netfilter: nf_nat: fix NAT issue in 2.6.30.4+
authorJozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Fri, 6 Nov 2009 08:43:42 +0000 (00:43 -0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 6 Nov 2009 08:43:42 +0000 (00:43 -0800)
Vitezslav Samel discovered that since 2.6.30.4+ active FTP can not work
over NAT. The "cause" of the problem was a fix of unacknowledged data
detection with NAT (commit a3a9f79e361e864f0e9d75ebe2a0cb43d17c4272).
However, actually, that fix uncovered a long standing bug in TCP conntrack:
when NAT was enabled, we simply updated the max of the right edge of
the segments we have seen (td_end), by the offset NAT produced with
changing IP/port in the data. However, we did not update the other parameter
(td_maxend) which is affected by the NAT offset. Thus that could drift
away from the correct value and thus resulted breaking active FTP.

The patch below fixes the issue by *not* updating the conntrack parameters
from NAT, but instead taking into account the NAT offsets in conntrack in a
consistent way. (Updating from NAT would be more harder and expensive because
it'd need to re-calculate parameters we already calculated in conntrack.)

Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/netfilter/nf_conntrack.h
include/net/netfilter/nf_nat_helper.h
net/ipv4/netfilter/nf_nat_core.c
net/ipv4/netfilter/nf_nat_helper.c
net/netfilter/nf_conntrack_core.c
net/netfilter/nf_conntrack_proto_tcp.c

index cbdd6284996d3cd29732c0ae406833a678a3fb80..5cf7270e3ffc3e94c6d80bd79168056e9c1e539f 100644 (file)
@@ -255,11 +255,9 @@ static inline bool nf_ct_kill(struct nf_conn *ct)
 }
 
 /* These are for NAT.  Icky. */
-/* Update TCP window tracking data when NAT mangles the packet */
-extern void nf_conntrack_tcp_update(const struct sk_buff *skb,
-                                   unsigned int dataoff,
-                                   struct nf_conn *ct, int dir,
-                                   s16 offset);
+extern s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
+                              enum ip_conntrack_dir dir,
+                              u32 seq);
 
 /* Fake conntrack entry for untracked connections */
 extern struct nf_conn nf_conntrack_untracked;
index 237a961f40e1fb23a20951ea166dacd8053790fc..4222220920a58cf616ddf0c26cba889f78d9b39d 100644 (file)
@@ -32,4 +32,8 @@ extern int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  * to port ct->master->saved_proto. */
 extern void nf_nat_follow_master(struct nf_conn *ct,
                                 struct nf_conntrack_expect *this);
+
+extern s16 nf_nat_get_offset(const struct nf_conn *ct,
+                            enum ip_conntrack_dir dir,
+                            u32 seq);
 #endif
index 68afc6ecd34341d976406204fba2807802f819bf..fe1a64479dd088de80cd51226db941609bfff658 100644 (file)
@@ -750,6 +750,8 @@ static int __init nf_nat_init(void)
        BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
        rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
                           nfnetlink_parse_nat_setup);
+       BUG_ON(nf_ct_nat_offset != NULL);
+       rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset);
        return 0;
 
  cleanup_extend:
@@ -764,6 +766,7 @@ static void __exit nf_nat_cleanup(void)
        nf_ct_extend_unregister(&nat_extend);
        rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
        rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
+       rcu_assign_pointer(nf_ct_nat_offset, NULL);
        synchronize_net();
 }
 
index 09172a65d9b61ea8de3cd52301e5c1e064d5b123..f9520fa3aba99d928fd656e40799c33a47fa40be 100644 (file)
@@ -73,6 +73,28 @@ adjust_tcp_sequence(u32 seq,
        DUMP_OFFSET(this_way);
 }
 
+/* Get the offset value, for conntrack */
+s16 nf_nat_get_offset(const struct nf_conn *ct,
+                     enum ip_conntrack_dir dir,
+                     u32 seq)
+{
+       struct nf_conn_nat *nat = nfct_nat(ct);
+       struct nf_nat_seq *this_way;
+       s16 offset;
+
+       if (!nat)
+               return 0;
+
+       this_way = &nat->seq[dir];
+       spin_lock_bh(&nf_nat_seqofs_lock);
+       offset = after(seq, this_way->correction_pos)
+                ? this_way->offset_after : this_way->offset_before;
+       spin_unlock_bh(&nf_nat_seqofs_lock);
+
+       return offset;
+}
+EXPORT_SYMBOL_GPL(nf_nat_get_offset);
+
 /* Frobs data inside this packet, which is linear. */
 static void mangle_contents(struct sk_buff *skb,
                            unsigned int dataoff,
@@ -189,11 +211,6 @@ nf_nat_mangle_tcp_packet(struct sk_buff *skb,
                adjust_tcp_sequence(ntohl(tcph->seq),
                                    (int)rep_len - (int)match_len,
                                    ct, ctinfo);
-               /* Tell TCP window tracking about seq change */
-               nf_conntrack_tcp_update(skb, ip_hdrlen(skb),
-                                       ct, CTINFO2DIR(ctinfo),
-                                       (int)rep_len - (int)match_len);
-
                nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
        }
        return 1;
@@ -415,12 +432,7 @@ nf_nat_seq_adjust(struct sk_buff *skb,
        tcph->seq = newseq;
        tcph->ack_seq = newack;
 
