linux/net/core/net_namespace.c
<<
>>
Prefs
   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/fs.h>
  14#include <linux/proc_ns.h>
  15#include <linux/file.h>
  16#include <linux/export.h>
  17#include <linux/user_namespace.h>
  18#include <linux/net_namespace.h>
  19#include <linux/sched/task.h>
  20
  21#include <net/sock.h>
  22#include <net/netlink.h>
  23#include <net/net_namespace.h>
  24#include <net/netns/generic.h>
  25
  26/*
  27 *      Our network namespace constructor/destructor lists
  28 */
  29
  30static LIST_HEAD(pernet_list);
  31static struct list_head *first_device = &pernet_list;
  32DEFINE_MUTEX(net_mutex);
  33
  34LIST_HEAD(net_namespace_list);
  35EXPORT_SYMBOL_GPL(net_namespace_list);
  36
  37struct net init_net = {
  38        .count          = ATOMIC_INIT(1),
  39        .dev_base_head  = LIST_HEAD_INIT(init_net.dev_base_head),
  40};
  41EXPORT_SYMBOL(init_net);
  42
  43static bool init_net_initialized;
  44
  45#define MIN_PERNET_OPS_ID       \
  46        ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
  47
  48#define INITIAL_NET_GEN_PTRS    13 /* +1 for len +2 for rcu_head */
  49
  50static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
  51
  52static struct net_generic *net_alloc_generic(void)
  53{
  54        struct net_generic *ng;
  55        unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
  56
  57        ng = kzalloc(generic_size, GFP_KERNEL);
  58        if (ng)
  59                ng->s.len = max_gen_ptrs;
  60
  61        return ng;
  62}
  63
  64static int net_assign_generic(struct net *net, unsigned int id, void *data)
  65{
  66        struct net_generic *ng, *old_ng;
  67
  68        BUG_ON(!mutex_is_locked(&net_mutex));
  69        BUG_ON(id < MIN_PERNET_OPS_ID);
  70
  71        old_ng = rcu_dereference_protected(net->gen,
  72                                           lockdep_is_held(&net_mutex));
  73        if (old_ng->s.len > id) {
  74                old_ng->ptr[id] = data;
  75                return 0;
  76        }
  77
  78        ng = net_alloc_generic();
  79        if (ng == NULL)
  80                return -ENOMEM;
  81
  82        /*
  83         * Some synchronisation notes:
  84         *
  85         * The net_generic explores the net->gen array inside rcu
  86         * read section. Besides once set the net->gen->ptr[x]
  87         * pointer never changes (see rules in netns/generic.h).
  88         *
  89         * That said, we simply duplicate this array and schedule
  90         * the old copy for kfree after a grace period.
  91         */
  92
  93        memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
  94               (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
  95        ng->ptr[id] = data;
  96
  97        rcu_assign_pointer(net->gen, ng);
  98        kfree_rcu(old_ng, s.rcu);
  99        return 0;
 100}
 101
 102static int ops_init(const struct pernet_operations *ops, struct net *net)
 103{
 104        int err = -ENOMEM;
 105        void *data = NULL;
 106
 107        if (ops->id && ops->size) {
 108                data = kzalloc(ops->size, GFP_KERNEL);
 109                if (!data)
 110                        goto out;
 111
 112                err = net_assign_generic(net, *ops->id, data);
 113                if (err)
 114                        goto cleanup;
 115        }
 116        err = 0;
 117        if (ops->init)
 118                err = ops->init(net);
 119        if (!err)
 120                return 0;
 121
 122cleanup:
 123        kfree(data);
 124
 125out:
 126        return err;
 127}
 128
 129static void ops_free(const struct pernet_operations *ops, struct net *net)
 130{
 131        if (ops->id && ops->size) {
 132                kfree(net_generic(net, *ops->id));
 133        }
 134}
 135
 136static void ops_exit_list(const struct pernet_operations *ops,
 137                          struct list_head *net_exit_list)
 138{
 139        struct net *net;
 140        if (ops->exit) {
 141                list_for_each_entry(net, net_exit_list, exit_list)
 142                        ops->exit(net);
 143        }
 144        if (ops->exit_batch)
 145                ops->exit_batch(net_exit_list);
 146}
 147
 148static void ops_free_list(const struct pernet_operations *ops,
 149                          struct list_head *net_exit_list)
 150{
 151        struct net *net;
 152        if (ops->size && ops->id) {
 153                list_for_each_entry(net, net_exit_list, exit_list)
 154                        ops_free(ops, net);
 155        }
 156}
 157
 158/* should be called with nsid_lock held */
 159static int alloc_netid(struct net *net, struct net *peer, int reqid)
 160{
 161        int min = 0, max = 0;
 162
 163        if (reqid >= 0) {
 164                min = reqid;
 165                max = reqid + 1;
 166        }
 167
 168        return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
 169}
 170
 171/* This function is used by idr_for_each(). If net is equal to peer, the
 172 * function returns the id so that idr_for_each() stops. Because we cannot
 173 * returns the id 0 (idr_for_each() will not stop), we return the magic value
 174 * NET_ID_ZERO (-1) for it.
