net/core/net_namespace.c

   1 #include <linux/workqueue.h>
   2 #include <linux/rtnetlink.h>
   3 #include <linux/cache.h>
   4 #include <linux/slab.h>
   5 #include <linux/list.h>
   6 #include <linux/delay.h>
   7 #include <linux/sched.h>
   8 #include <linux/idr.h>
   9 #include <net/net_namespace.h>
  10
  11 /*
  12  *      Our network namespace constructor/destructor lists
  13  */
  14
  15 static LIST_HEAD(pernet_list);
  16 static struct list_head *first_device = &pernet_list;
  17 static DEFINE_MUTEX(net_mutex);
  18
  19 LIST_HEAD(net_namespace_list);
  20
  21 struct net init_net;
  22 EXPORT_SYMBOL(init_net);
  23
  24 /*
  25  * setup_net runs the initializers for the network namespace object.
  26  */
  27 static __net_init int setup_net(struct net *net)
  28 {
  29         /* Must be called with net_mutex held */
  30         struct pernet_operations *ops;
  31         int error;
  32
  33         atomic_set(&net->count, 1);
  34         atomic_set(&net->use_count, 0);
  35
  36         error = 0;
  37         list_for_each_entry(ops, &pernet_list, list) {
  38                 if (ops->init) {
  39                         error = ops->init(net);
  40                         if (error < 0)
  41                                 goto out_undo;
  42                 }
  43         }
  44 out:
  45         return error;
  46
  47 out_undo:
  48         /* Walk through the list backwards calling the exit functions
  49          * for the pernet modules whose init functions did not fail.
  50          */
  51         list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
  52                 if (ops->exit)
  53                         ops->exit(net);
  54         }
  55
  56         rcu_barrier();
  57         goto out;
  58 }
  59
  60 #ifdef CONFIG_NET_NS
  61 static struct kmem_cache *net_cachep;
  62 static struct workqueue_struct *netns_wq;
  63
  64 static struct net *net_alloc(void)
  65 {
  66         return kmem_cache_zalloc(net_cachep, GFP_KERNEL);
  67 }
  68
  69 static void net_free(struct net *net)
  70 {
  71         if (!net)
  72                 return;
  73
  74         if (unlikely(atomic_read(&net->use_count) != 0)) {
  75                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
  76                         atomic_read(&net->use_count));
  77                 return;
  78         }
  79
  80         kmem_cache_free(net_cachep, net);
  81 }
  82
  83 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
  84 {
  85         struct net *new_net = NULL;
  86         int err;
  87
  88         get_net(old_net);
  89
  90         if (!(flags & CLONE_NEWNET))
  91                 return old_net;
  92
  93         err = -ENOMEM;
  94         new_net = net_alloc();
  95         if (!new_net)
  96                 goto out;
  97
  98         mutex_lock(&net_mutex);
  99         err = setup_net(new_net);
 100         if (err)
 101                 goto out_unlock;
 102
 103         rtnl_lock();
 104         list_add_tail(&new_net->list, &net_namespace_list);
 105         rtnl_unlock();
 106
 107
 108 out_unlock:
 109         mutex_unlock(&net_mutex);
 110 out:
 111         put_net(old_net);
 112         if (err) {
 113                 net_free(new_net);
 114                 new_net = ERR_PTR(err);
 115         }
 116         return new_net;
 117 }
 118
 119 static void cleanup_net(struct work_struct *work)
 120 {
 121         struct pernet_operations *ops;
 122         struct net *net;
 123
 124         net = container_of(work, struct net, work);
 125
 126         mutex_lock(&net_mutex);
 127
 128         /* Don't let anyone else find us. */
 129         rtnl_lock();
 130         list_del(&net->list);
 131         rtnl_unlock();
 132
 133         /* Run all of the network namespace exit methods */
 134         list_for_each_entry_reverse(ops, &pernet_list, list) {
 135                 if (ops->exit)
 136                         ops->exit(net);
 137         }
 138
 139         mutex_unlock(&net_mutex);
 140
 141         /* Ensure there are no outstanding rcu callbacks using this
 142          * network namespace.
