[DCCP]: Initial feature negotiation implementation

Still needs more work, but boots and doesn't crashes, even
does some negotiation!

18:38:52.174934  127.0.0.1.43458 > 127.0.0.1.5001: request <change_l ack_ratio 2, change_r ccid 2, change_l ccid 2>
18:38:52.218526  127.0.0.1.5001 > 127.0.0.1.43458: response <nop, nop, change_l ack_ratio 2, confirm_r ccid 2 2, confirm_l ccid 2 2, confirm_r ack_ratio 2>
18:38:52.185398  127.0.0.1.43458 > 127.0.0.1.5001: <nop, confirm_r ack_ratio 2, ack_vector0 0x00, elapsed_time 212>

:-)

Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

10 files changed, 813 insertions, 5 deletions

diff --git a/net/dccp/Makefile b/net/dccp/Makefile
index 87b27fff6e3..5736acea1c8 100644
--- a/net/dccp/Makefile
+++ b/net/dccp/Makefile
@@ -4,7 +4,7 @@ dccp_ipv6-y := ipv6.o
 
 obj-$(CONFIG_IP_DCCP) += dccp.o
 
-dccp-y := ccid.o input.o ipv4.o minisocks.o options.o output.o proto.o \
+dccp-y := ccid.o feat.o input.o ipv4.o minisocks.o options.o output.o proto.o \
 	  timer.o
 
