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 <linux/rculist.h>
  10 #include <linux/nsproxy.h>
  11 #include <net/net_namespace.h>
  12 #include <net/netns/generic.h>
  13
  14 /*
  15  *      Our network namespace constructor/destructor lists
  16  */
  17
  18 static LIST_HEAD(pernet_list);
  19 static struct list_head *first_device = &pernet_list;
  20 static DEFINE_MUTEX(net_mutex);
  21
  22 LIST_HEAD(net_namespace_list);
  23 EXPORT_SYMBOL_GPL(net_namespace_list);
  24
  25 struct net init_net;
  26 EXPORT_SYMBOL(init_net);
  27
  28 #define INITIAL_NET_GEN_PTRS    13 /* +1 for len +2 for rcu_head */
  29
  30 static void net_generic_release(struct rcu_head *rcu)
  31 {
  32         struct net_generic *ng;
  33
  34         ng = container_of(rcu, struct net_generic, rcu);
  35         kfree(ng);
  36 }
  37
  38 static int net_assign_generic(struct net *net, int id, void *data)
  39 {
  40         struct net_generic *ng, *old_ng;
  41
  42         BUG_ON(!mutex_is_locked(&net_mutex));
  43         BUG_ON(id == 0);
  44
  45         ng = old_ng = net->gen;
  46         if (old_ng->len >= id)
  47                 goto assign;
  48
  49         ng = kzalloc(sizeof(struct net_generic) +
  50                         id * sizeof(void *), GFP_KERNEL);
  51         if (ng == NULL)
  52                 return -ENOMEM;
  53
  54         /*
  55          * Some synchronisation notes:
  56          *
  57          * The net_generic explores the net->gen array inside rcu
  58          * read section. Besides once set the net->gen->ptr[x]
  59          * pointer never changes (see rules in netns/generic.h).
  60          *
  61          * That said, we simply duplicate this array and schedule
  62          * the old copy for kfree after a grace period.
  63          */
  64
  65         ng->len = id;
  66         memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
  67
  68         rcu_assign_pointer(net->gen, ng);
  69         call_rcu(&old_ng->rcu, net_generic_release);
  70 assign:
  71         ng->ptr[id - 1] = data;
  72         return 0;
  73 }
  74
  75 static int ops_init(const struct pernet_operations *ops, struct net *net)
  76 {
  77         int err;
  78         if (ops->id && ops->size) {
  79                 void *data = kzalloc(ops->size, GFP_KERNEL);
  80                 if (!data)
  81                         return -ENOMEM;
  82
  83                 err = net_assign_generic(net, *ops->id, data);
  84                 if (err) {
  85                         kfree(data);
  86                         return err;
  87                 }
  88         }
  89         if (ops->init)
  90                 return ops->init(net);
  91         return 0;
  92 }
  93
  94 static void ops_free(const struct pernet_operations *ops, struct net *net)
  95 {
  96         if (ops->id && ops->size) {
  97                 int id = *ops->id;
  98                 kfree(net_generic(net, id));
  99         }
 100 }
 101
 102 static void ops_exit_list(const struct pernet_operations *ops,
 103                           struct list_head *net_exit_list)
 104 {
 105         struct net *net;
 106         if (ops->exit) {
 107                 list_for_each_entry(net, net_exit_list, exit_list)
 108                         ops->exit(net);
 109         }
 110         if (ops->exit_batch)
 111                 ops->exit_batch(net_exit_list);
 112 }
 113
 114 static void ops_free_list(const struct pernet_operations *ops,
 115                           struct list_head *net_exit_list)
 116 {
 117         struct net *net;
 118         if (ops->size && ops->id) {
 119                 list_for_each_entry(net, net_exit_list, exit_list)
 120                         ops_free(ops, net);
 121         }
 122 }
 123
 124 /*
 125  * setup_net runs the initializers for the network namespace object.
 126  */
 127 static __net_init int setup_net(struct net *net)
 128 {
 129         /* Must be called with net_mutex held */
 130         const struct pernet_operations *ops, *saved_ops;
 131         int error = 0;
 132         LIST_HEAD(net_exit_list);
 133
 134         atomic_set(&net->count, 1);
 135
 136 #ifdef NETNS_REFCNT_DEBUG
 137         atomic_set(&net->use_count, 0);
 138 #endif
 139
 140         list_for_each_entry(ops, &pernet_list, list) {
 141                 error = ops_init(ops, net);
 142                 if (error < 0)
 143                         goto out_undo;
 144         }
 145 out:
 146         return error;
 147
 148 out_undo:
 149         /* Walk through the list backwards calling the exit functions
 150          * for the pernet modules whose init functions did not fail.
