ipvs: convert lblc scheduler to rcu
[linux-3.10.git] / net / netfilter / nf_conntrack_expect.c
index 990fa12..8c10e3d 100644 (file)
@@ -20,6 +20,8 @@
 #include <linux/percpu.h>
 #include <linux/kernel.h>
 #include <linux/jhash.h>
+#include <linux/moduleparam.h>
+#include <linux/export.h>
 #include <net/net_namespace.h>
 
 #include <net/netfilter/nf_conntrack.h>
 #include <net/netfilter/nf_conntrack_expect.h>
 #include <net/netfilter/nf_conntrack_helper.h>
 #include <net/netfilter/nf_conntrack_tuple.h>
-
-struct hlist_head *nf_ct_expect_hash __read_mostly;
-EXPORT_SYMBOL_GPL(nf_ct_expect_hash);
+#include <net/netfilter/nf_conntrack_zones.h>
 
 unsigned int nf_ct_expect_hsize __read_mostly;
 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
 
-static unsigned int nf_ct_expect_hash_rnd __read_mostly;
-static unsigned int nf_ct_expect_count;
 unsigned int nf_ct_expect_max __read_mostly;
-static int nf_ct_expect_hash_rnd_initted __read_mostly;
-static int nf_ct_expect_vmalloc;
 
 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
 
 /* nf_conntrack_expect helper functions */
-void nf_ct_unlink_expect(struct nf_conntrack_expect *exp)
+void nf_ct_unlink_expect_report(struct nf_conntrack_expect *exp,
+                               u32 pid, int report)
 {
        struct nf_conn_help *master_help = nfct_help(exp->master);
+       struct net *net = nf_ct_exp_net(exp);
 
        NF_CT_ASSERT(master_help);
        NF_CT_ASSERT(!timer_pending(&exp->timeout));
 
        hlist_del_rcu(&exp->hnode);
-       nf_ct_expect_count--;
+       net->ct.expect_count--;
 
        hlist_del(&exp->lnode);
        master_help->expecting[exp->class]--;
+
+       nf_ct_expect_event_report(IPEXP_DESTROY, exp, pid, report);
        nf_ct_expect_put(exp);
 
-       NF_CT_STAT_INC(expect_delete);
+       NF_CT_STAT_INC(net, expect_delete);
 }
-EXPORT_SYMBOL_GPL(nf_ct_unlink_expect);
+EXPORT_SYMBOL_GPL(nf_ct_unlink_expect_report);
 
 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
 {
@@ -75,30 +75,30 @@ static unsigned int nf_ct_expect_dst_hash(const struct nf_conntrack_tuple *tuple
 {
        unsigned int hash;
 
-       if (unlikely(!nf_ct_expect_hash_rnd_initted)) {
-               get_random_bytes(&nf_ct_expect_hash_rnd, 4);
-               nf_ct_expect_hash_rnd_initted = 1;
+       if (unlikely(!nf_conntrack_hash_rnd)) {
+               init_nf_conntrack_hash_rnd();
        }
 
        hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
                      (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
-                      (__force __u16)tuple->dst.u.all) ^ nf_ct_expect_hash_rnd);
+                      (__force __u16)tuple->dst.u.all) ^ nf_conntrack_hash_rnd);
        return ((u64)hash * nf_ct_expect_hsize) >> 32;
 }
 
 struct nf_conntrack_expect *
-__nf_ct_expect_find(const struct nf_conntrack_tuple *tuple)
+__nf_ct_expect_find(struct net *net, u16 zone,
+                   const struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_expect *i;
-       struct hlist_node *n;
        unsigned int h;
 
-       if (!nf_ct_expect_count)
+       if (!net->ct.expect_count)
                return NULL;
 
        h = nf_ct_expect_dst_hash(tuple);
-       hlist_for_each_entry_rcu(i, n, &nf_ct_expect_hash[h], hnode) {
-               if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask))
+       hlist_for_each_entry_rcu(i, &net->ct.expect_hash[h], hnode) {
+               if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
+                   nf_ct_zone(i->master) == zone)
                        return i;
        }
        return NULL;
@@ -107,12 +107,13 @@ EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
 
