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 <net/net_namespace.h>
  10 #include <net/netns/generic.h>
  11
  12 /*
  13  *      Our network namespace constructor/destructor lists
  14  */
  15
  16 static LIST_HEAD(pernet_list);
  17 static struct list_head *first_device = &pernet_list;
  18 static DEFINE_MUTEX(net_mutex);
  19
  20 LIST_HEAD(net_namespace_list);
  21
  22 struct net init_net;
  23 EXPORT_SYMBOL(init_net);
  24
  25 #define INITIAL_NET_GEN_PTRS    13 /* +1 for len +2 for rcu_head */
  26
  27 /*
  28  * setup_net runs the initializers for the network namespace object.
  29  */
  30 static __net_init int setup_net(struct net *net)
  31 {
  32         /* Must be called with net_mutex held */
  33         struct pernet_operations *ops;
  34         int error;
  35         struct net_generic *ng;
  36
  37         atomic_set(&net->count, 1);
  38 #ifdef NETNS_REFCNT_DEBUG
  39         atomic_set(&net->use_count, 0);
  40 #endif
  41
  42         error = -ENOMEM;
  43         ng = kzalloc(sizeof(struct net_generic) +
  44                         INITIAL_NET_GEN_PTRS * sizeof(void *), GFP_KERNEL);
  45         if (ng == NULL)
  46                 goto out;
  47
  48         ng->len = INITIAL_NET_GEN_PTRS;
  49         INIT_RCU_HEAD(&ng->rcu);
  50         rcu_assign_pointer(net->gen, ng);
  51
  52         error = 0;
  53         list_for_each_entry(ops, &pernet_list, list) {
  54                 if (ops->init) {
  55                         error = ops->init(net);
  56                         if (error < 0)
  57                                 goto out_undo;
  58                 }
  59         }
  60 out:
  61         return error;
  62
  63 out_undo:
  64         /* Walk through the list backwards calling the exit functions
  65          * for the pernet modules whose init functions did not fail.
  66          */
  67         list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
  68                 if (ops->exit)
  69                         ops->exit(net);
  70         }
  71
  72         rcu_barrier();
  73         kfree(ng);
  74         goto out;
  75 }
  76
  77 #ifdef CONFIG_NET_NS
  78 static struct kmem_cache *net_cachep;
  79 static struct workqueue_struct *netns_wq;
  80
  81 static struct net *net_alloc(void)
  82 {
  83         return kmem_cache_zalloc(net_cachep, GFP_KERNEL);
  84 }
  85
  86 static void net_free(struct net *net)
  87 {
  88         if (!net)
  89                 return;
  90
  91 #ifdef NETNS_REFCNT_DEBUG
  92         if (unlikely(atomic_read(&net->use_count) != 0)) {
  93                 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
  94                         atomic_read(&net->use_count));
  95                 return;
  96         }
  97 #endif
  98
  99         kmem_cache_free(net_cachep, net);
 100 }
 101
 102 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 103 {
 104         struct net *new_net = NULL;
 105         int err;
 106
 107         get_net(old_net);
 108
 109         if (!(flags & CLONE_NEWNET))
 110                 return old_net;
 111
 112         err = -ENOMEM;
 113         new_net = net_alloc();
 114         if (!new_net)
 115                 goto out;
 116
 117         mutex_lock(&net_mutex);
 118         err = setup_net(new_net);
 119         if (err)
 120                 goto out_unlock;
 121
 122         rtnl_lock();
 123         list_add_tail(&new_net->list, &net_namespace_list);
 124         rtnl_unlock();
 125
 126
 127 out_unlock:
 128         mutex_unlock(&net_mutex);
 129 out:
 130         put_net(old_net);
 131         if (err) {
 132                 net_free(new_net);
 133                 new_net = ERR_PTR(err);
 134         }
 135         return new_net;
 136 }
 137
 138 static void cleanup_net(struct work_struct *work)
 139 {
 140         struct pernet_operations *ops;
 141         struct net *net;
 142
 143         net = container_of(work, struct net, work);
 144
 145         mutex_lock(&net_mutex);
 146
 147         /* Don't let anyone else find us. */
 148         rtnl_lock();
 149         list_del(&net->list);
 150         rtnl_unlock();
 151
 152         /* Run all of the network namespace exit methods */
 153         list_for_each_entry_reverse(ops, &pernet_list, list) {
 154                 if (ops->exit)
 155                         ops->exit(net);
 156         }
 157
 158         mutex_unlock(&net_mutex);
 159
 160         /* Ensure there are no outstanding rcu callbacks using this
 161          * network namespace.
