net/core/net_namespace.c

   1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   2
   3 #include <linux/workqueue.h>
   4 #include <linux/rtnetlink.h>
   5 #include <linux/cache.h>
   6 #include <linux/slab.h>
   7 #include <linux/list.h>
   8 #include <linux/delay.h>
   9 #include <linux/sched.h>
  10 #include <linux/idr.h>
  11 #include <linux/rculist.h>
  12 #include <linux/nsproxy.h>
  13 #include <linux/fs.h>
  14 #include <linux/proc_ns.h>
  15 #include <linux/file.h>
  16 #include <linux/export.h>
  17 #include <linux/user_namespace.h>
  18 #include <net/net_namespace.h>
  19 #include <net/netns/generic.h>
  20
  21 /*
  22  *      Our network namespace constructor/destructor lists
  23  */
  24
  25 static LIST_HEAD(pernet_list);
  26 static struct list_head *first_device = &pernet_list;
  27 DEFINE_MUTEX(net_mutex);
  28
  29 LIST_HEAD(net_namespace_list);
  30 EXPORT_SYMBOL_GPL(net_namespace_list);
  31
  32 struct net init_net = {
  33         .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
  34 };
  35 EXPORT_SYMBOL(init_net);
  36
  37 #define INITIAL_NET_GEN_PTRS    13 /* +1 for len +2 for rcu_head */
  38
  39 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  40
  41 static struct net_generic *net_alloc_generic(void)
  42 {
  43         struct net_generic *ng;
  44         size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  45
  46         ng = kzalloc(generic_size, GFP_KERNEL);
  47         if (ng)
  48                 ng->len = max_gen_ptrs;
  49
  50         return ng;
  51 }
  52
  53 static int net_assign_generic(struct net *net, int id, void *data)
  54 {
  55         struct net_generic *ng, *old_ng;
  56
  57         BUG_ON(!mutex_is_locked(&net_mutex));
  58         BUG_ON(id == 0);
  59
  60         old_ng = rcu_dereference_protected(net->gen,
  61                                            lockdep_is_held(&net_mutex));
  62         ng = old_ng;
  63         if (old_ng->len >= id)
  64                 goto assign;
  65
  66         ng = net_alloc_generic();
  67         if (ng == NULL)
  68                 return -ENOMEM;
  69
  70         /*
  71          * Some synchronisation notes:
  72          *
  73          * The net_generic explores the net->gen array inside rcu
  74          * read section. Besides once set the net->gen->ptr[x]
  75          * pointer never changes (see rules in netns/generic.h).
  76          *
  77          * That said, we simply duplicate this array and schedule
  78          * the old copy for kfree after a grace period.
  79          */
  80
  81         memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
  82
  83         rcu_assign_pointer(net->gen, ng);
  84         kfree_rcu(old_ng, rcu);
  85 assign:
  86         ng->ptr[id - 1] = data;
  87         return 0;
  88 }
  89
  90 static int ops_init(const struct pernet_operations *ops, struct net *net)
  91 {
  92         int err = -ENOMEM;
  93         void *data = NULL;
  94
  95         if (ops->id && ops->size) {
  96                 data = kzalloc(ops->size, GFP_KERNEL);
  97                 if (!data)
  98                         goto out;
  99
 100                 err = net_assign_generic(net, *ops->id, data);
 101                 if (err)
 102                         goto cleanup;
 103         }
 104         err = 0;
 105         if (ops->init)
 106                 err = ops->init(net);
 107         if (!err)
 108                 return 0;
 109
 110 cleanup:
 111         kfree(data);
 112
 113 out:
 114         return err;
 115 }
 116
 117 static void ops_free(const struct pernet_operations *ops, struct net *net)
 118 {
 119         if (ops->id && ops->size) {
 120                 int id = *ops->id;
 121                 kfree(net_generic(net, id));
 122         }
 123 }
 124
 125 static void ops_exit_list(const struct pernet_operations *ops,
 126                           struct list_head *net_exit_list)
 127 {
 128         struct net *net;
 129         if (ops->exit) {
 130                 list_for_each_entry(net, net_exit_list, exit_list)
 131                         ops->exit(net);
 132         }
 133         if (ops->exit_batch)
 134                 ops->exit_batch(net_exit_list);
 135 }
 136
 137 static void ops_free_list(const struct pernet_operations *ops,
 138                           struct list_head *net_exit_list)
 139 {
 140         struct net *net;
 141         if (ops->size && ops->id) {
 142                 list_for_each_entry(net, net_exit_list, exit_list)
 143                         ops_free(ops, net);
 144         }
 145 }
 146
 147 /*
 148  * setup_net runs the initializers for the network namespace object.
