netfilter: ctnetlink: deliver events for conntracks changed from userspace

As for now, the creation and update of conntracks via ctnetlink do not
propagate an event to userspace. This can result in inconsistent situations
if several userspace processes modify the connection tracking table by means
of ctnetlink at the same time. Specifically, using the conntrack command
line tool and conntrackd at the same time can trigger unconsistencies.

This patch also modifies the event cache infrastructure to pass the
process PID and the ECHO flag to nfnetlink_send() to report back
to userspace if the process that triggered the change needs so.
Based on a suggestion from Patrick McHardy.

Signed-off-by: Pablo Neira Ayuso <[e-mail]>
Signed-off-by: Patrick McHardy <[e-mail]>

include/net/netfilter/nf_conntrack.h

···
 
 extern void nf_conntrack_hash_insert(struct nf_conn *ct);
 
-extern void nf_conntrack_flush(struct net *net);
+extern void nf_conntrack_flush(struct net *net, u32 pid, int report);
 
 extern bool nf_ct_get_tuplepr(const struct sk_buff *skb,
 			      unsigned int nhoff, u_int16_t l3num,

include/net/netfilter/nf_conntrack_ecache.h

···
 	unsigned int events;
 };
 
+/* This structure is passed to event handler */
+struct nf_ct_event {
+	struct nf_conn *ct;
+	u32 pid;
+	int report;
+};
+
 extern struct atomic_notifier_head nf_conntrack_chain;
 extern int nf_conntrack_register_notifier(struct notifier_block *nb);
 extern int nf_conntrack_unregister_notifier(struct notifier_block *nb);
···
 	local_bh_enable();
 }
 
-static inline void nf_conntrack_event(enum ip_conntrack_events event,
-				      struct nf_conn *ct)
+static inline void
+nf_conntrack_event_report(enum ip_conntrack_events event,
+			  struct nf_conn *ct,
+			  u32 pid,
+			  int report)
 {
+	struct nf_ct_event item = {
+		.ct 	= ct,
+		.pid	= pid,
+		.report = report
+	};
 	if (nf_ct_is_confirmed(ct) && !nf_ct_is_dying(ct))
-		atomic_notifier_call_chain(&nf_conntrack_chain, event, ct);
+		atomic_notifier_call_chain(&nf_conntrack_chain, event, &item);
 }
+
+static inline void
+nf_conntrack_event(enum ip_conntrack_events event, struct nf_conn *ct)
+{
+	nf_conntrack_event_report(event, ct, 0, 0);
+}
+
+struct nf_exp_event {
+	struct nf_conntrack_expect *exp;
+	u32 pid;
+	int report;
+};
 
 extern struct atomic_notifier_head nf_ct_expect_chain;
 extern int nf_ct_expect_register_notifier(struct notifier_block *nb);
 extern int nf_ct_expect_unregister_notifier(struct notifier_block *nb);
 
 static inline void
+nf_ct_expect_event_report(enum ip_conntrack_expect_events event,
+			  struct nf_conntrack_expect *exp,
+			  u32 pid,
+			  int report)
+{
+	struct nf_exp_event item = {
+		.exp	= exp,
+		.pid	= pid,
+		.report = report
+	};
+	atomic_notifier_call_chain(&nf_ct_expect_chain, event, &item);
+}
+
+static inline void
 nf_ct_expect_event(enum ip_conntrack_expect_events event,
 		   struct nf_conntrack_expect *exp)
 {
-	atomic_notifier_call_chain(&nf_ct_expect_chain, event, exp);
+	nf_ct_expect_event_report(event, exp, 0, 0);
 }
 
 extern int nf_conntrack_ecache_init(struct net *net);
···
 					    struct nf_conn *ct) {}
 static inline void nf_conntrack_event(enum ip_conntrack_events event,
 				      struct nf_conn *ct) {}
+static inline void nf_conntrack_event_report(enum ip_conntrack_events event,
+					     struct nf_conn *ct,
+					     u32 pid,
+					     int report) {}
 static inline void nf_ct_deliver_cached_events(const struct nf_conn *ct) {}
 static inline void nf_ct_expect_event(enum ip_conntrack_expect_events event,
 				      struct nf_conntrack_expect *exp) {}
+static inline void nf_ct_expect_event_report(enum ip_conntrack_expect_events e,
+					     struct nf_conntrack_expect *exp,
+ 					     u32 pid,
+ 					     int report) {}
 static inline void nf_ct_event_cache_flush(struct net *net) {}
 
