 
-	hci_unregister_sysfs(hdev);
-
 	write_lock_bh(&hci_dev_list_lock);
 	list_del(&hdev->list);
 	write_unlock_bh(&hci_dev_list_lock);
@@ -915,6 +913,8 @@ int hci_unregister_dev(struct hci_dev *hdev)
 
 	hci_notify(hdev, HCI_DEV_UNREG);
 
+	hci_unregister_sysfs(hdev);
+
 	__hci_dev_put(hdev);
 
 	return 0;
diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
index 1499132..b5d4019 100644
--- a/net/bluetooth/hci_sock.c
+++ b/net/bluetooth/hci_sock.c
@@ -734,7 +734,7 @@ error:
 	return err;
 }
 
-int __exit hci_sock_cleanup(void)
+void __exit hci_sock_cleanup(void)
 {
 	if (bt_sock_unregister(BTPROTO_HCI) < 0)
 		BT_ERR("HCI socket unregistration failed");
@@ -742,6 +742,4 @@ int __exit hci_sock_cleanup(void)
 	hci_unregister_notifier(&hci_sock_nblock);
 
 	proto_unregister(&hci_sk_proto);
-
-	return 0;
 }
diff --git a/net/core/sock.c b/net/core/sock.c
index 09cb3a7..2654c14 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -1621,7 +1621,7 @@ static void sock_def_readable(struct sock *sk, int len)
 {
 	read_lock(&sk->sk_callback_lock);
 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
-		wake_up_interruptible(sk->sk_sleep);
+		wake_up_interruptible_sync(sk->sk_sleep);
 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
 	read_unlock(&sk->sk_callback_lock);
 }
@@ -1635,7 +1635,7 @@ static void sock_def_write_space(struct sock *sk)
 	 */
 	if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
 		if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
-			wake_up_interruptible(sk->sk_sleep);
+			wake_up_interruptible_sync(sk->sk_sleep);
 
 		/* Should agree with poll, otherwise some programs break */
 		if (sock_writeable(sk))
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index ed750f9..01578f5 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1035,6 +1035,13 @@ static void tcp_cwnd_validate(struct sock *sk)
  * introducing MSS oddities to segment boundaries. In rare cases where
  * mss_now != mss_cache, we will request caller to create a small skb
  * per input skb which could be mostly avoided here (if desired).
+ *
+ * We explicitly want to create a request for splitting write queue tail
+ * to a small skb for Nagle purposes while avoiding unnecessary modulos,
+ * thus all the complexity (cwnd_len is always MSS multiple which we
+ * return whenever allowed by the other factors). Basically we need the
+ * modulo only when the receiver window alone is the limiting factor or
+ * when we would be allowed to send the split-due-to-Nagle skb fully.
  */
 static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb,
 					unsigned int mss_now, unsigned int cwnd)
@@ -1048,10 +1055,11 @@ static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb,
 	if (likely(cwnd_len <= window && skb != tcp_write_queue_tail(sk)))
 		return cwnd_len;
 
-	if (skb == tcp_write_queue_tail(sk) && cwnd_len <= skb->len)
+	needed = min(skb->len, window);
+
+	if (skb == tcp_write_queue_tail(sk) && cwnd_len <= needed)
 		return cwnd_len;
 
-	needed = min(skb->len, window);
 	return needed - needed % mss_now;
 }
 
diff --git a/net/netfilter/nf_conntrack_expect.c b/net/netfilter/nf_conntrack_expect.c
index e06bf00..684ec9c 100644
--- a/net/netfilter/nf_conntrack_expect.c
+++ b/net/netfilter/nf_conntrack_expect.c
@@ -381,7 +381,7 @@ int nf_ct_expect_related(struct nf_conntrack_expect *expect)
 	if (nf_ct_expect_count >= nf_ct_expect_max) {
 		if (net_ratelimit())
 			printk(KERN_WARNING
-			       "nf_conntrack: expectation table full");
+			       "nf_conntrack: expectation table full\n");
 		ret = -EMFILE;
 		goto out;
 	}
diff --git a/net/netfilter/nf_conntrack_extend.c b/net/netfilter/nf_conntrack_extend.c
index 8b9be1e..2bd9963 100644
--- a/net/netfilter/nf_conntrack_extend.c
+++ b/net/netfilter/nf_conntrack_extend.c
@@ -19,14 +19,6 @@
 static struct nf_ct_ext_type *nf_ct_ext_types[NF_CT_EXT_NUM];
 static DEFINE_MUTEX(nf_ct_ext_type_mutex);
 
