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         old_ng = rcu_dereference_protected(net->gen,
  46                                            lockdep_is_held(&net_mutex));
  47         ng = old_ng;
  48         if (old_ng->len >= id)
  49                 goto assign;
  50
  51         ng = kzalloc(sizeof(struct net_generic) +
  52                         id * sizeof(void *), GFP_KERNEL);
  53         if (ng == NULL)
  54                 return -ENOMEM;
  55
  56         /*
  57          * Some synchronisation notes:
  58          *
  59          * The net_generic explores the net->gen array inside rcu
  60          * read section. Besides once set the net->gen->ptr[x]
  61          * pointer never changes (see rules in netns/generic.h).
  62          *
  63          * That said, we simply duplicate this array and schedule
  64          * the old copy for kfree after a grace period.
  65          */
  66
  67         ng->len = id;
  68         memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
  69
  70         rcu_assign_pointer(net->gen, ng);
  71         call_rcu(&old_ng->rcu, net_generic_release);
  72 assign:
  73         ng->ptr[id - 1] = data;
  74         return 0;
  75 }
  76
  77 static int ops_init(const struct pernet_operations *ops, struct net *net)
  78 {
  79         int err;
  80         if (ops->id && ops->size) {
  81                 void *data = kzalloc(ops->size, GFP_KERNEL);
  82                 if (!data)
  83                         return -ENOMEM;
  84
  85                 err = net_assign_generic(net, *ops->id, data);
  86                 if (err) {
  87                         kfree(data);
  88                         return err;
  89                 }
  90         }
  91         if (ops->init)
  92                 return ops->init(net);
  93         return 0;
  94 }
  95
  96 static void ops_free(const struct pernet_operations *ops, struct net *net)
  97 {
  98         if (ops->id && ops->size) {
  99                 int id = *ops->id;
 100                 kfree(net_generic(net, id));
 101         }
 102 }
 103
 104 static void ops_exit_list(const struct pernet_operations *ops,
 105                           struct list_head *net_exit_list)
 106 {
 107         struct net *net;
 108         if (ops->exit) {
 109                 list_for_each_entry(net, net_exit_list, exit_list)
 110                         ops->exit(net);
 111         }
 112         if (ops->exit_batch)
 113                 ops->exit_batch(net_exit_list);
 114 }
 115
 116 static void ops_free_list(const struct pernet_operations *ops,
 117                           struct list_head *net_exit_list)
 118 {
 119         struct net *net;
 120         if (ops->size && ops->id) {
 121                 list_for_each_entry(net, net_exit_list, exit_list)
 122                         ops_free(ops, net);
 123         }
 124 }
 125
 126 /*
 127  * setup_net runs the initializers for the network namespace object.
 128  */
 129 static __net_init int setup_net(struct net *net)
 130 {
 131         /* Must be called with net_mutex held */
 132         const struct pernet_operations *ops, *saved_ops;
 133         int error = 0;
 134         LIST_HEAD(net_exit_list);
 135
 136         atomic_set(&net->count, 1);
 137
 138 #ifdef NETNS_REFCNT_DEBUG
 139         atomic_set(&net->use_count, 0);
 140 #endif
 141
 142         list_for_each_entry(ops, &pernet_list, list) {
 143                 error = ops_init(ops, net);
 144                 if (error < 0)
 145                         goto out_undo;
 146         }
 147 out:
 148         return error;
 149
 150 out_undo:
 151         /* Walk through the list backwards calling the exit functions
 152          * for the pernet modules whose init functions did not fail.