 151          */
 152         list_add(&net->exit_list, &net_exit_list);
 153         saved_ops = ops;
 154         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 155                 ops_exit_list(ops, &net_exit_list);
 156
 157         ops = saved_ops;
 158         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 159                 ops_free_list(ops, &net_exit_list);
 160
 161         rcu_barrier();
 162         goto out;
 163 }
 164
 165 static struct net_generic *net_alloc_generic(void)
 166 {
 167         struct net_generic *ng;
 168         size_t generic_size = sizeof(struct net_generic) +
 169                 INITIAL_NET_GEN_PTRS * sizeof(void *);
 170
 171         ng = kzalloc(generic_size, GFP_KERNEL);
 172         if (ng)
 173                 ng->len = INITIAL_NET_GEN_PTRS;
 174
 175         return ng;
 176 }
 177
 178 #ifdef CONFIG_NET_NS
 179 static struct kmem_cache *net_cachep;
 180 static struct workqueue_struct *netns_wq;
 181
 182 static struct net *net_alloc(void)
 183 {
 184         struct net *net = NULL;
 185         struct net_generic *ng;
 186
 187         ng = net_alloc_generic();
 188         if (!ng)
 189                 goto out;
 190
 191         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 192         if (!net)
 193                 goto out_free;
 194
 195         rcu_assign_pointer(net->gen, ng);
 196 out:
 197         return net;
 198
 199 out_free:
 200         kfree(ng);
 201         goto out;
 202 }
 203
 204 static void net_free(struct net *net)
 205 {
 206 #ifdef NETNS_REFCNT_DEBUG
 207         if (unlikely(atomic_read(&net->use_count) != 0)) {
 208                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
 209                         atomic_read(&net->use_count));
 210                 return;
 211         }
 212 #endif
 213         kfree(net->gen);
 214         kmem_cache_free(net_cachep, net);
 215 }
 216
 217 static struct net *net_create(void)
 218 {
 219         struct net *net;
 220         int rv;
 221
 222         net = net_alloc();
 223         if (!net)
 224                 return ERR_PTR(-ENOMEM);
 225         mutex_lock(&net_mutex);
 226         rv = setup_net(net);
 227         if (rv == 0) {
 228                 rtnl_lock();
 229                 list_add_tail_rcu(&net->list, &net_namespace_list);
 230                 rtnl_unlock();
 231         }
 232         mutex_unlock(&net_mutex);
 233         if (rv < 0) {
 234                 net_free(net);
 235                 return ERR_PTR(rv);
 236         }
 237         return net;
 238 }
 239
 240 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 241 {
 242         if (!(flags & CLONE_NEWNET))
 243                 return get_net(old_net);
 244         return net_create();
 245 }
 246
 247 static DEFINE_SPINLOCK(cleanup_list_lock);
 248 static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 249
 250 static void cleanup_net(struct work_struct *work)
 251 {
 252         const struct pernet_operations *ops;
 253         struct net *net, *tmp;
 254         LIST_HEAD(net_kill_list);
 255         LIST_HEAD(net_exit_list);
 256
 257         /* Atomically snapshot the list of namespaces to cleanup */
 258         spin_lock_irq(&cleanup_list_lock);
 259         list_replace_init(&cleanup_list, &net_kill_list);
 260         spin_unlock_irq(&cleanup_list_lock);
 261
 262         mutex_lock(&net_mutex);
 263
 264         /* Don't let anyone else find us. */
 265         rtnl_lock();
 266         list_for_each_entry(net, &net_kill_list, cleanup_list) {
 267                 list_del_rcu(&net->list);
 268                 list_add_tail(&net->exit_list, &net_exit_list);
 269         }
 270         rtnl_unlock();
 271
 272         /*
 273          * Another CPU might be rcu-iterating the list, wait for it.
 274          * This needs to be before calling the exit() notifiers, so
 275          * the rcu_barrier() below isn't sufficient alone.
 276          */
 277         synchronize_rcu();
 278
 279         /* Run all of the network namespace exit methods */
 280         list_for_each_entry_reverse(ops, &pernet_list, list)
 281                 ops_exit_list(ops, &net_exit_list);
 282
 283         /* Free the net generic variables */
 284         list_for_each_entry_reverse(ops, &pernet_list, list)
 285                 ops_free_list(ops, &net_exit_list);
 286
 287         mutex_unlock(&net_mutex);
 288
 289         /* Ensure there are no outstanding rcu callbacks using this
 290          * network namespace.
