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 = -ENOMEM;
  87         void *data = NULL;
  88
  89         if (ops->id && ops->size) {
  90                 data = kzalloc(ops->size, GFP_KERNEL);
  91                 if (!data)
  92                         goto out;
  93
  94                 err = net_assign_generic(net, *ops->id, data);
  95                 if (err)
  96                         goto cleanup;
  97         }
  98         err = 0;
  99         if (ops->init)
 100                 err = ops->init(net);
 101         if (!err)
 102                 return 0;
 103
 104 cleanup:
 105         kfree(data);
 106
 107 out:
 108         return err;
 109 }
 110
 111 static void ops_free(const struct pernet_operations *ops, struct net *net)
 112 {
 113         if (ops->id && ops->size) {
 114                 int id = *ops->id;
 115                 kfree(net_generic(net, id));
 116         }
 117 }
 118
 119 static void ops_exit_list(const struct pernet_operations *ops,
 120                           struct list_head *net_exit_list)
 121 {
 122         struct net *net;
 123         if (ops->exit) {
 124                 list_for_each_entry(net, net_exit_list, exit_list)
 125                         ops->exit(net);
 126         }
 127         if (ops->exit_batch)
 128                 ops->exit_batch(net_exit_list);
 129 }
 130
 131 static void ops_free_list(const struct pernet_operations *ops,
 132                           struct list_head *net_exit_list)
 133 {
 134         struct net *net;
 135         if (ops->size && ops->id) {
 136                 list_for_each_entry(net, net_exit_list, exit_list)
 137                         ops_free(ops, net);
 138         }
 139 }
 140
 141 /*
 142  * setup_net runs the initializers for the network namespace object.
 143  */
 144 static __net_init int setup_net(struct net *net)
 145 {
 146         /* Must be called with net_mutex held */
 147         const struct pernet_operations *ops, *saved_ops;
 148         int error = 0;
 149         LIST_HEAD(net_exit_list);
 150
 151         atomic_set(&net->count, 1);
 152         atomic_set(&net->passive, 1);
 153         net->dev_base_seq = 1;
 154
 155 #ifdef NETNS_REFCNT_DEBUG
 156         atomic_set(&net->use_count, 0);
 157 #endif
 158
 159         list_for_each_entry(ops, &pernet_list, list) {
 160                 error = ops_init(ops, net);
 161                 if (error < 0)
 162                         goto out_undo;
 163         }
 164 out:
 165         return error;
 166
 167 out_undo:
 168         /* Walk through the list backwards calling the exit functions
 169          * for the pernet modules whose init functions did not fail.
 170          */
 171         list_add(&net->exit_list, &net_exit_list);
 172         saved_ops = ops;
 173         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 174                 ops_exit_list(ops, &net_exit_list);
 175
 176         ops = saved_ops;
 177         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 178                 ops_free_list(ops, &net_exit_list);
 179
 180         rcu_barrier();
 181         goto out;
 182 }
 183
 184
 185 #ifdef CONFIG_NET_NS
 186 static struct kmem_cache *net_cachep;
 187 static struct workqueue_struct *netns_wq;
 188
 189 static struct net *net_alloc(void)
 190 {
 191         struct net *net = NULL;
 192         struct net_generic *ng;
 193
 194         ng = net_alloc_generic();
 195         if (!ng)
 196                 goto out;
 197
 198         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 199         if (!net)
 200                 goto out_free;
 201
 202         rcu_assign_pointer(net->gen, ng);
 203 out:
 204         return net;
 205
 206 out_free:
 207         kfree(ng);
 208         goto out;
 209 }
 210
 211 static void net_free(struct net *net)
 212 {
 213 #ifdef NETNS_REFCNT_DEBUG
 214         if (unlikely(atomic_read(&net->use_count) != 0)) {
 215                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
 216                         atomic_read(&net->use_count));
 217                 return;
 218         }
 219 #endif
 220         kfree(net->gen);
 221         kmem_cache_free(net_cachep, net);
 222 }
 223
 224 void net_drop_ns(void *p)
 225 {
 226         struct net *ns = p;
 227         if (ns && atomic_dec_and_test(&ns->passive))
 228                 net_free(ns);
 229 }
 230
 231 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 232 {
 233         struct net *net;
 234         int rv;
 235
 236         if (!(flags & CLONE_NEWNET))
 237                 return get_net(old_net);
 238
 239         net = net_alloc();
