net_namespace.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894
  1. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  2. #include <linux/workqueue.h>
  3. #include <linux/rtnetlink.h>
  4. #include <linux/cache.h>
  5. #include <linux/slab.h>
  6. #include <linux/list.h>
  7. #include <linux/delay.h>
  8. #include <linux/sched.h>
  9. #include <linux/idr.h>
  10. #include <linux/rculist.h>
  11. #include <linux/nsproxy.h>
  12. #include <linux/fs.h>
  13. #include <linux/proc_ns.h>
  14. #include <linux/file.h>
  15. #include <linux/export.h>
  16. #include <linux/user_namespace.h>
  17. #include <linux/net_namespace.h>
  18. #include <linux/rtnetlink.h>
  19. #include <net/sock.h>
  20. #include <net/netlink.h>
  21. #include <net/net_namespace.h>
  22. #include <net/netns/generic.h>
  23. /*
  24. * Our network namespace constructor/destructor lists
  25. */
  26. static LIST_HEAD(pernet_list);
  27. static struct list_head *first_device = &pernet_list;
  28. DEFINE_MUTEX(net_mutex);
  29. LIST_HEAD(net_namespace_list);
  30. EXPORT_SYMBOL_GPL(net_namespace_list);
  31. struct net init_net = {
  32. .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
  33. };
  34. EXPORT_SYMBOL(init_net);
  35. #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
  36. static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  37. static struct net_generic *net_alloc_generic(void)
  38. {
  39. struct net_generic *ng;
  40. size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  41. ng = kzalloc(generic_size, GFP_KERNEL);
  42. if (ng)
  43. ng->len = max_gen_ptrs;
  44. return ng;
  45. }
  46. static int net_assign_generic(struct net *net, int id, void *data)
  47. {
  48. struct net_generic *ng, *old_ng;
  49. BUG_ON(!mutex_is_locked(&net_mutex));
  50. BUG_ON(id == 0);
  51. old_ng = rcu_dereference_protected(net->gen,
  52. lockdep_is_held(&net_mutex));
  53. ng = old_ng;
  54. if (old_ng->len >= id)
  55. goto assign;
  56. ng = net_alloc_generic();
  57. if (ng == NULL)
  58. return -ENOMEM;
  59. /*
  60. * Some synchronisation notes:
  61. *
  62. * The net_generic explores the net->gen array inside rcu
  63. * read section. Besides once set the net->gen->ptr[x]
  64. * pointer never changes (see rules in netns/generic.h).
  65. *
  66. * That said, we simply duplicate this array and schedule
  67. * the old copy for kfree after a grace period.
  68. */
  69. memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
  70. rcu_assign_pointer(net->gen, ng);
  71. kfree_rcu(old_ng, rcu);
  72. assign:
  73. ng->ptr[id - 1] = data;
  74. return 0;
  75. }
  76. static int ops_init(const struct pernet_operations *ops, struct net *net)
  77. {
  78. int err = -ENOMEM;
  79. void *data = NULL;
  80. if (ops->id && ops->size) {
  81. data = kzalloc(ops->size, GFP_KERNEL);
  82. if (!data)
  83. goto out;
  84. err = net_assign_generic(net, *ops->id, data);
  85. if (err)
  86. goto cleanup;
  87. }
  88. err = 0;
  89. if (ops->init)
  90. err = ops->init(net);
  91. if (!err)
  92. return 0;
  93. cleanup:
  94. kfree(data);
  95. out:
  96. return err;
  97. }
  98. static void ops_free(const struct pernet_operations *ops, struct net *net)
  99. {
  100. if (ops->id && ops->size) {
  101. int id = *ops->id;
  102. kfree(net_generic(net, id));
  103. }
  104. }
  105. static void ops_exit_list(const struct pernet_operations *ops,
