net: spread __net_init, __net_exit
[safe/jmp/linux-2.6] / net / core / fib_rules.c
index 0d8bb2e..9a24377 100644 (file)
 #include <linux/types.h>
 #include <linux/kernel.h>
 #include <linux/list.h>
+#include <net/net_namespace.h>
+#include <net/sock.h>
 #include <net/fib_rules.h>
 
-static LIST_HEAD(rules_ops);
-static DEFINE_SPINLOCK(rules_mod_lock);
+int fib_default_rule_add(struct fib_rules_ops *ops,
+                        u32 pref, u32 table, u32 flags)
+{
+       struct fib_rule *r;
+
+       r = kzalloc(ops->rule_size, GFP_KERNEL);
+       if (r == NULL)
+               return -ENOMEM;
+
+       atomic_set(&r->refcnt, 1);
+       r->action = FR_ACT_TO_TBL;
+       r->pref = pref;
+       r->table = table;
+       r->flags = flags;
+       r->fr_net = hold_net(ops->fro_net);
+
+       /* The lock is not required here, the list in unreacheable
+        * at the moment this function is called */
+       list_add_tail(&r->list, &ops->rules_list);
+       return 0;
+}
+EXPORT_SYMBOL(fib_default_rule_add);
 
 static void notify_rule_change(int event, struct fib_rule *rule,
                               struct fib_rules_ops *ops, struct nlmsghdr *nlh,
                               u32 pid);
 
-static struct fib_rules_ops *lookup_rules_ops(int family)
+static struct fib_rules_ops *lookup_rules_ops(struct net *net, int family)
 {
        struct fib_rules_ops *ops;
 
        rcu_read_lock();
-       list_for_each_entry_rcu(ops, &rules_ops, list) {
+       list_for_each_entry_rcu(ops, &net->rules_ops, list) {
                if (ops->family == family) {
                        if (!try_module_get(ops->owner))
                                ops = NULL;
@@ -44,10 +66,19 @@ static void rules_ops_put(struct fib_rules_ops *ops)
                module_put(ops->owner);
 }
 
-int fib_rules_register(struct fib_rules_ops *ops)
+static void flush_route_cache(struct fib_rules_ops *ops)
+{
+       if (ops->flush_cache)
+               ops->flush_cache(ops);
+}
+
+static int __fib_rules_register(struct fib_rules_ops *ops)
 {
        int err = -EEXIST;
        struct fib_rules_ops *o;
+       struct net *net;
+
+       net = ops->fro_net;
 
        if (ops->rule_size < sizeof(struct fib_rule))
                return -EINVAL;
@@ -57,52 +88,74 @@ int fib_rules_register(struct fib_rules_ops *ops)
            ops->action == NULL)
                return -EINVAL;
 
-       spin_lock(&rules_mod_lock);
-       list_for_each_entry(o, &rules_ops, list)
+       spin_lock(&net->rules_mod_lock);
+       list_for_each_entry(o, &net->rules_ops, list)
                if (ops->family == o->family)
                        goto errout;
 
-       list_add_tail_rcu(&ops->list, &rules_ops);
+       hold_net(net);
+       list_add_tail_rcu(&ops->list, &net->rules_ops);
        err = 0;
 errout:
-       spin_unlock(&rules_mod_lock);
+       spin_unlock(&net->rules_mod_lock);
 
        return err;
 }
 
+struct fib_rules_ops *
+fib_rules_register(struct fib_rules_ops *tmpl, struct net *net)
+{
+       struct fib_rules_ops *ops;
+       int err;
+
+       ops = kmemdup(tmpl, sizeof (*ops), GFP_KERNEL);
+       if (ops == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       INIT_LIST_HEAD(&ops->rules_list);
+       ops->fro_net = net;
+
+       err = __fib_rules_register(ops);
+       if (err) {
+               kfree(ops);
+               ops = ERR_PTR(err);
+       }
+
+       return ops;
+}
+
 EXPORT_SYMBOL_GPL(fib_rules_register);
 
-static void cleanup_ops(struct fib_rules_ops *ops)
+void fib_rules_cleanup_ops(struct fib_rules_ops *ops)
 {
        struct fib_rule *rule, *tmp;
 
-       list_for_each_entry_safe(rule, tmp, ops->rules_list, list) {
+       list_for_each_entry_safe(rule, tmp, &ops->rules_list, list) {
                list_del_rcu(&rule->list);
                fib_rule_put(rule);
        }
 }
+EXPORT_SYMBOL_GPL(fib_rules_cleanup_ops);
 
