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 <linux/proc_fs.h>
  12 #include <linux/file.h>
  13 #include <linux/export.h>
  14 #include <net/net_namespace.h>
  15 #include <net/netns/generic.h>
  16
  17 /*
  18  *      Our network namespace constructor/destructor lists
  19  */
  20
  21 static LIST_HEAD(pernet_list);
  22 static struct list_head *first_device = &pernet_list;
  23 static DEFINE_MUTEX(net_mutex);
  24
  25 LIST_HEAD(net_namespace_list);
  26 EXPORT_SYMBOL_GPL(net_namespace_list);
  27
  28 struct net init_net;
  29 EXPORT_SYMBOL(init_net);
  30
  31 #define INITIAL_NET_GEN_PTRS    13 /* +1 for len +2 for rcu_head */
  32
  33 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  34
  35 static struct net_generic *net_alloc_generic(void)
  36 {
  37         struct net_generic *ng;
  38         size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  39
  40         ng = kzalloc(generic_size, GFP_KERNEL);
  41         if (ng)
  42                 ng->len = max_gen_ptrs;
  43
  44         return ng;
  45 }
  46
  47 static int net_assign_generic(struct net *net, int id, void *data)
  48 {
  49         struct net_generic *ng, *old_ng;
  50
  51         BUG_ON(!mutex_is_locked(&net_mutex));
  52         BUG_ON(id == 0);
  53
  54         old_ng = rcu_dereference_protected(net->gen,
  55                                            lockdep_is_held(&net_mutex));
  56         ng = old_ng;
  57         if (old_ng->len >= id)
  58                 goto assign;
  59
  60         ng = net_alloc_generic();
  61         if (ng == NULL)
  62                 return -ENOMEM;
  63
  64         /*
  65          * Some synchronisation notes:
  66          *
  67          * The net_generic explores the net->gen array inside rcu
  68          * read section. Besides once set the net->gen->ptr[x]
  69          * pointer never changes (see rules in netns/generic.h).
  70          *
  71          * That said, we simply duplicate this array and schedule
  72          * the old copy for kfree after a grace period.
  73          */
  74
  75         memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
  76
  77         rcu_assign_pointer(net->gen, ng);
  78         kfree_rcu(old_ng, rcu);
  79 assign:
  80         ng->ptr[id - 1] = data;
  81         return 0;
  82 }
  83
  84 static int ops_init(const struct pernet_operations *ops, struct net *net)
  85 {
  86         int err;
  87         if (ops->id && ops->size) {
  88                 void *data = kzalloc(ops->size, GFP_KERNEL);
  89                 if (!data)
  90                         return -ENOMEM;
  91
  92                 err = net_assign_generic(net, *ops->id, data);
  93                 if (err) {
  94                         kfree(data);
  95                         return err;
  96                 }
  97         }
  98         if (ops->init)
  99                 return ops->init(net);
 100         return 0;
 101 }
 102
 103 static void ops_free(const struct pernet_operations *ops, struct net *net)
 104 {
 105         if (ops->id && ops->size) {
 106                 int id = *ops->id;
 107                 kfree(net_generic(net, id));
 108         }
 109 }
 110
 111 static void ops_exit_list(const struct pernet_operations *ops,
 112                           struct list_head *net_exit_list)
 113 {
 114         struct net *net;
 115         if (ops->exit) {
 116                 list_for_each_entry(net, net_exit_list, exit_list)
 117                         ops->exit(net);
 118         }
 119         if (ops->exit_batch)
 120                 ops->exit_batch(net_exit_list);
 121 }
 122
 123 static void ops_free_list(const struct pernet_operations *ops,
 124                           struct list_head *net_exit_list)
 125 {
 126         struct net *net;
 127         if (ops->size && ops->id) {
 128                 list_for_each_entry(net, net_exit_list, exit_list)
 129                         ops_free(ops, net);
 130         }
 131 }
 132
 133 /*
 134  * setup_net runs the initializers for the network namespace object.