-/* Horrible trick to figure out smallest amount worth kmallocing. */
-#define CACHE(x) (x) + 0 *
-enum {
-	NF_CT_EXT_MIN_SIZE =
-#include <linux/kmalloc_sizes.h>
-	1 };
-#undef CACHE
-
 void __nf_ct_ext_destroy(struct nf_conn *ct)
 {
 	unsigned int i;
@@ -53,7 +45,7 @@ EXPORT_SYMBOL(__nf_ct_ext_destroy);
 static void *
 nf_ct_ext_create(struct nf_ct_ext **ext, enum nf_ct_ext_id id, gfp_t gfp)
 {
-	unsigned int off, len, real_len;
+	unsigned int off, len;
 	struct nf_ct_ext_type *t;
 
 	rcu_read_lock();
@@ -61,16 +53,14 @@ nf_ct_ext_create(struct nf_ct_ext **ext, enum nf_ct_ext_id id, gfp_t gfp)
 	BUG_ON(t == NULL);
 	off = ALIGN(sizeof(struct nf_ct_ext), t->align);
 	len = off + t->len;
-	real_len = t->alloc_size;
 	rcu_read_unlock();
 
-	*ext = kzalloc(real_len, gfp);
+	*ext = kzalloc(t->alloc_size, gfp);
 	if (!*ext)
 		return NULL;
 
 	(*ext)->offset[id] = off;
 	(*ext)->len = len;
-	(*ext)->real_len = real_len;
 
 	return (void *)(*ext) + off;
 }
@@ -95,7 +85,7 @@ void *__nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 	newlen = newoff + t->len;
 	rcu_read_unlock();
 
-	if (newlen >= ct->ext->real_len) {
+	if (newlen >= ksize(ct->ext)) {
 		new = kmalloc(newlen, gfp);
 		if (!new)
 			return NULL;
@@ -114,7 +104,6 @@ void *__nf_ct_ext_add(struct nf_conn *ct, enum nf_ct_ext_id id, gfp_t gfp)
 			rcu_read_unlock();
 		}
 		kfree(ct->ext);
-		new->real_len = newlen;
 		ct->ext = new;
 	}
 
@@ -156,8 +145,6 @@ static void update_alloc_size(struct nf_ct_ext_type *type)
 			t1->alloc_size = ALIGN(t1->alloc_size, t2->align)
 					 + t2->len;
 		}
-		if (t1->alloc_size < NF_CT_EXT_MIN_SIZE)
-			t1->alloc_size = NF_CT_EXT_MIN_SIZE;
 	}
 }
 
diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
index bfc2928..ddc80ea 100644
--- a/net/netfilter/nf_queue.c
+++ b/net/netfilter/nf_queue.c
@@ -51,7 +51,7 @@ int nf_unregister_queue_handler(int pf, const struct nf_queue_handler *qh)
 		return -EINVAL;
 
 	mutex_lock(&queue_handler_mutex);
-	if (queue_handler[pf] != qh) {
+	if (queue_handler[pf] && queue_handler[pf] != qh) {
 		mutex_unlock(&queue_handler_mutex);
 		return -EINVAL;
 	}
diff --git a/net/netfilter/nfnetlink_log.c b/net/netfilter/nfnetlink_log.c
index 7efa40d..bf3f19b 100644
--- a/net/netfilter/nfnetlink_log.c
+++ b/net/netfilter/nfnetlink_log.c
@@ -556,7 +556,7 @@ nfulnl_log_packet(unsigned int pf,
 	/* FIXME: do we want to make the size calculation conditional based on
 	 * what is actually present?  way more branches and checks, but more
 	 * memory efficient... */
-	size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
+	size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 		+ nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
@@ -702,20 +702,30 @@ nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 	struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 	u_int16_t group_num = ntohs(nfmsg->res_id);
 	struct nfulnl_instance *inst;
+	struct nfulnl_msg_config_cmd *cmd = NULL;
 	int ret = 0;
 
