net/ipv6/ip6_flowlabel.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      ip6_flowlabel.c         IPv6 flowlabel manager.
   4  *
   5  *      Authors:        Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   6  */
   7
   8 #include <linux/capability.h>
   9 #include <linux/errno.h>
  10 #include <linux/types.h>
  11 #include <linux/socket.h>
  12 #include <linux/net.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/in6.h>
  15 #include <linux/proc_fs.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/slab.h>
  18 #include <linux/export.h>
  19 #include <linux/pid_namespace.h>
  20
  21 #include <net/net_namespace.h>
  22 #include <net/sock.h>
  23
  24 #include <net/ipv6.h>
  25 #include <net/rawv6.h>
  26 #include <net/transp_v6.h>
  27
  28 #include <linux/uaccess.h>
  29
  30 #define FL_MIN_LINGER   6       /* Minimal linger. It is set to 6sec specified
  31                                    in old IPv6 RFC. Well, it was reasonable value.
  32                                  */
  33 #define FL_MAX_LINGER   150     /* Maximal linger timeout */
  34
  35 /* FL hash table */
  36
  37 #define FL_MAX_PER_SOCK 32
  38 #define FL_MAX_SIZE     4096
  39 #define FL_HASH_MASK    255
  40 #define FL_HASH(l)      (ntohl(l)&FL_HASH_MASK)
  41
  42 static atomic_t fl_size = ATOMIC_INIT(0);
  43 static struct ip6_flowlabel __rcu *fl_ht[FL_HASH_MASK+1];
  44
  45 static void ip6_fl_gc(struct timer_list *unused);
  46 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc);
  47
  48 /* FL hash table lock: it protects only of GC */
  49
  50 static DEFINE_SPINLOCK(ip6_fl_lock);
  51
  52 /* Big socket sock */
  53
  54 static DEFINE_SPINLOCK(ip6_sk_fl_lock);
  55
  56 #define for_each_fl_rcu(hash, fl)                               \
  57         for (fl = rcu_dereference_bh(fl_ht[(hash)]);            \
  58              fl != NULL;                                        \
  59              fl = rcu_dereference_bh(fl->next))
  60 #define for_each_fl_continue_rcu(fl)                            \
  61         for (fl = rcu_dereference_bh(fl->next);                 \
  62              fl != NULL;                                        \
  63              fl = rcu_dereference_bh(fl->next))
  64
  65 #define for_each_sk_fl_rcu(np, sfl)                             \
  66         for (sfl = rcu_dereference_bh(np->ipv6_fl_list);        \
  67              sfl != NULL;                                       \
  68              sfl = rcu_dereference_bh(sfl->next))
  69
  70 static inline struct ip6_flowlabel *__fl_lookup(struct net *net, __be32 label)
  71 {
  72         struct ip6_flowlabel *fl;
  73
  74         for_each_fl_rcu(FL_HASH(label), fl) {
  75                 if (fl->label == label && net_eq(fl->fl_net, net))
  76                         return fl;
  77         }
  78         return NULL;
  79 }
  80
  81 static struct ip6_flowlabel *fl_lookup(struct net *net, __be32 label)
  82 {
  83         struct ip6_flowlabel *fl;
  84
  85         rcu_read_lock_bh();
  86         fl = __fl_lookup(net, label);
  87         if (fl && !atomic_inc_not_zero(&fl->users))
  88                 fl = NULL;
  89         rcu_read_unlock_bh();
  90         return fl;
  91 }
  92
  93 static void fl_free_rcu(struct rcu_head *head)
  94 {
  95         struct ip6_flowlabel *fl = container_of(head, struct ip6_flowlabel, rcu);
  96
  97         if (fl->share == IPV6_FL_S_PROCESS)
  98                 put_pid(fl->owner.pid);
  99         kfree(fl->opt);
 100         kfree(fl);
 101 }
 102
 103
 104 static void fl_free(struct ip6_flowlabel *fl)
 105 {
 106         if (fl)
 107                 call_rcu(&fl->rcu, fl_free_rcu);
 108 }
 109
 110 static void fl_release(struct ip6_flowlabel *fl)
 111 {
 112         spin_lock_bh(&ip6_fl_lock);
 113
 114         fl->lastuse = jiffies;
 115         if (atomic_dec_and_test(&fl->users)) {
 116                 unsigned long ttd = fl->lastuse + fl->linger;
 117                 if (time_after(ttd, fl->expires))
 118                         fl->expires = ttd;
 119                 ttd = fl->expires;
 120                 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