-int fib_rules_unregister(struct fib_rules_ops *ops)
+static void fib_rules_put_rcu(struct rcu_head *head)
 {
-       int err = 0;
-       struct fib_rules_ops *o;
+       struct fib_rules_ops *ops = container_of(head, struct fib_rules_ops, rcu);
+       struct net *net = ops->fro_net;
 
-       spin_lock(&rules_mod_lock);
-       list_for_each_entry(o, &rules_ops, list) {
-               if (o == ops) {
-                       list_del_rcu(&o->list);
-                       cleanup_ops(ops);
-                       goto out;
-               }
-       }
+       release_net(net);
+       kfree(ops);
+}
 
-       err = -ENOENT;
-out:
-       spin_unlock(&rules_mod_lock);
+void fib_rules_unregister(struct fib_rules_ops *ops)
+{
+       struct net *net = ops->fro_net;
 
-       synchronize_rcu();
+       spin_lock(&net->rules_mod_lock);
+       list_del_rcu(&ops->list);
+       fib_rules_cleanup_ops(ops);
+       spin_unlock(&net->rules_mod_lock);
 
-       return err;
+       call_rcu(&ops->rcu, fib_rules_put_rcu);
 }
 
 EXPORT_SYMBOL_GPL(fib_rules_unregister);
@@ -112,7 +165,10 @@ static int fib_rule_match(struct fib_rule *rule, struct fib_rules_ops *ops,
 {
        int ret = 0;
 
-       if (rule->ifindex && (rule->ifindex != fl->iif))
+       if (rule->iifindex && (rule->iifindex != fl->iif))
+               goto out;
+
+       if (rule->oifindex && (rule->oifindex != fl->oif))
                goto out;
 
        if ((rule->mark ^ fl->mark) & rule->mark_mask)
@@ -131,7 +187,7 @@ int fib_rules_lookup(struct fib_rules_ops *ops, struct flowi *fl,
 
        rcu_read_lock();
 
-       list_for_each_entry_rcu(rule, ops->rules_list, list) {
+       list_for_each_entry_rcu(rule, &ops->rules_list, list) {
 jumped:
                if (!fib_rule_match(rule, ops, fl, flags))
                        continue;
@@ -146,7 +202,9 @@ jumped:
                                rule = target;
                                goto jumped;
                        }
-               } else
+               } else if (rule->action == FR_ACT_NOP)
+                       continue;
+               else
                        err = ops->action(rule, fl, flags, arg);
 
                if (err != -EAGAIN) {
@@ -189,6 +247,7 @@ errout:
 
 static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 {
+       struct net *net = sock_net(skb->sk);
        struct fib_rule_hdr *frh = nlmsg_data(nlh);
        struct fib_rules_ops *ops = NULL;
        struct fib_rule *rule, *r, *last = NULL;
@@ -198,9 +257,9 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
        if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
                goto errout;
 
-       ops = lookup_rules_ops(frh->family);
+       ops = lookup_rules_ops(net, frh->family);
        if (ops == NULL) {
-               err = EAFNOSUPPORT;
+               err = -EAFNOSUPPORT;
                goto errout;
        }
 
@@ -217,18 +276,29 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
                err = -ENOMEM;
                goto errout;
        }
+       rule->fr_net = hold_net(net);
 
        if (tb[FRA_PRIORITY])
                rule->pref = nla_get_u32(tb[FRA_PRIORITY]);
 
-       if (tb[FRA_IFNAME]) {
+       if (tb[FRA_IIFNAME]) {
+               struct net_device *dev;
+
+               rule->iifindex = -1;
+               nla_strlcpy(rule->iifname, tb[FRA_IIFNAME], IFNAMSIZ);
+               dev = __dev_get_by_name(net, rule->iifname);
+               if (dev)
+                       rule->iifindex = dev->ifindex;
+       }
+
+       if (tb[FRA_OIFNAME]) {
                struct net_device *dev;
 
-               rule->ifindex = -1;
-               nla_strlcpy(rule->ifname, tb[FRA_IFNAME], IFNAMSIZ);
-               dev = __dev_get_by_name(rule->ifname);
+               rule->oifindex = -1;
+               nla_strlcpy(rule->oifname, tb[FRA_OIFNAME], IFNAMSIZ);
+               dev = __dev_get_by_name(net, rule->oifname);
                if (dev)
-                       rule->ifindex = dev->ifindex;
+                       rule->oifindex = dev->ifindex;
        }
 
        if (tb[FRA_FWMARK]) {
@@ -247,8 +317,8 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
        rule->flags = frh->flags;
        rule->table = frh_get_table(frh, tb);
 