 153          */
 154         list_add(&net->exit_list, &net_exit_list);
 155         saved_ops = ops;
 156         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 157                 ops_exit_list(ops, &net_exit_list);
 158
 159         ops = saved_ops;
 160         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 161                 ops_free_list(ops, &net_exit_list);
 162
 163         rcu_barrier();
 164         goto out;
 165 }
 166
 167 static struct net_generic *net_alloc_generic(void)
 168 {
 169         struct net_generic *ng;
 170         size_t generic_size = sizeof(struct net_generic) +
 171                 INITIAL_NET_GEN_PTRS * sizeof(void *);
 172
 173         ng = kzalloc(generic_size, GFP_KERNEL);
 174         if (ng)
 175                 ng->len = INITIAL_NET_GEN_PTRS;
 176
 177         return ng;
 178 }
 179
 180 #ifdef CONFIG_NET_NS
 181 static struct kmem_cache *net_cachep;
 182 static struct workqueue_struct *netns_wq;
 183
 184 static struct net *net_alloc(void)
 185 {
 186         struct net *net = NULL;
 187         struct net_generic *ng;
 188
 189         ng = net_alloc_generic();
 190         if (!ng)
 191                 goto out;
 192
 193         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 194         if (!net)
 195                 goto out_free;
 196
 197         rcu_assign_pointer(net->gen, ng);
 198 out:
 199         return net;
 200
 201 out_free:
 202         kfree(ng);
 203         goto out;
 204 }
 205
 206 static void net_free(struct net *net)
 207 {
 208 #ifdef NETNS_REFCNT_DEBUG
 209         if (unlikely(atomic_read(&net->use_count) != 0)) {
 210                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
 211                         atomic_read(&net->use_count));
 212                 return;
 213         }
 214 #endif
 215         kfree(net->gen);
 216         kmem_cache_free(net_cachep, net);
 217 }
 218
 219 static struct net *net_create(void)
 220 {
 221         struct net *net;
 222         int rv;
 223
 224         net = net_alloc();
 225         if (!net)
 226                 return ERR_PTR(-ENOMEM);
 227         mutex_lock(&net_mutex);
 228         rv = setup_net(net);
 229         if (rv == 0) {
 230                 rtnl_lock();
 231                 list_add_tail_rcu(&net->list, &net_namespace_list);
 232                 rtnl_unlock();
 233         }
 234         mutex_unlock(&net_mutex);
 235         if (rv < 0) {
 236                 net_free(net);
 237                 return ERR_PTR(rv);
 238         }
 239         return net;
 240 }
 241
 242 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 243 {
 244         if (!(flags & CLONE_NEWNET))
 245                 return get_net(old_net);
 246         return net_create();
 247 }
 248
 249 static DEFINE_SPINLOCK(cleanup_list_lock);
 250 static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 251
 252 static void cleanup_net(struct work_struct *work)
 253 {
 254         const struct pernet_operations *ops;
 255         struct net *net, *tmp;
 256         LIST_HEAD(net_kill_list);
 257         LIST_HEAD(net_exit_list);
 258
 259         /* Atomically snapshot the list of namespaces to cleanup */
 260         spin_lock_irq(&cleanup_list_lock);
 261         list_replace_init(&cleanup_list, &net_kill_list);
 262         spin_unlock_irq(&cleanup_list_lock);
 263
 264         mutex_lock(&net_mutex);
 265
 266         /* Don't let anyone else find us. */
 267         rtnl_lock();
 268         list_for_each_entry(net, &net_kill_list, cleanup_list) {
 269                 list_del_rcu(&net->list);
 270                 list_add_tail(&net->exit_list, &net_exit_list);
 271         }
 272         rtnl_unlock();
 273
 274         /*
 275          * Another CPU might be rcu-iterating the list, wait for it.
 276          * This needs to be before calling the exit() notifiers, so
 277          * the rcu_barrier() below isn't sufficient alone.
 278          */
 279         synchronize_rcu();
 280
 281         /* Run all of the network namespace exit methods */
 282         list_for_each_entry_reverse(ops, &pernet_list, list)
 283                 ops_exit_list(ops, &net_exit_list);
 284
 285         /* Free the net generic variables */
 286         list_for_each_entry_reverse(ops, &pernet_list, list)
 287                 ops_free_list(ops, &net_exit_list);
 288
 289         mutex_unlock(&net_mutex);
 290
 291         /* Ensure there are no outstanding rcu callbacks using this
 292          * network namespace.
