net/ipv6/ip6_flowlabel.c

   1 /*
   2  *      ip6_flowlabel.c         IPv6 flowlabel manager.
   3  *
   4  *      This program is free software; you can redistribute it and/or
   5  *      modify it under the terms of the GNU General Public License
   6  *      as published by the Free Software Foundation; either version
   7  *      2 of the License, or (at your option) any later version.
   8  *
   9  *      Authors:        Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  */
  11
  12 #include <linux/capability.h>
  13 #include <linux/errno.h>
  14 #include <linux/types.h>
  15 #include <linux/socket.h>
  16 #include <linux/net.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/in6.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/seq_file.h>
  21 #include <linux/slab.h>
  22 #include <linux/export.h>
  23 #include <linux/pid_namespace.h>
  24
  25 #include <net/net_namespace.h>
  26 #include <net/sock.h>
  27
  28 #include <net/ipv6.h>
  29 #include <net/rawv6.h>
  30 #include <net/transp_v6.h>
  31
  32 #include <linux/uaccess.h>
  33
  34 #define FL_MIN_LINGER   6       /* Minimal linger. It is set to 6sec specified
  35                                    in old IPv6 RFC. Well, it was reasonable value.
  36                                  */
  37 #define FL_MAX_LINGER   150     /* Maximal linger timeout */
  38
  39 /* FL hash table */
  40
  41 #define FL_MAX_PER_SOCK 32
  42 #define FL_MAX_SIZE     4096
  43 #define FL_HASH_MASK    255
  44 #define FL_HASH(l)      (ntohl(l)&FL_HASH_MASK)
  45
  46 static atomic_t fl_size = ATOMIC_INIT(0);
  47 static struct ip6_flowlabel __rcu *fl_ht[FL_HASH_MASK+1];
  48
  49 static void ip6_fl_gc(unsigned long dummy);
  50 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc, 0, 0);
  51
  52 /* FL hash table lock: it protects only of GC */
  53
  54 static DEFINE_SPINLOCK(ip6_fl_lock);
  55
  56 /* Big socket sock */
  57
  58 static DEFINE_SPINLOCK(ip6_sk_fl_lock);
  59
  60 #define for_each_fl_rcu(hash, fl)                               \
  61         for (fl = rcu_dereference_bh(fl_ht[(hash)]);            \
  62              fl != NULL;                                        \
  63              fl = rcu_dereference_bh(fl->next))
  64 #define for_each_fl_continue_rcu(fl)                            \
  65         for (fl = rcu_dereference_bh(fl->next);                 \
  66              fl != NULL;                                        \
  67              fl = rcu_dereference_bh(fl->next))
  68
  69 #define for_each_sk_fl_rcu(np, sfl)                             \
  70         for (sfl = rcu_dereference_bh(np->ipv6_fl_list);        \
  71              sfl != NULL;                                       \
  72              sfl = rcu_dereference_bh(sfl->next))
  73
  74 static inline struct ip6_flowlabel *__fl_lookup(struct net *net, __be32 label)
  75 {
  76         struct ip6_flowlabel *fl;
  77
  78         for_each_fl_rcu(FL_HASH(label), fl) {
  79                 if (fl->label == label && net_eq(fl->fl_net, net))
  80                         return fl;
  81         }
  82         return NULL;
  83 }
  84
  85 static struct ip6_flowlabel *fl_lookup(struct net *net, __be32 label)
  86 {
  87         struct ip6_flowlabel *fl;
  88
  89         rcu_read_lock_bh();
  90         fl = __fl_lookup(net, label);
  91         if (fl && !atomic_inc_not_zero(&fl->users))
  92                 fl = NULL;
  93         rcu_read_unlock_bh();
  94         return fl;
  95 }
  96
  97 static void fl_free_rcu(struct rcu_head *head)
  98 {
  99         struct ip6_flowlabel *fl = container_of(head, struct ip6_flowlabel, rcu);
 100
 101         if (fl->share == IPV6_FL_S_PROCESS)
 102                 put_pid(fl->owner.pid);
 103         kfree(fl->opt);
 104         kfree(fl);
 105 }
 106
 107
 108 static void fl_free(struct ip6_flowlabel *fl)
 109 {
 110         if (fl)
 111                 call_rcu(&fl->rcu, fl_free_rcu);
 112 }
 113
 114 static void fl_release(struct ip6_flowlabel *fl)
 115 {
 116         spin_lock_bh(&ip6_fl_lock);
 117
 118         fl->lastuse = jiffies;
 119         if (atomic_dec_and_test(&fl->users)) {
 120                 unsigned long ttd = fl->lastuse + fl->linger;
 121                 if (time_after(ttd, fl->expires))