 240         if (!net)
 241                 return ERR_PTR(-ENOMEM);
 242         mutex_lock(&net_mutex);
 243         rv = setup_net(net);
 244         if (rv == 0) {
 245                 rtnl_lock();
 246                 list_add_tail_rcu(&net->list, &net_namespace_list);
 247                 rtnl_unlock();
 248         }
 249         mutex_unlock(&net_mutex);
 250         if (rv < 0) {
 251                 net_drop_ns(net);
 252                 return ERR_PTR(rv);
 253         }
 254         return net;
 255 }
 256
 257 static DEFINE_SPINLOCK(cleanup_list_lock);
 258 static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 259
 260 static void cleanup_net(struct work_struct *work)
 261 {
 262         const struct pernet_operations *ops;
 263         struct net *net, *tmp;
 264         LIST_HEAD(net_kill_list);
 265         LIST_HEAD(net_exit_list);
 266
 267         /* Atomically snapshot the list of namespaces to cleanup */
 268         spin_lock_irq(&cleanup_list_lock);
 269         list_replace_init(&cleanup_list, &net_kill_list);
 270         spin_unlock_irq(&cleanup_list_lock);
 271
 272         mutex_lock(&net_mutex);
 273
 274         /* Don't let anyone else find us. */
 275         rtnl_lock();
 276         list_for_each_entry(net, &net_kill_list, cleanup_list) {
 277                 list_del_rcu(&net->list);
 278                 list_add_tail(&net->exit_list, &net_exit_list);
 279         }
 280         rtnl_unlock();
 281
 282         /*
 283          * Another CPU might be rcu-iterating the list, wait for it.
 284          * This needs to be before calling the exit() notifiers, so
 285          * the rcu_barrier() below isn't sufficient alone.
 286          */
 287         synchronize_rcu();
 288
 289         /* Run all of the network namespace exit methods */
 290         list_for_each_entry_reverse(ops, &pernet_list, list)
 291                 ops_exit_list(ops, &net_exit_list);
 292
 293         /* Free the net generic variables */
 294         list_for_each_entry_reverse(ops, &pernet_list, list)
 295                 ops_free_list(ops, &net_exit_list);
 296
 297         mutex_unlock(&net_mutex);
 298
 299         /* Ensure there are no outstanding rcu callbacks using this
 300          * network namespace.
 301          */
 302         rcu_barrier();
 303
 304         /* Finally it is safe to free my network namespace structure */
 305         list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 306                 list_del_init(&net->exit_list);
 307                 net_drop_ns(net);
 308         }
 309 }
 310 static DECLARE_WORK(net_cleanup_work, cleanup_net);
 311
 312 void __put_net(struct net *net)
 313 {
 314         /* Cleanup the network namespace in process context */
 315         unsigned long flags;
 316
 317         spin_lock_irqsave(&cleanup_list_lock, flags);
 318         list_add(&net->cleanup_list, &cleanup_list);
 319         spin_unlock_irqrestore(&cleanup_list_lock, flags);
 320
 321         queue_work(netns_wq, &net_cleanup_work);
 322 }
 323 EXPORT_SYMBOL_GPL(__put_net);
 324
 325 struct net *get_net_ns_by_fd(int fd)
 326 {
 327         struct proc_inode *ei;
 328         struct file *file;
 329         struct net *net;
 330
 331         file = proc_ns_fget(fd);
 332         if (IS_ERR(file))
 333                 return ERR_CAST(file);
 334
 335         ei = PROC_I(file->f_dentry->d_inode);
 336         if (ei->ns_ops == &netns_operations)
 337                 net = get_net(ei->ns);
 338         else
 339                 net = ERR_PTR(-EINVAL);
 340
 341         fput(file);
 342         return net;
 343 }
 344
 345 #else
 346 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 347 {
 348         if (flags & CLONE_NEWNET)
 349                 return ERR_PTR(-EINVAL);
 350         return old_net;
 351 }
 352
 353 struct net *get_net_ns_by_fd(int fd)
 354 {
 355         return ERR_PTR(-EINVAL);
 356 }
 357 #endif
 358
 359 struct net *get_net_ns_by_pid(pid_t pid)
 360 {
 361         struct task_struct *tsk;
 362         struct net *net;
 363
 364         /* Lookup the network namespace */
 365         net = ERR_PTR(-ESRCH);
 366         rcu_read_lock();
 367         tsk = find_task_by_vpid(pid);
 368         if (tsk) {
 369                 struct nsproxy *nsproxy;
 370                 nsproxy = task_nsproxy(tsk);
 371                 if (nsproxy)
 372                         net = get_net(nsproxy->net_ns);
 373         }
 374         rcu_read_unlock();
 375         return net;
 376 }
 377 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 378
 379 static int __init net_ns_init(void)
 380 {