  106. struct list_head *net_exit_list)
  107. {
  108. struct net *net;
  109. if (ops->exit) {
  110. list_for_each_entry(net, net_exit_list, exit_list)
  111. ops->exit(net);
  112. }
  113. if (ops->exit_batch)
  114. ops->exit_batch(net_exit_list);
  115. }
  116. static void ops_free_list(const struct pernet_operations *ops,
  117. struct list_head *net_exit_list)
  118. {
  119. struct net *net;
  120. if (ops->size && ops->id) {
  121. list_for_each_entry(net, net_exit_list, exit_list)
  122. ops_free(ops, net);
  123. }
  124. }
  125. static int alloc_netid(struct net *net, struct net *peer, int reqid)
  126. {
  127. int min = 0, max = 0;
  128. ASSERT_RTNL();
  129. if (reqid >= 0) {
  130. min = reqid;
  131. max = reqid + 1;
  132. }
  133. return idr_alloc(&net->netns_ids, peer, min, max, GFP_KERNEL);
  134. }
  135. /* This function is used by idr_for_each(). If net is equal to peer, the
  136. * function returns the id so that idr_for_each() stops. Because we cannot
  137. * returns the id 0 (idr_for_each() will not stop), we return the magic value
  138. * NET_ID_ZERO (-1) for it.
  139. */
  140. #define NET_ID_ZERO -1
  141. static int net_eq_idr(int id, void *net, void *peer)
  142. {
  143. if (net_eq(net, peer))
  144. return id ? : NET_ID_ZERO;
  145. return 0;
  146. }
  147. static int __peernet2id(struct net *net, struct net *peer, bool alloc)
  148. {
  149. int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
  150. ASSERT_RTNL();
  151. /* Magic value for id 0. */
  152. if (id == NET_ID_ZERO)
  153. return 0;
  154. if (id > 0)
  155. return id;
  156. if (alloc)
  157. return alloc_netid(net, peer, -1);
  158. return -ENOENT;
  159. }
  160. /* This function returns the id of a peer netns. If no id is assigned, one will
  161. * be allocated and returned.
  162. */
  163. int peernet2id(struct net *net, struct net *peer)
  164. {
  165. bool alloc = atomic_read(&peer->count) == 0 ? false : true;
  166. int id;
  167. id = __peernet2id(net, peer, alloc);
  168. return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
  169. }
  170. EXPORT_SYMBOL(peernet2id);
  171. struct net *get_net_ns_by_id(struct net *net, int id)
  172. {
  173. struct net *peer;
  174. if (id < 0)
  175. return NULL;
  176. rcu_read_lock();
  177. peer = idr_find(&net->netns_ids, id);
  178. if (peer)
  179. get_net(peer);
  180. rcu_read_unlock();
  181. return peer;
  182. }
  183. /*
  184. * setup_net runs the initializers for the network namespace object.
  185. */
  186. static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
  187. {
  188. /* Must be called with net_mutex held */
  189. const struct pernet_operations *ops, *saved_ops;
  190. int error = 0;
  191. LIST_HEAD(net_exit_list);
  192. atomic_set(&net->count, 1);
  193. atomic_set(&net->passive, 1);
  194. net->dev_base_seq = 1;
  195. net->user_ns = user_ns;
  196. idr_init(&net->netns_ids);
  197. #ifdef NETNS_REFCNT_DEBUG
  198. atomic_set(&net->use_count, 0);
  199. #endif
  200. list_for_each_entry(ops, &pernet_list, list) {
  201. error = ops_init(ops, net);
  202. if (error < 0)
  203. goto out_undo;
  204. }
  205. out:
  206. return error;
  207. out_undo:
  208. /* Walk through the list backwards calling the exit functions
  209. * for the pernet modules whose init functions did not fail.