 149  */
 150 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 151 {
 152         /* Must be called with net_mutex held */
 153         const struct pernet_operations *ops, *saved_ops;
 154         int error = 0;
 155         LIST_HEAD(net_exit_list);
 156
 157         atomic_set(&net->count, 1);
 158         atomic_set(&net->passive, 1);
 159         net->dev_base_seq = 1;
 160         net->user_ns = user_ns;
 161
 162 #ifdef NETNS_REFCNT_DEBUG
 163         atomic_set(&net->use_count, 0);
 164 #endif
 165
 166         list_for_each_entry(ops, &pernet_list, list) {
 167                 error = ops_init(ops, net);
 168                 if (error < 0)
 169                         goto out_undo;
 170         }
 171 out:
 172         return error;
 173
 174 out_undo:
 175         /* Walk through the list backwards calling the exit functions
 176          * for the pernet modules whose init functions did not fail.
 177          */
 178         list_add(&net->exit_list, &net_exit_list);
 179         saved_ops = ops;
 180         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 181                 ops_exit_list(ops, &net_exit_list);
 182
 183         ops = saved_ops;
 184         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 185                 ops_free_list(ops, &net_exit_list);
 186
 187         rcu_barrier();
 188         goto out;
 189 }
 190
 191
 192 #ifdef CONFIG_NET_NS
 193 static struct kmem_cache *net_cachep;
 194 static struct workqueue_struct *netns_wq;
 195
 196 static struct net *net_alloc(void)
 197 {
 198         struct net *net = NULL;
 199         struct net_generic *ng;
 200
 201         ng = net_alloc_generic();
 202         if (!ng)
 203                 goto out;
 204
 205         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 206         if (!net)
 207                 goto out_free;
 208
 209         rcu_assign_pointer(net->gen, ng);
 210 out:
 211         return net;
 212
 213 out_free:
 214         kfree(ng);
 215         goto out;
 216 }
 217
 218 static void net_free(struct net *net)
 219 {
 220 #ifdef NETNS_REFCNT_DEBUG
 221         if (unlikely(atomic_read(&net->use_count) != 0)) {
 222                 pr_emerg("network namespace not free! Usage: %d\n",
 223                          atomic_read(&net->use_count));
 224                 return;
 225         }
 226 #endif
 227         kfree(rcu_access_pointer(net->gen));
 228         kmem_cache_free(net_cachep, net);
 229 }
 230
 231 void net_drop_ns(void *p)
 232 {
 233         struct net *ns = p;
 234         if (ns && atomic_dec_and_test(&ns->passive))
 235                 net_free(ns);
 236 }
 237
 238 struct net *copy_net_ns(unsigned long flags,
 239                         struct user_namespace *user_ns, struct net *old_net)
 240 {
 241         struct net *net;
 242         int rv;
 243
 244         if (!(flags & CLONE_NEWNET))
 245                 return get_net(old_net);
 246
 247         net = net_alloc();
 248         if (!net)
 249                 return ERR_PTR(-ENOMEM);
 250
 251         get_user_ns(user_ns);
 252
 253         mutex_lock(&net_mutex);
 254         rv = setup_net(net, user_ns);
 255         if (rv == 0) {
 256                 rtnl_lock();
 257                 list_add_tail_rcu(&net->list, &net_namespace_list);
 258                 rtnl_unlock();
 259         }
 260         mutex_unlock(&net_mutex);
 261         if (rv < 0) {
 262                 put_user_ns(user_ns);
 263                 net_drop_ns(net);
 264                 return ERR_PTR(rv);
 265         }
 266         return net;
 267 }
 268
 269 static DEFINE_SPINLOCK(cleanup_list_lock);
 270 static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 271
 272 static void cleanup_net(struct work_struct *work)
 273 {
 274         const struct pernet_operations *ops;
 275         struct net *net, *tmp;
 276         struct list_head net_kill_list;
 277         LIST_HEAD(net_exit_list);
 278
 279         /* Atomically snapshot the list of namespaces to cleanup */
 280         spin_lock_irq(&cleanup_list_lock);
 281         list_replace_init(&cleanup_list, &net_kill_list);
 282         spin_unlock_irq(&cleanup_list_lock);
 283
 284         mutex_lock(&net_mutex);
 285
 286         /* Don't let anyone else find us. */
 287         rtnl_lock();
 288         list_for_each_entry(net, &net_kill_list, cleanup_list) {
 289                 list_del_rcu(&net->list);
 290                 list_add_tail(&net->exit_list, &net_exit_list);
 291         }
 292         rtnl_unlock();
 293
 294         /*
 295          * Another CPU might be rcu-iterating the list, wait for it.