 175 */
 176#define NET_ID_ZERO -1
 177static int net_eq_idr(int id, void *net, void *peer)
 178{
 179        if (net_eq(net, peer))
 180                return id ? : NET_ID_ZERO;
 181        return 0;
 182}
 183
 184/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
 185 * is set to true, thus the caller knows that the new id must be notified via
 186 * rtnl.
 187 */
 188static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
 189{
 190        int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
 191        bool alloc_it = *alloc;
 192
 193        *alloc = false;
 194
 195        /* Magic value for id 0. */
 196        if (id == NET_ID_ZERO)
 197                return 0;
 198        if (id > 0)
 199                return id;
 200
 201        if (alloc_it) {
 202                id = alloc_netid(net, peer, -1);
 203                *alloc = true;
 204                return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
 205        }
 206
 207        return NETNSA_NSID_NOT_ASSIGNED;
 208}
 209
 210/* should be called with nsid_lock held */
 211static int __peernet2id(struct net *net, struct net *peer)
 212{
 213        bool no = false;
 214
 215        return __peernet2id_alloc(net, peer, &no);
 216}
 217
 218static void rtnl_net_notifyid(struct net *net, int cmd, int id);
 219/* This function returns the id of a peer netns. If no id is assigned, one will
 220 * be allocated and returned.
 221 */
 222int peernet2id_alloc(struct net *net, struct net *peer)
 223{
 224        bool alloc;
 225        int id;
 226
 227        if (atomic_read(&net->count) == 0)
 228                return NETNSA_NSID_NOT_ASSIGNED;
 229        spin_lock_bh(&net->nsid_lock);
 230        alloc = atomic_read(&peer->count) == 0 ? false : true;
 231        id = __peernet2id_alloc(net, peer, &alloc);
 232        spin_unlock_bh(&net->nsid_lock);
 233        if (alloc && id >= 0)
 234                rtnl_net_notifyid(net, RTM_NEWNSID, id);
 235        return id;
 236}
 237
 238/* This function returns, if assigned, the id of a peer netns. */
 239int peernet2id(struct net *net, struct net *peer)
 240{
 241        int id;
 242
 243        spin_lock_bh(&net->nsid_lock);
 244        id = __peernet2id(net, peer);
 245        spin_unlock_bh(&net->nsid_lock);
 246        return id;
 247}
 248EXPORT_SYMBOL(peernet2id);
 249
 250/* This function returns true is the peer netns has an id assigned into the
 251 * current netns.
 252 */
 253bool peernet_has_id(struct net *net, struct net *peer)
 254{
 255        return peernet2id(net, peer) >= 0;
 256}
 257
 258struct net *get_net_ns_by_id(struct net *net, int id)
 259{
 260        struct net *peer;
 261
 262        if (id < 0)
 263                return NULL;
 264
 265        rcu_read_lock();
 266        spin_lock_bh(&net->nsid_lock);
 267        peer = idr_find(&net->netns_ids, id);
 268        if (peer)
 269                get_net(peer);
 270        spin_unlock_bh(&net->nsid_lock);
 271        rcu_read_unlock();
 272
 273        return peer;
 274}
 275
 276/*
 277 * setup_net runs the initializers for the network namespace object.