-       if (!rule->pref && ops->default_pref)
-               rule->pref = ops->default_pref();
+       if (!tb[FRA_PRIORITY] && ops->default_pref)
+               rule->pref = ops->default_pref(ops);
 
        err = -EINVAL;
        if (tb[FRA_GOTO]) {
@@ -260,7 +330,7 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
                if (rule->target <= rule->pref)
                        goto errout_free;
 
-               list_for_each_entry(r, ops->rules_list, list) {
+               list_for_each_entry(r, &ops->rules_list, list) {
                        if (r->pref == rule->target) {
                                rule->ctarget = r;
                                break;
@@ -272,11 +342,11 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
        } else if (rule->action == FR_ACT_GOTO)
                goto errout_free;
 
-       err = ops->configure(rule, skb, nlh, frh, tb);
+       err = ops->configure(rule, skb, frh, tb);
        if (err < 0)
                goto errout_free;
 
-       list_for_each_entry(r, ops->rules_list, list) {
+       list_for_each_entry(r, &ops->rules_list, list) {
                if (r->pref > rule->pref)
                        break;
                last = r;
@@ -289,7 +359,7 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
                 * There are unresolved goto rules in the list, check if
                 * any of them are pointing to this new rule.
                 */
-               list_for_each_entry(r, ops->rules_list, list) {
+               list_for_each_entry(r, &ops->rules_list, list) {
                        if (r->action == FR_ACT_GOTO &&
                            r->target == rule->pref) {
                                BUG_ON(r->ctarget != NULL);
@@ -309,13 +379,15 @@ static int fib_nl_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
        if (last)
                list_add_rcu(&rule->list, &last->list);
        else
-               list_add_rcu(&rule->list, ops->rules_list);
+               list_add_rcu(&rule->list, &ops->rules_list);
 
        notify_rule_change(RTM_NEWRULE, rule, ops, nlh, NETLINK_CB(skb).pid);
+       flush_route_cache(ops);
        rules_ops_put(ops);
        return 0;
 
 errout_free:
+       release_net(rule->fr_net);
        kfree(rule);
 errout:
        rules_ops_put(ops);
@@ -324,6 +396,7 @@ errout:
 
 static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 {
+       struct net *net = sock_net(skb->sk);
        struct fib_rule_hdr *frh = nlmsg_data(nlh);
        struct fib_rules_ops *ops = NULL;
        struct fib_rule *rule, *tmp;
@@ -333,9 +406,9 @@ static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
        if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*frh)))
                goto errout;
 
-       ops = lookup_rules_ops(frh->family);
+       ops = lookup_rules_ops(net, frh->family);
        if (ops == NULL) {
-               err = EAFNOSUPPORT;
+               err = -EAFNOSUPPORT;
                goto errout;
        }
 
@@ -347,7 +420,7 @@ static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
        if (err < 0)
                goto errout;
 
-       list_for_each_entry(rule, ops->rules_list, list) {
+       list_for_each_entry(rule, &ops->rules_list, list) {
                if (frh->action && (frh->action != rule->action))
                        continue;
 
@@ -358,8 +431,12 @@ static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
                    (rule->pref != nla_get_u32(tb[FRA_PRIORITY])))
                        continue;
 
-               if (tb[FRA_IFNAME] &&
-                   nla_strcmp(tb[FRA_IFNAME], rule->ifname))
+               if (tb[FRA_IIFNAME] &&
+                   nla_strcmp(tb[FRA_IIFNAME], rule->iifname))
+                       continue;
+
+               if (tb[FRA_OIFNAME] &&
+                   nla_strcmp(tb[FRA_OIFNAME], rule->oifname))
                        continue;
 