 162          */
 163         rcu_barrier();
 164
 165         /* Finally it is safe to free my network namespace structure */
 166         net_free(net);
 167 }
 168
 169 void __put_net(struct net *net)
 170 {
 171         /* Cleanup the network namespace in process context */
 172         INIT_WORK(&net->work, cleanup_net);
 173         queue_work(netns_wq, &net->work);
 174 }
 175 EXPORT_SYMBOL_GPL(__put_net);
 176
 177 #else
 178 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
 179 {
 180         if (flags & CLONE_NEWNET)
 181                 return ERR_PTR(-EINVAL);
 182         return old_net;
 183 }
 184 #endif
 185
 186 static int __init net_ns_init(void)
 187 {
 188         int err;
 189
 190         printk(KERN_INFO "net_namespace: %zd bytes\n", sizeof(struct net));
 191 #ifdef CONFIG_NET_NS
 192         net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 193                                         SMP_CACHE_BYTES,
 194                                         SLAB_PANIC, NULL);
 195
 196         /* Create workqueue for cleanup */
 197         netns_wq = create_singlethread_workqueue("netns");
 198         if (!netns_wq)
 199                 panic("Could not create netns workq");
 200 #endif
 201
 202         mutex_lock(&net_mutex);
 203         err = setup_net(&init_net);
 204
 205         rtnl_lock();
 206         list_add_tail(&init_net.list, &net_namespace_list);
 207         rtnl_unlock();
 208
 209         mutex_unlock(&net_mutex);
 210         if (err)
 211                 panic("Could not setup the initial network namespace");
 212
 213         return 0;
 214 }
 215
 216 pure_initcall(net_ns_init);
 217
 218 #ifdef CONFIG_NET_NS
 219 static int register_pernet_operations(struct list_head *list,
 220                                       struct pernet_operations *ops)
 221 {
 222         struct net *net, *undo_net;
 223         int error;
 224
 225         list_add_tail(&ops->list, list);
 226         if (ops->init) {
 227                 for_each_net(net) {
 228                         error = ops->init(net);
 229                         if (error)
 230                                 goto out_undo;
 231                 }
 232         }
 233         return 0;
 234
 235 out_undo:
 236         /* If I have an error cleanup all namespaces I initialized */
 237         list_del(&ops->list);
 238         if (ops->exit) {
 239                 for_each_net(undo_net) {
 240                         if (undo_net == net)
 241                                 goto undone;
 242                         ops->exit(undo_net);
 243                 }
 244         }
 245 undone:
 246         return error;
 247 }
 248
 249 static void unregister_pernet_operations(struct pernet_operations *ops)
 250 {
 251         struct net *net;
 252
 253         list_del(&ops->list);
 254         if (ops->exit)
 255                 for_each_net(net)
 256                         ops->exit(net);
 257 }
 258
 259 #else
 260
 261 static int register_pernet_operations(struct list_head *list,
 262                                       struct pernet_operations *ops)
 263 {
 264         if (ops->init == NULL)
 265                 return 0;
 266         return ops->init(&init_net);
 267 }
 268
 269 static void unregister_pernet_operations(struct pernet_operations *ops)
 270 {
 271         if (ops->exit)
 272                 ops->exit(&init_net);
 273 }
 274 #endif
 275
 276 static DEFINE_IDA(net_generic_ids);
 277
 278 /**
 279  *      register_pernet_subsys - register a network namespace subsystem
 280  *      @ops:  pernet operations structure for the subsystem
 281  *
 282  *      Register a subsystem which has init and exit functions
 283  *      that are called when network namespaces are created and
 284  *      destroyed respectively.
 285  *
 286  *      When registered all network namespace init functions are
 287  *      called for every existing network namespace.  Allowing kernel
 288  *      modules to have a race free view of the set of network namespaces.
 289  *
 290  *      When a new network namespace is created all of the init
 291  *      methods are called in the order in which they were registered.
 292  *
 293  *      When a network namespace is destroyed all of the exit methods
 294  *      are called in the reverse of the order with which they were
 295  *      registered.
 296  */
 297 int register_pernet_subsys(struct pernet_operations *ops)
 298 {
 299         int error;
 300         mutex_lock(&net_mutex);
 301         error =  register_pernet_operations(first_device, ops);
 302         mutex_unlock(&net_mutex);
 303         return error;
 304 }
 305 EXPORT_SYMBOL_GPL(register_pernet_subsys);
 306
 307 /**
 308  *      unregister_pernet_subsys - unregister a network namespace subsystem
 309  *      @ops: pernet operations structure to manipulate
 310  *
 311  *      Remove the pernet operations structure from the list to be
 312  *      used when network namespaces are created or destroyed.  In
 313  *      addition run the exit method for all existing network
 314  *      namespaces.
 315  */
 316 void unregister_pernet_subsys(struct pernet_operations *module)
 317 {
 318         mutex_lock(&net_mutex);
 319         unregister_pernet_operations(module);
 320         mutex_unlock(&net_mutex);
 321 }
 322 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
 323
 324 /**
 325  *      register_pernet_device - register a network namespace device
 326  *      @ops:  pernet operations structure for the subsystem
 327  *
 328  *      Register a device which has init and exit functions
 329  *      that are called when network namespaces are created and
 330  *      destroyed respectively.