 278 */
 279static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 280{
 281        /* Must be called with net_mutex held */
 282        const struct pernet_operations *ops, *saved_ops;
 283        int error = 0;
 284        LIST_HEAD(net_exit_list);
 285
 286        atomic_set(&net->count, 1);
 287        refcount_set(&net->passive, 1);
 288        net->dev_base_seq = 1;
 289        net->user_ns = user_ns;
 290        idr_init(&net->netns_ids);
 291        spin_lock_init(&net->nsid_lock);
 292
 293        list_for_each_entry(ops, &pernet_list, list) {
 294                error = ops_init(ops, net);
 295                if (error < 0)
 296                        goto out_undo;
 297        }
 298out:
 299        return error;
 300
 301out_undo:
 302        /* Walk through the list backwards calling the exit functions
 303         * for the pernet modules whose init functions did not fail.
 304         */
 305        list_add(&net->exit_list, &net_exit_list);
 306        saved_ops = ops;
 307        list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 308                ops_exit_list(ops, &net_exit_list);
 309
 310        ops = saved_ops;
 311        list_for_each_entry_continue_reverse(ops, &pernet_list, list)
 312                ops_free_list(ops, &net_exit_list);
 313
 314        rcu_barrier();
 315        goto out;
 316}
 317
 318static int __net_init net_defaults_init_net(struct net *net)
 319{
 320        net->core.sysctl_somaxconn = SOMAXCONN;
 321        return 0;
 322}
 323
 324static struct pernet_operations net_defaults_ops = {
 325        .init = net_defaults_init_net,
 326};
 327
 328static __init int net_defaults_init(void)
 329{
 330        if (register_pernet_subsys(&net_defaults_ops))
 331                panic("Cannot initialize net default settings");
 332
 333        return 0;
 334}
 335
 336core_initcall(net_defaults_init);
 337
 338#ifdef CONFIG_NET_NS
 339static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
 340{
 341        return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
 342}
 343
 344static void dec_net_namespaces(struct ucounts *ucounts)
 345{
 346        dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
 347}
 348
 349static struct kmem_cache *net_cachep;
 350static struct workqueue_struct *netns_wq;
 351
 352static struct net *net_alloc(void)
 353{
 354        struct net *net = NULL;
 355        struct net_generic *ng;
 356
 357        ng = net_alloc_generic();
 358        if (!ng)
 359                goto out;
 360
 361        net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
 362        if (!net)
 363                goto out_free;
 364
 365        rcu_assign_pointer(net->gen, ng);
 366out:
 367        return net;
 368
 369out_free:
 370        kfree(ng);
 371        goto out;
 372}
 373
 374static void net_free(struct net *net)
 375{
 376        kfree(rcu_access_pointer(net->gen));
 377        kmem_cache_free(net_cachep, net);
 378}
 379
 380void net_drop_ns(void *p)
 381{
 382        struct net *ns = p;
 383        if (ns && refcount_dec_and_test(&ns->passive))
 384                net_free(ns);
 385}
 386
 387struct net *copy_net_ns(unsigned long flags,
 388                        struct user_namespace *user_ns, struct net *old_net)
 389{
 390        struct ucounts *ucounts;
 391        struct net *net;
 392        int rv;
 393
 394        if (!(flags & CLONE_NEWNET))
 395                return get_net(old_net);
 396
 397        ucounts = inc_net_namespaces(user_ns);
 398        if (!ucounts)
 399                return ERR_PTR(-ENOSPC);
 400
 401        net = net_alloc();
 402        if (!net) {
 403                dec_net_namespaces(ucounts);
 404                return ERR_PTR(-ENOMEM);
 405        }
 406
 407        get_user_ns(user_ns);
 408
 409        rv = mutex_lock_killable(&net_mutex);
 410        if (rv < 0) {
 411                net_free(net);
 412                dec_net_namespaces(ucounts);
 413                put_user_ns(user_ns);
 414                return ERR_PTR(rv);
 415        }
 416
 417        net->ucounts = ucounts;
 418        rv = setup_net(net, user_ns);
 419        if (rv == 0) {
 420                rtnl_lock();
 421                list_add_tail_rcu(&net->list, &net_namespace_list);
 422                rtnl_unlock();
 423        }
 424        mutex_unlock(&net_mutex);
 425        if (rv < 0) {
 426                dec_net_namespaces(ucounts);
 427                put_user_ns(user_ns);
 428                net_drop_ns(net);
 429                return ERR_PTR(rv);
 430        }
 431        return net;
 432}
 433
 434static DEFINE_SPINLOCK(cleanup_list_lock);
 435static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
