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
 133 #ifdef NETNS_REFCNT_DEBUG
 134         atomic_set(&net->use_count, 0);
 135 #endif
 136
 137         list_for_each_entry(ops, &pernet_list, list) {
 138                 error = ops_init(ops, net);
 139                 if (error < 0)
 140                         goto out_undo;
 141         }
 142 out:
 143         return error;
 144
 145 out_undo:
 146         /* Walk through the list backwards calling the exit functions
 147          * for the pernet modules whose init functions did not fail.
 148          */
 149         list_add(&net->exit_list, &net_exit_list);
 150         saved_ops = ops;
 151         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 152                 ops_exit_list(ops, &net_exit_list);
 153
 154         ops = saved_ops;
 155         list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 156                 ops_free_list(ops, &net_exit_list);
 157
 158         rcu_barrier();
 159         goto out;
 160 }
 161
 162 static struct net_generic *net_alloc_generic(void)
 163 {
 164         struct net_generic *ng;
 165         size_t generic_size = sizeof(struct net_generic) +
 166                 INITIAL_NET_GEN_PTRS * sizeof(void *);
 167
 168         ng = kzalloc(generic_size, GFP_KERNEL);
 169         if (ng)
 170                 ng->len = INITIAL_NET_GEN_PTRS;
 171
 172         return ng;
 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         }
 485         error = __register_pernet_operations(list, ops);
 486         if (error) {
 487                 rcu_barrier();
 488                 if (ops->id)
 489                         ida_remove(&net_generic_ids, *ops->id);
 490         }
 491
 492         return error;
 493 }
 494
 495 static void unregister_pernet_operations(struct pernet_operations *ops)
 496 {
 497
 498         __unregister_pernet_operations(ops);
 499         rcu_barrier();
 500         if (ops->id)
 501                 ida_remove(&net_generic_ids, *ops->id);
 502 }
 503
 504 /**
 505  *      register_pernet_subsys - register a network namespace subsystem
 506  *      @ops:  pernet operations structure for the subsystem
 507  *
 508  *      Register a subsystem which has init and exit functions
 509  *      that are called when network namespaces are created and
 510  *      destroyed respectively.
 511  *
 512  *      When registered all network namespace init functions are
 513  *      called for every existing network namespace.  Allowing kernel
 514  *      modules to have a race free view of the set of network namespaces.
 515  *
 516  *      When a new network namespace is created all of the init
 517  *      methods are called in the order in which they were registered.
 518  *
 519  *      When a network namespace is destroyed all of the exit methods
 520  *      are called in the reverse of the order with which they were
 521  *      registered.
 522  */
 523 int register_pernet_subsys(struct pernet_operations *ops)
 524 {
 525         int error;
 526         mutex_lock(&net_mutex);
 527         error =  register_pernet_operations(first_device, ops);
 528         mutex_unlock(&net_mutex);
 529         return error;
 530 }
 531 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 532
 533 /**
 534  *      unregister_pernet_subsys - unregister a network namespace subsystem
 535  *      @ops: pernet operations structure to manipulate
 536  *
 537  *      Remove the pernet operations structure from the list to be
 538  *      used when network namespaces are created or destroyed.  In
 539  *      addition run the exit method for all existing network
 540  *      namespaces.
 541  */
 542 void unregister_pernet_subsys(struct pernet_operations *ops)
 543 {
 544         mutex_lock(&net_mutex);
 545         unregister_pernet_operations(ops);
 546         mutex_unlock(&net_mutex);
 547 }
 548 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 549
 550 /**
 551  *      register_pernet_device - register a network namespace device
 552  *      @ops:  pernet operations structure for the subsystem
 553  *
 554  *      Register a device which has init and exit functions
 555  *      that are called when network namespaces are created and
 556  *      destroyed respectively.
 557  *
 558  *      When registered all network namespace init functions are
 559  *      called for every existing network namespace.  Allowing kernel
 560  *      modules to have a race free view of the set of network namespaces.
 561  *
 562  *      When a new network namespace is created all of the init
 563  *      methods are called in the order in which they were registered.
 564  *
 565  *      When a network namespace is destroyed all of the exit methods
 566  *      are called in the reverse of the order with which they were
 567  *      registered.
 568  */
 569 int register_pernet_device(struct pernet_operations *ops)
 570 {
 571         int error;
 572         mutex_lock(&net_mutex);
 573         error = register_pernet_operations(&pernet_list, ops);
 574         if (!error && (first_device == &pernet_list))
 575                 first_device = &ops->list;
 576         mutex_unlock(&net_mutex);
 577         return error;
 578 }
 579 EXPORT_SYMBOL_GPL(register_pernet_device);
 580
 581 /**
 582  *      unregister_pernet_device - unregister a network namespace netdevice
 583  *      @ops: pernet operations structure to manipulate
 584  *
 585  *      Remove the pernet operations structure from the list to be
 586  *      used when network namespaces are created or destroyed.  In
 587  *      addition run the exit method for all existing network
 588  *      namespaces.
 589  */
 590 void unregister_pernet_device(struct pernet_operations *ops)
 591 {
 592         mutex_lock(&net_mutex);
 593         if (&ops->list == first_device)
 594                 first_device = first_device->next;
 595         unregister_pernet_operations(ops);
 596         mutex_unlock(&net_mutex);
 597 }
 598 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 599
 600 #ifdef CONFIG_NET_NS
 601 static void *netns_get(struct task_struct *task)
 602 {
 603         struct net *net = NULL;
 604         struct nsproxy *nsproxy;
 605
 606         rcu_read_lock();
 607         nsproxy = task_nsproxy(task);
 608         if (nsproxy)
 609                 net = get_net(nsproxy->net_ns);
 610         rcu_read_unlock();
 611
 612         return net;
 613 }
 614
 615 static void netns_put(void *ns)
 616 {
 617         put_net(ns);
 618 }
 619
 620 static int netns_install(struct nsproxy *nsproxy, void *ns)
 621 {
 622         put_net(nsproxy->net_ns);
 623         nsproxy->net_ns = get_net(ns);
 624         return 0;
 625 }
 626
 627 const struct proc_ns_operations netns_operations = {
 628         .name           = "net",
 629         .type           = CLONE_NEWNET,
 630         .get            = netns_get,
 631         .put            = netns_put,
 632         .install        = netns_install,
 633 };
 634 #endif