 143          */
 144         rcu_barrier();
 145
 146         /* Finally it is safe to free my network namespace structure */
 147         net_free(net);
 148 }
 149
 150 void __put_net(struct net *net)
 151 {
 152         /* Cleanup the network namespace in process context */
 153         INIT_WORK(&net->work, cleanup_net);
 154         queue_work(netns_wq, &net->work);
 155 }
 156 EXPORT_SYMBOL_GPL(__put_net);
 157
 158 #else
 159 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 160 {
 161         if (flags & CLONE_NEWNET)
 162                 return ERR_PTR(-EINVAL);
 163         return old_net;
 164 }
 165 #endif
 166
 167 static int __init net_ns_init(void)
 168 {
 169         int err;
 170
 171         printk(KERN_INFO "net_namespace: %zd bytes\n", sizeof(struct net));
 172 #ifdef CONFIG_NET_NS
 173         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 174                                         SMP_CACHE_BYTES,
 175                                         SLAB_PANIC, NULL);
 176
 177         /* Create workqueue for cleanup */
 178         netns_wq = create_singlethread_workqueue("netns");
 179         if (!netns_wq)
 180                 panic("Could not create netns workq");
 181 #endif
 182
 183         mutex_lock(&net_mutex);
 184         err = setup_net(&init_net);
 185
 186         rtnl_lock();
 187         list_add_tail(&init_net.list, &net_namespace_list);
 188         rtnl_unlock();
 189
 190         mutex_unlock(&net_mutex);
 191         if (err)
 192                 panic("Could not setup the initial network namespace");
 193
 194         return 0;
 195 }
 196
 197 pure_initcall(net_ns_init);
 198
 199 #ifdef CONFIG_NET_NS
 200 static int register_pernet_operations(struct list_head *list,
 201                                       struct pernet_operations *ops)
 202 {
 203         struct net *net, *undo_net;
 204         int error;
 205
 206         list_add_tail(&ops->list, list);
 207         if (ops->init) {
 208                 for_each_net(net) {
 209                         error = ops->init(net);
 210                         if (error)
 211                                 goto out_undo;
 212                 }
 213         }
 214         return 0;
 215
 216 out_undo:
 217         /* If I have an error cleanup all namespaces I initialized */
 218         list_del(&ops->list);
 219         if (ops->exit) {
 220                 for_each_net(undo_net) {
 221                         if (undo_net == net)
 222                                 goto undone;
 223                         ops->exit(undo_net);
 224                 }
 225         }
 226 undone:
 227         return error;
 228 }
 229
 230 static void unregister_pernet_operations(struct pernet_operations *ops)
 231 {
 232         struct net *net;
 233
 234         list_del(&ops->list);
 235         if (ops->exit)
 236                 for_each_net(net)
 237                         ops->exit(net);
 238 }
 239
 240 #else
 241
 242 static int register_pernet_operations(struct list_head *list,
 243                                       struct pernet_operations *ops)
 244 {
 245         if (ops->init == NULL)
 246                 return 0;
 247         return ops->init(&init_net);
 248 }
 249
 250 static void unregister_pernet_operations(struct pernet_operations *ops)
 251 {
 252         if (ops->exit)
 253                 ops->exit(&init_net);
 254 }
 255 #endif
 256
 257 static DEFINE_IDA(net_generic_ids);
 258
 259 /**
 260  *      register_pernet_subsys - register a network namespace subsystem
 261  *      @ops:  pernet operations structure for the subsystem
 262  *
 263  *      Register a subsystem which has init and exit functions
 264  *      that are called when network namespaces are created and
 265  *      destroyed respectively.
 266  *
 267  *      When registered all network namespace init functions are
 268  *      called for every existing network namespace.  Allowing kernel
 269  *      modules to have a race free view of the set of network namespaces.
 270  *
 271  *      When a new network namespace is created all of the init
 272  *      methods are called in the order in which they were registered.
 273  *
 274  *      When a network namespace is destroyed all of the exit methods
 275  *      are called in the reverse of the order with which they were
 276  *      registered.
