OSDN Git Service

netfilter: ctnetlink: don't use conntrack/expect object addresses as id
authorFlorian Westphal <fw@strlen.de>
Fri, 16 Aug 2019 23:01:34 +0000 (00:01 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 25 Aug 2019 08:51:42 +0000 (10:51 +0200)
commit 3c79107631db1f7fd32cf3f7368e4672004a3010 upstream.

else, we leak the addresses to userspace via ctnetlink events
and dumps.

Compute an ID on demand based on the immutable parts of nf_conn struct.

Another advantage compared to using an address is that there is no
immediate re-use of the same ID in case the conntrack entry is freed and
reallocated again immediately.

Fixes: 3583240249ef ("[NETFILTER]: nf_conntrack_expect: kill unique ID")
Fixes: 7f85f914721f ("[NETFILTER]: nf_conntrack: kill unique ID")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/net/netfilter/nf_conntrack.h
net/netfilter/nf_conntrack_core.c
net/netfilter/nf_conntrack_netlink.c

index 9ae819e..b57a9f3 100644 (file)
@@ -336,6 +336,8 @@ struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
                                 gfp_t flags);
 void nf_ct_tmpl_free(struct nf_conn *tmpl);
 
+u32 nf_ct_get_id(const struct nf_conn *ct);
+
 #define NF_CT_STAT_INC(net, count)       __this_cpu_inc((net)->ct.stat->count)
 #define NF_CT_STAT_INC_ATOMIC(net, count) this_cpu_inc((net)->ct.stat->count)
 #define NF_CT_STAT_ADD_ATOMIC(net, count, v) this_cpu_add((net)->ct.stat->count, (v))
index df1d561..c3d3b10 100644 (file)
@@ -25,6 +25,7 @@
 #include <linux/slab.h>
 #include <linux/random.h>
 #include <linux/jhash.h>
+#include <linux/siphash.h>
 #include <linux/err.h>
 #include <linux/percpu.h>
 #include <linux/moduleparam.h>
@@ -301,6 +302,40 @@ nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 }
 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 
+/* Generate a almost-unique pseudo-id for a given conntrack.
+ *
+ * intentionally doesn't re-use any of the seeds used for hash
+ * table location, we assume id gets exposed to userspace.
+ *
+ * Following nf_conn items do not change throughout lifetime
+ * of the nf_conn after it has been committed to main hash table:
+ *
+ * 1. nf_conn address
+ * 2. nf_conn->ext address
+ * 3. nf_conn->master address (normally NULL)
+ * 4. tuple
+ * 5. the associated net namespace
+ */
+u32 nf_ct_get_id(const struct nf_conn *ct)
+{
+       static __read_mostly siphash_key_t ct_id_seed;
+       unsigned long a, b, c, d;
+
+       net_get_random_once(&ct_id_seed, sizeof(ct_id_seed));
+
+       a = (unsigned long)ct;
+       b = (unsigned long)ct->master ^ net_hash_mix(nf_ct_net(ct));
+       c = (unsigned long)ct->ext;
+       d = (unsigned long)siphash(&ct->tuplehash, sizeof(ct->tuplehash),
+                                  &ct_id_seed);
+#ifdef CONFIG_64BIT
+       return siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &ct_id_seed);
+#else
+       return siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &ct_id_seed);
+#endif
+}
+EXPORT_SYMBOL_GPL(nf_ct_get_id);
+
 static void
 clean_from_lists(struct nf_conn *ct)
 {
index 9e30fd0..deea281 100644 (file)
@@ -29,6 +29,7 @@
 #include <linux/spinlock.h>
 #include <linux/interrupt.h>
 #include <linux/slab.h>
+#include <linux/siphash.h>
 
 #include <linux/netfilter.h>
 #include <net/netlink.h>
@@ -441,7 +442,9 @@ static int ctnetlink_dump_ct_seq_adj(struct sk_buff *skb,
 
 static int ctnetlink_dump_id(struct sk_buff *skb, const struct nf_conn *ct)
 {
-       if (nla_put_be32(skb, CTA_ID, htonl((unsigned long)ct)))
+       __be32 id = (__force __be32)nf_ct_get_id(ct);
+
+       if (nla_put_be32(skb, CTA_ID, id))
                goto nla_put_failure;
        return 0;
 
@@ -1166,8 +1169,9 @@ static int ctnetlink_del_conntrack(struct net *net, struct sock *ctnl,
        ct = nf_ct_tuplehash_to_ctrack(h);
 
        if (cda[CTA_ID]) {
-               u_int32_t id = ntohl(nla_get_be32(cda[CTA_ID]));
-               if (id != (u32)(unsigned long)ct) {
+               __be32 id = nla_get_be32(cda[CTA_ID]);
+
+               if (id != (__force __be32)nf_ct_get_id(ct)) {
                        nf_ct_put(ct);
                        return -ENOENT;
                }
@@ -2472,6 +2476,25 @@ nla_put_failure:
 
 static const union nf_inet_addr any_addr;
 
+static __be32 nf_expect_get_id(const struct nf_conntrack_expect *exp)
+{
+       static __read_mostly siphash_key_t exp_id_seed;
+       unsigned long a, b, c, d;
+
+       net_get_random_once(&exp_id_seed, sizeof(exp_id_seed));
+
+       a = (unsigned long)exp;
+       b = (unsigned long)exp->helper;
+       c = (unsigned long)exp->master;
+       d = (unsigned long)siphash(&exp->tuple, sizeof(exp->tuple), &exp_id_seed);
+
+#ifdef CONFIG_64BIT
+       return (__force __be32)siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &exp_id_seed);
+#else
+       return (__force __be32)siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &exp_id_seed);
+#endif
+}
+
 static int
 ctnetlink_exp_dump_expect(struct sk_buff *skb,
                          const struct nf_conntrack_expect *exp)
@@ -2519,7 +2542,7 @@ ctnetlink_exp_dump_expect(struct sk_buff *skb,
        }
 #endif
        if (nla_put_be32(skb, CTA_EXPECT_TIMEOUT, htonl(timeout)) ||
-           nla_put_be32(skb, CTA_EXPECT_ID, htonl((unsigned long)exp)) ||
+           nla_put_be32(skb, CTA_EXPECT_ID, nf_expect_get_id(exp)) ||
            nla_put_be32(skb, CTA_EXPECT_FLAGS, htonl(exp->flags)) ||
            nla_put_be32(skb, CTA_EXPECT_CLASS, htonl(exp->class)))
                goto nla_put_failure;
@@ -2818,7 +2841,8 @@ static int ctnetlink_get_expect(struct net *net, struct sock *ctnl,
 
        if (cda[CTA_EXPECT_ID]) {
                __be32 id = nla_get_be32(cda[CTA_EXPECT_ID]);
-               if (ntohl(id) != (u32)(unsigned long)exp) {
+
+               if (id != nf_expect_get_id(exp)) {
                        nf_ct_expect_put(exp);
                        return -ENOENT;
                }