 121                         struct ipv6_txoptions *opt = fl->opt;
 122                         fl->opt = NULL;
 123                         kfree(opt);
 124                 }
 125                 if (!timer_pending(&ip6_fl_gc_timer) ||
 126                     time_after(ip6_fl_gc_timer.expires, ttd))
 127                         mod_timer(&ip6_fl_gc_timer, ttd);
 128         }
 129         spin_unlock_bh(&ip6_fl_lock);
 130 }
 131
 132 static void ip6_fl_gc(struct timer_list *unused)
 133 {
 134         int i;
 135         unsigned long now = jiffies;
 136         unsigned long sched = 0;
 137
 138         spin_lock(&ip6_fl_lock);
 139
 140         for (i = 0; i <= FL_HASH_MASK; i++) {
 141                 struct ip6_flowlabel *fl;
 142                 struct ip6_flowlabel __rcu **flp;
 143
 144                 flp = &fl_ht[i];
 145                 while ((fl = rcu_dereference_protected(*flp,
 146                                                        lockdep_is_held(&ip6_fl_lock))) != NULL) {
 147                         if (atomic_read(&fl->users) == 0) {
 148                                 unsigned long ttd = fl->lastuse + fl->linger;
 149                                 if (time_after(ttd, fl->expires))
 150                                         fl->expires = ttd;
 151                                 ttd = fl->expires;
 152                                 if (time_after_eq(now, ttd)) {
 153                                         *flp = fl->next;
 154                                         fl_free(fl);
 155                                         atomic_dec(&fl_size);
 156                                         continue;
 157                                 }
 158                                 if (!sched || time_before(ttd, sched))
 159                                         sched = ttd;
 160                         }
 161                         flp = &fl->next;
 162                 }
 163         }
 164         if (!sched && atomic_read(&fl_size))
 165                 sched = now + FL_MAX_LINGER;
 166         if (sched) {
 167                 mod_timer(&ip6_fl_gc_timer, sched);
 168         }
 169         spin_unlock(&ip6_fl_lock);
 170 }
 171
 172 static void __net_exit ip6_fl_purge(struct net *net)
 173 {
 174         int i;
 175
 176         spin_lock_bh(&ip6_fl_lock);
 177         for (i = 0; i <= FL_HASH_MASK; i++) {
 178                 struct ip6_flowlabel *fl;
 179                 struct ip6_flowlabel __rcu **flp;
 180
 181                 flp = &fl_ht[i];
 182                 while ((fl = rcu_dereference_protected(*flp,
 183                                                        lockdep_is_held(&ip6_fl_lock))) != NULL) {
 184                         if (net_eq(fl->fl_net, net) &&
 185                             atomic_read(&fl->users) == 0) {
 186                                 *flp = fl->next;
 187                                 fl_free(fl);
 188                                 atomic_dec(&fl_size);
 189                                 continue;
 190                         }
 191                         flp = &fl->next;
 192                 }
 193         }
 194         spin_unlock_bh(&ip6_fl_lock);
 195 }
 196
 197 static struct ip6_flowlabel *fl_intern(struct net *net,
 198                                        struct ip6_flowlabel *fl, __be32 label)
 199 {
 200         struct ip6_flowlabel *lfl;
 201
 202         fl->label = label & IPV6_FLOWLABEL_MASK;
 203
 204         spin_lock_bh(&ip6_fl_lock);
 205         if (label == 0) {
 206                 for (;;) {
 207                         fl->label = htonl(prandom_u32())&IPV6_FLOWLABEL_MASK;
 208                         if (fl->label) {
 209                                 lfl = __fl_lookup(net, fl->label);
 210                                 if (!lfl)
 211                                         break;
 212                         }
 213                 }
 214         } else {
 215                 /*
 216                  * we dropper the ip6_fl_lock, so this entry could reappear
 217                  * and we need to recheck with it.