 436
 437static void cleanup_net(struct work_struct *work)
 438{
 439        const struct pernet_operations *ops;
 440        struct net *net, *tmp;
 441        struct list_head net_kill_list;
 442        LIST_HEAD(net_exit_list);
 443
 444        /* Atomically snapshot the list of namespaces to cleanup */
 445        spin_lock_irq(&cleanup_list_lock);
 446        list_replace_init(&cleanup_list, &net_kill_list);
 447        spin_unlock_irq(&cleanup_list_lock);
 448
 449        mutex_lock(&net_mutex);
 450
 451        /* Don't let anyone else find us. */
 452        rtnl_lock();
 453        list_for_each_entry(net, &net_kill_list, cleanup_list) {
 454                list_del_rcu(&net->list);
 455                list_add_tail(&net->exit_list, &net_exit_list);
 456                for_each_net(tmp) {
 457                        int id;
 458
 459                        spin_lock_bh(&tmp->nsid_lock);
 460                        id = __peernet2id(tmp, net);
 461                        if (id >= 0)
 462                                idr_remove(&tmp->netns_ids, id);
 463                        spin_unlock_bh(&tmp->nsid_lock);
 464                        if (id >= 0)
 465                                rtnl_net_notifyid(tmp, RTM_DELNSID, id);
 466                }
 467                spin_lock_bh(&net->nsid_lock);
 468                idr_destroy(&net->netns_ids);
 469                spin_unlock_bh(&net->nsid_lock);
 470
 471        }
 472        rtnl_unlock();
 473
 474        /*
 475         * Another CPU might be rcu-iterating the list, wait for it.
 476         * This needs to be before calling the exit() notifiers, so
 477         * the rcu_barrier() below isn't sufficient alone.
 478         */
 479        synchronize_rcu();
 480
 481        /* Run all of the network namespace exit methods */
 482        list_for_each_entry_reverse(ops, &pernet_list, list)
 483                ops_exit_list(ops, &net_exit_list);
 484
 485        /* Free the net generic variables */
 486        list_for_each_entry_reverse(ops, &pernet_list, list)
 487                ops_free_list(ops, &net_exit_list);
 488
 489        mutex_unlock(&net_mutex);
 490
 491        /* Ensure there are no outstanding rcu callbacks using this
 492         * network namespace.
 493         */
 494        rcu_barrier();
 495
 496        /* Finally it is safe to free my network namespace structure */
 497        list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
 498                list_del_init(&net->exit_list);
 499                dec_net_namespaces(net->ucounts);
 500                put_user_ns(net->user_ns);
 501                net_drop_ns(net);
 502        }
 503}
 504
 505/**
 506 * net_ns_barrier - wait until concurrent net_cleanup_work is done
 507 *
 508 * cleanup_net runs from work queue and will first remove namespaces
 509 * from the global list, then run net exit functions.
 510 *
 511 * Call this in module exit path to make sure that all netns
 512 * ->exit ops have been invoked before the function is removed.
 513 */
 514void net_ns_barrier(void)
 515{
 516        mutex_lock(&net_mutex);
 517        mutex_unlock(&net_mutex);
 518}
 519EXPORT_SYMBOL(net_ns_barrier);
 520
 521static DECLARE_WORK(net_cleanup_work, cleanup_net);
 522
 523void __put_net(struct net *net)
 524{
 525        /* Cleanup the network namespace in process context */
 526        unsigned long flags;
 527
 528        spin_lock_irqsave(&cleanup_list_lock, flags);
 529        list_add(&net->cleanup_list, &cleanup_list);
 530        spin_unlock_irqrestore(&cleanup_list_lock, flags);
 531
 532        queue_work(netns_wq, &net_cleanup_work);
 533}
 534EXPORT_SYMBOL_GPL(__put_net);
 535
 536struct net *get_net_ns_by_fd(int fd)
 537{
 538        struct file *file;
 539        struct ns_common *ns;
 540        struct net *net;
 541
 542        file = proc_ns_fget(fd);
 543        if (IS_ERR(file))
 544                return ERR_CAST(file);
 545
 546        ns = get_proc_ns(file_inode(file));
 547        if (ns->ops == &netns_operations)
 548                net = get_net(container_of(ns, struct net, ns));
 549        else
 550                net = ERR_PTR(-EINVAL);
 551
 552        fput(file);
 553        return net;
 554}
 555
 556#else
 557struct net *get_net_ns_by_fd(int fd)
 558{
 559        return ERR_PTR(-EINVAL);
 560}
 561#endif
 562EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
 563
 564struct net *get_net_ns_by_pid(pid_t pid)
 565{
 566        struct task_struct *tsk;
 567        struct net *net;
 568
 569        /* Lookup the network namespace */
 570        net = ERR_PTR(-ESRCH);
 571        rcu_read_lock();