 291          */
 292         rcu_barrier();
 293
 294         /* Finally it is safe to free my network namespace structure */
 295         list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 296                 list_del_init(&net->exit_list);
 297                 net_free(net);
 298         }
 299 }
 300 static DECLARE_WORK(net_cleanup_work, cleanup_net);
 301
 302 void __put_net(struct net *net)
 303 {
 304         /* Cleanup the network namespace in process context */
 305         unsigned long flags;
 306
 307         spin_lock_irqsave(&cleanup_list_lock, flags);
 308         list_add(&net->cleanup_list, &cleanup_list);
 309         spin_unlock_irqrestore(&cleanup_list_lock, flags);
 310
 311         queue_work(netns_wq, &net_cleanup_work);
 312 }
 313 EXPORT_SYMBOL_GPL(__put_net);
 314
 315 #else
 316 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 317 {
 318         if (flags & CLONE_NEWNET)
 319                 return ERR_PTR(-EINVAL);
 320         return old_net;
 321 }
 322 #endif
 323
 324 struct net *get_net_ns_by_pid(pid_t pid)
 325 {
 326         struct task_struct *tsk;
 327         struct net *net;
 328
 329         /* Lookup the network namespace */
 330         net = ERR_PTR(-ESRCH);
 331         rcu_read_lock();
 332         tsk = find_task_by_vpid(pid);
 333         if (tsk) {
 334                 struct nsproxy *nsproxy;
 335                 nsproxy = task_nsproxy(tsk);
 336                 if (nsproxy)
 337                         net = get_net(nsproxy->net_ns);
 338         }
 339         rcu_read_unlock();
 340         return net;
 341 }
 342 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 343
 344 static int __init net_ns_init(void)
 345 {
 346         struct net_generic *ng;
 347
 348 #ifdef CONFIG_NET_NS
 349         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 350                                         SMP_CACHE_BYTES,
 351                                         SLAB_PANIC, NULL);
 352
 353         /* Create workqueue for cleanup */
 354         netns_wq = create_singlethread_workqueue("netns");
 355         if (!netns_wq)
 356                 panic("Could not create netns workq");
 357 #endif
 358
 359         ng = net_alloc_generic();
 360         if (!ng)
 361                 panic("Could not allocate generic netns");
 362
 363         rcu_assign_pointer(init_net.gen, ng);
 364
 365         mutex_lock(&net_mutex);
 366         if (setup_net(&init_net))
 367                 panic("Could not setup the initial network namespace");
 368
 369         rtnl_lock();
 370         list_add_tail_rcu(&init_net.list, &net_namespace_list);
 371         rtnl_unlock();
 372
 373         mutex_unlock(&net_mutex);
 374
 375         return 0;
 376 }
 377
 378 pure_initcall(net_ns_init);
 379
 380 #ifdef CONFIG_NET_NS
 381 static int __register_pernet_operations(struct list_head *list,
 382                                         struct pernet_operations *ops)
 383 {
 384         struct net *net;
 385         int error;
 386         LIST_HEAD(net_exit_list);
 387
 388         list_add_tail(&ops->list, list);
 389         if (ops->init || (ops->id && ops->size)) {
 390                 for_each_net(net) {
 391                         error = ops_init(ops, net);
 392                         if (error)
 393                                 goto out_undo;
 394                         list_add_tail(&net->exit_list, &net_exit_list);
 395                 }
 396         }
 397         return 0;
 398
 399 out_undo:
 400         /* If I have an error cleanup all namespaces I initialized */
 401         list_del(&ops->list);
 402         ops_exit_list(ops, &net_exit_list);
 403         ops_free_list(ops, &net_exit_list);
 404         return error;
 405 }
 406
 407 static void __unregister_pernet_operations(struct pernet_operations *ops)
 408 {
 409         struct net *net;
 410         LIST_HEAD(net_exit_list);
 411
 412         list_del(&ops->list);
 413         for_each_net(net)
 414                 list_add_tail(&net->exit_list, &net_exit_list);
 415         ops_exit_list(ops, &net_exit_list);
 416         ops_free_list(ops, &net_exit_list);
 417 }
 418
 419 #else
 420
 421 static int __register_pernet_operations(struct list_head *list,
 422                                         struct pernet_operations *ops)
 423 {
 424         int err = 0;
 425         err = ops_init(ops, &init_net);
 426         if (err)
 427                 ops_free(ops, &init_net);
 428         return err;
 429
 430 }
 431
 432 static void __unregister_pernet_operations(struct pernet_operations *ops)
 433 {
 434         LIST_HEAD(net_exit_list);
 435         list_add(&init_net.exit_list, &net_exit_list);
 436         ops_exit_list(ops, &net_exit_list);
 437         ops_free_list(ops, &net_exit_list);
 438 }
 439
 440 #endif /* CONFIG_NET_NS */
 441
 442 static DEFINE_IDA(net_generic_ids);
 443
 444 static int register_pernet_operations(struct list_head *list,
 445                                       struct pernet_operations *ops)
 446 {
 447         int error;
 448
 449         if (ops->id) {
 450 again:
 451                 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 452                 if (error < 0) {
 453                         if (error == -EAGAIN) {
 454                                 ida_pre_get(&net_generic_ids, GFP_KERNEL);
 455                                 goto again;
 456                         }
 457                         return error;
 458                 }
 459         }
 460         error = __register_pernet_operations(list, ops);
 461         if (error) {
 462                 rcu_barrier();
 463                 if (ops->id)
 464                         ida_remove(&net_generic_ids, *ops->id);
 465         }
 466
 467         return error;
 468 }
 469
 470 static void unregister_pernet_operations(struct pernet_operations *ops)
 471 {
 472
 473         __unregister_pernet_operations(ops);
 474         rcu_barrier();
 475         if (ops->id)
 476                 ida_remove(&net_generic_ids, *ops->id);
 477 }
 478
 479 /**
 480  *      register_pernet_subsys - register a network namespace subsystem
 481  *      @ops:  pernet operations structure for the subsystem
 482  *
 483  *      Register a subsystem which has init and exit functions
 484  *      that are called when network namespaces are created and
 485  *      destroyed respectively.