 static inline int nf_conntrack_ecache_init(struct net *net)

include/net/netfilter/nf_conntrack_expect.h

···
 		       u_int8_t, const __be16 *, const __be16 *);
 void nf_ct_expect_put(struct nf_conntrack_expect *exp);
 int nf_ct_expect_related(struct nf_conntrack_expect *expect);
+int nf_ct_expect_related_report(struct nf_conntrack_expect *expect, 
+				u32 pid, int report);
 
 #endif /*_NF_CONNTRACK_EXPECT_H*/
 

net/netfilter/nf_conntrack_core.c

···
 	NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 	NF_CT_ASSERT(!timer_pending(&ct->timeout));
 
-	nf_conntrack_event(IPCT_DESTROY, ct);
+	if (!test_bit(IPS_DYING_BIT, &ct->status))
+		nf_conntrack_event(IPCT_DESTROY, ct);
 	set_bit(IPS_DYING_BIT, &ct->status);
 
 	/* To make sure we don't get any weird locking issues here:
···
 }
 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 
+struct __nf_ct_flush_report {
+	u32 pid;
+	int report;
+};
+
 static int kill_all(struct nf_conn *i, void *data)
 {
+	struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
+
+	/* get_next_corpse sets the dying bit for us */
+	nf_conntrack_event_report(IPCT_DESTROY,
+				  i,
+				  fr->pid,
+				  fr->report);
 	return 1;
 }
 
···
 }
 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
 
-void nf_conntrack_flush(struct net *net)
+void nf_conntrack_flush(struct net *net, u32 pid, int report)
 {
-	nf_ct_iterate_cleanup(net, kill_all, NULL);
+	struct __nf_ct_flush_report fr = {
+		.pid 	= pid,
+		.report = report,
+	};
+	nf_ct_iterate_cleanup(net, kill_all, &fr);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
 
···
 	nf_ct_event_cache_flush(net);
 	nf_conntrack_ecache_fini(net);
  i_see_dead_people:
-	nf_conntrack_flush(net);
+	nf_conntrack_flush(net, 0, 0);
 	if (atomic_read(&net->ct.count) != 0) {
 		schedule();
 		goto i_see_dead_people;

net/netfilter/nf_conntrack_ecache.c

···
 __nf_ct_deliver_cached_events(struct nf_conntrack_ecache *ecache)
 {
 	if (nf_ct_is_confirmed(ecache->ct) && !nf_ct_is_dying(ecache->ct)
-	    && ecache->events)
-		atomic_notifier_call_chain(&nf_conntrack_chain, ecache->events,
-				    ecache->ct);
+	    && ecache->events) {
+		struct nf_ct_event item = {
+			.ct 	= ecache->ct,
+			.pid	= 0,
+			.report	= 0
+		};
+
+		atomic_notifier_call_chain(&nf_conntrack_chain,
+					   ecache->events,
+					   &item);
+	}
 
 	ecache->events = 0;
 	nf_ct_put(ecache->ct);

net/netfilter/nf_conntrack_expect.c

···
 	return 1;
 }
 
-int nf_ct_expect_related(struct nf_conntrack_expect *expect)
+static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
 {
 	const struct nf_conntrack_expect_policy *p;
 	struct nf_conntrack_expect *i;
···
 	struct net *net = nf_ct_exp_net(expect);
 	struct hlist_node *n;
 	unsigned int h;
-	int ret;
+	int ret = 0;
 
