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