+	if (nfula[NFULA_CFG_CMD]) {
+		u_int8_t pf = nfmsg->nfgen_family;
+		cmd = nla_data(nfula[NFULA_CFG_CMD]);
+
+		/* Commands without queue context */
+		switch (cmd->command) {
+		case NFULNL_CFG_CMD_PF_BIND:
+			return nf_log_register(pf, &nfulnl_logger);
+		case NFULNL_CFG_CMD_PF_UNBIND:
+			nf_log_unregister_pf(pf);
+			return 0;
+		}
+	}
+
 	inst = instance_lookup_get(group_num);
 	if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
 		ret = -EPERM;
 		goto out_put;
 	}
 
-	if (nfula[NFULA_CFG_CMD]) {
-		u_int8_t pf = nfmsg->nfgen_family;
-		struct nfulnl_msg_config_cmd *cmd;
-
-		cmd = nla_data(nfula[NFULA_CFG_CMD]);
-
+	if (cmd != NULL) {
 		switch (cmd->command) {
 		case NFULNL_CFG_CMD_BIND:
 			if (inst) {
@@ -738,14 +748,6 @@ nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 
 			instance_destroy(inst);
 			goto out;
-		case NFULNL_CFG_CMD_PF_BIND:
-			ret = nf_log_register(pf, &nfulnl_logger);
-			break;
-		case NFULNL_CFG_CMD_PF_UNBIND:
-			/* This is a bug and a feature.  We cannot unregister
-			 * other handlers, like nfnetlink_inst can */
-			nf_log_unregister_pf(pf);
-			break;
 		default:
 			ret = -ENOTSUPP;
 			break;
diff --git a/net/netfilter/nfnetlink_queue.c b/net/netfilter/nfnetlink_queue.c
index 0043d3a..012cb69 100644
--- a/net/netfilter/nfnetlink_queue.c
+++ b/net/netfilter/nfnetlink_queue.c
@@ -224,7 +224,7 @@ nfqnl_build_packet_message(struct nfqnl_instance *queue,
 	struct net_device *indev;
 	struct net_device *outdev;
 
-	size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
+	size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 		+ nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
 		+ nla_total_size(sizeof(u_int32_t))	/* ifindex */
@@ -703,19 +703,12 @@ nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 		/* Commands without queue context - might sleep */
 		switch (cmd->command) {
 		case NFQNL_CFG_CMD_PF_BIND:
-			ret = nf_register_queue_handler(ntohs(cmd->pf),
-							&nfqh);
-			break;
+			return nf_register_queue_handler(ntohs(cmd->pf),
+							 &nfqh);
 		case NFQNL_CFG_CMD_PF_UNBIND:
-			ret = nf_unregister_queue_handler(ntohs(cmd->pf),
-							  &nfqh);
-			break;
-		default:
-			break;
+			return nf_unregister_queue_handler(ntohs(cmd->pf),
+							   &nfqh);
 		}
-
-		if (ret < 0)
-			return ret;
 	}
 
 	rcu_read_lock();
diff --git a/net/netfilter/xt_time.c b/net/netfilter/xt_time.c
index e9a8794..9fa2e08 100644
--- a/net/netfilter/xt_time.c
+++ b/net/netfilter/xt_time.c
@@ -95,8 +95,11 @@ static inline void localtime_2(struct xtm *r, time_t time)
 	 */
 	r->dse = time / 86400;
 
-	/* 1970-01-01 (w=0) was a Thursday (4). */
-	r->weekday = (4 + r->dse) % 7;
+	/*
+	 * 1970-01-01 (w=0) was a Thursday (4).
+	 * -1 and +1 map Sunday properly onto 7.
+	 */
+	r->weekday = (4 + r->dse - 1) % 7 + 1;
 }
 