 122                         fl->expires = ttd;
 123                 ttd = fl->expires;
 124                 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
 125                         struct ipv6_txoptions *opt = fl->opt;
 126                         fl->opt = NULL;
 127                         kfree(opt);
 128                 }
 129                 if (!timer_pending(&ip6_fl_gc_timer) ||
 130                     time_after(ip6_fl_gc_timer.expires, ttd))
 131                         mod_timer(&ip6_fl_gc_timer, ttd);
 132         }
 133         spin_unlock_bh(&ip6_fl_lock);
 134 }
 135
 136 static void ip6_fl_gc(unsigned long dummy)
 137 {
 138         int i;
 139         unsigned long now = jiffies;
 140         unsigned long sched = 0;
 141
 142         spin_lock(&ip6_fl_lock);
 143
 144         for (i = 0; i <= FL_HASH_MASK; i++) {
 145                 struct ip6_flowlabel *fl;
 146                 struct ip6_flowlabel __rcu **flp;
 147
 148                 flp = &fl_ht[i];
 149                 while ((fl = rcu_dereference_protected(*flp,
 150                                                        lockdep_is_held(&ip6_fl_lock))) != NULL) {
 151                         if (atomic_read(&fl->users) == 0) {
 152                                 unsigned long ttd = fl->lastuse + fl->linger;
 153                                 if (time_after(ttd, fl->expires))
 154                                         fl->expires = ttd;
 155                                 ttd = fl->expires;
 156                                 if (time_after_eq(now, ttd)) {
 157                                         *flp = fl->next;
 158                                         fl_free(fl);
 159                                         atomic_dec(&fl_size);
 160                                         continue;
 161                                 }
 162                                 if (!sched || time_before(ttd, sched))
 163                                         sched = ttd;
 164                         }
 165                         flp = &fl->next;
 166                 }
 167         }
 168         if (!sched && atomic_read(&fl_size))
 169                 sched = now + FL_MAX_LINGER;
 170         if (sched) {
 171                 mod_timer(&ip6_fl_gc_timer, sched);
 172         }
 173         spin_unlock(&ip6_fl_lock);
 174 }
 175
 176 static void __net_exit ip6_fl_purge(struct net *net)
 177 {
 178         int i;
 179
 180         spin_lock_bh(&ip6_fl_lock);
 181         for (i = 0; i <= FL_HASH_MASK; i++) {
 182                 struct ip6_flowlabel *fl;
 183                 struct ip6_flowlabel __rcu **flp;
 184
 185                 flp = &fl_ht[i];
 186                 while ((fl = rcu_dereference_protected(*flp,
 187                                                        lockdep_is_held(&ip6_fl_lock))) != NULL) {
 188                         if (net_eq(fl->fl_net, net) &&
 189                             atomic_read(&fl->users) == 0) {
 190                                 *flp = fl->next;
 191                                 fl_free(fl);
 192                                 atomic_dec(&fl_size);
 193                                 continue;
 194                         }
 195                         flp = &fl->next;
 196                 }
 197         }
 198         spin_unlock_bh(&ip6_fl_lock);
 199 }
 200
 201 static struct ip6_flowlabel *fl_intern(struct net *net,
 202                                        struct ip6_flowlabel *fl, __be32 label)
 203 {
 204         struct ip6_flowlabel *lfl;
 205
 206         fl->label = label & IPV6_FLOWLABEL_MASK;
 207
 208         spin_lock_bh(&ip6_fl_lock);
 209         if (label == 0) {
 210                 for (;;) {
 211                         fl->label = htonl(prandom_u32())&IPV6_FLOWLABEL_MASK;
 212                         if (fl->label) {
 213                                 lfl = __fl_lookup(net, fl->label);
 214                                 if (!lfl)
 215                                         break;
 216                         }
 217                 }
 218         } else {
 219                 /*
 220                  * we dropper the ip6_fl_lock, so this entry could reappear
 221                  * and we need to recheck with it.