 572        tsk = find_task_by_vpid(pid);
 573        if (tsk) {
 574                struct nsproxy *nsproxy;
 575                task_lock(tsk);
 576                nsproxy = tsk->nsproxy;
 577                if (nsproxy)
 578                        net = get_net(nsproxy->net_ns);
 579                task_unlock(tsk);
 580        }
 581        rcu_read_unlock();
 582        return net;
 583}
 584EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
 585
 586static __net_init int net_ns_net_init(struct net *net)
 587{
 588#ifdef CONFIG_NET_NS
 589        net->ns.ops = &netns_operations;
 590#endif
 591        return ns_alloc_inum(&net->ns);
 592}
 593
 594static __net_exit void net_ns_net_exit(struct net *net)
 595{
 596        ns_free_inum(&net->ns);
 597}
 598
 599static struct pernet_operations __net_initdata net_ns_ops = {
 600        .init = net_ns_net_init,
 601        .exit = net_ns_net_exit,
 602};
 603
 604static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
 605        [NETNSA_NONE]           = { .type = NLA_UNSPEC },
 606        [NETNSA_NSID]           = { .type = NLA_S32 },
 607        [NETNSA_PID]            = { .type = NLA_U32 },
 608        [NETNSA_FD]             = { .type = NLA_U32 },
 609};
 610
 611static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
 612                          struct netlink_ext_ack *extack)
 613{
 614        struct net *net = sock_net(skb->sk);
 615        struct nlattr *tb[NETNSA_MAX + 1];
 616        struct nlattr *nla;
 617        struct net *peer;
 618        int nsid, err;
 619
 620        err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
 621                          rtnl_net_policy, extack);
 622        if (err < 0)
 623                return err;
 624        if (!tb[NETNSA_NSID]) {
 625                NL_SET_ERR_MSG(extack, "nsid is missing");
 626                return -EINVAL;
 627        }
 628        nsid = nla_get_s32(tb[NETNSA_NSID]);
 629
 630        if (tb[NETNSA_PID]) {
 631                peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 632                nla = tb[NETNSA_PID];
 633        } else if (tb[NETNSA_FD]) {
 634                peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 635                nla = tb[NETNSA_FD];
 636        } else {
 637                NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 638                return -EINVAL;
 639        }
 640        if (IS_ERR(peer)) {
 641                NL_SET_BAD_ATTR(extack, nla);
 642                NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 643                return PTR_ERR(peer);
 644        }
 645
 646        spin_lock_bh(&net->nsid_lock);
 647        if (__peernet2id(net, peer) >= 0) {
 648                spin_unlock_bh(&net->nsid_lock);
 649                err = -EEXIST;
 650                NL_SET_BAD_ATTR(extack, nla);
 651                NL_SET_ERR_MSG(extack,
 652                               "Peer netns already has a nsid assigned");
 653                goto out;
 654        }
 655
 656        err = alloc_netid(net, peer, nsid);
 657        spin_unlock_bh(&net->nsid_lock);
 658        if (err >= 0) {
 659                rtnl_net_notifyid(net, RTM_NEWNSID, err);
 660                err = 0;
 661        } else if (err == -ENOSPC && nsid >= 0) {
 662                err = -EEXIST;
 663                NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
 664                NL_SET_ERR_MSG(extack, "The specified nsid is already used");
 665        }
 666out:
 667        put_net(peer);
 668        return err;
 669}
 670
 671static int rtnl_net_get_size(void)
 672{
 673        return NLMSG_ALIGN(sizeof(struct rtgenmsg))
 674               + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
 675               ;
 676}
 677
 678static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
 679                         int cmd, struct net *net, int nsid)
 680{
 681        struct nlmsghdr *nlh;
 682        struct rtgenmsg *rth;
 683
 684        nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
 685        if (!nlh)
 686                return -EMSGSIZE;
 687
 688        rth = nlmsg_data(nlh);
 689        rth->rtgen_family = AF_UNSPEC;
 690
 691        if (nla_put_s32(skb, NETNSA_NSID, nsid))
 692                goto nla_put_failure;
 693
 694        nlmsg_end(skb, nlh);
 695        return 0;
 696
 697nla_put_failure:
 698        nlmsg_cancel(skb, nlh);
 699        return -EMSGSIZE;
 700}
 701
 702static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
 703                          struct netlink_ext_ack *extack)
 704{
 705        struct net *net = sock_net(skb->sk);
 706        struct nlattr *tb[NETNSA_MAX + 1];