 218                  *
 219                  * OTOH no need to search the active socket first, like it is
 220                  * done in ipv6_flowlabel_opt - sock is locked, so new entry
 221                  * with the same label can only appear on another sock
 222                  */
 223                 lfl = __fl_lookup(net, fl->label);
 224                 if (lfl) {
 225                         atomic_inc(&lfl->users);
 226                         spin_unlock_bh(&ip6_fl_lock);
 227                         return lfl;
 228                 }
 229         }
 230
 231         fl->lastuse = jiffies;
 232         fl->next = fl_ht[FL_HASH(fl->label)];
 233         rcu_assign_pointer(fl_ht[FL_HASH(fl->label)], fl);
 234         atomic_inc(&fl_size);
 235         spin_unlock_bh(&ip6_fl_lock);
 236         return NULL;
 237 }
 238
 239
 240
 241 /* Socket flowlabel lists */
 242
 243 struct ip6_flowlabel *fl6_sock_lookup(struct sock *sk, __be32 label)
 244 {
 245         struct ipv6_fl_socklist *sfl;
 246         struct ipv6_pinfo *np = inet6_sk(sk);
 247
 248         label &= IPV6_FLOWLABEL_MASK;
 249
 250         rcu_read_lock_bh();
 251         for_each_sk_fl_rcu(np, sfl) {
 252                 struct ip6_flowlabel *fl = sfl->fl;
 253                 if (fl->label == label) {
 254                         fl->lastuse = jiffies;
 255                         atomic_inc(&fl->users);
 256                         rcu_read_unlock_bh();
 257                         return fl;
 258                 }
 259         }
 260         rcu_read_unlock_bh();
 261         return NULL;
 262 }
 263 EXPORT_SYMBOL_GPL(fl6_sock_lookup);
 264
 265 void fl6_free_socklist(struct sock *sk)
 266 {
 267         struct ipv6_pinfo *np = inet6_sk(sk);
 268         struct ipv6_fl_socklist *sfl;
 269
 270         if (!rcu_access_pointer(np->ipv6_fl_list))
 271                 return;
 272
 273         spin_lock_bh(&ip6_sk_fl_lock);
 274         while ((sfl = rcu_dereference_protected(np->ipv6_fl_list,
 275                                                 lockdep_is_held(&ip6_sk_fl_lock))) != NULL) {
 276                 np->ipv6_fl_list = sfl->next;
 277                 spin_unlock_bh(&ip6_sk_fl_lock);
 278
 279                 fl_release(sfl->fl);
 280                 kfree_rcu(sfl, rcu);
 281
 282                 spin_lock_bh(&ip6_sk_fl_lock);
 283         }
 284         spin_unlock_bh(&ip6_sk_fl_lock);
 285 }
 286
 287 /* Service routines */
 288
 289
 290 /*
 291    It is the only difficult place. flowlabel enforces equal headers
 292    before and including routing header, however user may supply options
 293    following rthdr.
 294  */
 295
 296 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
 297                                          struct ip6_flowlabel *fl,
 298                                          struct ipv6_txoptions *fopt)
 299 {
 300         struct ipv6_txoptions *fl_opt = fl->opt;
 301
 302         if (!fopt || fopt->opt_flen == 0)
 303                 return fl_opt;
 304
 305         if (fl_opt) {
 306                 opt_space->hopopt = fl_opt->hopopt;
 307                 opt_space->dst0opt = fl_opt->dst0opt;
 308                 opt_space->srcrt = fl_opt->srcrt;
 309                 opt_space->opt_nflen = fl_opt->opt_nflen;
 310         } else {
 311                 if (fopt->opt_nflen == 0)
 312                         return fopt;
 313                 opt_space->hopopt = NULL;
 314                 opt_space->dst0opt = NULL;
 315                 opt_space->srcrt = NULL;
 316                 opt_space->opt_nflen = 0;
 317         }
 318         opt_space->dst1opt = fopt->dst1opt;
 319         opt_space->opt_flen = fopt->opt_flen;
 320         opt_space->tot_len = fopt->tot_len;
 321         return opt_space;
 322 }
 323 EXPORT_SYMBOL_GPL(fl6_merge_options);
 324
 325 static unsigned long check_linger(unsigned long ttl)
 326 {
 327         if (ttl < FL_MIN_LINGER)
 328                 return FL_MIN_LINGER*HZ;
 329         if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
 330                 return 0;
 331         return ttl*HZ;
 332 }
 333
 334 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
 335 {
 336         linger = check_linger(linger);
 337         if (!linger)
 338                 return -EPERM;
 339         expires = check_linger(expires);
 340         if (!expires)
 341                 return -EPERM;
 342
 343         spin_lock_bh(&ip6_fl_lock);
 344         fl->lastuse = jiffies;
 345         if (time_before(fl->linger, linger))
 346                 fl->linger = linger;
 347         if (time_before(expires, fl->linger))
 348                 expires = fl->linger;
 349         if (time_before(fl->expires, fl->lastuse + expires))
 350                 fl->expires = fl->lastuse + expires;
