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