 296          * This needs to be before calling the exit() notifiers, so
 297          * the rcu_barrier() below isn't sufficient alone.
 298          */
 299         synchronize_rcu();
 300
 301         /* Run all of the network namespace exit methods */
 302         list_for_each_entry_reverse(ops, &pernet_list, list)
 303                 ops_exit_list(ops, &net_exit_list);
 304
 305         /* Free the net generic variables */
 306         list_for_each_entry_reverse(ops, &pernet_list, list)
 307                 ops_free_list(ops, &net_exit_list);
 308
 309         mutex_unlock(&net_mutex);
 310
 311         /* Ensure there are no outstanding rcu callbacks using this
 312          * network namespace.
 313          */
 314         rcu_barrier();
 315
 316         /* Finally it is safe to free my network namespace structure */
 317         list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 318                 list_del_init(&net->exit_list);
 319                 put_user_ns(net->user_ns);
 320                 net_drop_ns(net);
 321         }
 322 }
 323 static DECLARE_WORK(net_cleanup_work, cleanup_net);
 324
 325 void __put_net(struct net *net)
 326 {
 327         /* Cleanup the network namespace in process context */
 328         unsigned long flags;
 329
 330         spin_lock_irqsave(&cleanup_list_lock, flags);
 331         list_add(&net->cleanup_list, &cleanup_list);
 332         spin_unlock_irqrestore(&cleanup_list_lock, flags);
 333
 334         queue_work(netns_wq, &net_cleanup_work);
 335 }
 336 EXPORT_SYMBOL_GPL(__put_net);
 337
 338 struct net *get_net_ns_by_fd(int fd)
 339 {
 340         struct proc_ns *ei;
 341         struct file *file;
 342         struct net *net;
 343
 344         file = proc_ns_fget(fd);
 345         if (IS_ERR(file))
 346                 return ERR_CAST(file);
 347
 348         ei = get_proc_ns(file_inode(file));
 349         if (ei->ns_ops == &netns_operations)
 350                 net = get_net(ei->ns);
 351         else
 352                 net = ERR_PTR(-EINVAL);
 353
 354         fput(file);
 355         return net;
 356 }
 357
 358 #else
 359 struct net *get_net_ns_by_fd(int fd)
 360 {
 361         return ERR_PTR(-EINVAL);
 362 }
 363 #endif
 364
 365 struct net *get_net_ns_by_pid(pid_t pid)
 366 {
 367         struct task_struct *tsk;
 368         struct net *net;
 369
 370         /* Lookup the network namespace */
 371         net = ERR_PTR(-ESRCH);
 372         rcu_read_lock();
 373         tsk = find_task_by_vpid(pid);
 374         if (tsk) {
 375                 struct nsproxy *nsproxy;
 376                 task_lock(tsk);
 377                 nsproxy = tsk->nsproxy;
 378                 if (nsproxy)
 379                         net = get_net(nsproxy->net_ns);
 380                 task_unlock(tsk);
 381         }
 382         rcu_read_unlock();
 383         return net;
 384 }
 385 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 386
 387 static __net_init int net_ns_net_init(struct net *net)
 388 {
 389         return proc_alloc_inum(&net->proc_inum);
 390 }
 391
 392 static __net_exit void net_ns_net_exit(struct net *net)
 393 {
 394         proc_free_inum(net->proc_inum);
 395 }
 396
 397 static struct pernet_operations __net_initdata net_ns_ops = {
 398         .init = net_ns_net_init,
 399         .exit = net_ns_net_exit,
 400 };
 401
 402 static int __init net_ns_init(void)
 403 {
 404         struct net_generic *ng;
 405
 406 #ifdef CONFIG_NET_NS
 407         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 408                                         SMP_CACHE_BYTES,
 409                                         SLAB_PANIC, NULL);
 410
 411         /* Create workqueue for cleanup */
 412         netns_wq = create_singlethread_workqueue("netns");
 413         if (!netns_wq)
 414                 panic("Could not create netns workq");
 415 #endif
 416
 417         ng = net_alloc_generic();
 418         if (!ng)
 419                 panic("Could not allocate generic netns");
 420
 421         rcu_assign_pointer(init_net.gen, ng);
 422
 423         mutex_lock(&net_mutex);
 424         if (setup_net(&init_net, &init_user_ns))