                if (tb[FRA_FWMARK] &&
@@ -390,7 +467,7 @@ static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
                 * actually been added.
                 */
                if (ops->nr_goto_rules > 0) {
-                       list_for_each_entry(tmp, ops->rules_list, list) {
+                       list_for_each_entry(tmp, &ops->rules_list, list) {
                                if (tmp->ctarget == rule) {
                                        rcu_assign_pointer(tmp->ctarget, NULL);
                                        ops->unresolved_rules++;
@@ -402,6 +479,7 @@ static int fib_nl_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
                notify_rule_change(RTM_DELRULE, rule, ops, nlh,
                                   NETLINK_CB(skb).pid);
                fib_rule_put(rule);
+               flush_route_cache(ops);
                rules_ops_put(ops);
                return 0;
        }
@@ -416,7 +494,8 @@ static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops *ops,
                                         struct fib_rule *rule)
 {
        size_t payload = NLMSG_ALIGN(sizeof(struct fib_rule_hdr))
-                        + nla_total_size(IFNAMSIZ) /* FRA_IFNAME */
+                        + nla_total_size(IFNAMSIZ) /* FRA_IIFNAME */
+                        + nla_total_size(IFNAMSIZ) /* FRA_OIFNAME */
                         + nla_total_size(4) /* FRA_PRIORITY */
                         + nla_total_size(4) /* FRA_TABLE */
                         + nla_total_size(4) /* FRA_FWMARK */
@@ -450,8 +529,19 @@ static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
        if (rule->action == FR_ACT_GOTO && rule->ctarget == NULL)
                frh->flags |= FIB_RULE_UNRESOLVED;
 
-       if (rule->ifname[0])
-               NLA_PUT_STRING(skb, FRA_IFNAME, rule->ifname);
+       if (rule->iifname[0]) {
+               NLA_PUT_STRING(skb, FRA_IIFNAME, rule->iifname);
+
+               if (rule->iifindex == -1)
+                       frh->flags |= FIB_RULE_IIF_DETACHED;
+       }
+
+       if (rule->oifname[0]) {
+               NLA_PUT_STRING(skb, FRA_OIFNAME, rule->oifname);
+
+               if (rule->oifindex == -1)
+                       frh->flags |= FIB_RULE_OIF_DETACHED;
+       }
 
        if (rule->pref)
                NLA_PUT_U32(skb, FRA_PRIORITY, rule->pref);
@@ -465,7 +555,7 @@ static int fib_nl_fill_rule(struct sk_buff *skb, struct fib_rule *rule,
        if (rule->target)
                NLA_PUT_U32(skb, FRA_GOTO, rule->target);
 
-       if (ops->fill(rule, skb, nlh, frh) < 0)
+       if (ops->fill(rule, skb, frh) < 0)
                goto nla_put_failure;
 
        return nlmsg_end(skb, nlh);
@@ -481,8 +571,7 @@ static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
        int idx = 0;
        struct fib_rule *rule;
 
-       rcu_read_lock();
-       list_for_each_entry_rcu(rule, ops->rules_list, list) {
+       list_for_each_entry(rule, &ops->rules_list, list) {
                if (idx < cb->args[1])
                        goto skip;
 
@@ -493,7 +582,6 @@ static int dump_rules(struct sk_buff *skb, struct netlink_callback *cb,
 skip:
                idx++;
        }
-       rcu_read_unlock();
        cb->args[1] = idx;
        rules_ops_put(ops);
 
@@ -502,13 +590,14 @@ skip:
 
 static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
 {
+       struct net *net = sock_net(skb->sk);
        struct fib_rules_ops *ops;
        int idx = 0, family;
 
        family = rtnl_msg_family(cb->nlh);
        if (family != AF_UNSPEC) {
                /* Protocol specific dump request */
-               ops = lookup_rules_ops(family);
+               ops = lookup_rules_ops(net, family);
                if (ops == NULL)
                        return -EAFNOSUPPORT;
 
@@ -516,7 +605,7 @@ static int fib_nl_dumprule(struct sk_buff *skb, struct netlink_callback *cb)
        }
 
        rcu_read_lock();
-       list_for_each_entry_rcu(ops, &rules_ops, list) {
+       list_for_each_entry_rcu(ops, &net->rules_ops, list) {
                if (idx < cb->args[0] || !try_module_get(ops->owner))
                        goto skip;
 
@@ -537,9 +626,11 @@ static void notify_rule_change(int event, struct fib_rule *rule,
                               struct fib_rules_ops *ops, struct nlmsghdr *nlh,
                               u32 pid)
 {
+       struct net *net;
        struct sk_buff *skb;
        int err = -ENOBUFS;
 
