]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/commitdiff
netfilter: kill the fake untracked conntrack objects
authorFlorian Westphal <fw@strlen.de>
Fri, 14 Apr 2017 18:31:08 +0000 (20:31 +0200)
committerPablo Neira Ayuso <pablo@netfilter.org>
Sat, 15 Apr 2017 09:47:57 +0000 (11:47 +0200)
resurrect an old patch from Pablo Neira to remove the untracked objects.

Currently, there are four possible states of an skb wrt. conntrack.

1. No conntrack attached, ct is NULL.
2. Normal (kmem cache allocated) ct attached.
3. a template (kmalloc'd), not in any hash tables at any point in time
4. the 'untracked' conntrack, a percpu nf_conn object, tagged via
   IPS_UNTRACKED_BIT in ct->status.

Untracked is supposed to be identical to case 1.  It exists only
so users can check

-m conntrack --ctstate UNTRACKED vs.
-m conntrack --ctstate INVALID

e.g. attempts to set connmark on INVALID or UNTRACKED conntracks is
supposed to be a no-op.

Thus currently we need to check
 ct == NULL || nf_ct_is_untracked(ct)

in a lot of places in order to avoid altering untracked objects.

The other consequence of the percpu untracked object is that all
-j NOTRACK (and, later, kfree_skb of such skbs) result in an atomic op
(inc/dec the untracked conntracks refcount).

This adds a new kernel-private ctinfo state, IP_CT_UNTRACKED, to
make the distinction instead.

The (few) places that care about packet invalid (ct is NULL) vs.
packet untracked now need to test ct == NULL vs. ctinfo == IP_CT_UNTRACKED,
but all other places can omit the nf_ct_is_untracked() check.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
12 files changed:
include/net/ip_vs.h
include/net/netfilter/nf_conntrack.h
include/uapi/linux/netfilter/nf_conntrack_common.h
net/ipv4/netfilter/nf_dup_ipv4.c
net/ipv6/netfilter/nf_conntrack_proto_icmpv6.c
net/ipv6/netfilter/nf_dup_ipv6.c
net/netfilter/nf_conntrack_core.c
net/netfilter/nf_nat_core.c
net/netfilter/nft_ct.c
net/netfilter/xt_CT.c
net/netfilter/xt_conntrack.c
net/netfilter/xt_state.c

index 8a4a57b887fb508c732b7e24cd4f9dd888a6941c..9a75d9933e6362cfb8f46ae853371f8524068713 100644 (file)
@@ -1556,12 +1556,8 @@ static inline void ip_vs_notrack(struct sk_buff *skb)
        struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 
        if (!ct || !nf_ct_is_untracked(ct)) {
-               struct nf_conn *untracked;
-
                nf_conntrack_put(&ct->ct_general);
-               untracked = nf_ct_untracked_get();
-               nf_conntrack_get(&untracked->ct_general);
-               nf_ct_set(skb, untracked, IP_CT_NEW);
+               nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
        }
 #endif
 }
index 19605878da4739d04f0642b20f8641ed8601d2eb..012b99f563e536ee60f536a89522a713cebcec38 100644 (file)
@@ -243,14 +243,6 @@ extern s32 (*nf_ct_nat_offset)(const struct nf_conn *ct,
                               enum ip_conntrack_dir dir,
                               u32 seq);
 
-/* Fake conntrack entry for untracked connections */
-DECLARE_PER_CPU_ALIGNED(struct nf_conn, nf_conntrack_untracked);
-static inline struct nf_conn *nf_ct_untracked_get(void)
-{
-       return raw_cpu_ptr(&nf_conntrack_untracked);
-}
-void nf_ct_untracked_status_or(unsigned long bits);
-
 /* Iterate over all conntracks: if iter returns true, it's deleted. */
 void nf_ct_iterate_cleanup(struct net *net,
                           int (*iter)(struct nf_conn *i, void *data),
@@ -283,7 +275,7 @@ static inline int nf_ct_is_dying(const struct nf_conn *ct)
 
 static inline int nf_ct_is_untracked(const struct nf_conn *ct)
 {
-       return test_bit(IPS_UNTRACKED_BIT, &ct->status);
+       return false;
 }
 
