netfilter: have ip*t REJECT set the sock err when an icmp is to be sent
[linux-3.10.git] / net / core / net_namespace.c
index ddd2cd2..f976520 100644 (file)
@@ -1,3 +1,5 @@
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
 #include <linux/workqueue.h>
 #include <linux/rtnetlink.h>
 #include <linux/cache.h>
@@ -8,6 +10,11 @@
 #include <linux/idr.h>
 #include <linux/rculist.h>
 #include <linux/nsproxy.h>
+#include <linux/fs.h>
+#include <linux/proc_ns.h>
+#include <linux/file.h>
+#include <linux/export.h>
+#include <linux/user_namespace.h>
 #include <net/net_namespace.h>
 #include <net/netns/generic.h>
 
@@ -22,32 +29,144 @@ static DEFINE_MUTEX(net_mutex);
 LIST_HEAD(net_namespace_list);
 EXPORT_SYMBOL_GPL(net_namespace_list);
 
-struct net init_net;
+struct net init_net = {
+       .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
+};
 EXPORT_SYMBOL(init_net);
 
 #define INITIAL_NET_GEN_PTRS   13 /* +1 for len +2 for rcu_head */
 
+static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
+
+static struct net_generic *net_alloc_generic(void)
+{
+       struct net_generic *ng;
+       size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
+
+       ng = kzalloc(generic_size, GFP_KERNEL);
+       if (ng)
+               ng->len = max_gen_ptrs;
+
+       return ng;
+}
+
+static int net_assign_generic(struct net *net, int id, void *data)
+{
+       struct net_generic *ng, *old_ng;
+
+       BUG_ON(!mutex_is_locked(&net_mutex));
+       BUG_ON(id == 0);
+
+       old_ng = rcu_dereference_protected(net->gen,
+                                          lockdep_is_held(&net_mutex));
+       ng = old_ng;
+       if (old_ng->len >= id)
+               goto assign;
+
+       ng = net_alloc_generic();
+       if (ng == NULL)
+               return -ENOMEM;
+
+       /*
+        * Some synchronisation notes:
+        *
+        * The net_generic explores the net->gen array inside rcu
+        * read section. Besides once set the net->gen->ptr[x]
+        * pointer never changes (see rules in netns/generic.h).
+        *
+        * That said, we simply duplicate this array and schedule
+        * the old copy for kfree after a grace period.
+        */
+
+       memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
+
+       rcu_assign_pointer(net->gen, ng);
+       kfree_rcu(old_ng, rcu);
+assign:
+       ng->ptr[id - 1] = data;
+       return 0;
+}
+
+static int ops_init(const struct pernet_operations *ops, struct net *net)
+{
+       int err = -ENOMEM;
+       void *data = NULL;
+
+       if (ops->id && ops->size) {
+               data = kzalloc(ops->size, GFP_KERNEL);
+               if (!data)
+                       goto out;
+
+               err = net_assign_generic(net, *ops->id, data);
+               if (err)
+                       goto cleanup;
+       }
+       err = 0;
+       if (ops->init)
+               err = ops->init(net);
+       if (!err)
+               return 0;
+
+cleanup:
+       kfree(data);
+
+out:
+       return err;
+}
+
+static void ops_free(const struct pernet_operations *ops, struct net *net)
+{
+       if (ops->id && ops->size) {
+               int id = *ops->id;
+               kfree(net_generic(net, id));
+       }
+}
+
+static void ops_exit_list(const struct pernet_operations *ops,
+                         struct list_head *net_exit_list)
+{
+       struct net *net;
+       if (ops->exit) {
+               list_for_each_entry(net, net_exit_list, exit_list)
+                       ops->exit(net);
+       }
+       if (ops->exit_batch)
+               ops->exit_batch(net_exit_list);
+}
+
+static void ops_free_list(const struct pernet_operations *ops,
+                         struct list_head *net_exit_list)
+{
+       struct net *net;
+       if (ops->size && ops->id) {
+               list_for_each_entry(net, net_exit_list, exit_list)
+                       ops_free(ops, net);
+       }
+}
+
 /*
  * setup_net runs the initializers for the network namespace object.
  */
-static __net_init int setup_net(struct net *net)
+static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 {
        /* Must be called with net_mutex held */
-       struct pernet_operations *ops;
+       const struct pernet_operations *ops, *saved_ops;
        int error = 0;
+       LIST_HEAD(net_exit_list);
 
        atomic_set(&net->count, 1);
+       atomic_set(&net->passive, 1);
+       net->dev_base_seq = 1;
+       net->user_ns = user_ns;
 
