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 #include <net/netns/generic.h>
  11
  12 /*
  13  *      Our network namespace constructor/destructor lists
  14  */
  15
  16 static LIST_HEAD(pernet_list);
  17 static struct list_head *first_device = &pernet_list;
  18 static DEFINE_MUTEX(net_mutex);
  19
  20 LIST_HEAD(net_namespace_list);
  21 EXPORT_SYMBOL_GPL(net_namespace_list);
  22
  23 struct net init_net;
  24 EXPORT_SYMBOL(init_net);
  25
  26 #define INITIAL_NET_GEN_PTRS    13 /* +1 for len +2 for rcu_head */
  27
  28 /*
  29  * setup_net runs the initializers for the network namespace object.
  30  */
  31 static __net_init int setup_net(struct net *net)
  32 {
  33         /* Must be called with net_mutex held */
  34         struct pernet_operations *ops;
  35         int error = 0;
  36
  37         atomic_set(&net->count, 1);
  38
  39 #ifdef NETNS_REFCNT_DEBUG
  40         atomic_set(&net->use_count, 0);
  41 #endif
  42
  43         list_for_each_entry(ops, &pernet_list, list) {
  44                 if (ops->init) {
  45                         error = ops->init(net);
  46                         if (error < 0)
  47                                 goto out_undo;
  48                 }
  49         }
  50 out:
  51         return error;
  52
  53 out_undo:
  54         /* Walk through the list backwards calling the exit functions
  55          * for the pernet modules whose init functions did not fail.
  56          */
  57         list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
  58                 if (ops->exit)
  59                         ops->exit(net);
  60         }
  61
  62         rcu_barrier();
  63         goto out;
  64 }
  65
  66 static struct net_generic *net_alloc_generic(void)
  67 {
  68         struct net_generic *ng;
  69         size_t generic_size = sizeof(struct net_generic) +
  70                 INITIAL_NET_GEN_PTRS * sizeof(void *);
  71
  72         ng = kzalloc(generic_size, GFP_KERNEL);
  73         if (ng)
  74                 ng->len = INITIAL_NET_GEN_PTRS;
  75
  76         return ng;
  77 }
  78
  79 #ifdef CONFIG_NET_NS
  80 static struct kmem_cache *net_cachep;
  81 static struct workqueue_struct *netns_wq;
  82
  83 static struct net *net_alloc(void)
  84 {
  85         struct net *net = NULL;
  86         struct net_generic *ng;
  87
  88         ng = net_alloc_generic();
  89         if (!ng)
  90                 goto out;
  91
  92         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
  93         if (!net)
  94                 goto out_free;
  95
  96         rcu_assign_pointer(net->gen, ng);
  97 out:
  98         return net;
  99
 100 out_free:
 101         kfree(ng);
 102         goto out;
 103 }
 104
 105 static void net_free(struct net *net)
 106 {
 107 #ifdef NETNS_REFCNT_DEBUG
 108         if (unlikely(atomic_read(&net->use_count) != 0)) {
 109                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
 110                         atomic_read(&net->use_count));
 111                 return;
 112         }
 113 #endif
 114         kfree(net->gen);
 115         kmem_cache_free(net_cachep, net);
 116 }
 117
 118 static struct net *net_create(void)
 119 {
 120         struct net *net;
 121         int rv;
 122
 123         net = net_alloc();
 124         if (!net)
 125                 return ERR_PTR(-ENOMEM);
 126         mutex_lock(&net_mutex);
 127         rv = setup_net(net);
 128         if (rv == 0) {
 129                 rtnl_lock();
 130                 list_add_tail(&net->list, &net_namespace_list);
 131                 rtnl_unlock();
 132         }
 133         mutex_unlock(&net_mutex);
 134         if (rv < 0) {
 135                 net_free(net);
 136                 return ERR_PTR(rv);
 137         }
 138         return net;
 139 }
 140
 141 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 142 {
 143         if (!(flags & CLONE_NEWNET))
 144                 return get_net(old_net);
 145         return net_create();
 146 }
 147
 148 static void cleanup_net(struct work_struct *work)
 149 {
 150         struct pernet_operations *ops;
 151         struct net *net;
 152
 153         net = container_of(work, struct net, work);
 154
 155         mutex_lock(&net_mutex);
 156
 157         /* Don't let anyone else find us. */
 158         rtnl_lock();
 159         list_del(&net->list);
 160         rtnl_unlock();
 161
 162         /* Run all of the network namespace exit methods */
 163         list_for_each_entry_reverse(ops, &pernet_list, list) {
 164                 if (ops->exit)
 165                         ops->exit(net);
 166         }
 167
 168         mutex_unlock(&net_mutex);
 169
 170         /* Ensure there are no outstanding rcu callbacks using this
 171          * network namespace.