 /* Just find a expectation corresponding to a tuple. */
 struct nf_conntrack_expect *
-nf_ct_expect_find_get(const struct nf_conntrack_tuple *tuple)
+nf_ct_expect_find_get(struct net *net, u16 zone,
+                     const struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_expect *i;
 
        rcu_read_lock();
-       i = __nf_ct_expect_find(tuple);
+       i = __nf_ct_expect_find(net, zone, tuple);
        if (i && !atomic_inc_not_zero(&i->use))
                i = NULL;
        rcu_read_unlock();
@@ -124,19 +125,20 @@ EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
 /* If an expectation for this connection is found, it gets delete from
  * global list then returned. */
 struct nf_conntrack_expect *
-nf_ct_find_expectation(const struct nf_conntrack_tuple *tuple)
+nf_ct_find_expectation(struct net *net, u16 zone,
+                      const struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_expect *i, *exp = NULL;
-       struct hlist_node *n;
        unsigned int h;
 
-       if (!nf_ct_expect_count)
+       if (!net->ct.expect_count)
                return NULL;
 
        h = nf_ct_expect_dst_hash(tuple);
-       hlist_for_each_entry(i, n, &nf_ct_expect_hash[h], hnode) {
+       hlist_for_each_entry(i, &net->ct.expect_hash[h], hnode) {
                if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
-                   nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)) {
+                   nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
+                   nf_ct_zone(i->master) == zone) {
                        exp = i;
                        break;
                }
@@ -168,13 +170,13 @@ void nf_ct_remove_expectations(struct nf_conn *ct)
 {
        struct nf_conn_help *help = nfct_help(ct);
        struct nf_conntrack_expect *exp;
-       struct hlist_node *n, *next;
+       struct hlist_node *next;
 
        /* Optimization: most connection never expect any others. */
        if (!help)
                return;
 
-       hlist_for_each_entry_safe(exp, n, next, &help->expectations, lnode) {
+       hlist_for_each_entry_safe(exp, next, &help->expectations, lnode) {
                if (del_timer(&exp->timeout)) {
                        nf_ct_unlink_expect(exp);
                        nf_ct_expect_put(exp);
@@ -205,9 +207,10 @@ static inline int expect_clash(const struct nf_conntrack_expect *a,
 static inline int expect_matches(const struct nf_conntrack_expect *a,
                                 const struct nf_conntrack_expect *b)
 {
-       return a->master == b->master && a->class == b->class
-               && nf_ct_tuple_equal(&a->tuple, &b->tuple)
-               && nf_ct_tuple_mask_equal(&a->mask, &b->mask);
+       return a->master == b->master && a->class == b->class &&
+               nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
+               nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
+               nf_ct_zone(a->master) == nf_ct_zone(b->master);
 }
 
 /* Generally a bad idea to call this: could have matched already. */
@@ -235,7 +238,6 @@ struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
 
        new->master = me;
        atomic_set(&new->use, 1);
-       INIT_RCU_HEAD(&new->rcu);
        return new;
 }
 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
@@ -308,28 +310,34 @@ void nf_ct_expect_put(struct nf_conntrack_expect *exp)
 }
 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
 
-static void nf_ct_expect_insert(struct nf_conntrack_expect *exp)
+static int nf_ct_expect_insert(struct nf_conntrack_expect *exp)
 {
        struct nf_conn_help *master_help = nfct_help(exp->master);
-       const struct nf_conntrack_expect_policy *p;
+       struct nf_conntrack_helper *helper;
+       struct net *net = nf_ct_exp_net(exp);
        unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
 
-       atomic_inc(&exp->use);
+       /* two references : one for hash insert, one for the timer */
+       atomic_add(2, &exp->use);
 
        hlist_add_head(&exp->lnode, &master_help->expectations);
        master_help->expecting[exp->class]++;
 