 #ifdef NETNS_REFCNT_DEBUG
        atomic_set(&net->use_count, 0);
 #endif
 
        list_for_each_entry(ops, &pernet_list, list) {
-               if (ops->init) {
-                       error = ops->init(net);
-                       if (error < 0)
-                               goto out_undo;
-               }
+               error = ops_init(ops, net);
+               if (error < 0)
+                       goto out_undo;
        }
 out:
        return error;
@@ -56,27 +175,19 @@ out_undo:
        /* Walk through the list backwards calling the exit functions
         * for the pernet modules whose init functions did not fail.
         */
-       list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
-               if (ops->exit)
-                       ops->exit(net);
-       }
+       list_add(&net->exit_list, &net_exit_list);
+       saved_ops = ops;
+       list_for_each_entry_continue_reverse(ops, &pernet_list, list)
+               ops_exit_list(ops, &net_exit_list);
+
+       ops = saved_ops;
+       list_for_each_entry_continue_reverse(ops, &pernet_list, list)
+               ops_free_list(ops, &net_exit_list);
 
        rcu_barrier();
        goto out;
 }
 
-static struct net_generic *net_alloc_generic(void)
-{
-       struct net_generic *ng;
-       size_t generic_size = sizeof(struct net_generic) +
-               INITIAL_NET_GEN_PTRS * sizeof(void *);
-
-       ng = kzalloc(generic_size, GFP_KERNEL);
-       if (ng)
-               ng->len = INITIAL_NET_GEN_PTRS;
-
-       return ng;
-}
 
 #ifdef CONFIG_NET_NS
 static struct kmem_cache *net_cachep;
@@ -108,8 +219,8 @@ static void net_free(struct net *net)
 {
 #ifdef NETNS_REFCNT_DEBUG
        if (unlikely(atomic_read(&net->use_count) != 0)) {
-               printk(KERN_EMERG "network namespace not free! Usage: %d\n",
-                       atomic_read(&net->use_count));
+               pr_emerg("network namespace not free! Usage: %d\n",
+                        atomic_read(&net->use_count));
                return;
        }
 #endif
@@ -117,16 +228,30 @@ static void net_free(struct net *net)
        kmem_cache_free(net_cachep, net);
 }
 
-static struct net *net_create(void)
+void net_drop_ns(void *p)
+{
+       struct net *ns = p;
+       if (ns && atomic_dec_and_test(&ns->passive))
+               net_free(ns);
+}
+
+struct net *copy_net_ns(unsigned long flags,
+                       struct user_namespace *user_ns, struct net *old_net)
 {
        struct net *net;
        int rv;
 
+       if (!(flags & CLONE_NEWNET))
+               return get_net(old_net);
+
        net = net_alloc();
        if (!net)
                return ERR_PTR(-ENOMEM);
+
+       get_user_ns(user_ns);
+
        mutex_lock(&net_mutex);
-       rv = setup_net(net);
+       rv = setup_net(net, user_ns);
        if (rv == 0) {
                rtnl_lock();
                list_add_tail_rcu(&net->list, &net_namespace_list);
@@ -134,31 +259,36 @@ static struct net *net_create(void)
        }
        mutex_unlock(&net_mutex);
        if (rv < 0) {
-               net_free(net);
+               put_user_ns(user_ns);
+               net_drop_ns(net);
                return ERR_PTR(rv);
        }
        return net;
 }
 
-struct net *copy_net_ns(unsigned long flags, struct net *old_net)
-{
-       if (!(flags & CLONE_NEWNET))
-               return get_net(old_net);
-       return net_create();
-}
+static DEFINE_SPINLOCK(cleanup_list_lock);
+static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 
 static void cleanup_net(struct work_struct *work)
 {
-       struct pernet_operations *ops;
-       struct net *net;
+       const struct pernet_operations *ops;
+       struct net *net, *tmp;
+       LIST_HEAD(net_kill_list);
+       LIST_HEAD(net_exit_list);
 
-       net = container_of(work, struct net, work);
+       /* Atomically snapshot the list of namespaces to cleanup */
+       spin_lock_irq(&cleanup_list_lock);
+       list_replace_init(&cleanup_list, &net_kill_list);
+       spin_unlock_irq(&cleanup_list_lock);
 
        mutex_lock(&net_mutex);
 