  210. */
  211. list_add(&net->exit_list, &net_exit_list);
  212. saved_ops = ops;
  213. list_for_each_entry_continue_reverse(ops, &pernet_list, list)
  214. ops_exit_list(ops, &net_exit_list);
  215. ops = saved_ops;
  216. list_for_each_entry_continue_reverse(ops, &pernet_list, list)
  217. ops_free_list(ops, &net_exit_list);
  218. rcu_barrier();
  219. goto out;
  220. }
  221. #ifdef CONFIG_NET_NS
  222. static struct kmem_cache *net_cachep;
  223. static struct workqueue_struct *netns_wq;
  224. static struct net *net_alloc(void)
  225. {
  226. struct net *net = NULL;
  227. struct net_generic *ng;
  228. ng = net_alloc_generic();
  229. if (!ng)
  230. goto out;
  231. net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
  232. if (!net)
  233. goto out_free;
  234. rcu_assign_pointer(net->gen, ng);
  235. out:
  236. return net;
  237. out_free:
  238. kfree(ng);
  239. goto out;
  240. }
  241. static void net_free(struct net *net)
  242. {
  243. #ifdef NETNS_REFCNT_DEBUG
  244. if (unlikely(atomic_read(&net->use_count) != 0)) {
  245. pr_emerg("network namespace not free! Usage: %d\n",
  246. atomic_read(&net->use_count));
  247. return;
  248. }
  249. #endif
  250. kfree(rcu_access_pointer(net->gen));
  251. kmem_cache_free(net_cachep, net);
  252. }
  253. void net_drop_ns(void *p)
  254. {
  255. struct net *ns = p;
  256. if (ns && atomic_dec_and_test(&ns->passive))
  257. net_free(ns);
  258. }
  259. struct net *copy_net_ns(unsigned long flags,
  260. struct user_namespace *user_ns, struct net *old_net)
  261. {
  262. struct net *net;
  263. int rv;
  264. if (!(flags & CLONE_NEWNET))
  265. return get_net(old_net);
  266. net = net_alloc();
  267. if (!net)
  268. return ERR_PTR(-ENOMEM);
  269. get_user_ns(user_ns);
  270. mutex_lock(&net_mutex);
  271. rv = setup_net(net, user_ns);
  272. if (rv == 0) {
  273. rtnl_lock();
  274. list_add_tail_rcu(&net->list, &net_namespace_list);
  275. rtnl_unlock();
  276. }
  277. mutex_unlock(&net_mutex);
  278. if (rv < 0) {
  279. put_user_ns(user_ns);
  280. net_drop_ns(net);
  281. return ERR_PTR(rv);
  282. }
  283. return net;
  284. }
  285. static DEFINE_SPINLOCK(cleanup_list_lock);
  286. static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
  287. static void cleanup_net(struct work_struct *work)
  288. {
  289. const struct pernet_operations *ops;
  290. struct net *net, *tmp;
  291. struct list_head net_kill_list;
  292. LIST_HEAD(net_exit_list);
  293. /* Atomically snapshot the list of namespaces to cleanup */
  294. spin_lock_irq(&cleanup_list_lock);
  295. list_replace_init(&cleanup_list, &net_kill_list);
  296. spin_unlock_irq(&cleanup_list_lock);
  297. mutex_lock(&net_mutex);
  298. /* Don't let anyone else find us. */
  299. rtnl_lock();
  300. list_for_each_entry(net, &net_kill_list, cleanup_list) {
  301. list_del_rcu(&net->list);
  302. list_add_tail(&net->exit_list, &net_exit_list);
  303. for_each_net(tmp) {
  304. int id = __peernet2id(tmp, net, false);
  305. if (id >= 0)
  306. idr_remove(&tmp->netns_ids, id);
  307. }
  308. idr_destroy(&net->netns_ids);
  309. }
  310. rtnl_unlock();
  311. /*
  312. * Another CPU might be rcu-iterating the list, wait for it.
  313. * This needs to be before calling the exit() notifiers, so
  314. * the rcu_barrier() below isn't sufficient alone.