 425                 panic("Could not setup the initial network namespace");
 426
 427         rtnl_lock();
 428         list_add_tail_rcu(&init_net.list, &net_namespace_list);
 429         rtnl_unlock();
 430
 431         mutex_unlock(&net_mutex);
 432
 433         register_pernet_subsys(&net_ns_ops);
 434
 435         return 0;
 436 }
 437
 438 pure_initcall(net_ns_init);
 439
 440 #ifdef CONFIG_NET_NS
 441 static int __register_pernet_operations(struct list_head *list,
 442                                         struct pernet_operations *ops)
 443 {
 444         struct net *net;
 445         int error;
 446         LIST_HEAD(net_exit_list);
 447
 448         list_add_tail(&ops->list, list);
 449         if (ops->init || (ops->id && ops->size)) {
 450                 for_each_net(net) {
 451                         error = ops_init(ops, net);
 452                         if (error)
 453                                 goto out_undo;
 454                         list_add_tail(&net->exit_list, &net_exit_list);
 455                 }
 456         }
 457         return 0;
 458
 459 out_undo:
 460         /* If I have an error cleanup all namespaces I initialized */
 461         list_del(&ops->list);
 462         ops_exit_list(ops, &net_exit_list);
 463         ops_free_list(ops, &net_exit_list);
 464         return error;
 465 }
 466
 467 static void __unregister_pernet_operations(struct pernet_operations *ops)
 468 {
 469         struct net *net;
 470         LIST_HEAD(net_exit_list);
 471
 472         list_del(&ops->list);
 473         for_each_net(net)
 474                 list_add_tail(&net->exit_list, &net_exit_list);
 475         ops_exit_list(ops, &net_exit_list);
 476         ops_free_list(ops, &net_exit_list);
 477 }
 478
 479 #else
 480
 481 static int __register_pernet_operations(struct list_head *list,
 482                                         struct pernet_operations *ops)
 483 {
 484         return ops_init(ops, &init_net);
 485 }
 486
 487 static void __unregister_pernet_operations(struct pernet_operations *ops)
 488 {
 489         LIST_HEAD(net_exit_list);
 490         list_add(&init_net.exit_list, &net_exit_list);
 491         ops_exit_list(ops, &net_exit_list);
 492         ops_free_list(ops, &net_exit_list);
 493 }
 494
 495 #endif /* CONFIG_NET_NS */
 496
 497 static DEFINE_IDA(net_generic_ids);
 498
 499 static int register_pernet_operations(struct list_head *list,
 500                                       struct pernet_operations *ops)
 501 {
 502         int error;
 503
 504         if (ops->id) {
 505 again:
 506                 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 507                 if (error < 0) {
 508                         if (error == -EAGAIN) {
 509                                 ida_pre_get(&net_generic_ids, GFP_KERNEL);
 510                                 goto again;
 511                         }
 512                         return error;
 513                 }
 514                 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
 515         }
 516         error = __register_pernet_operations(list, ops);
 517         if (error) {
 518                 rcu_barrier();
 519                 if (ops->id)
 520                         ida_remove(&net_generic_ids, *ops->id);
 521         }
 522
 523         return error;
 524 }
 525
 526 static void unregister_pernet_operations(struct pernet_operations *ops)
 527 {
 528
 529         __unregister_pernet_operations(ops);
 530         rcu_barrier();
 531         if (ops->id)
 532                 ida_remove(&net_generic_ids, *ops->id);
 533 }
 534
 535 /**
 536  *      register_pernet_subsys - register a network namespace subsystem
 537  *      @ops:  pernet operations structure for the subsystem
 538  *
 539  *      Register a subsystem which has init and exit functions
 540  *      that are called when network namespaces are created and
 541  *      destroyed respectively.
 542  *
 543  *      When registered all network namespace init functions are
 544  *      called for every existing network namespace.  Allowing kernel
 545  *      modules to have a race free view of the set of network namespaces.