 351         spin_unlock_bh(&ip6_fl_lock);
 352
 353         return 0;
 354 }
 355
 356 static struct ip6_flowlabel *
 357 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
 358           char __user *optval, int optlen, int *err_p)
 359 {
 360         struct ip6_flowlabel *fl = NULL;
 361         int olen;
 362         int addr_type;
 363         int err;
 364
 365         olen = optlen - CMSG_ALIGN(sizeof(*freq));
 366         err = -EINVAL;
 367         if (olen > 64 * 1024)
 368                 goto done;
 369
 370         err = -ENOMEM;
 371         fl = kzalloc(sizeof(*fl), GFP_KERNEL);
 372         if (!fl)
 373                 goto done;
 374
 375         if (olen > 0) {
 376                 struct msghdr msg;
 377                 struct flowi6 flowi6;
 378                 struct ipcm6_cookie ipc6;
 379
 380                 err = -ENOMEM;
 381                 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
 382                 if (!fl->opt)
 383                         goto done;
 384
 385                 memset(fl->opt, 0, sizeof(*fl->opt));
 386                 fl->opt->tot_len = sizeof(*fl->opt) + olen;
 387                 err = -EFAULT;
 388                 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
 389                         goto done;
 390
 391                 msg.msg_controllen = olen;
 392                 msg.msg_control = (void *)(fl->opt+1);
 393                 memset(&flowi6, 0, sizeof(flowi6));
 394
 395                 ipc6.opt = fl->opt;
 396                 err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, &ipc6);
 397                 if (err)
 398                         goto done;
 399                 err = -EINVAL;
 400                 if (fl->opt->opt_flen)
 401                         goto done;
 402                 if (fl->opt->opt_nflen == 0) {
 403                         kfree(fl->opt);
 404                         fl->opt = NULL;
 405                 }
 406         }
 407
 408         fl->fl_net = net;
 409         fl->expires = jiffies;
 410         err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
 411         if (err)
 412                 goto done;
 413         fl->share = freq->flr_share;
 414         addr_type = ipv6_addr_type(&freq->flr_dst);
 415         if ((addr_type & IPV6_ADDR_MAPPED) ||
 416             addr_type == IPV6_ADDR_ANY) {
 417                 err = -EINVAL;
 418                 goto done;
 419         }
 420         fl->dst = freq->flr_dst;
 421         atomic_set(&fl->users, 1);
 422         switch (fl->share) {
 423         case IPV6_FL_S_EXCL:
 424         case IPV6_FL_S_ANY:
 425                 break;
 426         case IPV6_FL_S_PROCESS:
 427                 fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
 428                 break;
 429         case IPV6_FL_S_USER:
 430                 fl->owner.uid = current_euid();
 431                 break;
 432         default:
 433                 err = -EINVAL;
 434                 goto done;
 435         }
 436         return fl;
 437
 438 done:
 439         fl_free(fl);
 440         *err_p = err;
 441         return NULL;
 442 }
 443
 444 static int mem_check(struct sock *sk)
 445 {
 446         struct ipv6_pinfo *np = inet6_sk(sk);
 447         struct ipv6_fl_socklist *sfl;
 448         int room = FL_MAX_SIZE - atomic_read(&fl_size);
 449         int count = 0;
 450
 451         if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
 452                 return 0;
 453
 454         rcu_read_lock_bh();
 455         for_each_sk_fl_rcu(np, sfl)
 456                 count++;
 457         rcu_read_unlock_bh();
 458
 459         if (room <= 0 ||
 460             ((count >= FL_MAX_PER_SOCK ||
 461               (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
 462              !capable(CAP_NET_ADMIN)))
 463                 return -ENOBUFS;
 464
 465         return 0;
 466 }
 467
 468 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
 469                 struct ip6_flowlabel *fl)
 470 {
 471         spin_lock_bh(&ip6_sk_fl_lock);
 472         sfl->fl = fl;
 473         sfl->next = np->ipv6_fl_list;
 474         rcu_assign_pointer(np->ipv6_fl_list, sfl);
 475         spin_unlock_bh(&ip6_sk_fl_lock);
 476 }
 477
 478 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
 479                            int flags)
 480 {
 481         struct ipv6_pinfo *np = inet6_sk(sk);
 482         struct ipv6_fl_socklist *sfl;
 483
 484         if (flags & IPV6_FL_F_REMOTE) {
 485                 freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
 486                 return 0;
 487         }
 488
 489         if (np->repflow) {
 490                 freq->flr_label = np->flow_label;
 491                 return 0;
 492         }
 493