-       hlist_add_head_rcu(&exp->hnode, &nf_ct_expect_hash[h]);
-       nf_ct_expect_count++;
+       hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
+       net->ct.expect_count++;
 
        setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
                    (unsigned long)exp);
-       p = &master_help->helper->expect_policy[exp->class];
-       exp->timeout.expires = jiffies + p->timeout * HZ;
+       helper = rcu_dereference_protected(master_help->helper,
+                                          lockdep_is_held(&nf_conntrack_lock));
+       if (helper) {
+               exp->timeout.expires = jiffies +
+                       helper->expect_policy[exp->class].timeout * HZ;
+       }
        add_timer(&exp->timeout);
 
-       atomic_inc(&exp->use);
-       NF_CT_STAT_INC(expect_create);
+       NF_CT_STAT_INC(net, expect_create);
+       return 0;
 }
 
 /* Race with expectations being used means we could have none to find; OK. */
@@ -338,9 +346,8 @@ static void evict_oldest_expect(struct nf_conn *master,
 {
        struct nf_conn_help *master_help = nfct_help(master);
        struct nf_conntrack_expect *exp, *last = NULL;
-       struct hlist_node *n;
 
-       hlist_for_each_entry(exp, n, &master_help->expectations, lnode) {
+       hlist_for_each_entry(exp, &master_help->expectations, lnode) {
                if (exp->class == new->class)
                        last = exp;
        }
@@ -351,44 +358,29 @@ static void evict_oldest_expect(struct nf_conn *master,
        }
 }
 
-static inline int refresh_timer(struct nf_conntrack_expect *i)
-{
-       struct nf_conn_help *master_help = nfct_help(i->master);
-       const struct nf_conntrack_expect_policy *p;
-
-       if (!del_timer(&i->timeout))
-               return 0;
-
-       p = &master_help->helper->expect_policy[i->class];
-       i->timeout.expires = jiffies + p->timeout * HZ;
-       add_timer(&i->timeout);
-       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 nf_conn *master = expect->master;
        struct nf_conn_help *master_help = nfct_help(master);
-       struct hlist_node *n;
+       struct nf_conntrack_helper *helper;
+       struct net *net = nf_ct_exp_net(expect);
+       struct hlist_node *next;
        unsigned int h;
-       int ret;
-
-       NF_CT_ASSERT(master_help);
+       int ret = 1;
 
-       spin_lock_bh(&nf_conntrack_lock);
-       if (!master_help->helper) {
+       if (!master_help) {
                ret = -ESHUTDOWN;
                goto out;
        }
        h = nf_ct_expect_dst_hash(&expect->tuple);
-       hlist_for_each_entry(i, n, &nf_ct_expect_hash[h], hnode) {
+       hlist_for_each_entry_safe(i, next, &net->ct.expect_hash[h], hnode) {
                if (expect_matches(i, expect)) {
-                       /* Refresh timer: if it's dying, ignore.. */
-                       if (refresh_timer(i)) {
-                               ret = 0;
-                               goto out;
+                       if (del_timer(&i->timeout)) {
+                               nf_ct_unlink_expect(i);
+                               nf_ct_expect_put(i);
+                               break;
                        }
                } else if (expect_clash(i, expect)) {
                        ret = -EBUSY;
@@ -396,45 +388,65 @@ int nf_ct_expect_related(struct nf_conntrack_expect *expect)
                }
        }
        /* Will be over limit? */
-       p = &master_help->helper->expect_policy[expect->class];
-       if (p->max_expected &&
-           master_help->expecting[expect->class] >= p->max_expected) {
-               evict_oldest_expect(master, expect);
-               if (master_help->expecting[expect->class] >= p->max_expected) {
-                       ret = -EMFILE;
-                       goto out;
+       helper = rcu_dereference_protected(master_help->helper,
+                                          lockdep_is_held(&nf_conntrack_lock));
+       if (helper) {
+               p = &helper->expect_policy[expect->class];
+               if (p->max_expected &&
+                   master_help->expecting[expect->class] >= p->max_expected) {
+                       evict_oldest_expect(master, expect);
+                       if (master_help->expecting[expect->class]
+                                               >= p->max_expected) {
+                               ret = -EMFILE;
+                               goto out;
+                       }
                }
        }
 
