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