 707        struct nlattr *nla;
 708        struct sk_buff *msg;
 709        struct net *peer;
 710        int err, id;
 711
 712        err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
 713                          rtnl_net_policy, extack);
 714        if (err < 0)
 715                return err;
 716        if (tb[NETNSA_PID]) {
 717                peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
 718                nla = tb[NETNSA_PID];
 719        } else if (tb[NETNSA_FD]) {
 720                peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
 721                nla = tb[NETNSA_FD];
 722        } else {
 723                NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
 724                return -EINVAL;
 725        }
 726
 727        if (IS_ERR(peer)) {
 728                NL_SET_BAD_ATTR(extack, nla);
 729                NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
 730                return PTR_ERR(peer);
 731        }
 732
 733        msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 734        if (!msg) {
 735                err = -ENOMEM;
 736                goto out;
 737        }
 738
 739        id = peernet2id(net, peer);
 740        err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
 741                            RTM_NEWNSID, net, id);
 742        if (err < 0)
 743                goto err_out;
 744
 745        err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
 746        goto out;
 747
 748err_out:
 749        nlmsg_free(msg);
 750out:
 751        put_net(peer);
 752        return err;
 753}
 754
 755struct rtnl_net_dump_cb {
 756        struct net *net;
 757        struct sk_buff *skb;
 758        struct netlink_callback *cb;
 759        int idx;
 760        int s_idx;
 761};
 762
 763static int rtnl_net_dumpid_one(int id, void *peer, void *data)
 764{
 765        struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
 766        int ret;
 767
 768        if (net_cb->idx < net_cb->s_idx)
 769                goto cont;
 770
 771        ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
 772                            net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
 773                            RTM_NEWNSID, net_cb->net, id);
 774        if (ret < 0)
 775                return ret;
 776
 777cont:
 778        net_cb->idx++;
 779        return 0;
 780}
 781
 782static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
 783{
 784        struct net *net = sock_net(skb->sk);
 785        struct rtnl_net_dump_cb net_cb = {
 786                .net = net,
 787                .skb = skb,
 788                .cb = cb,
 789                .idx = 0,
 790                .s_idx = cb->args[0],
 791        };
 792
 793        spin_lock_bh(&net->nsid_lock);
 794        idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
 795        spin_unlock_bh(&net->nsid_lock);
 796
 797        cb->args[0] = net_cb.idx;
 798        return skb->len;
 799}
 800
 801static void rtnl_net_notifyid(struct net *net, int cmd, int id)
 802{
 803        struct sk_buff *msg;
 804        int err = -ENOMEM;
 805
 806        msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
 807        if (!msg)
 808                goto out;
 809
 810        err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
 811        if (err < 0)
 812                goto err_out;
 813
 814        rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
 815        return;
 816
 817err_out:
 818        nlmsg_free(msg);
 819out:
 820        rtnl_set_sk_err(net, RTNLGRP_NSID, err);
 821}
 822
 823static int __init net_ns_init(void)
 824{
 825        struct net_generic *ng;
 826
 827#ifdef CONFIG_NET_NS
 828        net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
 829                                        SMP_CACHE_BYTES,
 830                                        SLAB_PANIC, NULL);
 831
 832        /* Create workqueue for cleanup */
 833        netns_wq = create_singlethread_workqueue("netns");
 834        if (!netns_wq)
 835                panic("Could not create netns workq");
 836#endif
 837
 838        ng = net_alloc_generic();
 839        if (!ng)
 840                panic("Could not allocate generic netns");
 841
 842        rcu_assign_pointer(init_net.gen, ng);
 843
 844        mutex_lock(&net_mutex);
 845        if (setup_net(&init_net, &init_user_ns))
 846                panic("Could not setup the initial network namespace");
 847
 848        init_net_initialized = true;
 849
 850        rtnl_lock();
 851        list_add_tail_rcu(&init_net.list, &net_namespace_list);
 852        rtnl_unlock();
 853
 854        mutex_unlock(&net_mutex);
