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         atomic_set(&net->passive, 1);
 132         net->dev_base_seq = 1;
 133
 134 #ifdef NETNS_REFCNT_DEBUG
 135         atomic_set(&net->use_count, 0);
 136 #endif
 137
 138         list_for_each_entry(ops, &pernet_list, list) {
 139                 error = ops_init(ops, net);
 140                 if (error < 0)
 141                         goto out_undo;
 142         }
 143 out:
 144         return error;
 145
 146 out_undo:
 147         /* Walk through the list backwards calling the exit functions
 148          * for the pernet modules whose init functions did not fail.
 149          */
 150         list_add(&net->exit_list, &net_exit_list);
 151         saved_ops = ops;
 152         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 153                 ops_exit_list(ops, &net_exit_list);
 154
 155         ops = saved_ops;
 156         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 157                 ops_free_list(ops, &net_exit_list);
 158
 159         rcu_barrier();
 160         goto out;
 161 }
 162
 163 static struct net_generic *net_alloc_generic(void)
 164 {
 165         struct net_generic *ng;
 166         size_t generic_size = sizeof(struct net_generic) +
 167                 INITIAL_NET_GEN_PTRS * sizeof(void *);
 168
 169         ng = kzalloc(generic_size, GFP_KERNEL);
 170         if (ng)
 171                 ng->len = INITIAL_NET_GEN_PTRS;
 172
 173         return ng;
 174 }
 175
 176 #ifdef CONFIG_NET_NS
 177 static struct kmem_cache *net_cachep;
 178 static struct workqueue_struct *netns_wq;
 179
 180 static struct net *net_alloc(void)
 181 {
 182         struct net *net = NULL;
 183         struct net_generic *ng;
 184
 185         ng = net_alloc_generic();
 186         if (!ng)
 187                 goto out;
 188
 189         net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 190         if (!net)
 191                 goto out_free;
 192
 193         rcu_assign_pointer(net->gen, ng);
 194 out:
 195         return net;
 196
 197 out_free:
 198         kfree(ng);
 199         goto out;
 200 }
 201
 202 static void net_free(struct net *net)
 203 {
 204 #ifdef NETNS_REFCNT_DEBUG
 205         if (unlikely(atomic_read(&net->use_count) != 0)) {
 206                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
 207                         atomic_read(&net->use_count));
 208                 return;
 209         }
 210 #endif
 211         kfree(net->gen);
 212         kmem_cache_free(net_cachep, net);
 213 }
 214
 215 void net_drop_ns(void *p)
 216 {
 217         struct net *ns = p;
 218         if (ns && atomic_dec_and_test(&ns->passive))
 219                 net_free(ns);
 220 }
 221
 222 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 223 {
 224         struct net *net;
 225         int rv;
 226
 227         if (!(flags & CLONE_NEWNET))
 228                 return get_net(old_net);
 229
 230         net = net_alloc();
 231         if (!net)
 232                 return ERR_PTR(-ENOMEM);
 233         mutex_lock(&net_mutex);
 234         rv = setup_net(net);
 235         if (rv == 0) {
 236                 rtnl_lock();
 237                 list_add_tail_rcu(&net->list, &net_namespace_list);
 238                 rtnl_unlock();
 239         }
 240         mutex_unlock(&net_mutex);
 241         if (rv < 0) {
 242                 net_drop_ns(net);
 243                 return ERR_PTR(rv);
 244         }
 245         return net;
 246 }
 247
 248 static DEFINE_SPINLOCK(cleanup_list_lock);
 249 static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 250
 251 static void cleanup_net(struct work_struct *work)
 252 {
 253         const struct pernet_operations *ops;
 254         struct net *net, *tmp;
 255         LIST_HEAD(net_kill_list);
 256         LIST_HEAD(net_exit_list);
 257
 258         /* Atomically snapshot the list of namespaces to cleanup */
 259         spin_lock_irq(&cleanup_list_lock);
 260         list_replace_init(&cleanup_list, &net_kill_list);
 261         spin_unlock_irq(&cleanup_list_lock);
 262
 263         mutex_lock(&net_mutex);
 264
 265         /* Don't let anyone else find us. */
 266         rtnl_lock();
 267         list_for_each_entry(net, &net_kill_list, cleanup_list) {
 268                 list_del_rcu(&net->list);
 269                 list_add_tail(&net->exit_list, &net_exit_list);
 270         }
 271         rtnl_unlock();
 272
 273         /*
 274          * Another CPU might be rcu-iterating the list, wait for it.
 275          * This needs to be before calling the exit() notifiers, so
 276          * the rcu_barrier() below isn't sufficient alone.
 277          */
 278         synchronize_rcu();
 279
 280         /* Run all of the network namespace exit methods */
 281         list_for_each_entry_reverse(ops, &pernet_list, list)
 282                 ops_exit_list(ops, &net_exit_list);
 283
 284         /* Free the net generic variables */
 285         list_for_each_entry_reverse(ops, &pernet_list, list)
 286                 ops_free_list(ops, &net_exit_list);
 287
 288         mutex_unlock(&net_mutex);
 289
 290         /* Ensure there are no outstanding rcu callbacks using this
 291          * network namespace.