 static void localtime_3(struct xtm *r, time_t time)
diff --git a/net/rxrpc/ar-recvmsg.c b/net/rxrpc/ar-recvmsg.c
index f19121d..a39bf97 100644
--- a/net/rxrpc/ar-recvmsg.c
+++ b/net/rxrpc/ar-recvmsg.c
@@ -143,7 +143,8 @@ int rxrpc_recvmsg(struct kiocb *iocb, struct socket *sock,
 		/* copy the peer address and timestamp */
 		if (!continue_call) {
 			if (msg->msg_name && msg->msg_namelen > 0)
-				memcpy(&msg->msg_name, &call->conn->trans->peer->srx,
+				memcpy(msg->msg_name,
+				       &call->conn->trans->peer->srx,
 				       sizeof(call->conn->trans->peer->srx));
 			sock_recv_timestamp(msg, &rx->sk, skb);
 		}
diff --git a/net/sctp/bind_addr.c b/net/sctp/bind_addr.c
index a27511e..ceefda0 100644
--- a/net/sctp/bind_addr.c
+++ b/net/sctp/bind_addr.c
@@ -209,6 +209,7 @@ int sctp_add_bind_addr(struct sctp_bind_addr *bp, union sctp_addr *new,
 int sctp_del_bind_addr(struct sctp_bind_addr *bp, union sctp_addr *del_addr)
 {
 	struct sctp_sockaddr_entry *addr, *temp;
+	int found = 0;
 
 	/* We hold the socket lock when calling this function,
 	 * and that acts as a writer synchronizing lock.
@@ -216,13 +217,14 @@ int sctp_del_bind_addr(struct sctp_bind_addr *bp, union sctp_addr *del_addr)
 	list_for_each_entry_safe(addr, temp, &bp->address_list, list) {
 		if (sctp_cmp_addr_exact(&addr->a, del_addr)) {
 			/* Found the exact match. */
+			found = 1;
 			addr->valid = 0;
 			list_del_rcu(&addr->list);
 			break;
 		}
 	}
 
-	if (addr && !addr->valid) {
+	if (found) {
 		call_rcu(&addr->rcu, sctp_local_addr_free);
 		SCTP_DBG_OBJCNT_DEC(addr);
 		return 0;
diff --git a/net/sctp/ipv6.c b/net/sctp/ipv6.c
index 87f9405..9aa0733 100644
--- a/net/sctp/ipv6.c
+++ b/net/sctp/ipv6.c
@@ -89,6 +89,7 @@ static int sctp_inet6addr_event(struct notifier_block *this, unsigned long ev,
 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
 	struct sctp_sockaddr_entry *addr = NULL;
 	struct sctp_sockaddr_entry *temp;
+	int found = 0;
 
 	switch (ev) {
 	case NETDEV_UP:
@@ -111,13 +112,14 @@ static int sctp_inet6addr_event(struct notifier_block *this, unsigned long ev,
 					&sctp_local_addr_list, list) {
 			if (ipv6_addr_equal(&addr->a.v6.sin6_addr,
 					     &ifa->addr)) {
+				found = 1;
 				addr->valid = 0;
 				list_del_rcu(&addr->list);
 				break;
 			}
 		}
 		spin_unlock_bh(&sctp_local_addr_lock);
-		if (addr && !addr->valid)
+		if (found)
 			call_rcu(&addr->rcu, sctp_local_addr_free);
 		break;
 	}
diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c
index 688546d..ad0a406 100644
--- a/net/sctp/protocol.c
+++ b/net/sctp/protocol.c
@@ -628,6 +628,7 @@ static int sctp_inetaddr_event(struct notifier_block *this, unsigned long ev,
 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
 	struct sctp_sockaddr_entry *addr = NULL;
 	struct sctp_sockaddr_entry *temp;
+	int found = 0;
 
 	switch (ev) {
 	case NETDEV_UP:
@@ -647,13 +648,14 @@ static int sctp_inetaddr_event(struct notifier_block *this, unsigned long ev,
 		list_for_each_entry_safe(addr, temp,
 					&sctp_local_addr_list, list) {
 			if (addr->a.v4.sin_addr.s_addr == ifa->ifa_local) {
+				found = 1;
 				addr->valid = 0;
 				list_del_rcu(&addr->list);
 				break;
 			}
 		}
 		spin_unlock_bh(&sctp_local_addr_lock);
-		if (addr && !addr->valid)
+		if (found)
 			call_rcu(&addr->rcu, sctp_local_addr_free);
 		break;
 	}
diff --git a/net/sctp/sm_make_chunk.c b/net/sctp/sm_make_chunk.c
index e45be4e..578630e 100644
--- a/net/sctp/sm_make_chunk.c
+++ b/net/sctp/sm_make_chunk.c
@@ -2375,6 +2375,14 @@ static int sctp_process_param(struct sctp_association *asoc,
 		asoc->peer.ipv4_address = 0;
 		asoc->peer.ipv6_address = 0;
 