-       if (nf_ct_expect_count >= nf_ct_expect_max) {
-               if (net_ratelimit())
-                       printk(KERN_WARNING
-                              "nf_conntrack: expectation table full\n");
+       if (net->ct.expect_count >= nf_ct_expect_max) {
+               net_warn_ratelimited("nf_conntrack: expectation table full\n");
                ret = -EMFILE;
-               goto out;
        }
+out:
+       return ret;
+}
+
+int nf_ct_expect_related_report(struct nf_conntrack_expect *expect, 
+                               u32 pid, int report)
+{
+       int ret;
 
-       nf_ct_expect_insert(expect);
-       nf_ct_expect_event(IPEXP_NEW, expect);
-       ret = 0;
+       spin_lock_bh(&nf_conntrack_lock);
+       ret = __nf_ct_expect_check(expect);
+       if (ret <= 0)
+               goto out;
+
+       ret = nf_ct_expect_insert(expect);
+       if (ret < 0)
+               goto out;
+       spin_unlock_bh(&nf_conntrack_lock);
+       nf_ct_expect_event_report(IPEXP_NEW, expect, pid, report);
+       return ret;
 out:
        spin_unlock_bh(&nf_conntrack_lock);
        return ret;
 }
-EXPORT_SYMBOL_GPL(nf_ct_expect_related);
+EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
 
-#ifdef CONFIG_PROC_FS
+#ifdef CONFIG_NF_CONNTRACK_PROCFS
 struct ct_expect_iter_state {
+       struct seq_net_private p;
        unsigned int bucket;
 };
 
 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
 {
+       struct net *net = seq_file_net(seq);
        struct ct_expect_iter_state *st = seq->private;
        struct hlist_node *n;
 
        for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
-               n = rcu_dereference(nf_ct_expect_hash[st->bucket].first);
+               n = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
                if (n)
                        return n;
        }
@@ -444,13 +456,14 @@ static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
                                             struct hlist_node *head)
 {
+       struct net *net = seq_file_net(seq);
        struct ct_expect_iter_state *st = seq->private;
 
-       head = rcu_dereference(head->next);
+       head = rcu_dereference(hlist_next_rcu(head));
        while (head == NULL) {
                if (++st->bucket >= nf_ct_expect_hsize)
                        return NULL;
-               head = rcu_dereference(nf_ct_expect_hash[st->bucket].first);
+               head = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
        }
        return head;
 }
@@ -487,6 +500,7 @@ static void exp_seq_stop(struct seq_file *seq, void *v)
 static int exp_seq_show(struct seq_file *s, void *v)
 {
        struct nf_conntrack_expect *expect;
+       struct nf_conntrack_helper *helper;
        struct hlist_node *n = v;
        char *delim = "";
 
@@ -509,8 +523,20 @@ static int exp_seq_show(struct seq_file *s, void *v)
                seq_printf(s, "PERMANENT");
                delim = ",";
        }
-       if (expect->flags & NF_CT_EXPECT_INACTIVE)
+       if (expect->flags & NF_CT_EXPECT_INACTIVE) {
                seq_printf(s, "%sINACTIVE", delim);
+               delim = ",";
+       }
+       if (expect->flags & NF_CT_EXPECT_USERSPACE)
+               seq_printf(s, "%sUSERSPACE", delim);
+
+       helper = rcu_dereference(nfct_help(expect->master)->helper);
+       if (helper) {
+               seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
+               if (helper->expect_policy[expect->class].name)
+                       seq_printf(s, "/%s",
+                                  helper->expect_policy[expect->class].name);
+       }
 
        return seq_putc(s, '\n');
 }
@@ -524,7 +550,7 @@ static const struct seq_operations exp_seq_ops = {
 
 static int exp_open(struct inode *inode, struct file *file)
 {
-       return seq_open_private(file, &exp_seq_ops,
+       return seq_open_net(inode, file, &exp_seq_ops,
                        sizeof(struct ct_expect_iter_state));
 }
 