 855
 856        register_pernet_subsys(&net_ns_ops);
 857
 858        rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
 859        rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
 860                      NULL);
 861
 862        return 0;
 863}
 864
 865pure_initcall(net_ns_init);
 866
 867#ifdef CONFIG_NET_NS
 868static int __register_pernet_operations(struct list_head *list,
 869                                        struct pernet_operations *ops)
 870{
 871        struct net *net;
 872        int error;
 873        LIST_HEAD(net_exit_list);
 874
 875        list_add_tail(&ops->list, list);
 876        if (ops->init || (ops->id && ops->size)) {
 877                for_each_net(net) {
 878                        error = ops_init(ops, net);
 879                        if (error)
 880                                goto out_undo;
 881                        list_add_tail(&net->exit_list, &net_exit_list);
 882                }
 883        }
 884        return 0;
 885
 886out_undo:
 887        /* If I have an error cleanup all namespaces I initialized */
 888        list_del(&ops->list);
 889        ops_exit_list(ops, &net_exit_list);
 890        ops_free_list(ops, &net_exit_list);
 891        return error;
 892}
 893
 894static void __unregister_pernet_operations(struct pernet_operations *ops)
 895{
 896        struct net *net;
 897        LIST_HEAD(net_exit_list);
 898
 899        list_del(&ops->list);
 900        for_each_net(net)
 901                list_add_tail(&net->exit_list, &net_exit_list);
 902        ops_exit_list(ops, &net_exit_list);
 903        ops_free_list(ops, &net_exit_list);
 904}
 905
 906#else
 907
 908static int __register_pernet_operations(struct list_head *list,
 909                                        struct pernet_operations *ops)
 910{
 911        if (!init_net_initialized) {
 912                list_add_tail(&ops->list, list);
 913                return 0;
 914        }
 915
 916        return ops_init(ops, &init_net);
 917}
 918
 919static void __unregister_pernet_operations(struct pernet_operations *ops)
 920{
 921        if (!init_net_initialized) {
 922                list_del(&ops->list);
 923        } else {
 924                LIST_HEAD(net_exit_list);
 925                list_add(&init_net.exit_list, &net_exit_list);
 926                ops_exit_list(ops, &net_exit_list);
 927                ops_free_list(ops, &net_exit_list);
 928        }
 929}
 930
 931#endif /* CONFIG_NET_NS */
 932
 933static DEFINE_IDA(net_generic_ids);
 934
 935static int register_pernet_operations(struct list_head *list,
 936                                      struct pernet_operations *ops)
 937{
 938        int error;
 939
 940        if (ops->id) {
 941again:
 942                error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id);
 943                if (error < 0) {
 944                        if (error == -EAGAIN) {
 945                                ida_pre_get(&net_generic_ids, GFP_KERNEL);
 946                                goto again;
 947                        }
 948                        return error;
 949                }
 950                max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
 951        }
 952        error = __register_pernet_operations(list, ops);
 953        if (error) {
 954                rcu_barrier();
 955                if (ops->id)
 956                        ida_remove(&net_generic_ids, *ops->id);
 957        }
 958
 959        return error;
 960}
 961
 962static void unregister_pernet_operations(struct pernet_operations *ops)
 963{
 964        
 965        __unregister_pernet_operations(ops);
 966        rcu_barrier();
 967        if (ops->id)
 968                ida_remove(&net_generic_ids, *ops->id);
 969}
 970
 971/**
 972 *      register_pernet_subsys - register a network namespace subsystem
 973 *      @ops:  pernet operations structure for the subsystem
 974 *
 975 *      Register a subsystem which has init and exit functions
 976 *      that are called when network namespaces are created and
 977 *      destroyed respectively.
 978 *
 979 *      When registered all network namespace init functions are
 980 *      called for every existing network namespace.  Allowing kernel
 981 *      modules to have a race free view of the set of network namespaces.
 982 *
 983 *      When a new network namespace is created all of the init
 984 *      methods are called in the order in which they were registered.
 985 *
 986 *      When a network namespace is destroyed all of the exit methods
 987 *      are called in the reverse of the order with which they were
 988 *      registered.