-	NF_CT_ASSERT(master_help);
-
-	spin_lock_bh(&nf_conntrack_lock);
 	if (!master_help->helper) {
 		ret = -ESHUTDOWN;
 		goto out;
···
 			printk(KERN_WARNING
 			       "nf_conntrack: expectation table full\n");
 		ret = -EMFILE;
-		goto out;
 	}
+out:
+	return ret;
+}
+
+int nf_ct_expect_related(struct nf_conntrack_expect *expect)
+{
+	int ret;
+
+	spin_lock_bh(&nf_conntrack_lock);
+	ret = __nf_ct_expect_check(expect);
+	if (ret < 0)
+		goto out;
 
 	nf_ct_expect_insert(expect);
+	atomic_inc(&expect->use);
+	spin_unlock_bh(&nf_conntrack_lock);
 	nf_ct_expect_event(IPEXP_NEW, expect);
-	ret = 0;
+	nf_ct_expect_put(expect);
+	return ret;
 out:
 	spin_unlock_bh(&nf_conntrack_lock);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(nf_ct_expect_related);
+
+int nf_ct_expect_related_report(struct nf_conntrack_expect *expect, 
+				u32 pid, int report)
+{
+	int ret;
+
+	spin_lock_bh(&nf_conntrack_lock);
+	ret = __nf_ct_expect_check(expect);
+	if (ret < 0)
+		goto out;
+	nf_ct_expect_insert(expect);
+out:
+	spin_unlock_bh(&nf_conntrack_lock);
+	if (ret == 0)
+		nf_ct_expect_event_report(IPEXP_NEW, expect, pid, report);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
 
 #ifdef CONFIG_PROC_FS
 struct ct_expect_iter_state {

net/netfilter/nf_conntrack_netlink.c

···
 	struct nlmsghdr *nlh;
 	struct nfgenmsg *nfmsg;
 	struct nlattr *nest_parms;
-	struct nf_conn *ct = (struct nf_conn *)ptr;
+	struct nf_ct_event *item = (struct nf_ct_event *)ptr;
+	struct nf_conn *ct = item->ct;
 	struct sk_buff *skb;
 	unsigned int type;
 	sk_buff_data_t b;
···
 	b = skb->tail;
 
 	type |= NFNL_SUBSYS_CTNETLINK << 8;
-	nlh   = NLMSG_PUT(skb, 0, 0, type, sizeof(struct nfgenmsg));
+	nlh   = NLMSG_PUT(skb, item->pid, 0, type, sizeof(struct nfgenmsg));
 	nfmsg = NLMSG_DATA(nlh);
 
 	nlh->nlmsg_flags    = flags;
···
 	rcu_read_unlock();
 
 	nlh->nlmsg_len = skb->tail - b;
-	nfnetlink_send(skb, 0, group, 0);
+	nfnetlink_send(skb, item->pid, group, item->report);
 	return NOTIFY_DONE;
 
 nla_put_failure:
···
 		err = ctnetlink_parse_tuple(cda, &tuple, CTA_TUPLE_REPLY, u3);
 	else {
 		/* Flush the whole table */
-		nf_conntrack_flush(&init_net);
+		nf_conntrack_flush(&init_net, 
+				   NETLINK_CB(skb).pid, 
+				   nlmsg_report(nlh));
 		return 0;
 	}
 
···
 			return -ENOENT;
 		}
 	}
+
+	nf_conntrack_event_report(IPCT_DESTROY,
+				  ct,
+				  NETLINK_CB(skb).pid,
+				  nlmsg_report(nlh));
+
+	/* death_by_timeout would report the event again */
+	set_bit(IPS_DYING_BIT, &ct->status);
 
 	nf_ct_kill(ct);
 	nf_ct_put(ct);
···
 	return 0;
 }
 
+static inline void
+ctnetlink_event_report(struct nf_conn *ct, u32 pid, int report)
+{
+	unsigned int events = 0;
+
+	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
+		events |= IPCT_RELATED;
+	else
+		events |= IPCT_NEW;
+
+	nf_conntrack_event_report(IPCT_STATUS |
+				  IPCT_HELPER |
+				  IPCT_REFRESH |
+				  IPCT_PROTOINFO |
+				  IPCT_NATSEQADJ |
+				  IPCT_MARK |
+				  events,
+				  ct,
+				  pid,
+				  report);
+}
+
 static int
 ctnetlink_create_conntrack(struct nlattr *cda[],
 			   struct nf_conntrack_tuple *otuple,
 			   struct nf_conntrack_tuple *rtuple,
-			   struct nf_conn *master_ct)
+			   struct nf_conn *master_ct,
+			   u32 pid,
+			   int report)
 {
 	struct nf_conn *ct;
 	int err = -EINVAL;
···
 		ct->master = master_ct;
 	}
 