 172          */
 173         rcu_barrier();
 174
 175         /* Finally it is safe to free my network namespace structure */
 176         net_free(net);
 177 }
 178
 179 void __put_net(struct net *net)
 180 {
 181         /* Cleanup the network namespace in process context */
 182         INIT_WORK(&net->work, cleanup_net);
 183         queue_work(netns_wq, &net->work);
 184 }
 185 EXPORT_SYMBOL_GPL(__put_net);
 186
 187 #else
 188 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 189 {
 190         if (flags & CLONE_NEWNET)
 191                 return ERR_PTR(-EINVAL);
 192         return old_net;
 193 }
 194 #endif
 195
 196 static int __init net_ns_init(void)
 197 {
 198         struct net_generic *ng;
 199
 200 #ifdef CONFIG_NET_NS
 201         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 202                                         SMP_CACHE_BYTES,
 203                                         SLAB_PANIC, NULL);
 204
 205         /* Create workqueue for cleanup */
 206         netns_wq = create_singlethread_workqueue("netns");
 207         if (!netns_wq)
 208                 panic("Could not create netns workq");
 209 #endif
 210
 211         ng = net_alloc_generic();
 212         if (!ng)
 213                 panic("Could not allocate generic netns");
 214
 215         rcu_assign_pointer(init_net.gen, ng);
 216
 217         mutex_lock(&net_mutex);
 218         if (setup_net(&init_net))
 219                 panic("Could not setup the initial network namespace");
 220
 221         rtnl_lock();
 222         list_add_tail(&init_net.list, &net_namespace_list);
 223         rtnl_unlock();
 224
 225         mutex_unlock(&net_mutex);
 226
 227         return 0;
 228 }
 229
 230 pure_initcall(net_ns_init);
 231
 232 #ifdef CONFIG_NET_NS
 233 static int register_pernet_operations(struct list_head *list,
 234                                       struct pernet_operations *ops)
 235 {
 236         struct net *net, *undo_net;
 237         int error;
 238
 239         list_add_tail(&ops->list, list);
 240         if (ops->init) {
 241                 for_each_net(net) {
 242                         error = ops->init(net);
 243                         if (error)
 244                                 goto out_undo;
 245                 }
 246         }
 247         return 0;
 248
 249 out_undo:
 250         /* If I have an error cleanup all namespaces I initialized */
 251         list_del(&ops->list);
 252         if (ops->exit) {
 253                 for_each_net(undo_net) {
 254                         if (undo_net == net)
 255                                 goto undone;
 256                         ops->exit(undo_net);
 257                 }
 258         }
 259 undone:
 260         return error;
 261 }
 262
 263 static void unregister_pernet_operations(struct pernet_operations *ops)
 264 {
 265         struct net *net;
 266
 267         list_del(&ops->list);
 268         if (ops->exit)
 269                 for_each_net(net)
 270                         ops->exit(net);
 271 }
 272
 273 #else
 274
 275 static int register_pernet_operations(struct list_head *list,
 276                                       struct pernet_operations *ops)
 277 {
 278         if (ops->init == NULL)
 279                 return 0;
 280         return ops->init(&init_net);
 281 }
 282
 283 static void unregister_pernet_operations(struct pernet_operations *ops)
 284 {
 285         if (ops->exit)
 286                 ops->exit(&init_net);
 287 }
 288 #endif
 289
 290 static DEFINE_IDA(net_generic_ids);
 291
 292 /**
 293  *      register_pernet_subsys - register a network namespace subsystem
 294  *      @ops:  pernet operations structure for the subsystem
 295  *
 296  *      Register a subsystem which has init and exit functions
 297  *      that are called when network namespaces are created and
 298  *      destroyed respectively.
