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