 /* Packet is received from loopback */
index 6a8e33dd4ecbd83b0d254d4cf2486dfe5cacfd34..b4a0a1940118fb0a9adf51b0a50c9f36846a5698 100644 (file)
@@ -28,12 +28,14 @@ enum ip_conntrack_info {
        /* only for userspace compatibility */
 #ifndef __KERNEL__
        IP_CT_NEW_REPLY = IP_CT_NUMBER,
+#else
+       IP_CT_UNTRACKED = 7,
 #endif
 };
 
 #define NF_CT_STATE_INVALID_BIT                        (1 << 0)
 #define NF_CT_STATE_BIT(ctinfo)                        (1 << ((ctinfo) % IP_CT_IS_REPLY + 1))
-#define NF_CT_STATE_UNTRACKED_BIT              (1 << (IP_CT_NUMBER + 1))
+#define NF_CT_STATE_UNTRACKED_BIT              (1 << (IP_CT_UNTRACKED + 1))
 
 /* Bitset representing status of connection. */
 enum ip_conntrack_status {
@@ -94,7 +96,7 @@ enum ip_conntrack_status {
        IPS_TEMPLATE_BIT = 11,
        IPS_TEMPLATE = (1 << IPS_TEMPLATE_BIT),
 
-       /* Conntrack is a fake untracked entry */
+       /* Conntrack is a fake untracked entry.  Obsolete and not used anymore */
        IPS_UNTRACKED_BIT = 12,
        IPS_UNTRACKED = (1 << IPS_UNTRACKED_BIT),
 
index f0dbff05fc28174de694ecad1d1adcde63c314ec..39895b9ddeb9601307f073298ab690ae1f9972d4 100644 (file)
@@ -69,8 +69,7 @@ void nf_dup_ipv4(struct net *net, struct sk_buff *skb, unsigned int hooknum,
 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
        /* Avoid counting cloned packets towards the original connection. */
        nf_reset(skb);
-       nf_ct_set(skb, nf_ct_untracked_get(), IP_CT_NEW);
-       nf_conntrack_get(skb_nfct(skb));
+       nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 #endif
        /*
         * If we are in PREROUTING/INPUT, decrease the TTL to mitigate potential
index d2c2ccbfbe728f8ff2d7bc42c5e7feb2ae4ad97b..d5f028e33f658bd62deb5b9b0737c33366a22a83 100644 (file)
@@ -221,8 +221,7 @@ icmpv6_error(struct net *net, struct nf_conn *tmpl,
        type = icmp6h->icmp6_type - 130;
        if (type >= 0 && type < sizeof(noct_valid_new) &&
            noct_valid_new[type]) {
-               nf_ct_set(skb, nf_ct_untracked_get(), IP_CT_NEW);
-               nf_conntrack_get(skb_nfct(skb));
+               nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
                return NF_ACCEPT;
        }
 
index 888ecd106e5f37ba6c9e93bf1d1f161ddbaa280c..4a7ddeddbaabf1866a65cfda5d27e8e9c1e75b57 100644 (file)
@@ -58,8 +58,7 @@ void nf_dup_ipv6(struct net *net, struct sk_buff *skb, unsigned int hooknum,
 
 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
        nf_reset(skb);
-       nf_ct_set(skb, nf_ct_untracked_get(), IP_CT_NEW);
-       nf_conntrack_get(skb_nfct(skb));
+       nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 #endif
        if (hooknum == NF_INET_PRE_ROUTING ||
            hooknum == NF_INET_LOCAL_IN) {
index bcf1d2a6539e8dbe34526b64052be8f842dd3586..03150f60714d453f42cc9b30b59537b818a98262 100644 (file)
@@ -180,14 +180,6 @@ EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 
 unsigned int nf_conntrack_max __read_mostly;
 seqcount_t nf_conntrack_generation __read_mostly;
-
-/* nf_conn must be 8 bytes aligned, as the 3 LSB bits are used
- * for the nfctinfo. We cheat by (ab)using the PER CPU cache line
- * alignment to enforce this.
- */
-DEFINE_PER_CPU_ALIGNED(struct nf_conn, nf_conntrack_untracked);
-EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
-
 static unsigned int nf_conntrack_hash_rnd __read_mostly;
 