 dccp-$(CONFIG_IP_DCCP_ACKVEC) += ackvec.o
diff --git a/net/dccp/feat.c b/net/dccp/feat.c
new file mode 100644
index 00000000000..99d7b7f9efa
--- /dev/null
+++ b/net/dccp/feat.c
@@ -0,0 +1,554 @@
+/*
+ *  net/dccp/feat.c
+ *
+ *  An implementation of the DCCP protocol
+ *  Andrea Bittau <a.bittau@cs.ucl.ac.uk>
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+
+#include "dccp.h"
+#include "feat.h"
+
+#define DCCP_FEAT_SP_NOAGREE (-123)
+
+int dccp_feat_change(struct sock *sk, u8 type, u8 feature, u8 *val, u8 len,
+		     gfp_t gfp)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	struct dccp_opt_pend *opt;
+
+	dccp_pr_debug("feat change type=%d feat=%d\n", type, feature);
+
+	/* check if that feature is already being negotiated */
+	list_for_each_entry(opt, &dp->dccps_options.dccpo_pending,
+			    dccpop_node) {
+		/* ok we found a negotiation for this option already */
+		if (opt->dccpop_feat == feature && opt->dccpop_type == type) {
+			dccp_pr_debug("Replacing old\n");
+			/* replace */
+			BUG_ON(opt->dccpop_val == NULL);
+			kfree(opt->dccpop_val);
+			opt->dccpop_val	 = val;
+			opt->dccpop_len	 = len;
+			opt->dccpop_conf = 0;
+			return 0;
+		}
+	}
+
+	/* negotiation for a new feature */
+	opt = kmalloc(sizeof(*opt), gfp);
+	if (opt == NULL)
+		return -ENOMEM;
+
+	opt->dccpop_type = type;
+	opt->dccpop_feat = feature;
+	opt->dccpop_len	 = len;
+	opt->dccpop_val	 = val;
+	opt->dccpop_conf = 0;
+	opt->dccpop_sc	 = NULL;
+
+	BUG_ON(opt->dccpop_val == NULL);
+
+	list_add_tail(&opt->dccpop_node, &dp->dccps_options.dccpo_pending);
+	return 0;
+}
+
+EXPORT_SYMBOL_GPL(dccp_feat_change);
+
+/* XXX taking only u8 vals */
+static int dccp_feat_update(struct sock *sk, u8 type, u8 feat, u8 val)
+{
+	/* FIXME implement */
+	dccp_pr_debug("changing [%d] feat %d to %d\n", type, feat, val);
+	return 0;
+}
+
+static int dccp_feat_reconcile(struct sock *sk, struct dccp_opt_pend *opt,
+			       u8 *rpref, u8 rlen)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	u8 *spref, slen, *res = NULL;
+	int i, j, rc, agree = 1;
+
+	BUG_ON(rpref == NULL);
+
+	/* check if we are the black sheep */
+	if (dp->dccps_role == DCCP_ROLE_CLIENT) {
+		spref = rpref;
+		slen  = rlen;
+		rpref = opt->dccpop_val;
+		rlen  = opt->dccpop_len;
+	} else {
+		spref = opt->dccpop_val;
+		slen  = opt->dccpop_len;
+	}
+	/*
+	 * Now we have server preference list in spref and client preference in
+	 * rpref
+	 */
+	BUG_ON(spref == NULL);
+	BUG_ON(rpref == NULL);
+
+	/* FIXME sanity check vals */
+
+	/* Are values in any order?  XXX Lame "algorithm" here */
+	/* XXX assume values are 1 byte */
+	for (i = 0; i < slen; i++) {
+		for (j = 0; j < rlen; j++) {
+			if (spref[i] == rpref[j]) {
+				res = &spref[i];
+				break;
+			}
+		}
+		if (res)
+			break;
+	}
+
+	/* we didn't agree on anything */
+	if (res == NULL) {
+		/* confirm previous value */
+		switch (opt->dccpop_feat) {
+		case DCCPF_CCID:
+			/* XXX did i get this right? =P */
+			if (opt->dccpop_type == DCCPO_CHANGE_L)
+				res = &dp->dccps_options.dccpo_tx_ccid;
+			else
+				res = &dp->dccps_options.dccpo_rx_ccid;
+			break;
+
+		default:
+			WARN_ON(1); /* XXX implement res */
+			return -EFAULT;
+		}
+
+		dccp_pr_debug("Don't agree... reconfirming %d\n", *res);
+		agree = 0; /* this is used for mandatory options... */
+	}
+
+	/* need to put result and our preference list */
+	/* XXX assume 1 byte vals */
+	rlen = 1 + opt->dccpop_len;
+	rpref = kmalloc(rlen, GFP_ATOMIC);
+	if (rpref == NULL)
+		return -ENOMEM;
+
+	*rpref = *res;
+	memcpy(&rpref[1], opt->dccpop_val, opt->dccpop_len);
+
+	/* put it in the "confirm queue" */
+	if (opt->dccpop_sc == NULL) {
+		opt->dccpop_sc = kmalloc(sizeof(*opt->dccpop_sc), GFP_ATOMIC);
+		if (opt->dccpop_sc == NULL) {
+			kfree(rpref);
+			return -ENOMEM;
+		}
+	} else {
+		/* recycle the confirm slot */
+		BUG_ON(opt->dccpop_sc->dccpoc_val == NULL);
+		kfree(opt->dccpop_sc->dccpoc_val);
+		dccp_pr_debug("recycling confirm slot\n");
+	}
+	memset(opt->dccpop_sc, 0, sizeof(*opt->dccpop_sc));
+
+	opt->dccpop_sc->dccpoc_val = rpref;
+	opt->dccpop_sc->dccpoc_len = rlen;
+
+	/* update the option on our side [we are about to send the confirm] */
+	rc = dccp_feat_update(sk, opt->dccpop_type, opt->dccpop_feat, *res);
+	if (rc) {
+		kfree(opt->dccpop_sc->dccpoc_val);
+		kfree(opt->dccpop_sc);
+		opt->dccpop_sc = 0;
+		return rc;
+	}
+
+	dccp_pr_debug("Will confirm %d\n", *rpref);
+
+	/* say we want to change to X but we just got a confirm X, suppress our
+	 * change
+	 */
+	if (!opt->dccpop_conf) {
+		if (*opt->dccpop_val == *res)
+			opt->dccpop_conf = 1;
+		dccp_pr_debug("won't ask for change of same feature\n");
+	}
+
+	return agree ? 0 : DCCP_FEAT_SP_NOAGREE; /* used for mandatory opts */
+}
+
+static int dccp_feat_sp(struct sock *sk, u8 type, u8 feature, u8 *val, u8 len)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	struct dccp_opt_pend *opt;
+	int rc = 1;
+	u8 t;
+
+	/*
+	 * We received a CHANGE.  We gotta match it against our own preference
+	 * list.  If we got a CHANGE_R it means it's a change for us, so we need
+	 * to compare our CHANGE_L list.
+	 */
+	if (type == DCCPO_CHANGE_L)
+		t = DCCPO_CHANGE_R;
+	else
+		t = DCCPO_CHANGE_L;