 135  */
 136 static __net_init int setup_net(struct net *net)
 137 {
 138         /* Must be called with net_mutex held */
 139         const struct pernet_operations *ops, *saved_ops;
 140         int error = 0;
 141         LIST_HEAD(net_exit_list);
 142
 143         atomic_set(&net->count, 1);
 144         atomic_set(&net->passive, 1);
 145         net->dev_base_seq = 1;
 146
 147 #ifdef NETNS_REFCNT_DEBUG
 148         atomic_set(&net->use_count, 0);
 149 #endif
 150
 151         list_for_each_entry(ops, &pernet_list, list) {
 152                 error = ops_init(ops, net);
 153                 if (error < 0)
 154                         goto out_undo;
 155         }
 156 out:
 157         return error;
 158
 159 out_undo:
 160         /* Walk through the list backwards calling the exit functions
 161          * for the pernet modules whose init functions did not fail.
 162          */
 163         list_add(&net->exit_list, &net_exit_list);
 164         saved_ops = ops;
 165         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 166                 ops_exit_list(ops, &net_exit_list);
 167
 168         ops = saved_ops;
 169         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 170                 ops_free_list(ops, &net_exit_list);
 171
 172         rcu_barrier();
 173         goto out;
 174 }
 175
 176
 177 #ifdef CONFIG_NET_NS
 178 static struct kmem_cache *net_cachep;
 179 static struct workqueue_struct *netns_wq;
 180
 181 static struct net *net_alloc(void)
 182 {
 183         struct net *net = NULL;
 184         struct net_generic *ng;
 185
 186         ng = net_alloc_generic();
 187         if (!ng)
 188                 goto out;
 189
 190         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 191         if (!net)
 192                 goto out_free;
 193
 194         rcu_assign_pointer(net->gen, ng);
 195 out:
 196         return net;
 197
 198 out_free:
 199         kfree(ng);
 200         goto out;
 201 }
 202
 203 static void net_free(struct net *net)
 204 {
 205 #ifdef NETNS_REFCNT_DEBUG
 206         if (unlikely(atomic_read(&net->use_count) != 0)) {
 207                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
 208                         atomic_read(&net->use_count));
 209                 return;
 210         }
 211 #endif
 212         kfree(net->gen);
 213         kmem_cache_free(net_cachep, net);
 214 }
 215
 216 void net_drop_ns(void *p)
 217 {
 218         struct net *ns = p;
 219         if (ns && atomic_dec_and_test(&ns->passive))
 220                 net_free(ns);
 221 }
 222
 223 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 224 {
 225         struct net *net;
 226         int rv;
 227
 228         if (!(flags & CLONE_NEWNET))
 229                 return get_net(old_net);
 230
 231         net = net_alloc();
 232         if (!net)
 233                 return ERR_PTR(-ENOMEM);
 234         mutex_lock(&net_mutex);
 235         rv = setup_net(net);
 236         if (rv == 0) {
 237                 rtnl_lock();
 238                 list_add_tail_rcu(&net->list, &net_namespace_list);
 239                 rtnl_unlock();
 240         }
 241         mutex_unlock(&net_mutex);
 242         if (rv < 0) {
 243                 net_drop_ns(net);
 244                 return ERR_PTR(rv);
 245         }
 246         return net;
 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_drop_ns(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 struct net *get_net_ns_by_fd(int fd)
 318 {
 319         struct proc_inode *ei;
 320         struct file *file;
 321         struct net *net;
 322
 323         file = proc_ns_fget(fd);
 324         if (IS_ERR(file))
 325                 return ERR_CAST(file);
 326
 327         ei = PROC_I(file->f_dentry->d_inode);
 328         if (ei->ns_ops == &netns_operations)
 329                 net = get_net(ei->ns);
 330         else
 331                 net = ERR_PTR(-EINVAL);
 332
 333         fput(file);
 334         return net;
 335 }
 336
 337 #else
 338 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 339 {
 340         if (flags & CLONE_NEWNET)
 341                 return ERR_PTR(-EINVAL);
 342         return old_net;
 343 }
 344
 345 struct net *get_net_ns_by_fd(int fd)
 346 {
 347         return ERR_PTR(-EINVAL);
 348 }
 349 #endif
 350
 351 struct net *get_net_ns_by_pid(pid_t pid)
 352 {
 353         struct task_struct *tsk;
 354         struct net *net;
 355
 356         /* Lookup the network namespace */
 357         net = ERR_PTR(-ESRCH);
 358         rcu_read_lock();
 359         tsk = find_task_by_vpid(pid);
 360         if (tsk) {
 361                 struct nsproxy *nsproxy;
 362                 nsproxy = task_nsproxy(tsk);
 363                 if (nsproxy)
 364                         net = get_net(nsproxy->net_ns);
 365         }
 366         rcu_read_unlock();
 367         return net;