 494         rcu_read_lock_bh();
 495
 496         for_each_sk_fl_rcu(np, sfl) {
 497                 if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
 498                         spin_lock_bh(&ip6_fl_lock);
 499                         freq->flr_label = sfl->fl->label;
 500                         freq->flr_dst = sfl->fl->dst;
 501                         freq->flr_share = sfl->fl->share;
 502                         freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
 503                         freq->flr_linger = sfl->fl->linger / HZ;
 504
 505                         spin_unlock_bh(&ip6_fl_lock);
 506                         rcu_read_unlock_bh();
 507                         return 0;
 508                 }
 509         }
 510         rcu_read_unlock_bh();
 511
 512         return -ENOENT;
 513 }
 514
 515 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
 516 {
 517         int uninitialized_var(err);
 518         struct net *net = sock_net(sk);
 519         struct ipv6_pinfo *np = inet6_sk(sk);
 520         struct in6_flowlabel_req freq;
 521         struct ipv6_fl_socklist *sfl1 = NULL;
 522         struct ipv6_fl_socklist *sfl;
 523         struct ipv6_fl_socklist __rcu **sflp;
 524         struct ip6_flowlabel *fl, *fl1 = NULL;
 525
 526
 527         if (optlen < sizeof(freq))
 528                 return -EINVAL;
 529
 530         if (copy_from_user(&freq, optval, sizeof(freq)))
 531                 return -EFAULT;
 532
 533         switch (freq.flr_action) {
 534         case IPV6_FL_A_PUT:
 535                 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
 536                         if (sk->sk_protocol != IPPROTO_TCP)
 537                                 return -ENOPROTOOPT;
 538                         if (!np->repflow)
 539                                 return -ESRCH;
 540                         np->flow_label = 0;
 541                         np->repflow = 0;
 542                         return 0;
 543                 }
 544                 spin_lock_bh(&ip6_sk_fl_lock);
 545                 for (sflp = &np->ipv6_fl_list;
 546                      (sfl = rcu_dereference_protected(*sflp,
 547                                                       lockdep_is_held(&ip6_sk_fl_lock))) != NULL;
 548                      sflp = &sfl->next) {
 549                         if (sfl->fl->label == freq.flr_label) {
 550                                 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
 551                                         np->flow_label &= ~IPV6_FLOWLABEL_MASK;
 552                                 *sflp = sfl->next;
 553                                 spin_unlock_bh(&ip6_sk_fl_lock);
 554                                 fl_release(sfl->fl);
 555                                 kfree_rcu(sfl, rcu);
 556                                 return 0;
 557                         }
 558                 }
 559                 spin_unlock_bh(&ip6_sk_fl_lock);
 560                 return -ESRCH;
 561
 562         case IPV6_FL_A_RENEW:
 563                 rcu_read_lock_bh();
 564                 for_each_sk_fl_rcu(np, sfl) {
 565                         if (sfl->fl->label == freq.flr_label) {
 566                                 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
 567                                 rcu_read_unlock_bh();
 568                                 return err;
 569                         }
 570                 }
 571                 rcu_read_unlock_bh();
 572
 573                 if (freq.flr_share == IPV6_FL_S_NONE &&
 574                     ns_capable(net->user_ns, CAP_NET_ADMIN)) {
 575                         fl = fl_lookup(net, freq.flr_label);
 576                         if (fl) {
 577                                 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
 578                                 fl_release(fl);
 579                                 return err;
 580                         }
 581                 }
 582                 return -ESRCH;
 583
 584         case IPV6_FL_A_GET:
 585                 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
 586                         struct net *net = sock_net(sk);
 587                         if (net->ipv6.sysctl.flowlabel_consistency) {
 588                                 net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
 589                                 return -EPERM;
 590                         }
 591
 592                         if (sk->sk_protocol != IPPROTO_TCP)
 593                                 return -ENOPROTOOPT;
 594
 595                         np->repflow = 1;
 596                         return 0;
 597                 }
 598
 599                 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
 600                         return -EINVAL;