+
+	/* find our preference list for this feature */
+	list_for_each_entry(opt, &dp->dccps_options.dccpo_pending,
+			    dccpop_node) {
+		if (opt->dccpop_type != t || opt->dccpop_feat != feature)
+			continue;
+
+		/* find the winner from the two preference lists */
+		rc = dccp_feat_reconcile(sk, opt, val, len);
+		break;
+	}
+
+	/* We didn't deal with the change.  This can happen if we have no
+	 * preference list for the feature.  In fact, it just shouldn't
+	 * happen---if we understand a feature, we should have a preference list
+	 * with at least the default value.
+	 */
+	BUG_ON(rc == 1);
+
+	return rc;
+}
+
+static int dccp_feat_nn(struct sock *sk, u8 type, u8 feature, u8 *val, u8 len)
+{
+	struct dccp_opt_pend *opt;
+	struct dccp_sock *dp = dccp_sk(sk);
+	u8 *copy;
+	int rc;
+
+	/* NN features must be change L */
+	if (type == DCCPO_CHANGE_R) {
+		dccp_pr_debug("received CHANGE_R %d for NN feat %d\n",
+			      type, feature);
+		return -EFAULT;
+	}
+
+	/* XXX sanity check opt val */
+
+	/* copy option so we can confirm it */
+	opt = kzalloc(sizeof(*opt), GFP_ATOMIC);
+	if (opt == NULL)
+		return -ENOMEM;
+
+	copy = kmalloc(len, GFP_ATOMIC);
+	if (copy == NULL) {
+		kfree(opt);
+		return -ENOMEM;
+	}
+	memcpy(copy, val, len);
+
+	opt->dccpop_type = DCCPO_CONFIRM_R; /* NN can only confirm R */
+	opt->dccpop_feat = feature;
+	opt->dccpop_val	 = copy;
+	opt->dccpop_len	 = len;
+
+	/* change feature */
+	rc = dccp_feat_update(sk, type, feature, *val);
+	if (rc) {
+		kfree(opt->dccpop_val);
+		kfree(opt);
+		return rc;
+	}
+
+	dccp_pr_debug("Confirming NN feature %d (val=%d)\n", feature, *copy);
+	list_add_tail(&opt->dccpop_node, &dp->dccps_options.dccpo_conf);
+
+	return 0;
+}
+
+static void dccp_feat_empty_confirm(struct sock *sk, u8 type, u8 feature)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	/* XXX check if other confirms for that are queued and recycle slot */
+	struct dccp_opt_pend *opt = kzalloc(sizeof(*opt), GFP_ATOMIC);
+
+	if (opt == NULL) {
+		/* XXX what do we do?  Ignoring should be fine.  It's a change
+		 * after all =P
+		 */
+		return;
+	}
+
+	opt->dccpop_type = type == DCCPO_CHANGE_L ? DCCPO_CONFIRM_R :
+						    DCCPO_CONFIRM_L;
+	opt->dccpop_feat = feature;
+	opt->dccpop_val	 = 0;
+	opt->dccpop_len	 = 0;
+
+	/* change feature */
+	dccp_pr_debug("Empty confirm feature %d type %d\n", feature, type);
+	list_add_tail(&opt->dccpop_node, &dp->dccps_options.dccpo_conf);
+}
+
+static void dccp_feat_flush_confirm(struct sock *sk)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	/* Check if there is anything to confirm in the first place */
+	int yes = !list_empty(&dp->dccps_options.dccpo_conf);
+
+	if (!yes) {
+		struct dccp_opt_pend *opt;
+
+		list_for_each_entry(opt, &dp->dccps_options.dccpo_pending,
+				    dccpop_node) {
+			if (opt->dccpop_conf) {
+				yes = 1;
+				break;
+			}
+		}
+	}
+
+	if (!yes)
+		return;
+
+	/* OK there is something to confirm... */
+	/* XXX check if packet is in flight?  Send delayed ack?? */
+	if (sk->sk_state == DCCP_OPEN)
+		dccp_send_ack(sk);
+}
+
+int dccp_feat_change_recv(struct sock *sk, u8 type, u8 feature, u8 *val, u8 len)
+{
+	int rc;
+
+	dccp_pr_debug("got feat change type=%d feat=%d\n", type, feature);
+
+	/* figure out if it's SP or NN feature */
+	switch (feature) {
+	/* deal with SP features */
+	case DCCPF_CCID:
+		rc = dccp_feat_sp(sk, type, feature, val, len);
+		break;
+
+	/* deal with NN features */
+	case DCCPF_ACK_RATIO:
+		rc = dccp_feat_nn(sk, type, feature, val, len);
+		break;
+
+	/* XXX implement other features */
+	default:
+		rc = -EFAULT;
+		break;
+	}
+
+	/* check if there were problems changing features */
+	if (rc) {
+		/* If we don't agree on SP, we sent a confirm for old value.
+		 * However we propagate rc to caller in case option was
+		 * mandatory
+		 */
+		if (rc != DCCP_FEAT_SP_NOAGREE)
+			dccp_feat_empty_confirm(sk, type, feature);
+	}
+
+	/* generate the confirm [if required] */
+	dccp_feat_flush_confirm(sk);
+
+	return rc;
+}
+
+EXPORT_SYMBOL_GPL(dccp_feat_change_recv);
+
+int dccp_feat_confirm_recv(struct sock *sk, u8 type, u8 feature,
+			   u8 *val, u8 len)
+{
+	u8 t;
+	struct dccp_opt_pend *opt;
+	struct dccp_sock *dp = dccp_sk(sk);
+	int rc = 1;
+	int all_confirmed = 1;
+
+	dccp_pr_debug("got feat confirm type=%d feat=%d\n", type, feature);
+
+	/* XXX sanity check type & feat */
+
+	/* locate our change request */
+	t = type == DCCPO_CONFIRM_L ? DCCPO_CHANGE_R : DCCPO_CHANGE_L;
+
+	list_for_each_entry(opt, &dp->dccps_options.dccpo_pending,
+			    dccpop_node) {
+		if (!opt->dccpop_conf && opt->dccpop_type == t &&
+		    opt->dccpop_feat == feature) {
+			/* we found it */
+			/* XXX do sanity check */
+
+			opt->dccpop_conf = 1;
+
+			/* We got a confirmation---change the option */
+			dccp_feat_update(sk, opt->dccpop_type,
+					 opt->dccpop_feat, *val);
+
+			dccp_pr_debug("feat %d type %d confirmed %d\n",
+				      feature, type, *val);
+			rc = 0;
+			break;
+		}
+
+		if (!opt->dccpop_conf)
+			all_confirmed = 0;
+	}
+
+	/* fix re-transmit timer */
+	/* XXX gotta make sure that no option negotiation occurs during
+	 * connection shutdown.  Consider that the CLOSEREQ is sent and timer is