 381         struct net_generic *ng;
 382
 383 #ifdef CONFIG_NET_NS
 384         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 385                                         SMP_CACHE_BYTES,
 386                                         SLAB_PANIC, NULL);
 387
 388         /* Create workqueue for cleanup */
 389         netns_wq = create_singlethread_workqueue("netns");
 390         if (!netns_wq)
 391                 panic("Could not create netns workq");
 392 #endif
 393
 394         ng = net_alloc_generic();
 395         if (!ng)
 396                 panic("Could not allocate generic netns");
 397
 398         rcu_assign_pointer(init_net.gen, ng);
 399
 400         mutex_lock(&net_mutex);
 401         if (setup_net(&init_net))
 402                 panic("Could not setup the initial network namespace");
 403
 404         rtnl_lock();
 405         list_add_tail_rcu(&init_net.list, &net_namespace_list);
 406         rtnl_unlock();
 407
 408         mutex_unlock(&net_mutex);
 409
 410         return 0;
 411 }
 412
 413 pure_initcall(net_ns_init);
 414
 415 #ifdef CONFIG_NET_NS
 416 static int __register_pernet_operations(struct list_head *list,
 417                                         struct pernet_operations *ops)
 418 {
 419         struct net *net;
 420         int error;
 421         LIST_HEAD(net_exit_list);
 422
 423         list_add_tail(&ops->list, list);
 424         if (ops->init || (ops->id && ops->size)) {
 425                 for_each_net(net) {
 426                         error = ops_init(ops, net);
 427                         if (error)
 428                                 goto out_undo;
 429                         list_add_tail(&net->exit_list, &net_exit_list);
 430                 }
 431         }
 432         return 0;
 433
 434 out_undo:
 435         /* If I have an error cleanup all namespaces I initialized */
 436         list_del(&ops->list);
 437         ops_exit_list(ops, &net_exit_list);
 438         ops_free_list(ops, &net_exit_list);
 439         return error;
 440 }
 441
 442 static void __unregister_pernet_operations(struct pernet_operations *ops)
 443 {
 444         struct net *net;
 445         LIST_HEAD(net_exit_list);
 446
 447         list_del(&ops->list);
 448         for_each_net(net)
 449                 list_add_tail(&net->exit_list, &net_exit_list);
 450         ops_exit_list(ops, &net_exit_list);
 451         ops_free_list(ops, &net_exit_list);
 452 }
 453
 454 #else
 455
 456 static int __register_pernet_operations(struct list_head *list,
 457                                         struct pernet_operations *ops)
 458 {
 459         return ops_init(ops, &init_net);
 460 }
 461
 462 static void __unregister_pernet_operations(struct pernet_operations *ops)
 463 {
 464         LIST_HEAD(net_exit_list);
 465         list_add(&init_net.exit_list, &net_exit_list);
 466         ops_exit_list(ops, &net_exit_list);
 467         ops_free_list(ops, &net_exit_list);
 468 }
 469
 470 #endif /* CONFIG_NET_NS */
 471
 472 static DEFINE_IDA(net_generic_ids);
 473
 474 static int register_pernet_operations(struct list_head *list,
 475                                       struct pernet_operations *ops)
 476 {
 477         int error;
 478
 479         if (ops->id) {
 480 again:
 481                 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 482                 if (error < 0) {
 483                         if (error == -EAGAIN) {
 484                                 ida_pre_get(&net_generic_ids, GFP_KERNEL);
 485                                 goto again;
 486                         }
 487                         return error;
 488                 }
 489                 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
 490         }
 491         error = __register_pernet_operations(list, ops);
 492         if (error) {
 493                 rcu_barrier();
 494                 if (ops->id)
 495                         ida_remove(&net_generic_ids, *ops->id);
 496         }
 497
 498         return error;
 499 }
 500
 501 static void unregister_pernet_operations(struct pernet_operations *ops)
 502 {
 503
 504         __unregister_pernet_operations(ops);
 505         rcu_barrier();
 506         if (ops->id)
 507                 ida_remove(&net_generic_ids, *ops->id);
 508 }
 509
 510 /**
 511  *      register_pernet_subsys - register a network namespace subsystem
 512  *      @ops:  pernet operations structure for the subsystem
 513  *
 514  *      Register a subsystem which has init and exit functions
 515  *      that are called when network namespaces are created and
 516  *      destroyed respectively.