@@ -533,72 +559,76 @@ static const struct file_operations exp_file_ops = {
        .open    = exp_open,
        .read    = seq_read,
        .llseek  = seq_lseek,
-       .release = seq_release_private,
+       .release = seq_release_net,
 };
-#endif /* CONFIG_PROC_FS */
+#endif /* CONFIG_NF_CONNTRACK_PROCFS */
 
-static int __init exp_proc_init(void)
+static int exp_proc_init(struct net *net)
 {
-#ifdef CONFIG_PROC_FS
+#ifdef CONFIG_NF_CONNTRACK_PROCFS
        struct proc_dir_entry *proc;
 
-       proc = proc_net_fops_create(&init_net, "nf_conntrack_expect", 0440, &exp_file_ops);
+       proc = proc_create("nf_conntrack_expect", 0440, net->proc_net,
+                          &exp_file_ops);
        if (!proc)
                return -ENOMEM;
-#endif /* CONFIG_PROC_FS */
+#endif /* CONFIG_NF_CONNTRACK_PROCFS */
        return 0;
 }
 
-static void exp_proc_remove(void)
+static void exp_proc_remove(struct net *net)
 {
-#ifdef CONFIG_PROC_FS
-       proc_net_remove(&init_net, "nf_conntrack_expect");
-#endif /* CONFIG_PROC_FS */
+#ifdef CONFIG_NF_CONNTRACK_PROCFS
+       remove_proc_entry("nf_conntrack_expect", net->proc_net);
+#endif /* CONFIG_NF_CONNTRACK_PROCFS */
 }
 
-module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0600);
+module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
 
-int __init nf_conntrack_expect_init(void)
+int nf_conntrack_expect_pernet_init(struct net *net)
 {
        int err = -ENOMEM;
 
+       net->ct.expect_count = 0;
+       net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
+       if (net->ct.expect_hash == NULL)
+               goto err1;
+
+       err = exp_proc_init(net);
+       if (err < 0)
+               goto err2;
+
+       return 0;
+err2:
+       nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
+err1:
+       return err;
+}
+
+void nf_conntrack_expect_pernet_fini(struct net *net)
+{
+       exp_proc_remove(net);
+       nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
+}
+
+int nf_conntrack_expect_init(void)
+{
        if (!nf_ct_expect_hsize) {
                nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
                if (!nf_ct_expect_hsize)
                        nf_ct_expect_hsize = 1;
        }
        nf_ct_expect_max = nf_ct_expect_hsize * 4;
-
-       nf_ct_expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
-                                                 &nf_ct_expect_vmalloc);
-       if (nf_ct_expect_hash == NULL)
-               goto err1;
-
        nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
-                                       sizeof(struct nf_conntrack_expect),
-                                       0, 0, NULL);
+                               sizeof(struct nf_conntrack_expect),
+                               0, 0, NULL);
        if (!nf_ct_expect_cachep)
-               goto err2;
-
-       err = exp_proc_init();
-       if (err < 0)
-               goto err3;
-
+               return -ENOMEM;
        return 0;
-
-err3:
-       kmem_cache_destroy(nf_ct_expect_cachep);
-err2:
-       nf_ct_free_hashtable(nf_ct_expect_hash, nf_ct_expect_vmalloc,
-                            nf_ct_expect_hsize);
-err1:
-       return err;
 }
 
 void nf_conntrack_expect_fini(void)
 {
-       exp_proc_remove();
+       rcu_barrier(); /* Wait for call_rcu() before destroy */
        kmem_cache_destroy(nf_ct_expect_cachep);
-       nf_ct_free_hashtable(nf_ct_expect_hash, nf_ct_expect_vmalloc,
-                            nf_ct_expect_hsize);
 }