+	 * on.  if all options are confirmed it might kill timer which should
+	 * remain alive until close is received.
+	 */
+	if (all_confirmed) {
+		dccp_pr_debug("clear feat negotiation timer %p\n", sk);
+		inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
+	}
+
+	if (rc)
+		dccp_pr_debug("feat %d type %d never requested\n",
+			      feature, type);
+	return 0;
+}
+
+EXPORT_SYMBOL_GPL(dccp_feat_confirm_recv);
+
+void dccp_feat_clean(struct sock *sk)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	struct dccp_opt_pend *opt, *next;
+
+	list_for_each_entry_safe(opt, next, &dp->dccps_options.dccpo_pending,
+				 dccpop_node) {
+                BUG_ON(opt->dccpop_val == NULL);
+                kfree(opt->dccpop_val);
+
+		if (opt->dccpop_sc != NULL) {
+			BUG_ON(opt->dccpop_sc->dccpoc_val == NULL);
+			kfree(opt->dccpop_sc->dccpoc_val);
+			kfree(opt->dccpop_sc);
+		}
+
+                kfree(opt);
+        }
+	INIT_LIST_HEAD(&dp->dccps_options.dccpo_pending);
+
+	list_for_each_entry_safe(opt, next, &dp->dccps_options.dccpo_conf,
+				 dccpop_node) {
+		BUG_ON(opt == NULL);
+		if (opt->dccpop_val != NULL)
+			kfree(opt->dccpop_val);
+		kfree(opt);
+	}
+	INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
+}
+
+EXPORT_SYMBOL_GPL(dccp_feat_clean);
+
+/* this is to be called only when a listening sock creates its child.  It is
+ * assumed by the function---the confirm is not duplicated, but rather it is
+ * "passed on".
+ */
+int dccp_feat_clone(struct sock *oldsk, struct sock *newsk)
+{
+	struct dccp_sock *olddp = dccp_sk(oldsk);
+	struct dccp_sock *newdp = dccp_sk(newsk);
+	struct dccp_opt_pend *opt;
+	int rc = 0;
+
+	INIT_LIST_HEAD(&newdp->dccps_options.dccpo_pending);
+	INIT_LIST_HEAD(&newdp->dccps_options.dccpo_conf);
+
+	list_for_each_entry(opt, &olddp->dccps_options.dccpo_pending,
+			    dccpop_node) {
+		struct dccp_opt_pend *newopt;
+		/* copy the value of the option */
+		u8 *val = kmalloc(opt->dccpop_len, GFP_ATOMIC);
+
+		if (val == NULL)
+			goto out_clean;
+		memcpy(val, opt->dccpop_val, opt->dccpop_len);
+
+		newopt = kmalloc(sizeof(*newopt), GFP_ATOMIC);
+		if (newopt == NULL) {
+			kfree(val);
+			goto out_clean;
+		}
+
+		/* insert the option */
+		memcpy(newopt, opt, sizeof(*newopt));
+		newopt->dccpop_val = val;
+		list_add_tail(&newopt->dccpop_node,
+			      &newdp->dccps_options.dccpo_pending);
+
+		/* XXX what happens with backlogs and multiple connections at
+		 * once...
+		 */
+		/* the master socket no longer needs to worry about confirms */
+		opt->dccpop_sc = 0; /* it's not a memleak---new socket has it */
+
+		/* reset state for a new socket */
+		opt->dccpop_conf = 0;
+	}
+
+	/* XXX not doing anything about the conf queue */
+
+out:
+	return rc;
+
+out_clean:
+	dccp_feat_clean(newsk);
+	rc = -ENOMEM;
+	goto out;
+}
+
+EXPORT_SYMBOL_GPL(dccp_feat_clone);
+
+static int __dccp_feat_init(struct sock *sk, u8 type, u8 feat, u8 *val, u8 len)
+{
+	int rc = -ENOMEM;
+	u8 *copy = kmalloc(len, GFP_KERNEL);
+
+	if (copy != NULL) {
+		memcpy(copy, val, len);
+		rc = dccp_feat_change(sk, type, feat, copy, len, GFP_KERNEL);
+		if (rc)
+			kfree(copy);
+	}
+	return rc;
+}
+
+int dccp_feat_init(struct sock *sk)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	int rc;
+
+	INIT_LIST_HEAD(&dp->dccps_options.dccpo_pending);
+	INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
+
+	/* CCID L */
+	rc = __dccp_feat_init(sk, DCCPO_CHANGE_L, DCCPF_CCID,
+			      &dp->dccps_options.dccpo_tx_ccid, 1);
+	if (rc)
+		goto out;
+
+	/* CCID R */
+	rc = __dccp_feat_init(sk, DCCPO_CHANGE_R, DCCPF_CCID,
+			      &dp->dccps_options.dccpo_rx_ccid, 1);
+	if (rc)
+		goto out;
+
+	/* Ack ratio */
+	rc = __dccp_feat_init(sk, DCCPO_CHANGE_L, DCCPF_ACK_RATIO,
+			      &dp->dccps_options.dccpo_ack_ratio, 1);
+out:
+	return rc;
+}
+
+EXPORT_SYMBOL_GPL(dccp_feat_init);
diff --git a/net/dccp/feat.h b/net/dccp/feat.h
new file mode 100644
index 00000000000..67da06265f1
--- /dev/null
+++ b/net/dccp/feat.h
@@ -0,0 +1,28 @@
+#ifndef _DCCP_FEAT_H
+#define _DCCP_FEAT_H
+/*
+ *  net/dccp/feat.h
+ *
+ *  An implementation of the DCCP protocol
+ *  Copyright (c) 2005 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
+ *
+ *	This program is free software; you can redistribute it and/or modify it
+ *	under the terms of the GNU General Public License version 2 as
+ *	published by the Free Software Foundation.
+ */
+
+#include <linux/types.h>
+
+struct sock;
+
+extern int  dccp_feat_change(struct sock *sk, u8 type, u8 feature,
+			     u8 *val, u8 len, gfp_t gfp);
+extern int  dccp_feat_change_recv(struct sock *sk, u8 type, u8 feature,
+				  u8 *val, u8 len);
+extern int  dccp_feat_confirm_recv(struct sock *sk, u8 type, u8 feature,
+				   u8 *val, u8 len);
+extern void dccp_feat_clean(struct sock *sk);
+extern int  dccp_feat_clone(struct sock *oldsk, struct sock *newsk);
+extern int  dccp_feat_init(struct sock *sk);
+
+#endif /* _DCCP_FEAT_H */
diff --git a/net/dccp/input.c b/net/dccp/input.c
index b6cba72b44e..4b6d43d8b92 100644
--- a/net/dccp/input.c
+++ b/net/dccp/input.c
@@ -300,6 +300,9 @@ static int dccp_rcv_request_sent_state_process(struct sock *sk,
 			goto out_invalid_packet;
 		}
 