 601
 602                 if (net->ipv6.sysctl.flowlabel_state_ranges &&
 603                     (freq.flr_label & IPV6_FLOWLABEL_STATELESS_FLAG))
 604                         return -ERANGE;
 605
 606                 fl = fl_create(net, sk, &freq, optval, optlen, &err);
 607                 if (!fl)
 608                         return err;
 609                 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
 610
 611                 if (freq.flr_label) {
 612                         err = -EEXIST;
 613                         rcu_read_lock_bh();
 614                         for_each_sk_fl_rcu(np, sfl) {
 615                                 if (sfl->fl->label == freq.flr_label) {
 616                                         if (freq.flr_flags&IPV6_FL_F_EXCL) {
 617                                                 rcu_read_unlock_bh();
 618                                                 goto done;
 619                                         }
 620                                         fl1 = sfl->fl;
 621                                         atomic_inc(&fl1->users);
 622                                         break;
 623                                 }
 624                         }
 625                         rcu_read_unlock_bh();
 626
 627                         if (!fl1)
 628                                 fl1 = fl_lookup(net, freq.flr_label);
 629                         if (fl1) {
 630 recheck:
 631                                 err = -EEXIST;
 632                                 if (freq.flr_flags&IPV6_FL_F_EXCL)
 633                                         goto release;
 634                                 err = -EPERM;
 635                                 if (fl1->share == IPV6_FL_S_EXCL ||
 636                                     fl1->share != fl->share ||
 637                                     ((fl1->share == IPV6_FL_S_PROCESS) &&
 638                                      (fl1->owner.pid != fl->owner.pid)) ||
 639                                     ((fl1->share == IPV6_FL_S_USER) &&
 640                                      !uid_eq(fl1->owner.uid, fl->owner.uid)))
 641                                         goto release;
 642
 643                                 err = -ENOMEM;
 644                                 if (!sfl1)
 645                                         goto release;
 646                                 if (fl->linger > fl1->linger)
 647                                         fl1->linger = fl->linger;
 648                                 if ((long)(fl->expires - fl1->expires) > 0)
 649                                         fl1->expires = fl->expires;
 650                                 fl_link(np, sfl1, fl1);
 651                                 fl_free(fl);
 652                                 return 0;
 653
 654 release:
 655                                 fl_release(fl1);
 656                                 goto done;
 657                         }
 658                 }
 659                 err = -ENOENT;
 660                 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
 661                         goto done;
 662
 663                 err = -ENOMEM;
 664                 if (!sfl1)
 665                         goto done;
 666
 667                 err = mem_check(sk);
 668                 if (err != 0)
 669                         goto done;
 670
 671                 fl1 = fl_intern(net, fl, freq.flr_label);
 672                 if (fl1)
 673                         goto recheck;
 674
 675                 if (!freq.flr_label) {
 676                         if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
 677                                          &fl->label, sizeof(fl->label))) {
 678                                 /* Intentionally ignore fault. */
 679                         }
 680                 }
 681
 682                 fl_link(np, sfl1, fl);
 683                 return 0;
 684
 685         default:
 686                 return -EINVAL;
 687         }
 688
 689 done:
 690         fl_free(fl);
 691         kfree(sfl1);
 692         return err;
 693 }
 694
 695 #ifdef CONFIG_PROC_FS
 696
 697 struct ip6fl_iter_state {
 698         struct seq_net_private p;
 699         struct pid_namespace *pid_ns;
 700         int bucket;
 701 };
 702
 703 #define ip6fl_seq_private(seq)  ((struct ip6fl_iter_state *)(seq)->private)
 704
 705 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
 706 {
 707         struct ip6_flowlabel *fl = NULL;
 708         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 709         struct net *net = seq_file_net(seq);
 710
 711         for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
 712                 for_each_fl_rcu(state->bucket, fl) {
 713                         if (net_eq(fl->fl_net, net))