+		/* Assume that peer supports the address family
+		 * by which it sends a packet.
+		 */
+		if (peer_addr->sa.sa_family == AF_INET6)
+			asoc->peer.ipv6_address = 1;
+		else if (peer_addr->sa.sa_family == AF_INET)
+			asoc->peer.ipv4_address = 1;
+
 		/* Cycle through address types; avoid divide by 0. */
 		sat = ntohs(param.p->length) - sizeof(sctp_paramhdr_t);
 		if (sat)
diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 9398926..d994d82 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -2933,17 +2933,39 @@ static int sctp_setsockopt_maxburst(struct sock *sk,
 				    char __user *optval,
 				    int optlen)
 {
+	struct sctp_assoc_value params;
+	struct sctp_sock *sp;
+	struct sctp_association *asoc;
 	int val;
+	int assoc_id = 0;
 
-	if (optlen != sizeof(int))
+	if (optlen < sizeof(int))
 		return -EINVAL;
-	if (get_user(val, (int __user *)optval))
-		return -EFAULT;
 
-	if (val < 0)
+	if (optlen == sizeof(int)) {
+		printk(KERN_WARNING
+		   "SCTP: Use of int in max_burst socket option deprecated\n");
+		printk(KERN_WARNING
+		   "SCTP: Use struct sctp_assoc_value instead\n");
+		if (copy_from_user(&val, optval, optlen))
+			return -EFAULT;
+	} else if (optlen == sizeof(struct sctp_assoc_value)) {
+		if (copy_from_user(&params, optval, optlen))
+			return -EFAULT;
+		val = params.assoc_value;
+		assoc_id = params.assoc_id;
+	} else
 		return -EINVAL;
 
-	sctp_sk(sk)->max_burst = val;
+	sp = sctp_sk(sk);
+
+	if (assoc_id != 0) {
+		asoc = sctp_id2assoc(sk, assoc_id);
+		if (!asoc)
+			return -EINVAL;
+		asoc->max_burst = val;
+	} else
+		sp->max_burst = val;
 
 	return 0;
 }
@@ -5005,20 +5027,45 @@ static int sctp_getsockopt_maxburst(struct sock *sk, int len,
 				    char __user *optval,
 				    int __user *optlen)
 {
-	int val;
+	struct sctp_assoc_value params;
+	struct sctp_sock *sp;
+	struct sctp_association *asoc;
 
 	if (len < sizeof(int))
 		return -EINVAL;
 
-	len = sizeof(int);
+	if (len == sizeof(int)) {
+		printk(KERN_WARNING
+		   "SCTP: Use of int in max_burst socket option deprecated\n");
+		printk(KERN_WARNING
+		   "SCTP: Use struct sctp_assoc_value instead\n");
+		params.assoc_id = 0;
+	} else if (len == sizeof (struct sctp_assoc_value)) {
+		if (copy_from_user(&params, optval, len))
+			return -EFAULT;
+	} else
+		return -EINVAL;
 
-	val = sctp_sk(sk)->max_burst;
-	if (put_user(len, optlen))
-		return -EFAULT;
-	if (copy_to_user(optval, &val, len))
-		return -EFAULT;
+	sp = sctp_sk(sk);
+
+	if (params.assoc_id != 0) {
+		asoc = sctp_id2assoc(sk, params.assoc_id);
+		if (!asoc)
+			return -EINVAL;
+		params.assoc_value = asoc->max_burst;
+	} else
+		params.assoc_value = sp->max_burst;
+
+	if (len == sizeof(int)) {
+		if (copy_to_user(optval, &params.assoc_value, len))
+			return -EFAULT;
+	} else {
+		if (copy_to_user(optval, &params, len))
+			return -EFAULT;
+	}
+
+	return 0;
 
-	return -ENOTSUPP;
 }
 
 static int sctp_getsockopt_hmac_ident(struct sock *sk, int len,
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 0598b22..981f190 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -156,7 +156,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
 	struct svc_rdma_op_ctxt *ctxt;
 	int ret = 0;
 