 222                  *
 223                  * OTOH no need to search the active socket first, like it is
 224                  * done in ipv6_flowlabel_opt - sock is locked, so new entry
 225                  * with the same label can only appear on another sock
 226                  */
 227                 lfl = __fl_lookup(net, fl->label);
 228                 if (lfl) {
 229                         atomic_inc(&lfl->users);
 230                         spin_unlock_bh(&ip6_fl_lock);
 231                         return lfl;
 232                 }
 233         }
 234
 235         fl->lastuse = jiffies;
 236         fl->next = fl_ht[FL_HASH(fl->label)];
 237         rcu_assign_pointer(fl_ht[FL_HASH(fl->label)], fl);
 238         atomic_inc(&fl_size);
 239         spin_unlock_bh(&ip6_fl_lock);
 240         return NULL;
 241 }
 242
 243
 244
 245 /* Socket flowlabel lists */
 246
 247 struct ip6_flowlabel *fl6_sock_lookup(struct sock *sk, __be32 label)
 248 {
 249         struct ipv6_fl_socklist *sfl;
 250         struct ipv6_pinfo *np = inet6_sk(sk);
 251
 252         label &= IPV6_FLOWLABEL_MASK;
 253
 254         rcu_read_lock_bh();
 255         for_each_sk_fl_rcu(np, sfl) {
 256                 struct ip6_flowlabel *fl = sfl->fl;
 257
 258                 if (fl->label == label && atomic_inc_not_zero(&fl->users)) {
 259                         fl->lastuse = jiffies;
 260                         rcu_read_unlock_bh();
 261                         return fl;
 262                 }
 263         }
 264         rcu_read_unlock_bh();
 265         return NULL;
 266 }
 267 EXPORT_SYMBOL_GPL(fl6_sock_lookup);
 268
 269 void fl6_free_socklist(struct sock *sk)
 270 {
 271         struct ipv6_pinfo *np = inet6_sk(sk);
 272         struct ipv6_fl_socklist *sfl;
 273
 274         if (!rcu_access_pointer(np->ipv6_fl_list))
 275                 return;
 276
 277         spin_lock_bh(&ip6_sk_fl_lock);
 278         while ((sfl = rcu_dereference_protected(np->ipv6_fl_list,
 279                                                 lockdep_is_held(&ip6_sk_fl_lock))) != NULL) {
 280                 np->ipv6_fl_list = sfl->next;
 281                 spin_unlock_bh(&ip6_sk_fl_lock);
 282
 283                 fl_release(sfl->fl);
 284                 kfree_rcu(sfl, rcu);
 285
 286                 spin_lock_bh(&ip6_sk_fl_lock);
 287         }
 288         spin_unlock_bh(&ip6_sk_fl_lock);
 289 }
 290
 291 /* Service routines */
 292
 293
 294 /*
 295    It is the only difficult place. flowlabel enforces equal headers
 296    before and including routing header, however user may supply options
 297    following rthdr.