 292          */
 293         rcu_barrier();
 294
 295         /* Finally it is safe to free my network namespace structure */
 296         list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 297                 list_del_init(&net->exit_list);
 298                 net_drop_ns(net);
 299         }
 300 }
 301 static DECLARE_WORK(net_cleanup_work, cleanup_net);
 302
 303 void __put_net(struct net *net)
 304 {
 305         /* Cleanup the network namespace in process context */
 306         unsigned long flags;
 307
 308         spin_lock_irqsave(&cleanup_list_lock, flags);
 309         list_add(&net->cleanup_list, &cleanup_list);
 310         spin_unlock_irqrestore(&cleanup_list_lock, flags);
 311
 312         queue_work(netns_wq, &net_cleanup_work);
 313 }
 314 EXPORT_SYMBOL_GPL(__put_net);
 315
 316 struct net *get_net_ns_by_fd(int fd)
 317 {
 318         struct proc_inode *ei;
 319         struct file *file;
 320         struct net *net;
 321
 322         file = proc_ns_fget(fd);
 323         if (IS_ERR(file))
 324                 return ERR_CAST(file);
 325
 326         ei = PROC_I(file->f_dentry->d_inode);
 327         if (ei->ns_ops == &netns_operations)
 328                 net = get_net(ei->ns);
 329         else
 330                 net = ERR_PTR(-EINVAL);
 331
 332         fput(file);
 333         return net;
 334 }
 335
 336 #else
 337 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 338 {
 339         if (flags & CLONE_NEWNET)
 340                 return ERR_PTR(-EINVAL);
 341         return old_net;
 342 }
 343
 344 struct net *get_net_ns_by_fd(int fd)
 345 {
 346         return ERR_PTR(-EINVAL);
 347 }
 348 #endif
 349
 350 struct net *get_net_ns_by_pid(pid_t pid)
 351 {
 352         struct task_struct *tsk;
 353         struct net *net;
 354
 355         /* Lookup the network namespace */
 356         net = ERR_PTR(-ESRCH);
 357         rcu_read_lock();
 358         tsk = find_task_by_vpid(pid);
 359         if (tsk) {
 360                 struct nsproxy *nsproxy;
 361                 nsproxy = task_nsproxy(tsk);
 362                 if (nsproxy)
 363                         net = get_net(nsproxy->net_ns);
 364         }
 365         rcu_read_unlock();
 366         return net;
 367 }
 368 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 369
 370 static int __init net_ns_init(void)
 371 {
 372         struct net_generic *ng;
 373
 374 #ifdef CONFIG_NET_NS
 375         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 376                                         SMP_CACHE_BYTES,
 377                                         SLAB_PANIC, NULL);
 378
 379         /* Create workqueue for cleanup */
 380         netns_wq = create_singlethread_workqueue("netns");
 381         if (!netns_wq)
 382                 panic("Could not create netns workq");
 383 #endif
 384
 385         ng = net_alloc_generic();
 386         if (!ng)
 387                 panic("Could not allocate generic netns");
 388
 389         rcu_assign_pointer(init_net.gen, ng);
 390
 391         mutex_lock(&net_mutex);
 392         if (setup_net(&init_net))
 393                 panic("Could not setup the initial network namespace");
 394
 395         rtnl_lock();
 396         list_add_tail_rcu(&init_net.list, &net_namespace_list);
 397         rtnl_unlock();
 398
 399         mutex_unlock(&net_mutex);
 400
 401         return 0;
 402 }
 403
 404 pure_initcall(net_ns_init);
 405
 406 #ifdef CONFIG_NET_NS
 407 static int __register_pernet_operations(struct list_head *list,
 408                                         struct pernet_operations *ops)
 409 {
 410         struct net *net;
 411         int error;
 412         LIST_HEAD(net_exit_list);
 413
 414         list_add_tail(&ops->list, list);
 415         if (ops->init || (ops->id && ops->size)) {
 416                 for_each_net(net) {
 417                         error = ops_init(ops, net);
 418                         if (error)
 419                                 goto out_undo;
 420                         list_add_tail(&net->exit_list, &net_exit_list);
 421                 }
 422         }
 423         return 0;
 424
 425 out_undo:
 426         /* If I have an error cleanup all namespaces I initialized */
 427         list_del(&ops->list);
 428         ops_exit_list(ops, &net_exit_list);
 429         ops_free_list(ops, &net_exit_list);
 430         return error;
 431 }
 432
 433 static void __unregister_pernet_operations(struct pernet_operations *ops)
 434 {
 435         struct net *net;
 436         LIST_HEAD(net_exit_list);
 437
 438         list_del(&ops->list);
 439         for_each_net(net)
 440                 list_add_tail(&net->exit_list, &net_exit_list);
 441         ops_exit_list(ops, &net_exit_list);
 442         ops_free_list(ops, &net_exit_list);
 443 }
 444
 445 #else
 446
 447 static int __register_pernet_operations(struct list_head *list,
 448                                         struct pernet_operations *ops)
 449 {
 450         int err = 0;
 451         err = ops_init(ops, &init_net);
 452         if (err)
 453                 ops_free(ops, &init_net);
 454         return err;
 455
 456 }
 457
 458 static void __unregister_pernet_operations(struct pernet_operations *ops)
 459 {
 460         LIST_HEAD(net_exit_list);
 461         list_add(&init_net.exit_list, &net_exit_list);
 462         ops_exit_list(ops, &net_exit_list);
 463         ops_free_list(ops, &net_exit_list);
 464 }
 465
 466 #endif /* CONFIG_NET_NS */
 467
 468 static DEFINE_IDA(net_generic_ids);
 469
 470 static int register_pernet_operations(struct list_head *list,
 471                                       struct pernet_operations *ops)
 472 {
 473         int error;
 474
 475         if (ops->id) {
 476 again:
 477                 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
 478                 if (error < 0) {
 479                         if (error == -EAGAIN) {
 480                                 ida_pre_get(&net_generic_ids, GFP_KERNEL);
 481                                 goto again;
 482                         }
 483                         return error;
 484                 }
 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