 368 }
 369 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 370
 371 static int __init net_ns_init(void)
 372 {
 373         struct net_generic *ng;
 374
 375 #ifdef CONFIG_NET_NS
 376         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 377                                         SMP_CACHE_BYTES,
 378                                         SLAB_PANIC, NULL);
 379
 380         /* Create workqueue for cleanup */
 381         netns_wq = create_singlethread_workqueue("netns");
 382         if (!netns_wq)
 383                 panic("Could not create netns workq");
 384 #endif
 385
 386         ng = net_alloc_generic();
 387         if (!ng)
 388                 panic("Could not allocate generic netns");
 389
 390         rcu_assign_pointer(init_net.gen, ng);
 391
 392         mutex_lock(&net_mutex);
 393         if (setup_net(&init_net))
 394                 panic("Could not setup the initial network namespace");
 395
 396         rtnl_lock();
 397         list_add_tail_rcu(&init_net.list, &net_namespace_list);
 398         rtnl_unlock();
 399
 400         mutex_unlock(&net_mutex);
 401
 402         return 0;
 403 }
 404
 405 pure_initcall(net_ns_init);
 406
 407 #ifdef CONFIG_NET_NS
 408 static int __register_pernet_operations(struct list_head *list,
 409                                         struct pernet_operations *ops)
 410 {
 411         struct net *net;
 412         int error;
 413         LIST_HEAD(net_exit_list);
 414
 415         list_add_tail(&ops->list, list);
 416         if (ops->init || (ops->id && ops->size)) {
 417                 for_each_net(net) {
 418                         error = ops_init(ops, net);
 419                         if (error)
 420                                 goto out_undo;
 421                         list_add_tail(&net->exit_list, &net_exit_list);
 422                 }
 423         }
 424         return 0;
 425
 426 out_undo:
 427         /* If I have an error cleanup all namespaces I initialized */
 428         list_del(&ops->list);
 429         ops_exit_list(ops, &net_exit_list);
 430         ops_free_list(ops, &net_exit_list);
 431         return error;
 432 }
 433
 434 static void __unregister_pernet_operations(struct pernet_operations *ops)
 435 {
 436         struct net *net;
 437         LIST_HEAD(net_exit_list);
 438
 439         list_del(&ops->list);
 440         for_each_net(net)
 441                 list_add_tail(&net->exit_list, &net_exit_list);
 442         ops_exit_list(ops, &net_exit_list);
 443         ops_free_list(ops, &net_exit_list);
 444 }
 445
 446 #else
 447
 448 static int __register_pernet_operations(struct list_head *list,
 449                                         struct pernet_operations *ops)
 450 {
 451         int err = 0;
 452         err = ops_init(ops, &init_net);
 453         if (err)
 454                 ops_free(ops, &init_net);
 455         return err;
 456
 457 }
 458
 459 static void __unregister_pernet_operations(struct pernet_operations *ops)
 460 {
 461         LIST_HEAD(net_exit_list);
 462         list_add(&init_net.exit_list, &net_exit_list);
 463         ops_exit_list(ops, &net_exit_list);
 464         ops_free_list(ops, &net_exit_list);
 465 }
 466
 467 #endif /* CONFIG_NET_NS */
 468
 469 static DEFINE_IDA(net_generic_ids);
 470
 471 static int register_pernet_operations(struct list_head *list,
 472                                       struct pernet_operations *ops)
 473 {
 474         int error;
 475
 476         if (ops->id) {
 477 again:
 478                 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 479                 if (error < 0) {
 480                         if (error == -EAGAIN) {
 481                                 ida_pre_get(&net_generic_ids, GFP_KERNEL);
 482                                 goto again;
 483                         }
 484                         return error;
 485                 }
 486                 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
 487         }
 488         error = __register_pernet_operations(list, ops);
 489         if (error) {
 490                 rcu_barrier();
 491                 if (ops->id)
 492                         ida_remove(&net_generic_ids, *ops->id);
 493         }
 494
 495         return error;
 496 }
 497
 498 static void unregister_pernet_operations(struct pernet_operations *ops)
 499 {
 500
 501         __unregister_pernet_operations(ops);
 502         rcu_barrier();
 503         if (ops->id)
 504                 ida_remove(&net_generic_ids, *ops->id);
 505 }
 506
 507 /**
 508  *      register_pernet_subsys - register a network namespace subsystem
 509  *      @ops:  pernet operations structure for the subsystem
 510  *
 511  *      Register a subsystem which has init and exit functions
 512  *      that are called when network namespaces are created and
 513  *      destroyed respectively.