 298  */
 299
 300 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
 301                                          struct ip6_flowlabel *fl,
 302                                          struct ipv6_txoptions *fopt)
 303 {
 304         struct ipv6_txoptions *fl_opt = fl->opt;
 305
 306         if (!fopt || fopt->opt_flen == 0)
 307                 return fl_opt;
 308
 309         if (fl_opt) {
 310                 opt_space->hopopt = fl_opt->hopopt;
 311                 opt_space->dst0opt = fl_opt->dst0opt;
 312                 opt_space->srcrt = fl_opt->srcrt;
 313                 opt_space->opt_nflen = fl_opt->opt_nflen;
 314         } else {
 315                 if (fopt->opt_nflen == 0)
 316                         return fopt;
 317                 opt_space->hopopt = NULL;
 318                 opt_space->dst0opt = NULL;
 319                 opt_space->srcrt = NULL;
 320                 opt_space->opt_nflen = 0;
 321         }
 322         opt_space->dst1opt = fopt->dst1opt;
 323         opt_space->opt_flen = fopt->opt_flen;
 324         opt_space->tot_len = fopt->tot_len;
 325         return opt_space;
 326 }
 327 EXPORT_SYMBOL_GPL(fl6_merge_options);
 328
 329 static unsigned long check_linger(unsigned long ttl)
 330 {
 331         if (ttl < FL_MIN_LINGER)
 332                 return FL_MIN_LINGER*HZ;
 333         if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
 334                 return 0;
 335         return ttl*HZ;
 336 }
 337
 338 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
 339 {
 340         linger = check_linger(linger);
 341         if (!linger)
 342                 return -EPERM;
 343         expires = check_linger(expires);
 344         if (!expires)
 345                 return -EPERM;
 346
 347         spin_lock_bh(&ip6_fl_lock);
 348         fl->lastuse = jiffies;
 349         if (time_before(fl->linger, linger))
 350                 fl->linger = linger;
 351         if (time_before(expires, fl->linger))
 352                 expires = fl->linger;
 353         if (time_before(fl->expires, fl->lastuse + expires))
 354                 fl->expires = fl->lastuse + expires;
 355         spin_unlock_bh(&ip6_fl_lock);
 356
 357         return 0;
 358 }
 359
 360 static struct ip6_flowlabel *
 361 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
 362           char __user *optval, int optlen, int *err_p)
 363 {
 364         struct ip6_flowlabel *fl = NULL;
 365         int olen;
 366         int addr_type;
 367         int err;
 368
 369         olen = optlen - CMSG_ALIGN(sizeof(*freq));
 370         err = -EINVAL;
 371         if (olen > 64 * 1024)
 372                 goto done;
 373
 374         err = -ENOMEM;
 375         fl = kzalloc(sizeof(*fl), GFP_KERNEL);
 376         if (!fl)
 377                 goto done;
 378
 379         if (olen > 0) {
 380                 struct msghdr msg;
 381                 struct flowi6 flowi6;
 382                 struct sockcm_cookie sockc_junk;
 383                 struct ipcm6_cookie ipc6;
 384
 385                 err = -ENOMEM;
 386                 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
 387                 if (!fl->opt)
 388                         goto done;
 389
 390                 memset(fl->opt, 0, sizeof(*fl->opt));
 391                 fl->opt->tot_len = sizeof(*fl->opt) + olen;
 392                 err = -EFAULT;
 393                 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
 394                         goto done;
 395
 396                 msg.msg_controllen = olen;
 397                 msg.msg_control = (void *)(fl->opt+1);
 398                 memset(&flowi6, 0, sizeof(flowi6));
 399
 400                 ipc6.opt = fl->opt;
 401                 err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, &ipc6, &sockc_junk);
 402                 if (err)
 403                         goto done;
 404                 err = -EINVAL;
 405                 if (fl->opt->opt_flen)
 406                         goto done;
 407                 if (fl->opt->opt_nflen == 0) {
 408                         kfree(fl->opt);
 409                         fl->opt = NULL;
 410                 }
 411         }
 412
 413         fl->fl_net = net;
 414         fl->expires = jiffies;
 415         err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
 416         if (err)
 417                 goto done;
 418         fl->share = freq->flr_share;
 419         addr_type = ipv6_addr_type(&freq->flr_dst);
 420         if ((addr_type & IPV6_ADDR_MAPPED) ||
 421             addr_type == IPV6_ADDR_ANY) {
 422                 err = -EINVAL;
 423                 goto done;
 424         }
 425         fl->dst = freq->flr_dst;
 426         atomic_set(&fl->users, 1);
 427         switch (fl->share) {
 428         case IPV6_FL_S_EXCL:
 429         case IPV6_FL_S_ANY:
 430                 break;
 431         case IPV6_FL_S_PROCESS:
 432                 fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
 433                 break;
 434         case IPV6_FL_S_USER:
 435                 fl->owner.uid = current_euid();
 436                 break;
 437         default:
 438                 err = -EINVAL;
 439                 goto done;
 440         }
 441         return fl;
 442
 443 done:
 444         fl_free(fl);
 445         *err_p = err;
 446         return NULL;
 447 }
 448
 449 static int mem_check(struct sock *sk)
 450 {
 451         struct ipv6_pinfo *np = inet6_sk(sk);
 452         struct ipv6_fl_socklist *sfl;
 453         int room = FL_MAX_SIZE - atomic_read(&fl_size);
 454         int count = 0;
 455
 456         if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
 457                 return 0;
 458
 459         rcu_read_lock_bh();
 460         for_each_sk_fl_rcu(np, sfl)
 461                 count++;
 462         rcu_read_unlock_bh();
 463
 464         if (room <= 0 ||
 465             ((count >= FL_MAX_PER_SOCK ||
 466               (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
 467              !capable(CAP_NET_ADMIN)))
 468                 return -ENOBUFS;
 469
 470         return 0;
 471 }
 472
 473 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
 474                 struct ip6_flowlabel *fl)
 475 {
 476         spin_lock_bh(&ip6_sk_fl_lock);
 477         sfl->fl = fl;
 478         sfl->next = np->ipv6_fl_list;
 479         rcu_assign_pointer(np->ipv6_fl_list, sfl);
 480         spin_unlock_bh(&ip6_sk_fl_lock);
 481 }
 482
 483 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
 484                            int flags)
 485 {
 486         struct ipv6_pinfo *np = inet6_sk(sk);
 487         struct ipv6_fl_socklist *sfl;
 488
 489         if (flags & IPV6_FL_F_REMOTE) {
 490                 freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
 491                 return 0;
 492         }
 493
 494         if (np->repflow) {
 495                 freq->flr_label = np->flow_label;
 496                 return 0;
 497         }
 498
 499         rcu_read_lock_bh();
 500
 501         for_each_sk_fl_rcu(np, sfl) {
 502                 if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
 503                         spin_lock_bh(&ip6_fl_lock);
 504                         freq->flr_label = sfl->fl->label;
 505                         freq->flr_dst = sfl->fl->dst;
 506                         freq->flr_share = sfl->fl->share;
 507                         freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
 508                         freq->flr_linger = sfl->fl->linger / HZ;
 509
 510                         spin_unlock_bh(&ip6_fl_lock);
 511                         rcu_read_unlock_bh();
 512                         return 0;
 513                 }
 514         }
 515         rcu_read_unlock_bh();
 516
 517         return -ENOENT;
 518 }
 519
 520 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
 521 {
 522         int err;
 523         struct net *net = sock_net(sk);
 524         struct ipv6_pinfo *np = inet6_sk(sk);
 525         struct in6_flowlabel_req freq;
 526         struct ipv6_fl_socklist *sfl1 = NULL;
 527         struct ipv6_fl_socklist *sfl;
 528         struct ipv6_fl_socklist __rcu **sflp;
 529         struct ip6_flowlabel *fl, *fl1 = NULL;
 530
 531
 532         if (optlen < sizeof(freq))
 533                 return -EINVAL;
 534
 535         if (copy_from_user(&freq, optval, sizeof(freq)))
 536                 return -EFAULT;
 537
 538         switch (freq.flr_action) {
 539         case IPV6_FL_A_PUT:
 540                 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
 541                         if (sk->sk_protocol != IPPROTO_TCP)
 542                                 return -ENOPROTOOPT;
 543                         if (!np->repflow)
 544                                 return -ESRCH;
 545                         np->flow_label = 0;
 546                         np->repflow = 0;
 547                         return 0;
 548                 }
 549                 spin_lock_bh(&ip6_sk_fl_lock);
 550                 for (sflp = &np->ipv6_fl_list;
 551                      (sfl = rcu_dereference_protected(*sflp,
 552                                                       lockdep_is_held(&ip6_sk_fl_lock))) != NULL;
 553                      sflp = &sfl->next) {
 554                         if (sfl->fl->label == freq.flr_label) {
 555                                 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
 556                                         np->flow_label &= ~IPV6_FLOWLABEL_MASK;
 557                                 *sflp = sfl->next;