 299  *
 300  *      When registered all network namespace init functions are
 301  *      called for every existing network namespace.  Allowing kernel
 302  *      modules to have a race free view of the set of network namespaces.
 303  *
 304  *      When a new network namespace is created all of the init
 305  *      methods are called in the order in which they were registered.
 306  *
 307  *      When a network namespace is destroyed all of the exit methods
 308  *      are called in the reverse of the order with which they were
 309  *      registered.
 310  */
 311 int register_pernet_subsys(struct pernet_operations *ops)
 312 {
 313         int error;
 314         mutex_lock(&net_mutex);
 315         error =  register_pernet_operations(first_device, ops);
 316         mutex_unlock(&net_mutex);
 317         return error;
 318 }
 319 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 320
 321 /**
 322  *      unregister_pernet_subsys - unregister a network namespace subsystem
 323  *      @ops: pernet operations structure to manipulate
 324  *
 325  *      Remove the pernet operations structure from the list to be
 326  *      used when network namespaces are created or destroyed.  In
 327  *      addition run the exit method for all existing network
 328  *      namespaces.
 329  */
 330 void unregister_pernet_subsys(struct pernet_operations *module)
 331 {
 332         mutex_lock(&net_mutex);
 333         unregister_pernet_operations(module);
 334         mutex_unlock(&net_mutex);
 335 }
 336 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 337
 338 int register_pernet_gen_subsys(int *id, struct pernet_operations *ops)
 339 {
 340         int rv;
 341
 342         mutex_lock(&net_mutex);
 343 again:
 344         rv = ida_get_new_above(&net_generic_ids, 1, id);
 345         if (rv < 0) {
 346                 if (rv == -EAGAIN) {
 347                         ida_pre_get(&net_generic_ids, GFP_KERNEL);
 348                         goto again;
 349                 }
 350                 goto out;
 351         }
 352         rv = register_pernet_operations(first_device, ops);
 353         if (rv < 0)
 354                 ida_remove(&net_generic_ids, *id);
 355 out:
 356         mutex_unlock(&net_mutex);
 357         return rv;
 358 }
 359 EXPORT_SYMBOL_GPL(register_pernet_gen_subsys);
 360
 361 void unregister_pernet_gen_subsys(int id, struct pernet_operations *ops)
 362 {
 363         mutex_lock(&net_mutex);
 364         unregister_pernet_operations(ops);
 365         ida_remove(&net_generic_ids, id);
 366         mutex_unlock(&net_mutex);
 367 }
 368 EXPORT_SYMBOL_GPL(unregister_pernet_gen_subsys);
 369
 370 /**
 371  *      register_pernet_device - register a network namespace device
 372  *      @ops:  pernet operations structure for the subsystem
 373  *
 374  *      Register a device which has init and exit functions
 375  *      that are called when network namespaces are created and
 376  *      destroyed respectively.
 377  *
 378  *      When registered all network namespace init functions are
 379  *      called for every existing network namespace.  Allowing kernel
 380  *      modules to have a race free view of the set of network namespaces.
 381  *
 382  *      When a new network namespace is created all of the init
 383  *      methods are called in the order in which they were registered.
 384  *
 385  *      When a network namespace is destroyed all of the exit methods
 386  *      are called in the reverse of the order with which they were
 387  *      registered.