  315. */
  316. synchronize_rcu();
  317. /* Run all of the network namespace exit methods */
  318. list_for_each_entry_reverse(ops, &pernet_list, list)
  319. ops_exit_list(ops, &net_exit_list);
  320. /* Free the net generic variables */
  321. list_for_each_entry_reverse(ops, &pernet_list, list)
  322. ops_free_list(ops, &net_exit_list);
  323. mutex_unlock(&net_mutex);
  324. /* Ensure there are no outstanding rcu callbacks using this
  325. * network namespace.
  326. */
  327. rcu_barrier();
  328. /* Finally it is safe to free my network namespace structure */
  329. list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
  330. list_del_init(&net->exit_list);
  331. put_user_ns(net->user_ns);
  332. net_drop_ns(net);
  333. }
  334. }
  335. static DECLARE_WORK(net_cleanup_work, cleanup_net);
  336. void __put_net(struct net *net)
  337. {
  338. /* Cleanup the network namespace in process context */
  339. unsigned long flags;
  340. spin_lock_irqsave(&cleanup_list_lock, flags);
  341. list_add(&net->cleanup_list, &cleanup_list);
  342. spin_unlock_irqrestore(&cleanup_list_lock, flags);
  343. queue_work(netns_wq, &net_cleanup_work);
  344. }
  345. EXPORT_SYMBOL_GPL(__put_net);
  346. struct net *get_net_ns_by_fd(int fd)
  347. {
  348. struct file *file;
  349. struct ns_common *ns;
  350. struct net *net;
  351. file = proc_ns_fget(fd);
  352. if (IS_ERR(file))
  353. return ERR_CAST(file);
  354. ns = get_proc_ns(file_inode(file));
  355. if (ns->ops == &netns_operations)
  356. net = get_net(container_of(ns, struct net, ns));
  357. else
  358. net = ERR_PTR(-EINVAL);
  359. fput(file);
  360. return net;
  361. }
  362. #else
  363. struct net *get_net_ns_by_fd(int fd)
  364. {
  365. return ERR_PTR(-EINVAL);
  366. }
  367. #endif
  368. EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
  369. struct net *get_net_ns_by_pid(pid_t pid)
  370. {
  371. struct task_struct *tsk;
  372. struct net *net;
  373. /* Lookup the network namespace */
  374. net = ERR_PTR(-ESRCH);
  375. rcu_read_lock();
  376. tsk = find_task_by_vpid(pid);
  377. if (tsk) {
  378. struct nsproxy *nsproxy;
  379. task_lock(tsk);
  380. nsproxy = tsk->nsproxy;
  381. if (nsproxy)
  382. net = get_net(nsproxy->net_ns);
  383. task_unlock(tsk);
  384. }
  385. rcu_read_unlock();
  386. return net;
  387. }
  388. EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
  389. static __net_init int net_ns_net_init(struct net *net)
  390. {
  391. #ifdef CONFIG_NET_NS
  392. net->ns.ops = &netns_operations;
  393. #endif
  394. return ns_alloc_inum(&net->ns);
  395. }
  396. static __net_exit void net_ns_net_exit(struct net *net)
  397. {
  398. ns_free_inum(&net->ns);
  399. }
  400. static struct pernet_operations __net_initdata net_ns_ops = {
  401. .init = net_ns_net_init,
  402. .exit = net_ns_net_exit,
  403. };
  404. static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
  405. [NETNSA_NONE] = { .type = NLA_UNSPEC },
  406. [NETNSA_NSID] = { .type = NLA_S32 },
  407. [NETNSA_PID] = { .type = NLA_U32 },
  408. [NETNSA_FD] = { .type = NLA_U32 },
  409. };
  410. static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
  411. {
  412. struct net *net = sock_net(skb->sk);
  413. struct nlattr *tb[NETNSA_MAX + 1];
  414. struct net *peer;
  415. int nsid, err;
  416. err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
  417. rtnl_net_policy);
  418. if (err < 0)
  419. return err;
  420. if (!tb[NETNSA_NSID])
  421. return -EINVAL;