 558                                 spin_unlock_bh(&ip6_sk_fl_lock);
 559                                 fl_release(sfl->fl);
 560                                 kfree_rcu(sfl, rcu);
 561                                 return 0;
 562                         }
 563                 }
 564                 spin_unlock_bh(&ip6_sk_fl_lock);
 565                 return -ESRCH;
 566
 567         case IPV6_FL_A_RENEW:
 568                 rcu_read_lock_bh();
 569                 for_each_sk_fl_rcu(np, sfl) {
 570                         if (sfl->fl->label == freq.flr_label) {
 571                                 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
 572                                 rcu_read_unlock_bh();
 573                                 return err;
 574                         }
 575                 }
 576                 rcu_read_unlock_bh();
 577
 578                 if (freq.flr_share == IPV6_FL_S_NONE &&
 579                     ns_capable(net->user_ns, CAP_NET_ADMIN)) {
 580                         fl = fl_lookup(net, freq.flr_label);
 581                         if (fl) {
 582                                 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
 583                                 fl_release(fl);
 584                                 return err;
 585                         }
 586                 }
 587                 return -ESRCH;
 588
 589         case IPV6_FL_A_GET:
 590                 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
 591                         struct net *net = sock_net(sk);
 592                         if (net->ipv6.sysctl.flowlabel_consistency) {
 593                                 net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
 594                                 return -EPERM;
 595                         }
 596
 597                         if (sk->sk_protocol != IPPROTO_TCP)
 598                                 return -ENOPROTOOPT;
 599
 600                         np->repflow = 1;
 601                         return 0;
 602                 }
 603
 604                 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
 605                         return -EINVAL;
 606
 607                 if (net->ipv6.sysctl.flowlabel_state_ranges &&
 608                     (freq.flr_label & IPV6_FLOWLABEL_STATELESS_FLAG))
 609                         return -ERANGE;
 610
 611                 fl = fl_create(net, sk, &freq, optval, optlen, &err);
 612                 if (!fl)
 613                         return err;
 614                 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
 615
 616                 if (freq.flr_label) {
 617                         err = -EEXIST;
 618                         rcu_read_lock_bh();
 619                         for_each_sk_fl_rcu(np, sfl) {
 620                                 if (sfl->fl->label == freq.flr_label) {
 621                                         if (freq.flr_flags&IPV6_FL_F_EXCL) {
 622                                                 rcu_read_unlock_bh();
 623                                                 goto done;
 624                                         }
 625                                         fl1 = sfl->fl;
 626                                         if (!atomic_inc_not_zero(&fl1->users))
 627                                                 fl1 = NULL;
 628                                         break;
 629                                 }
 630                         }
 631                         rcu_read_unlock_bh();
 632
 633                         if (!fl1)
 634                                 fl1 = fl_lookup(net, freq.flr_label);
 635                         if (fl1) {
 636 recheck:
 637                                 err = -EEXIST;
 638                                 if (freq.flr_flags&IPV6_FL_F_EXCL)
 639                                         goto release;
 640                                 err = -EPERM;
 641                                 if (fl1->share == IPV6_FL_S_EXCL ||
 642                                     fl1->share != fl->share ||
 643                                     ((fl1->share == IPV6_FL_S_PROCESS) &&
 644                                      (fl1->owner.pid != fl->owner.pid)) ||
 645                                     ((fl1->share == IPV6_FL_S_USER) &&
 646                                      !uid_eq(fl1->owner.uid, fl->owner.uid)))
 647                                         goto release;
 648
 649                                 err = -ENOMEM;
 650                                 if (!sfl1)
 651                                         goto release;
 652                                 if (fl->linger > fl1->linger)
 653                                         fl1->linger = fl->linger;