 388  */
 389 int register_pernet_device(struct pernet_operations *ops)
 390 {
 391         int error;
 392         mutex_lock(&net_mutex);
 393         error = register_pernet_operations(&pernet_list, ops);
 394         if (!error && (first_device == &pernet_list))
 395                 first_device = &ops->list;
 396         mutex_unlock(&net_mutex);
 397         return error;
 398 }
 399 EXPORT_SYMBOL_GPL(register_pernet_device);
 400
 401 int register_pernet_gen_device(int *id, struct pernet_operations *ops)
 402 {
 403         int error;
 404         mutex_lock(&net_mutex);
 405 again:
 406         error = ida_get_new_above(&net_generic_ids, 1, id);
 407         if (error) {
 408                 if (error == -EAGAIN) {
 409                         ida_pre_get(&net_generic_ids, GFP_KERNEL);
 410                         goto again;
 411                 }
 412                 goto out;
 413         }
 414         error = register_pernet_operations(&pernet_list, ops);
 415         if (error)
 416                 ida_remove(&net_generic_ids, *id);
 417         else if (first_device == &pernet_list)
 418                 first_device = &ops->list;
 419 out:
 420         mutex_unlock(&net_mutex);
 421         return error;
 422 }
 423 EXPORT_SYMBOL_GPL(register_pernet_gen_device);
 424
 425 /**
 426  *      unregister_pernet_device - unregister a network namespace netdevice
 427  *      @ops: pernet operations structure to manipulate
 428  *
 429  *      Remove the pernet operations structure from the list to be
 430  *      used when network namespaces are created or destroyed.  In
 431  *      addition run the exit method for all existing network
 432  *      namespaces.
 433  */
 434 void unregister_pernet_device(struct pernet_operations *ops)
 435 {
 436         mutex_lock(&net_mutex);
 437         if (&ops->list == first_device)
 438                 first_device = first_device->next;
 439         unregister_pernet_operations(ops);
 440         mutex_unlock(&net_mutex);
 441 }
 442 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 443
 444 void unregister_pernet_gen_device(int id, struct pernet_operations *ops)
 445 {
 446         mutex_lock(&net_mutex);
 447         if (&ops->list == first_device)
 448                 first_device = first_device->next;
 449         unregister_pernet_operations(ops);
 450         ida_remove(&net_generic_ids, id);
 451         mutex_unlock(&net_mutex);
 452 }
 453 EXPORT_SYMBOL_GPL(unregister_pernet_gen_device);
 454
 455 static void net_generic_release(struct rcu_head *rcu)
 456 {
 457         struct net_generic *ng;
 458
 459         ng = container_of(rcu, struct net_generic, rcu);
 460         kfree(ng);
 461 }
 462
 463 int net_assign_generic(struct net *net, int id, void *data)
 464 {
 465         struct net_generic *ng, *old_ng;
 466
 467         BUG_ON(!mutex_is_locked(&net_mutex));
 468         BUG_ON(id == 0);
 469
 470         ng = old_ng = net->gen;
 471         if (old_ng->len >= id)
 472                 goto assign;
 473
 474         ng = kzalloc(sizeof(struct net_generic) +
 475                         id * sizeof(void *), GFP_KERNEL);
 476         if (ng == NULL)
 477                 return -ENOMEM;
 478
 479         /*
 480          * Some synchronisation notes:
 481          *
 482          * The net_generic explores the net->gen array inside rcu
 483          * read section. Besides once set the net->gen->ptr[x]
 484          * pointer never changes (see rules in netns/generic.h).
 485          *
 486          * That said, we simply duplicate this array and schedule
 487          * the old copy for kfree after a grace period.
 488          */
 489
 490         ng->len = id;
 491         memcpy(&ng->ptr, &old_ng->ptr, old_ng->len);
 492
 493         rcu_assign_pointer(net->gen, ng);
 494         call_rcu(&old_ng->rcu, net_generic_release);
 495 assign:
 496         ng->ptr[id - 1] = data;
 497         return 0;
 498 }
 499 EXPORT_SYMBOL_GPL(net_assign_generic);