  422. nsid = nla_get_s32(tb[NETNSA_NSID]);
  423. if (tb[NETNSA_PID])
  424. peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
  425. else if (tb[NETNSA_FD])
  426. peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
  427. else
  428. return -EINVAL;
  429. if (IS_ERR(peer))
  430. return PTR_ERR(peer);
  431. if (__peernet2id(net, peer, false) >= 0) {
  432. err = -EEXIST;
  433. goto out;
  434. }
  435. err = alloc_netid(net, peer, nsid);
  436. if (err > 0)
  437. err = 0;
  438. out:
  439. put_net(peer);
  440. return err;
  441. }
  442. static int rtnl_net_get_size(void)
  443. {
  444. return NLMSG_ALIGN(sizeof(struct rtgenmsg))
  445. + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
  446. ;
  447. }
  448. static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
  449. int cmd, struct net *net, struct net *peer)
  450. {
  451. struct nlmsghdr *nlh;
  452. struct rtgenmsg *rth;
  453. int id;
  454. ASSERT_RTNL();
  455. nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
  456. if (!nlh)
  457. return -EMSGSIZE;
  458. rth = nlmsg_data(nlh);
  459. rth->rtgen_family = AF_UNSPEC;
  460. id = __peernet2id(net, peer, false);
  461. if (id < 0)
  462. id = NETNSA_NSID_NOT_ASSIGNED;
  463. if (nla_put_s32(skb, NETNSA_NSID, id))
  464. goto nla_put_failure;
  465. nlmsg_end(skb, nlh);
  466. return 0;
  467. nla_put_failure:
  468. nlmsg_cancel(skb, nlh);
  469. return -EMSGSIZE;
  470. }
  471. static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
  472. {
  473. struct net *net = sock_net(skb->sk);
  474. struct nlattr *tb[NETNSA_MAX + 1];
  475. struct sk_buff *msg;
  476. int err = -ENOBUFS;
  477. struct net *peer;
  478. err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
  479. rtnl_net_policy);
  480. if (err < 0)
  481. return err;
  482. if (tb[NETNSA_PID])
  483. peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
  484. else if (tb[NETNSA_FD])
  485. peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
  486. else
  487. return -EINVAL;
  488. if (IS_ERR(peer))
  489. return PTR_ERR(peer);
  490. msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
  491. if (!msg) {
  492. err = -ENOMEM;
  493. goto out;
  494. }
  495. err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
  496. RTM_GETNSID, net, peer);
  497. if (err < 0)
  498. goto err_out;
  499. err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
  500. goto out;
  501. err_out:
  502. nlmsg_free(msg);
  503. out:
  504. put_net(peer);
  505. return err;
  506. }
  507. static int __init net_ns_init(void)
  508. {
  509. struct net_generic *ng;
  510. #ifdef CONFIG_NET_NS
  511. net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
  512. SMP_CACHE_BYTES,
  513. SLAB_PANIC, NULL);
  514. /* Create workqueue for cleanup */
  515. netns_wq = create_singlethread_workqueue("netns");
  516. if (!netns_wq)
  517. panic("Could not create netns workq");
  518. #endif
  519. ng = net_alloc_generic();
  520. if (!ng)
  521. panic("Could not allocate generic netns");
  522. rcu_assign_pointer(init_net.gen, ng);
  523. mutex_lock(&net_mutex);
  524. if (setup_net(&init_net, &init_user_ns))
  525. panic("Could not setup the initial network namespace");
  526. rtnl_lock();
  527. list_add_tail_rcu(&init_net.list, &net_namespace_list);
  528. rtnl_unlock();
  529. mutex_unlock(&net_mutex);
  530. register_pernet_subsys(&net_ns_ops);
  531. rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