 517  *
 518  *      When registered all network namespace init functions are
 519  *      called for every existing network namespace.  Allowing kernel
 520  *      modules to have a race free view of the set of network namespaces.
 521  *
 522  *      When a new network namespace is created all of the init
 523  *      methods are called in the order in which they were registered.
 524  *
 525  *      When a network namespace is destroyed all of the exit methods
 526  *      are called in the reverse of the order with which they were
 527  *      registered.
 528  */
 529 int register_pernet_subsys(struct pernet_operations *ops)
 530 {
 531         int error;
 532         mutex_lock(&net_mutex);
 533         error =  register_pernet_operations(first_device, ops);
 534         mutex_unlock(&net_mutex);
 535         return error;
 536 }
 537 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 538
 539 /**
 540  *      unregister_pernet_subsys - unregister a network namespace subsystem
 541  *      @ops: pernet operations structure to manipulate
 542  *
 543  *      Remove the pernet operations structure from the list to be
 544  *      used when network namespaces are created or destroyed.  In
 545  *      addition run the exit method for all existing network
 546  *      namespaces.
 547  */
 548 void unregister_pernet_subsys(struct pernet_operations *ops)
 549 {
 550         mutex_lock(&net_mutex);
 551         unregister_pernet_operations(ops);
 552         mutex_unlock(&net_mutex);
 553 }
 554 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 555
 556 /**
 557  *      register_pernet_device - register a network namespace device
 558  *      @ops:  pernet operations structure for the subsystem
 559  *
 560  *      Register a device which has init and exit functions
 561  *      that are called when network namespaces are created and
 562  *      destroyed respectively.
 563  *
 564  *      When registered all network namespace init functions are
 565  *      called for every existing network namespace.  Allowing kernel
 566  *      modules to have a race free view of the set of network namespaces.
 567  *
 568  *      When a new network namespace is created all of the init
 569  *      methods are called in the order in which they were registered.
 570  *
 571  *      When a network namespace is destroyed all of the exit methods
 572  *      are called in the reverse of the order with which they were
 573  *      registered.
 574  */
 575 int register_pernet_device(struct pernet_operations *ops)
 576 {
 577         int error;
 578         mutex_lock(&net_mutex);
 579         error = register_pernet_operations(&pernet_list, ops);
 580         if (!error && (first_device == &pernet_list))
 581                 first_device = &ops->list;
 582         mutex_unlock(&net_mutex);
 583         return error;
 584 }
 585 EXPORT_SYMBOL_GPL(register_pernet_device);
 586
 587 /**
 588  *      unregister_pernet_device - unregister a network namespace netdevice
 589  *      @ops: pernet operations structure to manipulate
 590  *
 591  *      Remove the pernet operations structure from the list to be
 592  *      used when network namespaces are created or destroyed.  In
 593  *      addition run the exit method for all existing network
 594  *      namespaces.
 595  */
 596 void unregister_pernet_device(struct pernet_operations *ops)
 597 {
 598         mutex_lock(&net_mutex);
 599         if (&ops->list == first_device)
 600                 first_device = first_device->next;
 601         unregister_pernet_operations(ops);
 602         mutex_unlock(&net_mutex);
 603 }
 604 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 605
 606 #ifdef CONFIG_NET_NS
 607 static void *netns_get(struct task_struct *task)
 608 {
 609         struct net *net = NULL;
 610         struct nsproxy *nsproxy;
 611
 612         rcu_read_lock();
 613         nsproxy = task_nsproxy(task);
 614         if (nsproxy)
 615                 net = get_net(nsproxy->net_ns);
 616         rcu_read_unlock();
 617
 618         return net;
 619 }
 620
 621 static void netns_put(void *ns)
 622 {
 623         put_net(ns);
 624 }
 625
 626 static int netns_install(struct nsproxy *nsproxy, void *ns)
 627 {
 628         put_net(nsproxy->net_ns);
 629         nsproxy->net_ns = get_net(ns);
 630         return 0;
 631 }
 632
 633 const struct proc_ns_operations netns_operations = {
 634         .name           = "net",
 635         .type           = CLONE_NEWNET,
 636         .get            = netns_get,
 637         .put            = netns_put,
 638         .install        = netns_install,
 639 };
 640 #endif