        /* Don't let anyone else find us. */
        rtnl_lock();
-       list_del_rcu(&net->list);
+       list_for_each_entry(net, &net_kill_list, cleanup_list) {
+               list_del_rcu(&net->list);
+               list_add_tail(&net->exit_list, &net_exit_list);
+       }
        rtnl_unlock();
 
        /*
@@ -169,10 +299,12 @@ static void cleanup_net(struct work_struct *work)
        synchronize_rcu();
 
        /* Run all of the network namespace exit methods */
-       list_for_each_entry_reverse(ops, &pernet_list, list) {
-               if (ops->exit)
-                       ops->exit(net);
-       }
+       list_for_each_entry_reverse(ops, &pernet_list, list)
+               ops_exit_list(ops, &net_exit_list);
+
+       /* Free the net generic variables */
+       list_for_each_entry_reverse(ops, &pernet_list, list)
+               ops_free_list(ops, &net_exit_list);
 
        mutex_unlock(&net_mutex);
 
@@ -182,23 +314,51 @@ static void cleanup_net(struct work_struct *work)
        rcu_barrier();
 
        /* Finally it is safe to free my network namespace structure */
-       net_free(net);
+       list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
+               list_del_init(&net->exit_list);
+               put_user_ns(net->user_ns);
+               net_drop_ns(net);
+       }
 }
+static DECLARE_WORK(net_cleanup_work, cleanup_net);
 
 void __put_net(struct net *net)
 {
        /* Cleanup the network namespace in process context */
-       INIT_WORK(&net->work, cleanup_net);
-       queue_work(netns_wq, &net->work);
+       unsigned long flags;
+
+       spin_lock_irqsave(&cleanup_list_lock, flags);
+       list_add(&net->cleanup_list, &cleanup_list);
+       spin_unlock_irqrestore(&cleanup_list_lock, flags);
+
+       queue_work(netns_wq, &net_cleanup_work);
 }
 EXPORT_SYMBOL_GPL(__put_net);
 
+struct net *get_net_ns_by_fd(int fd)
+{
+       struct proc_ns *ei;
+       struct file *file;
+       struct net *net;
+
+       file = proc_ns_fget(fd);
+       if (IS_ERR(file))
+               return ERR_CAST(file);
+
+       ei = get_proc_ns(file_inode(file));
+       if (ei->ns_ops == &netns_operations)
+               net = get_net(ei->ns);
+       else
+               net = ERR_PTR(-EINVAL);
+
+       fput(file);
+       return net;
+}
+
 #else
-struct net *copy_net_ns(unsigned long flags, struct net *old_net)
+struct net *get_net_ns_by_fd(int fd)
 {
-       if (flags & CLONE_NEWNET)
-               return ERR_PTR(-EINVAL);
-       return old_net;
+       return ERR_PTR(-EINVAL);
 }
 #endif
 
@@ -222,6 +382,21 @@ struct net *get_net_ns_by_pid(pid_t pid)
 }
 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 
+static __net_init int net_ns_net_init(struct net *net)
+{
+       return proc_alloc_inum(&net->proc_inum);
+}
+
+static __net_exit void net_ns_net_exit(struct net *net)
+{
+       proc_free_inum(net->proc_inum);
+}
+
+static struct pernet_operations __net_initdata net_ns_ops = {
+       .init = net_ns_net_init,
+       .exit = net_ns_net_exit,
+};
+
 static int __init net_ns_init(void)
 {
        struct net_generic *ng;
@@ -244,7 +419,7 @@ static int __init net_ns_init(void)
        rcu_assign_pointer(init_net.gen, ng);
 
        mutex_lock(&net_mutex);
-       if (setup_net(&init_net))
+       if (setup_net(&init_net, &init_user_ns))
                panic("Could not setup the initial network namespace");
 
        rtnl_lock();
@@ -253,24 +428,28 @@ static int __init net_ns_init(void)
 
        mutex_unlock(&net_mutex);
 
+       register_pernet_subsys(&net_ns_ops);
+
        return 0;
 }
 
 pure_initcall(net_ns_init);
 