  532. rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, NULL, NULL);
  533. return 0;
  534. }
  535. pure_initcall(net_ns_init);
  536. #ifdef CONFIG_NET_NS
  537. static int __register_pernet_operations(struct list_head *list,
  538. struct pernet_operations *ops)
  539. {
  540. struct net *net;
  541. int error;
  542. LIST_HEAD(net_exit_list);
  543. list_add_tail(&ops->list, list);
  544. if (ops->init || (ops->id && ops->size)) {
  545. for_each_net(net) {
  546. error = ops_init(ops, net);
  547. if (error)
  548. goto out_undo;
  549. list_add_tail(&net->exit_list, &net_exit_list);
  550. }
  551. }
  552. return 0;
  553. out_undo:
  554. /* If I have an error cleanup all namespaces I initialized */
  555. list_del(&ops->list);
  556. ops_exit_list(ops, &net_exit_list);
  557. ops_free_list(ops, &net_exit_list);
  558. return error;
  559. }
  560. static void __unregister_pernet_operations(struct pernet_operations *ops)
  561. {
  562. struct net *net;
  563. LIST_HEAD(net_exit_list);
  564. list_del(&ops->list);
  565. for_each_net(net)
  566. list_add_tail(&net->exit_list, &net_exit_list);
  567. ops_exit_list(ops, &net_exit_list);
  568. ops_free_list(ops, &net_exit_list);
  569. }
  570. #else
  571. static int __register_pernet_operations(struct list_head *list,
  572. struct pernet_operations *ops)
  573. {
  574. return ops_init(ops, &init_net);
  575. }
  576. static void __unregister_pernet_operations(struct pernet_operations *ops)
  577. {
  578. LIST_HEAD(net_exit_list);
  579. list_add(&init_net.exit_list, &net_exit_list);
  580. ops_exit_list(ops, &net_exit_list);
  581. ops_free_list(ops, &net_exit_list);
  582. }
  583. #endif /* CONFIG_NET_NS */
  584. static DEFINE_IDA(net_generic_ids);
  585. static int register_pernet_operations(struct list_head *list,
  586. struct pernet_operations *ops)
  587. {
  588. int error;
  589. if (ops->id) {
  590. again:
  591. error = ida_get_new_above(&net_generic_ids, 1, ops->id);
  592. if (error < 0) {
  593. if (error == -EAGAIN) {
  594. ida_pre_get(&net_generic_ids, GFP_KERNEL);
  595. goto again;
  596. }
  597. return error;
  598. }
  599. max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
  600. }
  601. error = __register_pernet_operations(list, ops);
  602. if (error) {
  603. rcu_barrier();
  604. if (ops->id)
  605. ida_remove(&net_generic_ids, *ops->id);
  606. }
  607. return error;
  608. }
  609. static void unregister_pernet_operations(struct pernet_operations *ops)
  610. {
  611. __unregister_pernet_operations(ops);
  612. rcu_barrier();
  613. if (ops->id)
  614. ida_remove(&net_generic_ids, *ops->id);
  615. }
  616. /**
  617. * register_pernet_subsys - register a network namespace subsystem
  618. * @ops: pernet operations structure for the subsystem
  619. *
  620. * Register a subsystem which has init and exit functions
  621. * that are called when network namespaces are created and
  622. * destroyed respectively.
  623. *
  624. * When registered all network namespace init functions are
  625. * called for every existing network namespace. Allowing kernel
  626. * modules to have a race free view of the set of network namespaces.
  627. *
  628. * When a new network namespace is created all of the init
  629. * methods are called in the order in which they were registered.
  630. *
  631. * When a network namespace is destroyed all of the exit methods
  632. * are called in the reverse of the order with which they were
  633. * registered.