 514  *
 515  *      When registered all network namespace init functions are
 516  *      called for every existing network namespace.  Allowing kernel
 517  *      modules to have a race free view of the set of network namespaces.
 518  *
 519  *      When a new network namespace is created all of the init
 520  *      methods are called in the order in which they were registered.
 521  *
 522  *      When a network namespace is destroyed all of the exit methods
 523  *      are called in the reverse of the order with which they were
 524  *      registered.
 525  */
 526 int register_pernet_subsys(struct pernet_operations *ops)
 527 {
 528         int error;
 529         mutex_lock(&net_mutex);
 530         error =  register_pernet_operations(first_device, ops);
 531         mutex_unlock(&net_mutex);
 532         return error;
 533 }
 534 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 535
 536 /**
 537  *      unregister_pernet_subsys - unregister a network namespace subsystem
 538  *      @ops: pernet operations structure to manipulate
 539  *
 540  *      Remove the pernet operations structure from the list to be
 541  *      used when network namespaces are created or destroyed.  In
 542  *      addition run the exit method for all existing network
 543  *      namespaces.
 544  */
 545 void unregister_pernet_subsys(struct pernet_operations *ops)
 546 {
 547         mutex_lock(&net_mutex);
 548         unregister_pernet_operations(ops);
 549         mutex_unlock(&net_mutex);
 550 }
 551 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 552
 553 /**
 554  *      register_pernet_device - register a network namespace device
 555  *      @ops:  pernet operations structure for the subsystem
 556  *
 557  *      Register a device which has init and exit functions
 558  *      that are called when network namespaces are created and
 559  *      destroyed respectively.
 560  *
 561  *      When registered all network namespace init functions are
 562  *      called for every existing network namespace.  Allowing kernel
 563  *      modules to have a race free view of the set of network namespaces.
 564  *
 565  *      When a new network namespace is created all of the init
 566  *      methods are called in the order in which they were registered.
 567  *
 568  *      When a network namespace is destroyed all of the exit methods
 569  *      are called in the reverse of the order with which they were
 570  *      registered.
 571  */
 572 int register_pernet_device(struct pernet_operations *ops)
 573 {
 574         int error;
 575         mutex_lock(&net_mutex);
 576         error = register_pernet_operations(&pernet_list, ops);
 577         if (!error && (first_device == &pernet_list))
 578                 first_device = &ops->list;
 579         mutex_unlock(&net_mutex);
 580         return error;
 581 }
 582 EXPORT_SYMBOL_GPL(register_pernet_device);
 583
 584 /**
 585  *      unregister_pernet_device - unregister a network namespace netdevice
 586  *      @ops: pernet operations structure to manipulate
 587  *
 588  *      Remove the pernet operations structure from the list to be
 589  *      used when network namespaces are created or destroyed.  In
 590  *      addition run the exit method for all existing network
 591  *      namespaces.
 592  */
 593 void unregister_pernet_device(struct pernet_operations *ops)
 594 {
 595         mutex_lock(&net_mutex);
 596         if (&ops->list == first_device)
 597                 first_device = first_device->next;
 598         unregister_pernet_operations(ops);
 599         mutex_unlock(&net_mutex);
 600 }
 601 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 602
 603 #ifdef CONFIG_NET_NS
 604 static void *netns_get(struct task_struct *task)
 605 {
 606         struct net *net = NULL;
 607         struct nsproxy *nsproxy;
 608
 609         rcu_read_lock();
 610         nsproxy = task_nsproxy(task);
 611         if (nsproxy)
 612                 net = get_net(nsproxy->net_ns);
 613         rcu_read_unlock();
 614
 615         return net;
 616 }
 617
 618 static void netns_put(void *ns)
 619 {
 620         put_net(ns);
 621 }
 622
 623 static int netns_install(struct nsproxy *nsproxy, void *ns)
 624 {
 625         put_net(nsproxy->net_ns);
 626         nsproxy->net_ns = get_net(ns);
 627         return 0;
 628 }
 629
 630 const struct proc_ns_operations netns_operations = {
 631         .name           = "net",
 632         .type           = CLONE_NEWNET,
 633         .get            = netns_get,
 634         .put            = netns_put,
 635         .install        = netns_install,
 636 };
 637 #endif