 #ifdef CONFIG_NET_NS
-static int register_pernet_operations(struct list_head *list,
-                                     struct pernet_operations *ops)
+static int __register_pernet_operations(struct list_head *list,
+                                       struct pernet_operations *ops)
 {
-       struct net *net, *undo_net;
+       struct net *net;
        int error;
+       LIST_HEAD(net_exit_list);
 
        list_add_tail(&ops->list, list);
-       if (ops->init) {
+       if (ops->init || (ops->id && ops->size)) {
                for_each_net(net) {
-                       error = ops->init(net);
+                       error = ops_init(ops, net);
                        if (error)
                                goto out_undo;
+                       list_add_tail(&net->exit_list, &net_exit_list);
                }
        }
        return 0;
@@ -278,45 +457,78 @@ static int register_pernet_operations(struct list_head *list,
 out_undo:
        /* If I have an error cleanup all namespaces I initialized */
        list_del(&ops->list);
-       if (ops->exit) {
-               for_each_net(undo_net) {
-                       if (undo_net == net)
-                               goto undone;
-                       ops->exit(undo_net);
-               }
-       }
-undone:
+       ops_exit_list(ops, &net_exit_list);
+       ops_free_list(ops, &net_exit_list);
        return error;
 }
 
-static void unregister_pernet_operations(struct pernet_operations *ops)
+static void __unregister_pernet_operations(struct pernet_operations *ops)
 {
        struct net *net;
+       LIST_HEAD(net_exit_list);
 
        list_del(&ops->list);
-       if (ops->exit)
-               for_each_net(net)
-                       ops->exit(net);
+       for_each_net(net)
+               list_add_tail(&net->exit_list, &net_exit_list);
+       ops_exit_list(ops, &net_exit_list);
+       ops_free_list(ops, &net_exit_list);
 }
 
 #else
 
+static int __register_pernet_operations(struct list_head *list,
+                                       struct pernet_operations *ops)
+{
+       return ops_init(ops, &init_net);
+}
+
+static void __unregister_pernet_operations(struct pernet_operations *ops)
+{
+       LIST_HEAD(net_exit_list);
+       list_add(&init_net.exit_list, &net_exit_list);
+       ops_exit_list(ops, &net_exit_list);
+       ops_free_list(ops, &net_exit_list);
+}
+
+#endif /* CONFIG_NET_NS */
+
+static DEFINE_IDA(net_generic_ids);
+
 static int register_pernet_operations(struct list_head *list,
                                      struct pernet_operations *ops)
 {
-       if (ops->init == NULL)
-               return 0;
-       return ops->init(&init_net);
+       int error;
+
+       if (ops->id) {
+again:
+               error = ida_get_new_above(&net_generic_ids, 1, ops->id);
+               if (error < 0) {
+                       if (error == -EAGAIN) {
+                               ida_pre_get(&net_generic_ids, GFP_KERNEL);
+                               goto again;
+                       }
+                       return error;
+               }
+               max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
+       }
+       error = __register_pernet_operations(list, ops);
+       if (error) {
+               rcu_barrier();
+               if (ops->id)
+                       ida_remove(&net_generic_ids, *ops->id);
+       }
+
+       return error;
 }
 
 static void unregister_pernet_operations(struct pernet_operations *ops)
 {
-       if (ops->exit)
-               ops->exit(&init_net);
+       
+       __unregister_pernet_operations(ops);
+       rcu_barrier();
+       if (ops->id)
+               ida_remove(&net_generic_ids, *ops->id);
 }
-#endif
-
-static DEFINE_IDA(net_generic_ids);
 
 /**
  *      register_pernet_subsys - register a network namespace subsystem
@@ -356,45 +568,13 @@ EXPORT_SYMBOL_GPL(register_pernet_subsys);
  *     addition run the exit method for all existing network
  *     namespaces.
  */
-void unregister_pernet_subsys(struct pernet_operations *module)
-{
-       mutex_lock(&net_mutex);
-       unregister_pernet_operations(module);
-       mutex_unlock(&net_mutex);
-}
-EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
-
-int register_pernet_gen_subsys(int *id, struct pernet_operations *ops)
-{
-       int rv;
-
-       mutex_lock(&net_mutex);
-again:
-       rv = ida_get_new_above(&net_generic_ids, 1, id);
-       if (rv < 0) {
-               if (rv == -EAGAIN) {
-                       ida_pre_get(&net_generic_ids, GFP_KERNEL);
-                       goto again;
-               }
-               goto out;
-       }
-       rv = register_pernet_operations(first_device, ops);
-       if (rv < 0)
-               ida_remove(&net_generic_ids, *id);
-out:
-       mutex_unlock(&net_mutex);
-       return rv;
-}
-EXPORT_SYMBOL_GPL(register_pernet_gen_subsys);
-
-void unregister_pernet_gen_subsys(int id, struct pernet_operations *ops)
+void unregister_pernet_subsys(struct pernet_operations *ops)
 {
        mutex_lock(&net_mutex);
        unregister_pernet_operations(ops);
-       ida_remove(&net_generic_ids, id);
        mutex_unlock(&net_mutex);
 }
-EXPORT_SYMBOL_GPL(unregister_pernet_gen_subsys);
+EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 
 /**
  *      register_pernet_device - register a network namespace device
@@ -427,30 +607,6 @@ int register_pernet_device(struct pernet_operations *ops)
 }
 EXPORT_SYMBOL_GPL(register_pernet_device);
 