+       net = ops->fro_net;
        skb = nlmsg_new(fib_rule_nlmsg_size(ops, rule), GFP_KERNEL);
        if (skb == NULL)
                goto errout;
@@ -551,10 +642,12 @@ static void notify_rule_change(int event, struct fib_rule *rule,
                kfree_skb(skb);
                goto errout;
        }
-       err = rtnl_notify(skb, pid, ops->nlgroup, nlh, GFP_KERNEL);
+
+       rtnl_notify(skb, net, pid, ops->nlgroup, nlh, GFP_KERNEL);
+       return;
 errout:
        if (err < 0)
-               rtnl_set_sk_err(ops->nlgroup, err);
+               rtnl_set_sk_err(net, ops->nlgroup, err);
 }
 
 static void attach_rules(struct list_head *rules, struct net_device *dev)
@@ -562,9 +655,12 @@ static void attach_rules(struct list_head *rules, struct net_device *dev)
        struct fib_rule *rule;
 
        list_for_each_entry(rule, rules, list) {
-               if (rule->ifindex == -1 &&
-                   strcmp(dev->name, rule->ifname) == 0)
-                       rule->ifindex = dev->ifindex;
+               if (rule->iifindex == -1 &&
+                   strcmp(dev->name, rule->iifname) == 0)
+                       rule->iifindex = dev->ifindex;
+               if (rule->oifindex == -1 &&
+                   strcmp(dev->name, rule->oifname) == 0)
+                       rule->oifindex = dev->ifindex;
        }
 }
 
@@ -572,9 +668,12 @@ static void detach_rules(struct list_head *rules, struct net_device *dev)
 {
        struct fib_rule *rule;
 
-       list_for_each_entry(rule, rules, list)
-               if (rule->ifindex == dev->ifindex)
-                       rule->ifindex = -1;
+       list_for_each_entry(rule, rules, list) {
+               if (rule->iifindex == dev->ifindex)
+                       rule->iifindex = -1;
+               if (rule->oifindex == dev->ifindex)
+                       rule->oifindex = -1;
+       }
 }
 
 
@@ -582,6 +681,7 @@ static int fib_rules_event(struct notifier_block *this, unsigned long event,
                            void *ptr)
 {
        struct net_device *dev = ptr;
+       struct net *net = dev_net(dev);
        struct fib_rules_ops *ops;
 
        ASSERT_RTNL();
@@ -589,13 +689,13 @@ static int fib_rules_event(struct notifier_block *this, unsigned long event,
 
        switch (event) {
        case NETDEV_REGISTER:
-               list_for_each_entry(ops, &rules_ops, list)
-                       attach_rules(ops->rules_list, dev);
+               list_for_each_entry(ops, &net->rules_ops, list)
+                       attach_rules(&ops->rules_list, dev);
                break;
 
        case NETDEV_UNREGISTER:
-               list_for_each_entry(ops, &rules_ops, list)
-                       detach_rules(ops->rules_list, dev);
+               list_for_each_entry(ops, &net->rules_ops, list)
+                       detach_rules(&ops->rules_list, dev);
                break;
        }
 
@@ -608,13 +708,41 @@ static struct notifier_block fib_rules_notifier = {
        .notifier_call = fib_rules_event,
 };
 
+static int __net_init fib_rules_net_init(struct net *net)
+{
+       INIT_LIST_HEAD(&net->rules_ops);
+       spin_lock_init(&net->rules_mod_lock);
+       return 0;
+}
+
+static struct pernet_operations fib_rules_net_ops = {
+       .init = fib_rules_net_init,
+};
+
 static int __init fib_rules_init(void)
 {
+       int err;
        rtnl_register(PF_UNSPEC, RTM_NEWRULE, fib_nl_newrule, NULL);
        rtnl_register(PF_UNSPEC, RTM_DELRULE, fib_nl_delrule, NULL);
        rtnl_register(PF_UNSPEC, RTM_GETRULE, NULL, fib_nl_dumprule);
 
-       return register_netdevice_notifier(&fib_rules_notifier);
+       err = register_pernet_subsys(&fib_rules_net_ops);
+       if (err < 0)
+               goto fail;
+
+       err = register_netdevice_notifier(&fib_rules_notifier);
+       if (err < 0)
+               goto fail_unregister;
+
+       return 0;
+
+fail_unregister:
+       unregister_pernet_subsys(&fib_rules_net_ops);
+fail:
+       rtnl_unregister(PF_UNSPEC, RTM_NEWRULE);
+       rtnl_unregister(PF_UNSPEC, RTM_DELRULE);
+       rtnl_unregister(PF_UNSPEC, RTM_GETRULE);
+       return err;
 }
 
 subsys_initcall(fib_rules_init);