 654                                 if ((long)(fl->expires - fl1->expires) > 0)
 655                                         fl1->expires = fl->expires;
 656                                 fl_link(np, sfl1, fl1);
 657                                 fl_free(fl);
 658                                 return 0;
 659
 660 release:
 661                                 fl_release(fl1);
 662                                 goto done;
 663                         }
 664                 }
 665                 err = -ENOENT;
 666                 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
 667                         goto done;
 668
 669                 err = -ENOMEM;
 670                 if (!sfl1)
 671                         goto done;
 672
 673                 err = mem_check(sk);
 674                 if (err != 0)
 675                         goto done;
 676
 677                 fl1 = fl_intern(net, fl, freq.flr_label);
 678                 if (fl1)
 679                         goto recheck;
 680
 681                 if (!freq.flr_label) {
 682                         if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
 683                                          &fl->label, sizeof(fl->label))) {
 684                                 /* Intentionally ignore fault. */
 685                         }
 686                 }
 687
 688                 fl_link(np, sfl1, fl);
 689                 return 0;
 690
 691         default:
 692                 return -EINVAL;
 693         }
 694
 695 done:
 696         fl_free(fl);
 697         kfree(sfl1);
 698         return err;
 699 }
 700
 701 #ifdef CONFIG_PROC_FS
 702
 703 struct ip6fl_iter_state {
 704         struct seq_net_private p;
 705         struct pid_namespace *pid_ns;
 706         int bucket;
 707 };
 708
 709 #define ip6fl_seq_private(seq)  ((struct ip6fl_iter_state *)(seq)->private)
 710
 711 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
 712 {
 713         struct ip6_flowlabel *fl = NULL;
 714         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 715         struct net *net = seq_file_net(seq);
 716
 717         for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
 718                 for_each_fl_rcu(state->bucket, fl) {
 719                         if (net_eq(fl->fl_net, net))
 720                                 goto out;
 721                 }
 722         }
 723         fl = NULL;
 724 out:
 725         return fl;
 726 }
 727
 728 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
 729 {
 730         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 731         struct net *net = seq_file_net(seq);
 732
 733         for_each_fl_continue_rcu(fl) {
 734                 if (net_eq(fl->fl_net, net))
 735                         goto out;
 736         }
 737
 738 try_again:
 739         if (++state->bucket <= FL_HASH_MASK) {
 740                 for_each_fl_rcu(state->bucket, fl) {
 741                         if (net_eq(fl->fl_net, net))
 742                                 goto out;
 743                 }
 744                 goto try_again;
 745         }
 746         fl = NULL;
 747
 748 out:
 749         return fl;
 750 }
 751
 752 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
 753 {
 754         struct ip6_flowlabel *fl = ip6fl_get_first(seq);
 755         if (fl)
 756                 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
 757                         --pos;
 758         return pos ? NULL : fl;
 759 }
 760
 761 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
 762         __acquires(RCU)
 763 {
 764         rcu_read_lock_bh();
 765         return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 766 }
 767
 768 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 769 {
 770         struct ip6_flowlabel *fl;
 771
 772         if (v == SEQ_START_TOKEN)
 773                 fl = ip6fl_get_first(seq);
 774         else
 775                 fl = ip6fl_get_next(seq, v);
 776         ++*pos;
 777         return fl;
 778 }
 779
 780 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
 781         __releases(RCU)
 782 {
 783         rcu_read_unlock_bh();
 784 }
 785
 786 static int ip6fl_seq_show(struct seq_file *seq, void *v)
 787 {
 788         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 789         if (v == SEQ_START_TOKEN) {
 790                 seq_puts(seq, "Label S Owner  Users  Linger Expires  Dst                              Opt\n");
 791         } else {
 792                 struct ip6_flowlabel *fl = v;
 793                 seq_printf(seq,
 794                            "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
 795                            (unsigned int)ntohl(fl->label),
 796                            fl->share,
 797                            ((fl->share == IPV6_FL_S_PROCESS) ?