-int register_pernet_gen_device(int *id, struct pernet_operations *ops)
-{
-       int error;
-       mutex_lock(&net_mutex);
-again:
-       error = ida_get_new_above(&net_generic_ids, 1, id);
-       if (error) {
-               if (error == -EAGAIN) {
-                       ida_pre_get(&net_generic_ids, GFP_KERNEL);
-                       goto again;
-               }
-               goto out;
-       }
-       error = register_pernet_operations(&pernet_list, ops);
-       if (error)
-               ida_remove(&net_generic_ids, *id);
-       else if (first_device == &pernet_list)
-               first_device = &ops->list;
-out:
-       mutex_unlock(&net_mutex);
-       return error;
-}
-EXPORT_SYMBOL_GPL(register_pernet_gen_device);
-
 /**
  *      unregister_pernet_device - unregister a network namespace netdevice
  *     @ops: pernet operations structure to manipulate
@@ -470,59 +626,51 @@ void unregister_pernet_device(struct pernet_operations *ops)
 }
 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 
-void unregister_pernet_gen_device(int id, struct pernet_operations *ops)
+#ifdef CONFIG_NET_NS
+static void *netns_get(struct task_struct *task)
 {
-       mutex_lock(&net_mutex);
-       if (&ops->list == first_device)
-               first_device = first_device->next;
-       unregister_pernet_operations(ops);
-       ida_remove(&net_generic_ids, id);
-       mutex_unlock(&net_mutex);
-}
-EXPORT_SYMBOL_GPL(unregister_pernet_gen_device);
+       struct net *net = NULL;
+       struct nsproxy *nsproxy;
 
-static void net_generic_release(struct rcu_head *rcu)
-{
-       struct net_generic *ng;
+       rcu_read_lock();
+       nsproxy = task_nsproxy(task);
+       if (nsproxy)
+               net = get_net(nsproxy->net_ns);
+       rcu_read_unlock();
 
-       ng = container_of(rcu, struct net_generic, rcu);
-       kfree(ng);
+       return net;
 }
 
-int net_assign_generic(struct net *net, int id, void *data)
+static void netns_put(void *ns)
 {
-       struct net_generic *ng, *old_ng;
-
-       BUG_ON(!mutex_is_locked(&net_mutex));
-       BUG_ON(id == 0);
-
-       ng = old_ng = net->gen;
-       if (old_ng->len >= id)
-               goto assign;
-
-       ng = kzalloc(sizeof(struct net_generic) +
-                       id * sizeof(void *), GFP_KERNEL);
-       if (ng == NULL)
-               return -ENOMEM;
+       put_net(ns);
+}
 
-       /*
-        * Some synchronisation notes:
-        *
-        * The net_generic explores the net->gen array inside rcu
-        * read section. Besides once set the net->gen->ptr[x]
-        * pointer never changes (see rules in netns/generic.h).
-        *
-        * That said, we simply duplicate this array and schedule
-        * the old copy for kfree after a grace period.
-        */
+static int netns_install(struct nsproxy *nsproxy, void *ns)
+{
+       struct net *net = ns;
 
-       ng->len = id;
-       memcpy(&ng->ptr, &old_ng->ptr, old_ng->len);
+       if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
+           !nsown_capable(CAP_SYS_ADMIN))
+               return -EPERM;
 
-       rcu_assign_pointer(net->gen, ng);
-       call_rcu(&old_ng->rcu, net_generic_release);
-assign:
-       ng->ptr[id - 1] = data;
+       put_net(nsproxy->net_ns);
+       nsproxy->net_ns = get_net(net);
        return 0;
 }
-EXPORT_SYMBOL_GPL(net_assign_generic);
+
+static unsigned int netns_inum(void *ns)
+{
+       struct net *net = ns;
+       return net->proc_inum;
+}
+
+const struct proc_ns_operations netns_operations = {
+       .name           = "net",
+       .type           = CLONE_NEWNET,
+       .get            = netns_get,
+       .put            = netns_put,
+       .install        = netns_install,
+       .inum           = netns_inum,
+};
+#endif