 989 */
 990int register_pernet_subsys(struct pernet_operations *ops)
 991{
 992        int error;
 993        mutex_lock(&net_mutex);
 994        error =  register_pernet_operations(first_device, ops);
 995        mutex_unlock(&net_mutex);
 996        return error;
 997}
 998EXPORT_SYMBOL_GPL(register_pernet_subsys);
 999
1000/**
1001 *      unregister_pernet_subsys - unregister a network namespace subsystem
1002 *      @ops: pernet operations structure to manipulate
1003 *
1004 *      Remove the pernet operations structure from the list to be
1005 *      used when network namespaces are created or destroyed.  In
1006 *      addition run the exit method for all existing network
1007 *      namespaces.
1008 */
1009void unregister_pernet_subsys(struct pernet_operations *ops)
1010{
1011        mutex_lock(&net_mutex);
1012        unregister_pernet_operations(ops);
1013        mutex_unlock(&net_mutex);
1014}
1015EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1016
1017/**
1018 *      register_pernet_device - register a network namespace device
1019 *      @ops:  pernet operations structure for the subsystem
1020 *
1021 *      Register a device which has init and exit functions
1022 *      that are called when network namespaces are created and
1023 *      destroyed respectively.
1024 *
1025 *      When registered all network namespace init functions are
1026 *      called for every existing network namespace.  Allowing kernel
1027 *      modules to have a race free view of the set of network namespaces.
1028 *
1029 *      When a new network namespace is created all of the init
1030 *      methods are called in the order in which they were registered.
1031 *
1032 *      When a network namespace is destroyed all of the exit methods
1033 *      are called in the reverse of the order with which they were
1034 *      registered.
1035 */
1036int register_pernet_device(struct pernet_operations *ops)
1037{
1038        int error;
1039        mutex_lock(&net_mutex);
1040        error = register_pernet_operations(&pernet_list, ops);
1041        if (!error && (first_device == &pernet_list))
1042                first_device = &ops->list;
1043        mutex_unlock(&net_mutex);
1044        return error;
1045}
1046EXPORT_SYMBOL_GPL(register_pernet_device);
1047
1048/**
1049 *      unregister_pernet_device - unregister a network namespace netdevice
1050 *      @ops: pernet operations structure to manipulate
1051 *
1052 *      Remove the pernet operations structure from the list to be
1053 *      used when network namespaces are created or destroyed.  In
1054 *      addition run the exit method for all existing network
1055 *      namespaces.
1056 */
1057void unregister_pernet_device(struct pernet_operations *ops)
1058{
1059        mutex_lock(&net_mutex);
1060        if (&ops->list == first_device)
1061                first_device = first_device->next;
1062        unregister_pernet_operations(ops);
1063        mutex_unlock(&net_mutex);
1064}
1065EXPORT_SYMBOL_GPL(unregister_pernet_device);
1066
1067#ifdef CONFIG_NET_NS
1068static struct ns_common *netns_get(struct task_struct *task)
1069{
1070        struct net *net = NULL;
1071        struct nsproxy *nsproxy;
1072
1073        task_lock(task);
1074        nsproxy = task->nsproxy;
1075        if (nsproxy)
1076                net = get_net(nsproxy->net_ns);
1077        task_unlock(task);
1078
1079        return net ? &net->ns : NULL;
1080}
1081
1082static inline struct net *to_net_ns(struct ns_common *ns)
1083{
1084        return container_of(ns, struct net, ns);
1085}
1086
1087static void netns_put(struct ns_common *ns)
1088{
1089        put_net(to_net_ns(ns));
1090}
1091
1092static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1093{
1094        struct net *net = to_net_ns(ns);
1095
1096        if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1097            !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1098                return -EPERM;
1099
1100        put_net(nsproxy->net_ns);
1101        nsproxy->net_ns = get_net(net);
1102        return 0;
1103}
1104
1105static struct user_namespace *netns_owner(struct ns_common *ns)
1106{
1107        return to_net_ns(ns)->user_ns;
1108}
1109
1110const struct proc_ns_operations netns_operations = {
1111        .name           = "net",
1112        .type           = CLONE_NEWNET,
1113        .get            = netns_get,
1114        .put            = netns_put,
1115        .install        = netns_install,
1116        .owner          = netns_owner,
1117};
1118#endif
1119