  634. */
  635. int register_pernet_subsys(struct pernet_operations *ops)
  636. {
  637. int error;
  638. mutex_lock(&net_mutex);
  639. error = register_pernet_operations(first_device, ops);
  640. mutex_unlock(&net_mutex);
  641. return error;
  642. }
  643. EXPORT_SYMBOL_GPL(register_pernet_subsys);
  644. /**
  645. * unregister_pernet_subsys - unregister a network namespace subsystem
  646. * @ops: pernet operations structure to manipulate
  647. *
  648. * Remove the pernet operations structure from the list to be
  649. * used when network namespaces are created or destroyed. In
  650. * addition run the exit method for all existing network
  651. * namespaces.
  652. */
  653. void unregister_pernet_subsys(struct pernet_operations *ops)
  654. {
  655. mutex_lock(&net_mutex);
  656. unregister_pernet_operations(ops);
  657. mutex_unlock(&net_mutex);
  658. }
  659. EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
  660. /**
  661. * register_pernet_device - register a network namespace device
  662. * @ops: pernet operations structure for the subsystem
  663. *
  664. * Register a device which has init and exit functions
  665. * that are called when network namespaces are created and
  666. * destroyed respectively.
  667. *
  668. * When registered all network namespace init functions are
  669. * called for every existing network namespace. Allowing kernel
  670. * modules to have a race free view of the set of network namespaces.
  671. *
  672. * When a new network namespace is created all of the init
  673. * methods are called in the order in which they were registered.
  674. *
  675. * When a network namespace is destroyed all of the exit methods
  676. * are called in the reverse of the order with which they were
  677. * registered.
  678. */
  679. int register_pernet_device(struct pernet_operations *ops)
  680. {
  681. int error;
  682. mutex_lock(&net_mutex);
  683. error = register_pernet_operations(&pernet_list, ops);
  684. if (!error && (first_device == &pernet_list))
  685. first_device = &ops->list;
  686. mutex_unlock(&net_mutex);
  687. return error;
  688. }
  689. EXPORT_SYMBOL_GPL(register_pernet_device);
  690. /**
  691. * unregister_pernet_device - unregister a network namespace netdevice
  692. * @ops: pernet operations structure to manipulate
  693. *
  694. * Remove the pernet operations structure from the list to be
  695. * used when network namespaces are created or destroyed. In
  696. * addition run the exit method for all existing network
  697. * namespaces.
  698. */
  699. void unregister_pernet_device(struct pernet_operations *ops)
  700. {
  701. mutex_lock(&net_mutex);
  702. if (&ops->list == first_device)
  703. first_device = first_device->next;
  704. unregister_pernet_operations(ops);
  705. mutex_unlock(&net_mutex);
  706. }
  707. EXPORT_SYMBOL_GPL(unregister_pernet_device);
  708. #ifdef CONFIG_NET_NS
  709. static struct ns_common *netns_get(struct task_struct *task)
  710. {
  711. struct net *net = NULL;
  712. struct nsproxy *nsproxy;
  713. task_lock(task);
  714. nsproxy = task->nsproxy;
  715. if (nsproxy)
  716. net = get_net(nsproxy->net_ns);
  717. task_unlock(task);
  718. return net ? &net->ns : NULL;
  719. }
  720. static inline struct net *to_net_ns(struct ns_common *ns)
  721. {
  722. return container_of(ns, struct net, ns);
  723. }
  724. static void netns_put(struct ns_common *ns)
  725. {
  726. put_net(to_net_ns(ns));
  727. }
  728. static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
  729. {
  730. struct net *net = to_net_ns(ns);
  731. if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
  732. !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
  733. return -EPERM;
  734. put_net(nsproxy->net_ns);
  735. nsproxy->net_ns = get_net(net);
  736. return 0;
  737. }
  738. const struct proc_ns_operations netns_operations = {
  739. .name = "net",
  740. .type = CLONE_NEWNET,
  741. .get = netns_get,
  742. .put = netns_put,
  743. .install = netns_install,
  744. };
  745. #endif