+		if (dccp_parse_options(sk, skb))
+			goto out_invalid_packet;
+
                 if (dp->dccps_options.dccpo_send_ack_vector &&
                     dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
                                     DCCP_SKB_CB(skb)->dccpd_seq,
diff --git a/net/dccp/ipv4.c b/net/dccp/ipv4.c
index 38321ad8187..fcfb486f90c 100644
--- a/net/dccp/ipv4.c
+++ b/net/dccp/ipv4.c
@@ -28,6 +28,7 @@
 #include "ackvec.h"
 #include "ccid.h"
 #include "dccp.h"
+#include "feat.h"
 
 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
 	.lhash_lock	= RW_LOCK_UNLOCKED,
@@ -535,7 +536,8 @@ int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 	if (req == NULL)
 		goto drop;
 
-	/* FIXME: process options */
+	if (dccp_parse_options(sk, skb))
+		goto drop;
 
 	dccp_openreq_init(req, &dp, skb);
 
@@ -1049,6 +1051,8 @@ int dccp_v4_init_sock(struct sock *sk)
 	 * setsockopt(CCIDs-I-want/accept). -acme
 	 */
 	if (likely(!dccp_ctl_socket_init)) {
+		int rc;
+
 		if (dp->dccps_options.dccpo_send_ack_vector) {
 			dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
 			if (dp->dccps_hc_rx_ackvec == NULL)
@@ -1069,8 +1073,16 @@ int dccp_v4_init_sock(struct sock *sk)
 			dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
 			return -ENOMEM;
 		}
-	} else
+
+		rc = dccp_feat_init(sk);
+		if (rc)
+			return rc;
+	} else {
+		/* control socket doesn't need feat nego */
+		INIT_LIST_HEAD(&dp->dccps_options.dccpo_pending);
+		INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
 		dccp_ctl_socket_init = 0;
+	}
 
 	dccp_init_xmit_timers(sk);
 	icsk->icsk_rto = DCCP_TIMEOUT_INIT;
@@ -1118,6 +1130,9 @@ int dccp_v4_destroy_sock(struct sock *sk)
 	ccid_exit(dp->dccps_hc_tx_ccid, sk);
 	dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
 