 277  */
 278 int register_pernet_subsys(struct pernet_operations *ops)
 279 {
 280         int error;
 281         mutex_lock(&net_mutex);
 282         error =  register_pernet_operations(first_device, ops);
 283         mutex_unlock(&net_mutex);
 284         return error;
 285 }
 286 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 287
 288 /**
 289  *      unregister_pernet_subsys - unregister a network namespace subsystem
 290  *      @ops: pernet operations structure to manipulate
 291  *
 292  *      Remove the pernet operations structure from the list to be
 293  *      used when network namespaces are created or destroyed.  In
 294  *      addition run the exit method for all existing network
 295  *      namespaces.
 296  */
 297 void unregister_pernet_subsys(struct pernet_operations *module)
 298 {
 299         mutex_lock(&net_mutex);
 300         unregister_pernet_operations(module);
 301         mutex_unlock(&net_mutex);
 302 }
 303 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 304
 305 /**
 306  *      register_pernet_device - register a network namespace device
 307  *      @ops:  pernet operations structure for the subsystem
 308  *
 309  *      Register a device which has init and exit functions
 310  *      that are called when network namespaces are created and
 311  *      destroyed respectively.
 312  *
 313  *      When registered all network namespace init functions are
 314  *      called for every existing network namespace.  Allowing kernel
 315  *      modules to have a race free view of the set of network namespaces.
 316  *
 317  *      When a new network namespace is created all of the init
 318  *      methods are called in the order in which they were registered.
 319  *
 320  *      When a network namespace is destroyed all of the exit methods
 321  *      are called in the reverse of the order with which they were
 322  *      registered.
 323  */
 324 int register_pernet_device(struct pernet_operations *ops)
 325 {
 326         int error;
 327         mutex_lock(&net_mutex);
 328         error = register_pernet_operations(&pernet_list, ops);
 329         if (!error && (first_device == &pernet_list))
 330                 first_device = &ops->list;
 331         mutex_unlock(&net_mutex);
 332         return error;
 333 }
 334 EXPORT_SYMBOL_GPL(register_pernet_device);
 335
 336 int register_pernet_gen_device(int *id, struct pernet_operations *ops)
 337 {
 338         int error;
 339         mutex_lock(&net_mutex);
 340 again:
 341         error = ida_get_new_above(&net_generic_ids, 1, id);
 342         if (error) {
 343                 if (error == -EAGAIN) {
 344                         ida_pre_get(&net_generic_ids, GFP_KERNEL);
 345                         goto again;
 346                 }
 347                 goto out;
 348         }
 349         error = register_pernet_operations(&pernet_list, ops);
 350         if (error)
 351                 ida_remove(&net_generic_ids, *id);
 352         else if (first_device == &pernet_list)
 353                 first_device = &ops->list;
 354 out:
 355         mutex_unlock(&net_mutex);
 356         return error;
 357 }
 358 EXPORT_SYMBOL_GPL(register_pernet_gen_device);
 359
 360 /**
 361  *      unregister_pernet_device - unregister a network namespace netdevice
 362  *      @ops: pernet operations structure to manipulate
 363  *
 364  *      Remove the pernet operations structure from the list to be
 365  *      used when network namespaces are created or destroyed.  In
 366  *      addition run the exit method for all existing network
 367  *      namespaces.
 368  */
 369 void unregister_pernet_device(struct pernet_operations *ops)
 370 {
 371         mutex_lock(&net_mutex);
 372         if (&ops->list == first_device)
 373                 first_device = first_device->next;
 374         unregister_pernet_operations(ops);
 375         mutex_unlock(&net_mutex);
 376 }
 377 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 378
 379 void unregister_pernet_gen_device(int id, struct pernet_operations *ops)
 380 {
 381         mutex_lock(&net_mutex);
 382         if (&ops->list == first_device)
 383                 first_device = first_device->next;
 384         unregister_pernet_operations(ops);
 385         ida_remove(&net_generic_ids, id);
 386         mutex_unlock(&net_mutex);
 387 }
 388 EXPORT_SYMBOL_GPL(unregister_pernet_gen_device);