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