 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
@@ -1314,9 +1306,10 @@ nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
        int ret;
 
        tmpl = nf_ct_get(skb, &ctinfo);
-       if (tmpl) {
+       if (tmpl || ctinfo == IP_CT_UNTRACKED) {
                /* Previously seen (loopback or untracked)?  Ignore. */
-               if (!nf_ct_is_template(tmpl)) {
+               if ((tmpl && !nf_ct_is_template(tmpl)) ||
+                    ctinfo == IP_CT_UNTRACKED) {
                        NF_CT_STAT_INC_ATOMIC(net, ignore);
                        return NF_ACCEPT;
                }
@@ -1629,18 +1622,6 @@ void nf_ct_free_hashtable(void *hash, unsigned int size)
 }
 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
 
-static int untrack_refs(void)
-{
-       int cnt = 0, cpu;
-
-       for_each_possible_cpu(cpu) {
-               struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
-
-               cnt += atomic_read(&ct->ct_general.use) - 1;
-       }
-       return cnt;
-}
-
 void nf_conntrack_cleanup_start(void)
 {
        conntrack_gc_work.exiting = true;
@@ -1650,8 +1631,6 @@ void nf_conntrack_cleanup_start(void)
 void nf_conntrack_cleanup_end(void)
 {
        RCU_INIT_POINTER(nf_ct_destroy, NULL);
-       while (untrack_refs() > 0)
-               schedule();
 
        cancel_delayed_work_sync(&conntrack_gc_work.dwork);
        nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
@@ -1825,20 +1804,11 @@ EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
                  &nf_conntrack_htable_size, 0600);
 
-void nf_ct_untracked_status_or(unsigned long bits)
-{
-       int cpu;
-
-       for_each_possible_cpu(cpu)
-               per_cpu(nf_conntrack_untracked, cpu).status |= bits;
-}
-EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
-
 int nf_conntrack_init_start(void)
 {
        int max_factor = 8;
        int ret = -ENOMEM;
-       int i, cpu;
+       int i;
 
        seqcount_init(&nf_conntrack_generation);
 
@@ -1921,15 +1891,6 @@ int nf_conntrack_init_start(void)
        if (ret < 0)
                goto err_proto;
 
-       /* Set up fake conntrack: to never be deleted, not in any hashes */
-       for_each_possible_cpu(cpu) {
-               struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
-               write_pnet(&ct->ct_net, &init_net);
-               atomic_set(&ct->ct_general.use, 1);
-       }
-       /*  - and look it like as a confirmed connection */
-       nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
-
        conntrack_gc_work_init(&conntrack_gc_work);
        queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, HZ);
 
@@ -1977,6 +1938,7 @@ int nf_conntrack_init_net(struct net *net)
        int ret = -ENOMEM;
        int cpu;
 
+       BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
        atomic_set(&net->ct.count, 0);
 
        net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
index 5e35643da6508d55937a2229ee39aa311852a4ec..9cbf49f9c1b712de6ff3f94b26ab2da10c9b684c 100644 (file)
@@ -861,9 +861,6 @@ static int __init nf_nat_init(void)
 
        nf_ct_helper_expectfn_register(&follow_master_nat);
 
-       /* Initialize fake conntrack so that NAT will skip it */
-       nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
-
        BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
        RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
                           nfnetlink_parse_nat_setup);
index 6e23dbbedd7f4eb4d19300817ee2284e508fd827..6c6fd48b024c2d4379179e2936f242f6d7730aa6 100644 (file)
@@ -72,12 +72,12 @@ static void nft_ct_get_eval(const struct nft_expr *expr,
 
        switch (priv->key) {
        case NFT_CT_STATE:
-               if (ct == NULL)
-                       state = NF_CT_STATE_INVALID_BIT;
-               else if (nf_ct_is_untracked(ct))
+               if (ct)
+                       state = NF_CT_STATE_BIT(ctinfo);
+               else if (ctinfo == IP_CT_UNTRACKED)
                        state = NF_CT_STATE_UNTRACKED_BIT;
                else
-                       state = NF_CT_STATE_BIT(ctinfo);
+                       state = NF_CT_STATE_INVALID_BIT;
                *dest = state;
                return;
        default:
@@ -718,12 +718,10 @@ static void nft_notrack_eval(const struct nft_expr *expr,
 
        ct = nf_ct_get(pkt->skb, &ctinfo);
        /* Previously seen (loopback or untracked)?  Ignore. */
-       if (ct)
+       if (ct || ctinfo == IP_CT_UNTRACKED)
                return;
 