-	BUG_ON(sge_count >= 32);
+	BUG_ON(sge_count > RPCSVC_MAXPAGES);
 	dprintk("svcrdma: RDMA_WRITE rmr=%x, to=%llx, xdr_off=%d, "
 		"write_len=%d, xdr_sge=%p, sge_count=%d\n",
 		rmr, (unsigned long long)to, xdr_off,
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index f09444c..16fd3f6 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -54,7 +54,6 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 					int flags);
 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
 static void svc_rdma_release_rqst(struct svc_rqst *);
-static void rdma_destroy_xprt(struct svcxprt_rdma *xprt);
 static void dto_tasklet_func(unsigned long data);
 static void svc_rdma_detach(struct svc_xprt *xprt);
 static void svc_rdma_free(struct svc_xprt *xprt);
@@ -247,6 +246,7 @@ static void dto_tasklet_func(unsigned long data)
 			sq_cq_reap(xprt);
 		}
 
+		svc_xprt_put(&xprt->sc_xprt);
 		spin_lock_irqsave(&dto_lock, flags);
 	}
 	spin_unlock_irqrestore(&dto_lock, flags);
@@ -275,8 +275,10 @@ static void rq_comp_handler(struct ib_cq *cq, void *cq_context)
 	 * add it
 	 */
 	spin_lock_irqsave(&dto_lock, flags);
-	if (list_empty(&xprt->sc_dto_q))
+	if (list_empty(&xprt->sc_dto_q)) {
+		svc_xprt_get(&xprt->sc_xprt);
 		list_add_tail(&xprt->sc_dto_q, &dto_xprt_q);
+	}
 	spin_unlock_irqrestore(&dto_lock, flags);
 
 	/* Tasklet does all the work to avoid irqsave locks. */
@@ -386,8 +388,10 @@ static void sq_comp_handler(struct ib_cq *cq, void *cq_context)
 	 * add it
 	 */
 	spin_lock_irqsave(&dto_lock, flags);
-	if (list_empty(&xprt->sc_dto_q))
+	if (list_empty(&xprt->sc_dto_q)) {
+		svc_xprt_get(&xprt->sc_xprt);
 		list_add_tail(&xprt->sc_dto_q, &dto_xprt_q);
+	}
 	spin_unlock_irqrestore(&dto_lock, flags);
 
 	/* Tasklet does all the work to avoid irqsave locks. */
@@ -611,6 +615,7 @@ static int rdma_cma_handler(struct rdma_cm_id *cma_id,
 	switch (event->event) {
 	case RDMA_CM_EVENT_ESTABLISHED:
 		/* Accept complete */
+		svc_xprt_get(xprt);
 		dprintk("svcrdma: Connection completed on DTO xprt=%p, "
 			"cm_id=%p\n", xprt, cma_id);
 		clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
@@ -661,15 +666,15 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 
 	listen_id = rdma_create_id(rdma_listen_handler, cma_xprt, RDMA_PS_TCP);
 	if (IS_ERR(listen_id)) {
-		rdma_destroy_xprt(cma_xprt);
+		svc_xprt_put(&cma_xprt->sc_xprt);
 		dprintk("svcrdma: rdma_create_id failed = %ld\n",
 			PTR_ERR(listen_id));
 		return (void *)listen_id;
 	}
 	ret = rdma_bind_addr(listen_id, sa);
 	if (ret) {
-		rdma_destroy_xprt(cma_xprt);
 		rdma_destroy_id(listen_id);
+		svc_xprt_put(&cma_xprt->sc_xprt);
 		dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
 		return ERR_PTR(ret);
 	}
@@ -678,8 +683,9 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 	ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
 	if (ret) {
 		rdma_destroy_id(listen_id);
-		rdma_destroy_xprt(cma_xprt);
+		svc_xprt_put(&cma_xprt->sc_xprt);
 		dprintk("svcrdma: rdma_listen failed = %d\n", ret);
+		return ERR_PTR(ret);
 	}
 
 	/*
@@ -820,6 +826,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 		newxprt->sc_sq_depth = qp_attr.cap.max_send_wr;
 		newxprt->sc_max_requests = qp_attr.cap.max_recv_wr;
 	}
+	svc_xprt_get(&newxprt->sc_xprt);
 	newxprt->sc_qp = newxprt->sc_cm_id->qp;
 