-       if (!nf_nat_sack_adjust(skb, tcph, ct, ctinfo))
-               return 0;
-
-       nf_conntrack_tcp_update(skb, ip_hdrlen(skb), ct, dir, seqoff);
-
-       return 1;
+       return nf_nat_sack_adjust(skb, tcph, ct, ctinfo);
 }
 
 /* Setup NAT on this expected conntrack so it follows master. */
index 7c9ec3dee96e3d17687087d6baf0fd2e594ccfe2..0cdfb388a191e57c41b6ac56e78adf99f88c8b88 100644 (file)
@@ -1350,6 +1350,11 @@ err_stat:
        return ret;
 }
 
+s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
+                       enum ip_conntrack_dir dir,
+                       u32 seq);
+EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
+
 int nf_conntrack_init(struct net *net)
 {
        int ret;
@@ -1367,6 +1372,9 @@ int nf_conntrack_init(struct net *net)
                /* For use by REJECT target */
                rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
                rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
+
+               /* Howto get NAT offsets */
+               rcu_assign_pointer(nf_ct_nat_offset, NULL);
        }
        return 0;
 
index 97a82ba75376aff182fecb1ac37b97e6ca1bf021..ba2b769372834dafcb10bc124027842a86027850 100644 (file)
@@ -492,6 +492,21 @@ static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
        }
 }
 
+#ifdef CONFIG_NF_NAT_NEEDED
+static inline s16 nat_offset(const struct nf_conn *ct,
+                            enum ip_conntrack_dir dir,
+                            u32 seq)
+{
+       typeof(nf_ct_nat_offset) get_offset = rcu_dereference(nf_ct_nat_offset);
+
+       return get_offset != NULL ? get_offset(ct, dir, seq) : 0;
+}
+#define NAT_OFFSET(pf, ct, dir, seq) \
+       (pf == NFPROTO_IPV4 ? nat_offset(ct, dir, seq) : 0)
+#else
+#define NAT_OFFSET(pf, ct, dir, seq)   0
+#endif
+
 static bool tcp_in_window(const struct nf_conn *ct,
                          struct ip_ct_tcp *state,
                          enum ip_conntrack_dir dir,
@@ -506,6 +521,7 @@ static bool tcp_in_window(const struct nf_conn *ct,
        struct ip_ct_tcp_state *receiver = &state->seen[!dir];
        const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
        __u32 seq, ack, sack, end, win, swin;
+       s16 receiver_offset;
        bool res;
 
        /*
@@ -519,11 +535,16 @@ static bool tcp_in_window(const struct nf_conn *ct,
        if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
                tcp_sack(skb, dataoff, tcph, &sack);
 
+       /* Take into account NAT sequence number mangling */
+       receiver_offset = NAT_OFFSET(pf, ct, !dir, ack - 1);
+       ack -= receiver_offset;
+       sack -= receiver_offset;
+
        pr_debug("tcp_in_window: START\n");
        pr_debug("tcp_in_window: ");
        nf_ct_dump_tuple(tuple);
-       pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
-                seq, ack, sack, win, end);
+       pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
+                seq, ack, receiver_offset, sack, receiver_offset, win, end);
        pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
                 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
                 sender->td_end, sender->td_maxend, sender->td_maxwin,
@@ -613,8 +634,8 @@ static bool tcp_in_window(const struct nf_conn *ct,
 
        pr_debug("tcp_in_window: ");
        nf_ct_dump_tuple(tuple);
-       pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
-                seq, ack, sack, win, end);
+       pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
+                seq, ack, receiver_offset, sack, receiver_offset, win, end);
        pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
                 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
                 sender->td_end, sender->td_maxend, sender->td_maxwin,
@@ -700,7 +721,7 @@ static bool tcp_in_window(const struct nf_conn *ct,
                        before(seq, sender->td_maxend + 1) ?
                        after(end, sender->td_end - receiver->td_maxwin - 1) ?
                        before(sack, receiver->td_end + 1) ?
-                       after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
+                       after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
                        : "ACK is under the lower bound (possible overly delayed ACK)"
                        : "ACK is over the upper bound (ACKed data not seen yet)"
                        : "SEQ is under the lower bound (already ACKed data retransmitted)"
@@ -715,39 +736,6 @@ static bool tcp_in_window(const struct nf_conn *ct,
        return res;
 }
 
-#ifdef CONFIG_NF_NAT_NEEDED
-/* Update sender->td_end after NAT successfully mangled the packet */
-/* Caller must linearize skb at tcp header. */
-void nf_conntrack_tcp_update(const struct sk_buff *skb,
-                            unsigned int dataoff,
-                            struct nf_conn *ct, int dir,
-                            s16 offset)
-{
-       const struct tcphdr *tcph = (const void *)skb->data + dataoff;
-       const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
-       const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
-       __u32 end;
-
-       end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
-
-       spin_lock_bh(&ct->lock);
-       /*
-        * We have to worry for the ack in the reply packet only...
-        */
-       if (ct->proto.tcp.seen[dir].td_end + offset == end)
-               ct->proto.tcp.seen[dir].td_end = end;
-       ct->proto.tcp.last_end = end;
-       spin_unlock_bh(&ct->lock);
-       pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
-                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
-                sender->td_end, sender->td_maxend, sender->td_maxwin,
-                sender->td_scale,
-                receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
-                receiver->td_scale);
-}
-EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
-#endif
-
 #define        TH_FIN  0x01
 #define        TH_SYN  0x02
 #define        TH_RST  0x04