 331  *
 332  *      When registered all network namespace init functions are
 333  *      called for every existing network namespace.  Allowing kernel
 334  *      modules to have a race free view of the set of network namespaces.
 335  *
 336  *      When a new network namespace is created all of the init
 337  *      methods are called in the order in which they were registered.
 338  *
 339  *      When a network namespace is destroyed all of the exit methods
 340  *      are called in the reverse of the order with which they were
 341  *      registered.
 342  */
 343 int register_pernet_device(struct pernet_operations *ops)
 344 {
 345         int error;
 346         mutex_lock(&net_mutex);
 347         error = register_pernet_operations(&pernet_list, ops);
 348         if (!error && (first_device == &pernet_list))
 349                 first_device = &ops->list;
 350         mutex_unlock(&net_mutex);
 351         return error;
 352 }
 353 EXPORT_SYMBOL_GPL(register_pernet_device);
 354
 355 int register_pernet_gen_device(int *id, struct pernet_operations *ops)
 356 {
 357         int error;
 358         mutex_lock(&net_mutex);
 359 again:
 360         error = ida_get_new_above(&net_generic_ids, 1, id);
 361         if (error) {
 362                 if (error == -EAGAIN) {
 363                         ida_pre_get(&net_generic_ids, GFP_KERNEL);
 364                         goto again;
 365                 }
 366                 goto out;
 367         }
 368         error = register_pernet_operations(&pernet_list, ops);
 369         if (error)
 370                 ida_remove(&net_generic_ids, *id);
 371         else if (first_device == &pernet_list)
 372                 first_device = &ops->list;
 373 out:
 374         mutex_unlock(&net_mutex);
 375         return error;
 376 }
 377 EXPORT_SYMBOL_GPL(register_pernet_gen_device);
 378
 379 /**
 380  *      unregister_pernet_device - unregister a network namespace netdevice
 381  *      @ops: pernet operations structure to manipulate
 382  *
 383  *      Remove the pernet operations structure from the list to be
 384  *      used when network namespaces are created or destroyed.  In
 385  *      addition run the exit method for all existing network
 386  *      namespaces.
 387  */
 388 void unregister_pernet_device(struct pernet_operations *ops)
 389 {
 390         mutex_lock(&net_mutex);
 391         if (&ops->list == first_device)
 392                 first_device = first_device->next;
 393         unregister_pernet_operations(ops);
 394         mutex_unlock(&net_mutex);
 395 }
 396 EXPORT_SYMBOL_GPL(unregister_pernet_device);
 397
 398 void unregister_pernet_gen_device(int id, struct pernet_operations *ops)
 399 {
 400         mutex_lock(&net_mutex);
 401         if (&ops->list == first_device)
 402                 first_device = first_device->next;
 403         unregister_pernet_operations(ops);
 404         ida_remove(&net_generic_ids, id);
 405         mutex_unlock(&net_mutex);
 406 }
 407 EXPORT_SYMBOL_GPL(unregister_pernet_gen_device);
 408
 409 static void net_generic_release(struct rcu_head *rcu)
 410 {
 411         struct net_generic *ng;
 412
 413         ng = container_of(rcu, struct net_generic, rcu);
 414         kfree(ng);
 415 }
 416
 417 int net_assign_generic(struct net *net, int id, void *data)
 418 {
 419         struct net_generic *ng, *old_ng;
 420
 421         BUG_ON(!mutex_is_locked(&net_mutex));
 422         BUG_ON(id == 0);
 423
 424         ng = old_ng = net->gen;
 425         if (old_ng->len >= id)
 426                 goto assign;
 427
 428         ng = kzalloc(sizeof(struct net_generic) +
 429                         id * sizeof(void *), GFP_KERNEL);
 430         if (ng == NULL)
 431                 return -ENOMEM;
 432
 433         /*
 434          * Some synchronisation notes:
 435          *
 436          * The net_generic explores the net->gen array inside rcu
 437          * read section. Besides once set the net->gen->ptr[x]
 438          * pointer never changes (see rules in netns/generic.h).
 439          *
 440          * That said, we simply duplicate this array and schedule
 441          * the old copy for kfree after a grace period.
 442          */
 443
 444         ng->len = id;
 445         INIT_RCU_HEAD(&ng->rcu);
 446         memcpy(&ng->ptr, &old_ng->ptr, old_ng->len);
 447
 448         rcu_assign_pointer(net->gen, ng);
 449         call_rcu(&old_ng->rcu, net_generic_release);
 450 assign:
 451         ng->ptr[id - 1] = data;
 452         return 0;
 453 }
 454 EXPORT_SYMBOL_GPL(net_assign_generic);