+	/* clean up feature negotiation state */
+	dccp_feat_clean(sk);
+
 	return 0;
 }
 
diff --git a/net/dccp/minisocks.c b/net/dccp/minisocks.c
index a60a3e948c3..9e1de5919ee 100644
--- a/net/dccp/minisocks.c
+++ b/net/dccp/minisocks.c
@@ -22,6 +22,7 @@
 #include "ackvec.h"
 #include "ccid.h"
 #include "dccp.h"
+#include "feat.h"
 
 struct inet_timewait_death_row dccp_death_row = {
 	.sysctl_max_tw_buckets = NR_FILE * 2,
@@ -114,6 +115,9 @@ struct sock *dccp_create_openreq_child(struct sock *sk,
 		newicsk->icsk_rto	   = DCCP_TIMEOUT_INIT;
 		do_gettimeofday(&newdp->dccps_epoch);
 
+		if (dccp_feat_clone(sk, newsk))
+			goto out_free;
+
 		if (newdp->dccps_options.dccpo_send_ack_vector) {
 			newdp->dccps_hc_rx_ackvec =
 						dccp_ackvec_alloc(GFP_ATOMIC);
diff --git a/net/dccp/options.c b/net/dccp/options.c
index 0a76426c9ae..7f99306c8e9 100644
--- a/net/dccp/options.c
+++ b/net/dccp/options.c
@@ -21,12 +21,14 @@
 #include "ackvec.h"
 #include "ccid.h"
 #include "dccp.h"
+#include "feat.h"
 
 /* stores the default values for new connection. may be changed with sysctl */
 static const struct dccp_options dccpo_default_values = {
 	.dccpo_sequence_window	  = DCCPF_INITIAL_SEQUENCE_WINDOW,
 	.dccpo_rx_ccid		  = DCCPF_INITIAL_CCID,
 	.dccpo_tx_ccid		  = DCCPF_INITIAL_CCID,
+	.dccpo_ack_ratio	  = DCCPF_INITIAL_ACK_RATIO,
 	.dccpo_send_ack_vector	  = DCCPF_INITIAL_SEND_ACK_VECTOR,
 	.dccpo_send_ndp_count	  = DCCPF_INITIAL_SEND_NDP_COUNT,
 };
@@ -69,6 +71,8 @@ int dccp_parse_options(struct sock *sk, struct sk_buff *skb)
 	unsigned char opt, len;
 	unsigned char *value;
 	u32 elapsed_time;
+	int rc;
+	int mandatory = 0;
 
 	memset(opt_recv, 0, sizeof(*opt_recv));
 
@@ -100,6 +104,11 @@ int dccp_parse_options(struct sock *sk, struct sk_buff *skb)
 		switch (opt) {
 		case DCCPO_PADDING:
 			break;
+		case DCCPO_MANDATORY:
+			if (mandatory)
+				goto out_invalid_option;
+			mandatory = 1;
+			break;
 		case DCCPO_NDP_COUNT:
 			if (len > 3)
 				goto out_invalid_option;
@@ -108,6 +117,31 @@ int dccp_parse_options(struct sock *sk, struct sk_buff *skb)
 			dccp_pr_debug("%sNDP count=%d\n", debug_prefix,
 				      opt_recv->dccpor_ndp);
 			break;
+		case DCCPO_CHANGE_L:
+			/* fall through */
+		case DCCPO_CHANGE_R:
+			if (len < 2)
+				goto out_invalid_option;
+			rc = dccp_feat_change_recv(sk, opt, *value, value + 1,
+						   len - 1);
+			/*
+			 * When there is a change error, change_recv is
+			 * responsible for dealing with it.  i.e. reply with an
+			 * empty confirm.
+			 * If the change was mandatory, then we need to die.
+			 */
+			if (rc && mandatory)
+				goto out_invalid_option;
+			break;
+		case DCCPO_CONFIRM_L:
+			/* fall through */
+		case DCCPO_CONFIRM_R:
+			if (len < 2)
+				goto out_invalid_option;
+			if (dccp_feat_confirm_recv(sk, opt, *value,
+						   value + 1, len - 1))
+				goto out_invalid_option;
+			break;
 		case DCCPO_ACK_VECTOR_0:
 		case DCCPO_ACK_VECTOR_1:
 			if (pkt_type == DCCP_PKT_DATA)
@@ -208,6 +242,9 @@ int dccp_parse_options(struct sock *sk, struct sk_buff *skb)
 				sk, opt, len);
 			break;
 	        }
+
+		if (opt != DCCPO_MANDATORY)
+			mandatory = 0;
 	}
 