 714                                 goto out;
 715                 }
 716         }
 717         fl = NULL;
 718 out:
 719         return fl;
 720 }
 721
 722 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
 723 {
 724         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 725         struct net *net = seq_file_net(seq);
 726
 727         for_each_fl_continue_rcu(fl) {
 728                 if (net_eq(fl->fl_net, net))
 729                         goto out;
 730         }
 731
 732 try_again:
 733         if (++state->bucket <= FL_HASH_MASK) {
 734                 for_each_fl_rcu(state->bucket, fl) {
 735                         if (net_eq(fl->fl_net, net))
 736                                 goto out;
 737                 }
 738                 goto try_again;
 739         }
 740         fl = NULL;
 741
 742 out:
 743         return fl;
 744 }
 745
 746 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
 747 {
 748         struct ip6_flowlabel *fl = ip6fl_get_first(seq);
 749         if (fl)
 750                 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
 751                         --pos;
 752         return pos ? NULL : fl;
 753 }
 754
 755 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
 756         __acquires(RCU)
 757 {
 758         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 759
 760         state->pid_ns = proc_pid_ns(file_inode(seq->file));
 761
 762         rcu_read_lock_bh();
 763         return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 764 }
 765
 766 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 767 {
 768         struct ip6_flowlabel *fl;
 769
 770         if (v == SEQ_START_TOKEN)
 771                 fl = ip6fl_get_first(seq);
 772         else
 773                 fl = ip6fl_get_next(seq, v);
 774         ++*pos;
 775         return fl;
 776 }
 777
 778 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
 779         __releases(RCU)
 780 {
 781         rcu_read_unlock_bh();
 782 }
 783
 784 static int ip6fl_seq_show(struct seq_file *seq, void *v)
 785 {
 786         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 787         if (v == SEQ_START_TOKEN) {
 788                 seq_puts(seq, "Label S Owner  Users  Linger Expires  Dst                              Opt\n");
 789         } else {
 790                 struct ip6_flowlabel *fl = v;
 791                 seq_printf(seq,
 792                            "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
 793                            (unsigned int)ntohl(fl->label),
 794                            fl->share,
 795                            ((fl->share == IPV6_FL_S_PROCESS) ?
 796                             pid_nr_ns(fl->owner.pid, state->pid_ns) :
 797                             ((fl->share == IPV6_FL_S_USER) ?
 798                              from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
 799                              0)),
 800                            atomic_read(&fl->users),
 801                            fl->linger/HZ,
 802                            (long)(fl->expires - jiffies)/HZ,
 803                            &fl->dst,
 804                            fl->opt ? fl->opt->opt_nflen : 0);
 805         }
 806         return 0;
 807 }
 808
 809 static const struct seq_operations ip6fl_seq_ops = {
 810         .start  =       ip6fl_seq_start,
 811         .next   =       ip6fl_seq_next,
 812         .stop   =       ip6fl_seq_stop,
 813         .show   =       ip6fl_seq_show,
 814 };
 815
 816 static int __net_init ip6_flowlabel_proc_init(struct net *net)
 817 {
 818         if (!proc_create_net("ip6_flowlabel", 0444, net->proc_net,
 819                         &ip6fl_seq_ops, sizeof(struct ip6fl_iter_state)))
 820                 return -ENOMEM;
 821         return 0;
 822 }
 823
 824 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
 825 {
 826         remove_proc_entry("ip6_flowlabel", net->proc_net);
 827 }
 828 #else
 829 static inline int ip6_flowlabel_proc_init(struct net *net)
 830 {
 831         return 0;
 832 }
 833 static inline void ip6_flowlabel_proc_fini(struct net *net)
 834 {
 835 }
 836 #endif
 837
 838 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
 839 {
 840         ip6_fl_purge(net);
 841         ip6_flowlabel_proc_fini(net);
 842 }
 843
 844 static struct pernet_operations ip6_flowlabel_net_ops = {
 845         .init = ip6_flowlabel_proc_init,
 846         .exit = ip6_flowlabel_net_exit,
 847 };
 848
 849 int ip6_flowlabel_init(void)
 850 {
 851         return register_pernet_subsys(&ip6_flowlabel_net_ops);
 852 }
 853
 854 void ip6_flowlabel_cleanup(void)
 855 {
 856         del_timer(&ip6_fl_gc_timer);
 857         unregister_pernet_subsys(&ip6_flowlabel_net_ops);
 858 }