 798                             pid_nr_ns(fl->owner.pid, state->pid_ns) :
 799                             ((fl->share == IPV6_FL_S_USER) ?
 800                              from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
 801                              0)),
 802                            atomic_read(&fl->users),
 803                            fl->linger/HZ,
 804                            (long)(fl->expires - jiffies)/HZ,
 805                            &fl->dst,
 806                            fl->opt ? fl->opt->opt_nflen : 0);
 807         }
 808         return 0;
 809 }
 810
 811 static const struct seq_operations ip6fl_seq_ops = {
 812         .start  =       ip6fl_seq_start,
 813         .next   =       ip6fl_seq_next,
 814         .stop   =       ip6fl_seq_stop,
 815         .show   =       ip6fl_seq_show,
 816 };
 817
 818 static int ip6fl_seq_open(struct inode *inode, struct file *file)
 819 {
 820         struct seq_file *seq;
 821         struct ip6fl_iter_state *state;
 822         int err;
 823
 824         err = seq_open_net(inode, file, &ip6fl_seq_ops,
 825                            sizeof(struct ip6fl_iter_state));
 826
 827         if (!err) {
 828                 seq = file->private_data;
 829                 state = ip6fl_seq_private(seq);
 830                 rcu_read_lock();
 831                 state->pid_ns = get_pid_ns(task_active_pid_ns(current));
 832                 rcu_read_unlock();
 833         }
 834         return err;
 835 }
 836
 837 static int ip6fl_seq_release(struct inode *inode, struct file *file)
 838 {
 839         struct seq_file *seq = file->private_data;
 840         struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
 841         put_pid_ns(state->pid_ns);
 842         return seq_release_net(inode, file);
 843 }
 844
 845 static const struct file_operations ip6fl_seq_fops = {
 846         .owner          =       THIS_MODULE,
 847         .open           =       ip6fl_seq_open,
 848         .read           =       seq_read,
 849         .llseek         =       seq_lseek,
 850         .release        =       ip6fl_seq_release,
 851 };
 852
 853 static int __net_init ip6_flowlabel_proc_init(struct net *net)
 854 {
 855         if (!proc_create("ip6_flowlabel", S_IRUGO, net->proc_net,
 856                          &ip6fl_seq_fops))
 857                 return -ENOMEM;
 858         return 0;
 859 }
 860
 861 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
 862 {
 863         remove_proc_entry("ip6_flowlabel", net->proc_net);
 864 }
 865 #else
 866 static inline int ip6_flowlabel_proc_init(struct net *net)
 867 {
 868         return 0;
 869 }
 870 static inline void ip6_flowlabel_proc_fini(struct net *net)
 871 {
 872 }
 873 #endif
 874
 875 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
 876 {
 877         ip6_fl_purge(net);
 878         ip6_flowlabel_proc_fini(net);
 879 }
 880
 881 static struct pernet_operations ip6_flowlabel_net_ops = {
 882         .init = ip6_flowlabel_proc_init,
 883         .exit = ip6_flowlabel_net_exit,
 884 };
 885
 886 int ip6_flowlabel_init(void)
 887 {
 888         return register_pernet_subsys(&ip6_flowlabel_net_ops);
 889 }
 890
 891 void ip6_flowlabel_cleanup(void)
 892 {
 893         del_timer(&ip6_fl_gc_timer);
 894         unregister_pernet_subsys(&ip6_flowlabel_net_ops);
 895 }