 486  *
 487  *      When registered all network namespace init functions are
 488  *      called for every existing network namespace.  Allowing kernel
 489  *      modules to have a race free view of the set of network namespaces.
 490  *
 491  *      When a new network namespace is created all of the init
 492  *      methods are called in the order in which they were registered.
 493  *
 494  *      When a network namespace is destroyed all of the exit methods
 495  *      are called in the reverse of the order with which they were
 496  *      registered.
 497  */
 498 int register_pernet_subsys(struct pernet_operations *ops)
 499 {
 500         int error;
 501         mutex_lock(&net_mutex);
 502         error =  register_pernet_operations(first_device, ops);
 503         mutex_unlock(&net_mutex);
 504         return error;
 505 }
 506 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 507
 508 /**
 509  *      unregister_pernet_subsys - unregister a network namespace subsystem
 510  *      @ops: pernet operations structure to manipulate
 511  *
 512  *      Remove the pernet operations structure from the list to be
 513  *      used when network namespaces are created or destroyed.  In
 514  *      addition run the exit method for all existing network
 515  *      namespaces.
 516  */
 517 void unregister_pernet_subsys(struct pernet_operations *ops)
 518 {
 519         mutex_lock(&net_mutex);
 520         unregister_pernet_operations(ops);
 521         mutex_unlock(&net_mutex);
 522 }
 523 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 524
 525 /**
 526  *      register_pernet_device - register a network namespace device
 527  *      @ops:  pernet operations structure for the subsystem
 528  *
 529  *      Register a device which has init and exit functions
 530  *      that are called when network namespaces are created and
 531  *      destroyed respectively.
 532  *
 533  *      When registered all network namespace init functions are
 534  *      called for every existing network namespace.  Allowing kernel
 535  *      modules to have a race free view of the set of network namespaces.
 536  *
 537  *      When a new network namespace is created all of the init
 538  *      methods are called in the order in which they were registered.
 539  *
 540  *      When a network namespace is destroyed all of the exit methods
 541  *      are called in the reverse of the order with which they were
 542  *      registered.
 543  */
 544 int register_pernet_device(struct pernet_operations *ops)
 545 {
 546         int error;
 547         mutex_lock(&net_mutex);
 548         error = register_pernet_operations(&pernet_list, ops);
 549         if (!error && (first_device == &pernet_list))
 550                 first_device = &ops->list;
 551         mutex_unlock(&net_mutex);
 552         return error;
 553 }
 554 EXPORT_SYMBOL_GPL(register_pernet_device);
 555
 556 /**
 557  *      unregister_pernet_device - unregister a network namespace netdevice
 558  *      @ops: pernet operations structure to manipulate
 559  *
 560  *      Remove the pernet operations structure from the list to be
 561  *      used when network namespaces are created or destroyed.  In
 562  *      addition run the exit method for all existing network
 563  *      namespaces.
 564  */
 565 void unregister_pernet_device(struct pernet_operations *ops)
 566 {
 567         mutex_lock(&net_mutex);
 568         if (&ops->list == first_device)
 569                 first_device = first_device->next;
 570         unregister_pernet_operations(ops);
 571         mutex_unlock(&net_mutex);
 572 }
 573 EXPORT_SYMBOL_GPL(unregister_pernet_device);