 293          */
 294         rcu_barrier();
 295
 296         /* Finally it is safe to free my network namespace structure */
 297         list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 298                 list_del_init(&net->exit_list);
 299                 net_free(net);
 300         }
 301 }
 302 static DECLARE_WORK(net_cleanup_work, cleanup_net);
 303
 304 void __put_net(struct net *net)
 305 {
 306         /* Cleanup the network namespace in process context */
 307         unsigned long flags;
 308
 309         spin_lock_irqsave(&cleanup_list_lock, flags);
 310         list_add(&net->cleanup_list, &cleanup_list);
 311         spin_unlock_irqrestore(&cleanup_list_lock, flags);
 312
 313         queue_work(netns_wq, &net_cleanup_work);
 314 }
 315 EXPORT_SYMBOL_GPL(__put_net);
 316
 317 #else
 318 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 319 {
 320         if (flags & CLONE_NEWNET)
 321                 return ERR_PTR(-EINVAL);
 322         return old_net;
 323 }
 324 #endif
 325
 326 struct net *get_net_ns_by_pid(pid_t pid)
 327 {
 328         struct task_struct *tsk;
 329         struct net *net;
 330
 331         /* Lookup the network namespace */
 332         net = ERR_PTR(-ESRCH);
 333         rcu_read_lock();
 334         tsk = find_task_by_vpid(pid);
 335         if (tsk) {
 336                 struct nsproxy *nsproxy;
 337                 nsproxy = task_nsproxy(tsk);
 338                 if (nsproxy)
 339                         net = get_net(nsproxy->net_ns);
 340         }
 341         rcu_read_unlock();
 342         return net;
 343 }
 344 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 345
 346 static int __init net_ns_init(void)
 347 {
 348         struct net_generic *ng;
 349
 350 #ifdef CONFIG_NET_NS
 351         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 352                                         SMP_CACHE_BYTES,
 353                                         SLAB_PANIC, NULL);
 354
 355         /* Create workqueue for cleanup */
 356         netns_wq = create_singlethread_workqueue("netns");
 357         if (!netns_wq)
 358                 panic("Could not create netns workq");
 359 #endif
 360
 361         ng = net_alloc_generic();
 362         if (!ng)
 363                 panic("Could not allocate generic netns");
 364
 365         rcu_assign_pointer(init_net.gen, ng);
 366
 367         mutex_lock(&net_mutex);
 368         if (setup_net(&init_net))
 369                 panic("Could not setup the initial network namespace");
 370
 371         rtnl_lock();
 372         list_add_tail_rcu(&init_net.list, &net_namespace_list);
 373         rtnl_unlock();
 374
 375         mutex_unlock(&net_mutex);
 376
 377         return 0;
 378 }
 379
 380 pure_initcall(net_ns_init);
 381
 382 #ifdef CONFIG_NET_NS
 383 static int __register_pernet_operations(struct list_head *list,
 384                                         struct pernet_operations *ops)
 385 {
 386         struct net *net;
 387         int error;
 388         LIST_HEAD(net_exit_list);
 389
 390         list_add_tail(&ops->list, list);
 391         if (ops->init || (ops->id && ops->size)) {
 392                 for_each_net(net) {
 393                         error = ops_init(ops, net);
 394                         if (error)
 395                                 goto out_undo;
 396                         list_add_tail(&net->exit_list, &net_exit_list);
 397                 }
 398         }
 399         return 0;
 400
 401 out_undo:
 402         /* If I have an error cleanup all namespaces I initialized */
 403         list_del(&ops->list);
 404         ops_exit_list(ops, &net_exit_list);
 405         ops_free_list(ops, &net_exit_list);
 406         return error;
 407 }
 408
 409 static void __unregister_pernet_operations(struct pernet_operations *ops)
 410 {
 411         struct net *net;
 412         LIST_HEAD(net_exit_list);
 413
 414         list_del(&ops->list);
 415         for_each_net(net)
 416                 list_add_tail(&net->exit_list, &net_exit_list);
 417         ops_exit_list(ops, &net_exit_list);
 418         ops_free_list(ops, &net_exit_list);
 419 }
 420
 421 #else
 422
 423 static int __register_pernet_operations(struct list_head *list,
 424                                         struct pernet_operations *ops)
 425 {
 426         int err = 0;
 427         err = ops_init(ops, &init_net);
 428         if (err)
 429                 ops_free(ops, &init_net);
 430         return err;
 431
 432 }
 433
 434 static void __unregister_pernet_operations(struct pernet_operations *ops)
 435 {
 436         LIST_HEAD(net_exit_list);
 437         list_add(&init_net.exit_list, &net_exit_list);
 438         ops_exit_list(ops, &net_exit_list);
 439         ops_free_list(ops, &net_exit_list);
 440 }
 441
 442 #endif /* CONFIG_NET_NS */
 443
 444 static DEFINE_IDA(net_generic_ids);
 445
 446 static int register_pernet_operations(struct list_head *list,
 447                                       struct pernet_operations *ops)
 448 {
 449         int error;
 450
 451         if (ops->id) {
 452 again:
 453                 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 454                 if (error < 0) {
 455                         if (error == -EAGAIN) {
 456                                 ida_pre_get(&net_generic_ids, GFP_KERNEL);
 457                                 goto again;
 458                         }
 459                         return error;
 460                 }
 461         }
 462         error = __register_pernet_operations(list, ops);
 463         if (error) {
 464                 rcu_barrier();
 465                 if (ops->id)
 466                         ida_remove(&net_generic_ids, *ops->id);
 467         }
 468
 469         return error;
 470 }
 471
 472 static void unregister_pernet_operations(struct pernet_operations *ops)
 473 {
 474
 475         __unregister_pernet_operations(ops);
 476         rcu_barrier();
 477         if (ops->id)
 478                 ida_remove(&net_generic_ids, *ops->id);
 479 }
 480
 481 /**
 482  *      register_pernet_subsys - register a network namespace subsystem
 483  *      @ops:  pernet operations structure for the subsystem
 484  *
 485  *      Register a subsystem which has init and exit functions
 486  *      that are called when network namespaces are created and
 487  *      destroyed respectively.