+	nf_conntrack_get(&ct->ct_general);
 	add_timer(&ct->timeout);
 	nf_conntrack_hash_insert(ct);
 	rcu_read_unlock();
+	ctnetlink_event_report(ct, pid, report);
+	nf_ct_put(ct);
 
 	return 0;
 
···
 			err = ctnetlink_create_conntrack(cda,
 							 &otuple,
 							 &rtuple,
-							 master_ct);
+							 master_ct,
+							 NETLINK_CB(skb).pid,
+							 nlmsg_report(nlh));
 		if (err < 0 && master_ct)
 			nf_ct_put(master_ct);
 
···
 	 * so there's no need to increase the refcount */
 	err = -EEXIST;
 	if (!(nlh->nlmsg_flags & NLM_F_EXCL)) {
+		struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
+
 		/* we only allow nat config for new conntracks */
 		if (cda[CTA_NAT_SRC] || cda[CTA_NAT_DST]) {
 			err = -EOPNOTSUPP;
···
 			err = -EOPNOTSUPP;
 			goto out_unlock;
 		}
-		err = ctnetlink_change_conntrack(nf_ct_tuplehash_to_ctrack(h),
-						 cda);
+
+		err = ctnetlink_change_conntrack(ct, cda);
+		if (err == 0) {
+			nf_conntrack_get(&ct->ct_general);
+			spin_unlock_bh(&nf_conntrack_lock);
+			ctnetlink_event_report(ct,
+					       NETLINK_CB(skb).pid,
+					       nlmsg_report(nlh));
+			nf_ct_put(ct);
+		} else
+			spin_unlock_bh(&nf_conntrack_lock);
+
+		return err;
 	}
 
 out_unlock:
···
 {
 	struct nlmsghdr *nlh;
 	struct nfgenmsg *nfmsg;
-	struct nf_conntrack_expect *exp = (struct nf_conntrack_expect *)ptr;
+	struct nf_exp_event *item = (struct nf_exp_event *)ptr;
+	struct nf_conntrack_expect *exp = item->exp;
 	struct sk_buff *skb;
 	unsigned int type;
 	sk_buff_data_t b;
···
 	b = skb->tail;
 
 	type |= NFNL_SUBSYS_CTNETLINK_EXP << 8;
-	nlh   = NLMSG_PUT(skb, 0, 0, type, sizeof(struct nfgenmsg));
+	nlh   = NLMSG_PUT(skb, item->pid, 0, type, sizeof(struct nfgenmsg));
 	nfmsg = NLMSG_DATA(nlh);
 
 	nlh->nlmsg_flags    = flags;
···
 	rcu_read_unlock();
 
 	nlh->nlmsg_len = skb->tail - b;
-	nfnetlink_send(skb, 0, NFNLGRP_CONNTRACK_EXP_NEW, 0);
+	nfnetlink_send(skb, item->pid, NFNLGRP_CONNTRACK_EXP_NEW, item->report);
 	return NOTIFY_DONE;
 
 nla_put_failure:
···
 }
 
 static int
-ctnetlink_create_expect(struct nlattr *cda[], u_int8_t u3)
+ctnetlink_create_expect(struct nlattr *cda[], u_int8_t u3, u32 pid, int report)
 {
 	struct nf_conntrack_tuple tuple, mask, master_tuple;
 	struct nf_conntrack_tuple_hash *h = NULL;
···
 	memcpy(&exp->mask.src.u3, &mask.src.u3, sizeof(exp->mask.src.u3));
 	exp->mask.src.u.all = mask.src.u.all;
 
-	err = nf_ct_expect_related(exp);
+	err = nf_ct_expect_related_report(exp, pid, report);
 	nf_ct_expect_put(exp);
 
 out:
···
 	if (!exp) {
 		spin_unlock_bh(&nf_conntrack_lock);
 		err = -ENOENT;
-		if (nlh->nlmsg_flags & NLM_F_CREATE)
-			err = ctnetlink_create_expect(cda, u3);
+		if (nlh->nlmsg_flags & NLM_F_CREATE) {
+			err = ctnetlink_create_expect(cda,
+						      u3,
+						      NETLINK_CB(skb).pid,
+						      nlmsg_report(nlh));
+		}
 		return err;
 	}