 	return 0;
@@ -356,7 +393,7 @@ void dccp_insert_option_timestamp(struct sock *sk, struct sk_buff *skb)
 {
 	struct timeval tv;
 	u32 now;
-	
+
 	dccp_timestamp(sk, &tv);
 	now = timeval_usecs(&tv) / 10;
 	/* yes this will overflow but that is the point as we want a
@@ -402,7 +439,7 @@ static void dccp_insert_option_timestamp_echo(struct sock *sk,
 	tstamp_echo = htonl(dp->dccps_timestamp_echo);
 	memcpy(to, &tstamp_echo, 4);
 	to += 4;
-	
+
 	if (elapsed_time_len == 2) {
 		const u16 var16 = htons((u16)elapsed_time);
 		memcpy(to, &var16, 2);
@@ -421,6 +458,93 @@ static void dccp_insert_option_timestamp_echo(struct sock *sk,
 	dp->dccps_timestamp_time.tv_usec = 0;
 }
 
+static int dccp_insert_feat_opt(struct sk_buff *skb, u8 type, u8 feat,
+			        u8 *val, u8 len)
+{
+	u8 *to;
+
+	if (DCCP_SKB_CB(skb)->dccpd_opt_len + len + 3 > DCCP_MAX_OPT_LEN) {
+		LIMIT_NETDEBUG(KERN_INFO "DCCP: packet too small"
+			       " to insert feature %d option!\n", feat);
+		return -1;
+	}
+
+	DCCP_SKB_CB(skb)->dccpd_opt_len += len + 3;
+
+	to    = skb_push(skb, len + 3);
+	*to++ = type;
+	*to++ = len + 3;
+	*to++ = feat;
+
+	if (len)
+		memcpy(to, val, len);
+	dccp_pr_debug("option %d feat %d len %d\n", type, feat, len);
+
+	return 0;
+}
+
+static void dccp_insert_feat(struct sock *sk, struct sk_buff *skb)
+{
+	struct dccp_sock *dp = dccp_sk(sk);
+	struct dccp_opt_pend *opt, *next;
+	int change = 0;
+
+	/* confirm any options [NN opts] */
+	list_for_each_entry_safe(opt, next, &dp->dccps_options.dccpo_conf,
+				 dccpop_node) {
+		dccp_insert_feat_opt(skb, opt->dccpop_type,
+				     opt->dccpop_feat, opt->dccpop_val,
+				     opt->dccpop_len);
+		/* fear empty confirms */
+		if (opt->dccpop_val)
+			kfree(opt->dccpop_val);
+		kfree(opt);
+	}
+	INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
+
+	/* see which features we need to send */
+	list_for_each_entry(opt, &dp->dccps_options.dccpo_pending,
+			    dccpop_node) {
+		/* see if we need to send any confirm */
+		if (opt->dccpop_sc) {
+			dccp_insert_feat_opt(skb, opt->dccpop_type + 1,
+					     opt->dccpop_feat,
+					     opt->dccpop_sc->dccpoc_val,
+					     opt->dccpop_sc->dccpoc_len);
+
+			BUG_ON(!opt->dccpop_sc->dccpoc_val);
+			kfree(opt->dccpop_sc->dccpoc_val);
+			kfree(opt->dccpop_sc);
+			opt->dccpop_sc = NULL;
+		}
+
+		/* any option not confirmed, re-send it */
+		if (!opt->dccpop_conf) {
+			dccp_insert_feat_opt(skb, opt->dccpop_type,
+					     opt->dccpop_feat, opt->dccpop_val,
+					     opt->dccpop_len);
+			change++;
+		}
+	}
+
+	/* Retransmit timer.
+	 * If this is the master listening sock, we don't set a timer on it.  It
+	 * should be fine because if the dude doesn't receive our RESPONSE
+	 * [which will contain the CHANGE] he will send another REQUEST which
+	 * will "retrnasmit" the change.
+	 */
+	if (change && dp->dccps_role != DCCP_ROLE_LISTEN) {
+		dccp_pr_debug("reset feat negotiation timer %p\n", sk);
+
+		/* XXX don't reset the timer on re-transmissions.  I.e. reset it
+		 * only when sending new stuff i guess.  Currently the timer
+		 * never backs off because on re-transmission it just resets it!
+		 */
+		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+					  inet_csk(sk)->icsk_rto, DCCP_RTO_MAX);
+	}
+}
+
 void dccp_insert_options(struct sock *sk, struct sk_buff *skb)
 {
 	struct dccp_sock *dp = dccp_sk(sk);
@@ -447,6 +571,17 @@ void dccp_insert_options(struct sock *sk, struct sk_buff *skb)
 		dp->dccps_hc_tx_insert_options = 0;
 	}
 