 488  *
 489  *      When registered all network namespace init functions are
 490  *      called for every existing network namespace.  Allowing kernel
 491  *      modules to have a race free view of the set of network namespaces.
 492  *
 493  *      When a new network namespace is created all of the init
 494  *      methods are called in the order in which they were registered.
 495  *
 496  *      When a network namespace is destroyed all of the exit methods
 497  *      are called in the reverse of the order with which they were
 498  *      registered.
 499  */
 500 int register_pernet_subsys(struct pernet_operations *ops)
 501 {
 502         int error;
 503         mutex_lock(&net_mutex);
 504         error =  register_pernet_operations(first_device, ops);
 505         mutex_unlock(&net_mutex);
 506         return error;
 507 }
 508 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 509
 510 /**
 511  *      unregister_pernet_subsys - unregister a network namespace subsystem
 512  *      @ops: pernet operations structure to manipulate
 513  *
 514  *      Remove the pernet operations structure from the list to be
 515  *      used when network namespaces are created or destroyed.  In
 516  *      addition run the exit method for all existing network
 517  *      namespaces.
 518  */
 519 void unregister_pernet_subsys(struct pernet_operations *ops)
 520 {
 521         mutex_lock(&net_mutex);
 522         unregister_pernet_operations(ops);
 523         mutex_unlock(&net_mutex);
 524 }
 525 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 526
 527 /**
 528  *      register_pernet_device - register a network namespace device
 529  *      @ops:  pernet operations structure for the subsystem
 530  *
 531  *      Register a device which has init and exit functions
 532  *      that are called when network namespaces are created and
 533  *      destroyed respectively.
 534  *
 535  *      When registered all network namespace init functions are
 536  *      called for every existing network namespace.  Allowing kernel
 537  *      modules to have a race free view of the set of network namespaces.
 538  *
 539  *      When a new network namespace is created all of the init
 540  *      methods are called in the order in which they were registered.
 541  *
 542  *      When a network namespace is destroyed all of the exit methods
 543  *      are called in the reverse of the order with which they were
 544  *      registered.
 545  */
 546 int register_pernet_device(struct pernet_operations *ops)
 547 {
 548         int error;
 549         mutex_lock(&net_mutex);
 550         error = register_pernet_operations(&pernet_list, ops);
 551         if (!error && (first_device == &pernet_list))
 552                 first_device = &ops->list;
 553         mutex_unlock(&net_mutex);
 554         return error;
 555 }
 556 EXPORT_SYMBOL_GPL(register_pernet_device);
 557
 558 /**
 559  *      unregister_pernet_device - unregister a network namespace netdevice
 560  *      @ops: pernet operations structure to manipulate
 561  *
 562  *      Remove the pernet operations structure from the list to be
 563  *      used when network namespaces are created or destroyed.  In
 564  *      addition run the exit method for all existing network
 565  *      namespaces.
 566  */
 567 void unregister_pernet_device(struct pernet_operations *ops)
 568 {
 569         mutex_lock(&net_mutex);
 570         if (&ops->list == first_device)
 571                 first_device = first_device->next;
 572         unregister_pernet_operations(ops);
 573         mutex_unlock(&net_mutex);
 574 }
 575 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 576
 577 #ifdef CONFIG_NET_NS
 578 static void *netns_get(struct task_struct *task)
 579 {
 580         struct net *net;
 581         rcu_read_lock();
 582         net = get_net(task->nsproxy->net_ns);
 583         rcu_read_unlock();
 584         return net;
 585 }
 586
 587 static void netns_put(void *ns)
 588 {
 589         put_net(ns);
 590 }
 591
 592 static int netns_install(struct nsproxy *nsproxy, void *ns)
 593 {
 594         put_net(nsproxy->net_ns);
 595         nsproxy->net_ns = get_net(ns);
 596         return 0;
 597 }
 598
 599 const struct proc_ns_operations netns_operations = {
 600         .name           = "net",
 601         .type           = CLONE_NEWNET,
 602         .get            = netns_get,
 603         .put            = netns_put,
 604         .install        = netns_install,
 605 };
 606 #endif