 	/* Register all of physical memory */
@@ -891,8 +898,15 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 
  errout:
 	dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
+	/* Take a reference in case the DTO handler runs */
+	svc_xprt_get(&newxprt->sc_xprt);
+	if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp)) {
+		ib_destroy_qp(newxprt->sc_qp);
+		svc_xprt_put(&newxprt->sc_xprt);
+	}
 	rdma_destroy_id(newxprt->sc_cm_id);
-	rdma_destroy_xprt(newxprt);
+	/* This call to put will destroy the transport */
+	svc_xprt_put(&newxprt->sc_xprt);
 	return NULL;
 }
 
@@ -919,54 +933,60 @@ static void svc_rdma_release_rqst(struct svc_rqst *rqstp)
 	rqstp->rq_xprt_ctxt = NULL;
 }
 
-/* Disable data ready events for this connection */
+/*
+ * When connected, an svc_xprt has at least three references:
+ *
+ * - A reference held by the QP. We still hold that here because this
+ *   code deletes the QP and puts the reference.
+ *
+ * - A reference held by the cm_id between the ESTABLISHED and
+ *   DISCONNECTED events. If the remote peer disconnected first, this
+ *   reference could be gone.
+ *
+ * - A reference held by the svc_recv code that called this function
+ *   as part of close processing.
+ *
+ * At a minimum two references should still be held.
+ */
 static void svc_rdma_detach(struct svc_xprt *xprt)
 {
 	struct svcxprt_rdma *rdma =
 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
-	unsigned long flags;
-
 	dprintk("svc: svc_rdma_detach(%p)\n", xprt);
-	/*
-	 * Shutdown the connection. This will ensure we don't get any
-	 * more events from the provider.
-	 */
+
+	/* Disconnect and flush posted WQE */
 	rdma_disconnect(rdma->sc_cm_id);
-	rdma_destroy_id(rdma->sc_cm_id);
 
-	/* We may already be on the DTO list */
-	spin_lock_irqsave(&dto_lock, flags);
-	if (!list_empty(&rdma->sc_dto_q))
-		list_del_init(&rdma->sc_dto_q);
-	spin_unlock_irqrestore(&dto_lock, flags);
+	/* Destroy the QP if present (not a listener) */
+	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp)) {
+		ib_destroy_qp(rdma->sc_qp);
+		svc_xprt_put(xprt);
+	}
+
+	/* Destroy the CM ID */
+	rdma_destroy_id(rdma->sc_cm_id);
 }
 
 static void svc_rdma_free(struct svc_xprt *xprt)
 {
 	struct svcxprt_rdma *rdma = (struct svcxprt_rdma *)xprt;
 	dprintk("svcrdma: svc_rdma_free(%p)\n", rdma);
-	rdma_destroy_xprt(rdma);
-	kfree(rdma);
-}
-
-static void rdma_destroy_xprt(struct svcxprt_rdma *xprt)
-{
-	if (xprt->sc_qp && !IS_ERR(xprt->sc_qp))
-		ib_destroy_qp(xprt->sc_qp);
-
-	if (xprt->sc_sq_cq && !IS_ERR(xprt->sc_sq_cq))
-		ib_destroy_cq(xprt->sc_sq_cq);
+	/* We should only be called from kref_put */
+	BUG_ON(atomic_read(&xprt->xpt_ref.refcount) != 0);
+	if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
+		ib_destroy_cq(rdma->sc_sq_cq);
 
-	if (xprt->sc_rq_cq && !IS_ERR(xprt->sc_rq_cq))
-		ib_destroy_cq(xprt->sc_rq_cq);
+	if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
+		ib_destroy_cq(rdma->sc_rq_cq);
 
-	if (xprt->sc_phys_mr && !IS_ERR(xprt->sc_phys_mr))
-		ib_dereg_mr(xprt->sc_phys_mr);
+	if (rdma->sc_phys_mr && !IS_ERR(rdma->sc_phys_mr))
+		ib_dereg_mr(rdma->sc_phys_mr);
 
-	if (xprt->sc_pd && !IS_ERR(xprt->sc_pd))
-		ib_dealloc_pd(xprt->sc_pd);
+	if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
+		ib_dealloc_pd(rdma->sc_pd);
 
-	destroy_context_cache(xprt->sc_ctxt_head);
+	destroy_context_cache(rdma->sc_ctxt_head);
+	kfree(rdma);
 }
 
 static int svc_rdma_has_wspace(struct svc_xprt *xprt)