-       ct = nf_ct_untracked_get();
-       atomic_inc(&ct->ct_general.use);
-       nf_ct_set(skb, ct, IP_CT_NEW);
+       nf_ct_set(skb, ct, IP_CT_UNTRACKED);
 }
 
 static struct nft_expr_type nft_notrack_type;
index b008db0184b8a2a1679d5d4737427a672c36b560..3cbe1bcf6a742c6d74fa455daa7f1cb7bf56ba57 100644 (file)
@@ -26,11 +26,12 @@ static inline int xt_ct_target(struct sk_buff *skb, struct nf_conn *ct)
        if (skb->_nfct != 0)
                return XT_CONTINUE;
 
-       /* special case the untracked ct : we want the percpu object */
-       if (!ct)
-               ct = nf_ct_untracked_get();
-       atomic_inc(&ct->ct_general.use);
-       nf_ct_set(skb, ct, IP_CT_NEW);
+       if (ct) {
+               atomic_inc(&ct->ct_general.use);
+               nf_ct_set(skb, ct, IP_CT_NEW);
+       } else {
+               nf_ct_set(skb, ct, IP_CT_UNTRACKED);
+       }
 
        return XT_CONTINUE;
 }
@@ -335,7 +336,7 @@ static void xt_ct_tg_destroy(const struct xt_tgdtor_param *par,
        struct nf_conn *ct = info->ct;
        struct nf_conn_help *help;
 
-       if (ct && !nf_ct_is_untracked(ct)) {
+       if (ct) {
                help = nfct_help(ct);
                if (help)
                        module_put(help->helper->me);
@@ -412,8 +413,7 @@ notrack_tg(struct sk_buff *skb, const struct xt_action_param *par)
        if (skb->_nfct != 0)
                return XT_CONTINUE;
 
-       nf_ct_set(skb, nf_ct_untracked_get(), IP_CT_NEW);
-       nf_conntrack_get(skb_nfct(skb));
+       nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
 
        return XT_CONTINUE;
 }
index c0fb217bc64969bd0da2c656d4418f8c80b29d14..39cf1d019240e61f504bc8ced96a63c41c9bd839 100644 (file)
@@ -172,12 +172,11 @@ conntrack_mt(const struct sk_buff *skb, struct xt_action_param *par,
 
        ct = nf_ct_get(skb, &ctinfo);
 
-       if (ct) {
-               if (nf_ct_is_untracked(ct))
-                       statebit = XT_CONNTRACK_STATE_UNTRACKED;
-               else
-                       statebit = XT_CONNTRACK_STATE_BIT(ctinfo);
-       } else
+       if (ct)
+               statebit = XT_CONNTRACK_STATE_BIT(ctinfo);
+       else if (ctinfo == IP_CT_UNTRACKED)
+               statebit = XT_CONNTRACK_STATE_UNTRACKED;
+       else
                statebit = XT_CONNTRACK_STATE_INVALID;
 
        if (info->match_flags & XT_CONNTRACK_STATE) {
index 5746a33789a50ced02746d57403970872f2cf29f..5fbd79194d21ee1f26fa669162b8de92d965f8c8 100644 (file)
@@ -28,14 +28,13 @@ state_mt(const struct sk_buff *skb, struct xt_action_param *par)
        unsigned int statebit;
        struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 
-       if (!ct)
+       if (ct)
+               statebit = XT_STATE_BIT(ctinfo);
+       else if (ctinfo == IP_CT_UNTRACKED)
+               statebit = XT_STATE_UNTRACKED;
+       else
                statebit = XT_STATE_INVALID;
-       else {
-               if (nf_ct_is_untracked(ct))
-                       statebit = XT_STATE_UNTRACKED;
-               else
-                       statebit = XT_STATE_BIT(ctinfo);
-       }
+
        return (sinfo->statemask & statebit);
 }