 546  *
 547  *      When a new network namespace is created all of the init
 548  *      methods are called in the order in which they were registered.
 549  *
 550  *      When a network namespace is destroyed all of the exit methods
 551  *      are called in the reverse of the order with which they were
 552  *      registered.
 553  */
 554 int register_pernet_subsys(struct pernet_operations *ops)
 555 {
 556         int error;
 557         mutex_lock(&net_mutex);
 558         error =  register_pernet_operations(first_device, ops);
 559         mutex_unlock(&net_mutex);
 560         return error;
 561 }
 562 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 563
 564 /**
 565  *      unregister_pernet_subsys - unregister a network namespace subsystem
 566  *      @ops: pernet operations structure to manipulate
 567  *
 568  *      Remove the pernet operations structure from the list to be
 569  *      used when network namespaces are created or destroyed.  In
 570  *      addition run the exit method for all existing network
 571  *      namespaces.
 572  */
 573 void unregister_pernet_subsys(struct pernet_operations *ops)
 574 {
 575         mutex_lock(&net_mutex);
 576         unregister_pernet_operations(ops);
 577         mutex_unlock(&net_mutex);
 578 }
 579 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 580
 581 /**
 582  *      register_pernet_device - register a network namespace device
 583  *      @ops:  pernet operations structure for the subsystem
 584  *
 585  *      Register a device which has init and exit functions
 586  *      that are called when network namespaces are created and
 587  *      destroyed respectively.
 588  *
 589  *      When registered all network namespace init functions are
 590  *      called for every existing network namespace.  Allowing kernel
 591  *      modules to have a race free view of the set of network namespaces.
 592  *
 593  *      When a new network namespace is created all of the init
 594  *      methods are called in the order in which they were registered.
 595  *
 596  *      When a network namespace is destroyed all of the exit methods
 597  *      are called in the reverse of the order with which they were
 598  *      registered.
 599  */
 600 int register_pernet_device(struct pernet_operations *ops)
 601 {
 602         int error;
 603         mutex_lock(&net_mutex);
 604         error = register_pernet_operations(&pernet_list, ops);
 605         if (!error && (first_device == &pernet_list))
 606                 first_device = &ops->list;
 607         mutex_unlock(&net_mutex);
 608         return error;
 609 }
 610 EXPORT_SYMBOL_GPL(register_pernet_device);
 611
 612 /**
 613  *      unregister_pernet_device - unregister a network namespace netdevice
 614  *      @ops: pernet operations structure to manipulate
 615  *
 616  *      Remove the pernet operations structure from the list to be
 617  *      used when network namespaces are created or destroyed.  In
 618  *      addition run the exit method for all existing network
 619  *      namespaces.
 620  */
 621 void unregister_pernet_device(struct pernet_operations *ops)
 622 {
 623         mutex_lock(&net_mutex);
 624         if (&ops->list == first_device)
 625                 first_device = first_device->next;
 626         unregister_pernet_operations(ops);
 627         mutex_unlock(&net_mutex);
 628 }
 629 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 630
 631 #ifdef CONFIG_NET_NS
 632 static void *netns_get(struct task_struct *task)
 633 {
 634         struct net *net = NULL;
 635         struct nsproxy *nsproxy;
 636
 637         task_lock(task);
 638         nsproxy = task->nsproxy;
 639         if (nsproxy)
 640                 net = get_net(nsproxy->net_ns);
 641         task_unlock(task);
 642
 643         return net;
 644 }
 645
 646 static void netns_put(void *ns)
 647 {
 648         put_net(ns);
 649 }
 650
 651 static int netns_install(struct nsproxy *nsproxy, void *ns)
 652 {
 653         struct net *net = ns;
 654
 655         if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
 656             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 657                 return -EPERM;
 658
 659         put_net(nsproxy->net_ns);
 660         nsproxy->net_ns = get_net(net);
 661         return 0;
 662 }
 663
 664 static unsigned int netns_inum(void *ns)
 665 {
 666         struct net *net = ns;
 667         return net->proc_inum;
 668 }
 669
 670 const struct proc_ns_operations netns_operations = {
 671         .name           = "net",
 672         .type           = CLONE_NEWNET,
 673         .get            = netns_get,
 674         .put            = netns_put,
 675         .install        = netns_install,
 676         .inum           = netns_inum,
 677 };
 678 #endif