+	/* Feature negotiation */
+	switch(DCCP_SKB_CB(skb)->dccpd_type) {
+		/* Data packets can't do feat negotiation */
+	case DCCP_PKT_DATA:
+	case DCCP_PKT_DATAACK:
+		break;
+	default:
+		dccp_insert_feat(sk, skb);
+		break;
+	}
+
 	/* XXX: insert other options when appropriate */
 
 	if (DCCP_SKB_CB(skb)->dccpd_opt_len != 0) {
diff --git a/net/dccp/output.c b/net/dccp/output.c
index efd7ffb903a..0cc2bcf5652 100644
--- a/net/dccp/output.c
+++ b/net/dccp/output.c
@@ -64,6 +64,10 @@ static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 		case DCCP_PKT_DATAACK:
 			break;
 
+		case DCCP_PKT_REQUEST:
+			set_ack = 0;
+			/* fall through */
+
 		case DCCP_PKT_SYNC:
 		case DCCP_PKT_SYNCACK:
 			ackno = dcb->dccpd_seq;
diff --git a/net/dccp/proto.c b/net/dccp/proto.c
index 81ad2495371..1a8cf8ecfe6 100644
--- a/net/dccp/proto.c
+++ b/net/dccp/proto.c
@@ -37,6 +37,7 @@
 
 #include "ccid.h"
 #include "dccp.h"
+#include "feat.h"
 
 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
 
@@ -255,6 +256,39 @@ static int dccp_setsockopt_service(struct sock *sk, const u32 service,
 	return 0;
 }
 
+/* byte 1 is feature.  the rest is the preference list */
+static int dccp_setsockopt_change(struct sock *sk, int type,
+				  struct dccp_so_feat __user *optval)
+{
+	struct dccp_so_feat opt;
+	u8 *val;
+	int rc;
+
+	if (copy_from_user(&opt, optval, sizeof(opt)))
+		return -EFAULT;
+
+	val = kmalloc(opt.dccpsf_len, GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	if (copy_from_user(val, opt.dccpsf_val, opt.dccpsf_len)) {
+		rc = -EFAULT;
+		goto out_free_val;
+	}
+
+	rc = dccp_feat_change(sk, type, opt.dccpsf_feat, val, opt.dccpsf_len,
+			      GFP_KERNEL);
+	if (rc)
+		goto out_free_val;
+
+out:
+	return rc;
+
+out_free_val:
+	kfree(val);
+	goto out;
+}
+
 int dccp_setsockopt(struct sock *sk, int level, int optname,
 		    char __user *optval, int optlen)
 {
@@ -284,6 +318,25 @@ int dccp_setsockopt(struct sock *sk, int level, int optname,
 	case DCCP_SOCKOPT_PACKET_SIZE:
 		dp->dccps_packet_size = val;
 		break;
+
+	case DCCP_SOCKOPT_CHANGE_L:
+		if (optlen != sizeof(struct dccp_so_feat))
+			err = -EINVAL;
+		else
+			err = dccp_setsockopt_change(sk, DCCPO_CHANGE_L,
+					             (struct dccp_so_feat *)
+						     optval);
+		break;
+
+	case DCCP_SOCKOPT_CHANGE_R:
+		if (optlen != sizeof(struct dccp_so_feat))
+			err = -EINVAL;
+		else
+			err = dccp_setsockopt_change(sk, DCCPO_CHANGE_R,
+						     (struct dccp_so_feat *)
+						     optval);
+		break;
+
 	default:
 		err = -ENOPROTOOPT;
 		break;
diff --git a/net/dccp/timer.c b/net/dccp/timer.c
index aa34b576e22..d7c786608ec 100644
--- a/net/dccp/timer.c
+++ b/net/dccp/timer.c
@@ -141,6 +141,17 @@ static void dccp_retransmit_timer(struct sock *sk)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
 
+	/* retransmit timer is used for feature negotiation throughout
+	 * connection.  In this case, no packet is re-transmitted, but rather an
+	 * ack is generated and pending changes are splaced into its options.
+	 */
+	if (sk->sk_send_head == NULL) {
+		dccp_pr_debug("feat negotiation retransmit timeout %p\n", sk);
+		if (sk->sk_state == DCCP_OPEN)
+			dccp_send_ack(sk);
+		goto backoff;
+	}
+
 	/*
 	 * sk->sk_send_head has to have one skb with
 	 * DCCP_SKB_CB(skb)->dccpd_type set to one of the retransmittable DCCP
@@ -177,6 +188,7 @@ static void dccp_retransmit_timer(struct sock *sk)
 		goto out;
 	}
 
+backoff:
 	icsk->icsk_backoff++;
 	icsk->icsk_retransmits++;