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/config.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/if_arp.h>
  19 #include <linux/in6.h>
  20 #include <linux/route.h>
  21 #include <linux/proc_fs.h>
  22
  23 #include <net/sock.h>
  24
  25 #include <net/ipv6.h>
  26 #include <net/ndisc.h>
  27 #include <net/protocol.h>
  28 #include <net/ip6_route.h>
  29 #include <net/addrconf.h>
  30 #include <net/rawv6.h>
  31 #include <net/icmp.h>
  32 #include <net/transp_v6.h>
  33
  34 #include <asm/uaccess.h>
  35
  36 #define FL_MIN_LINGER   6       /* Minimal linger. It is set to 6sec specified
  37                                    in old IPv6 RFC. Well, it was reasonable value.
  38                                  */
  39 #define FL_MAX_LINGER   60      /* Maximal linger timeout */
  40
  41 /* FL hash table */
  42
  43 #define FL_MAX_PER_SOCK 32
  44 #define FL_MAX_SIZE     4096
  45 #define FL_HASH_MASK    255
  46 #define FL_HASH(l)      (ntohl(l)&FL_HASH_MASK)
  47
  48 static atomic_t fl_size = ATOMIC_INIT(0);
  49 static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1];
  50
  51 static struct timer_list ip6_fl_gc_timer;
  52
  53 /* FL hash table lock: it protects only of GC */
  54
  55 static rwlock_t ip6_fl_lock = RW_LOCK_UNLOCKED;
  56
  57 /* Big socket sock */
  58
  59 static rwlock_t ip6_sk_fl_lock = RW_LOCK_UNLOCKED;
  60
  61
  62 static __inline__ struct ip6_flowlabel * __fl_lookup(u32 label)
  63 {
  64         struct ip6_flowlabel *fl;
  65
  66         for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) {
  67                 if (fl->label == label)
  68                         return fl;
  69         }
  70         return NULL;
  71 }
  72
  73 static struct ip6_flowlabel * fl_lookup(u32 label)
  74 {
  75         struct ip6_flowlabel *fl;
  76
  77         read_lock_bh(&ip6_fl_lock);
  78         fl = __fl_lookup(label);
  79         if (fl)
  80                 atomic_inc(&fl->users);
  81         read_unlock_bh(&ip6_fl_lock);
  82         return fl;
  83 }
  84
  85
  86 static void fl_free(struct ip6_flowlabel *fl)
  87 {
  88         if (fl->opt)
  89                 kfree(fl->opt);
  90         kfree(fl);
  91 }
  92
  93 static void fl_release(struct ip6_flowlabel *fl)
  94 {
  95         fl->lastuse = jiffies;
  96         if (atomic_dec_and_test(&fl->users)) {
  97                 unsigned long ttd = fl->lastuse + fl->linger;
  98                 if ((long)(ttd - fl->expires) > 0)
  99                         fl->expires = ttd;
 100                 ttd = fl->expires;
 101                 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
 102                         struct ipv6_txoptions *opt = fl->opt;
 103                         fl->opt = NULL;
 104                         kfree(opt);
 105                 }
 106                 if (!del_timer(&ip6_fl_gc_timer) ||
 107                     (long)(ip6_fl_gc_timer.expires - ttd) > 0)
 108                         ip6_fl_gc_timer.expires = ttd;
 109                 add_timer(&ip6_fl_gc_timer);
 110         }
 111 }
 112
 113 static void ip6_fl_gc(unsigned long dummy)
 114 {
 115         int i;
 116         unsigned long now = jiffies;
 117         unsigned long sched = 0;
 118
 119         write_lock(&ip6_fl_lock);
 120
 121         for (i=0; i<=FL_HASH_MASK; i++) {
 122                 struct ip6_flowlabel *fl, **flp;
 123                 flp = &fl_ht[i];
 124                 while ((fl=*flp) != NULL) {
 125                         if (atomic_read(&fl->users) == 0) {
 126                                 unsigned long ttd = fl->lastuse + fl->linger;
 127                                 if ((long)(ttd - fl->expires) > 0)
 128                                         fl->expires = ttd;
 129                                 ttd = fl->expires;
 130                                 if ((long)(now - ttd) >= 0) {
 131                                         *flp = fl->next;
 132                                         fl_free(fl);
 133                                         atomic_dec(&fl_size);
 134                                         continue;
 135                                 }
 136                                 if (!sched || (long)(ttd - sched) < 0)
 137                                         sched = ttd;
 138                         }
 139                         flp = &fl->next;
 140                 }
 141         }
 142         if (!sched && atomic_read(&fl_size))
 143                 sched = now + FL_MAX_LINGER;
 144         if (sched) {
 145                 ip6_fl_gc_timer.expires = sched;
 146                 add_timer(&ip6_fl_gc_timer);
 147         }
 148         write_unlock(&ip6_fl_lock);
 149 }
 150
 151 static int fl_intern(struct ip6_flowlabel *fl, __u32 label)
 152 {
 153         fl->label = label & IPV6_FLOWLABEL_MASK;
 154
 155         write_lock_bh(&ip6_fl_lock);
 156         if (label == 0) {
 157                 for (;;) {
 158                         fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK;
 159                         if (fl->label) {
 160                                 struct ip6_flowlabel *lfl;
 161                                 lfl = __fl_lookup(fl->label);
 162                                 if (lfl == NULL)
 163                                         break;
 164                         }
 165                 }
 166         }
 167
 168         fl->lastuse = jiffies;
 169         fl->next = fl_ht[FL_HASH(fl->label)];
 170         fl_ht[FL_HASH(fl->label)] = fl;
 171         atomic_inc(&fl_size);
 172         write_unlock_bh(&ip6_fl_lock);
 173         return 0;
 174 }
 175
 176
 177
 178 /* Socket flowlabel lists */
 179
 180 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, u32 label)
 181 {
 182         struct ipv6_fl_socklist *sfl;
 183         struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
 184
 185         label &= IPV6_FLOWLABEL_MASK;
 186
 187         for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) {
 188                 struct ip6_flowlabel *fl = sfl->fl;
 189                 if (fl->label == label) {
 190                         fl->lastuse = jiffies;
 191                         atomic_inc(&fl->users);
 192                         return fl;
 193                 }
 194         }
 195         return NULL;
 196 }
 197
 198 void fl6_free_socklist(struct sock *sk)
 199 {
 200         struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
 201         struct ipv6_fl_socklist *sfl;
 202
 203         while ((sfl = np->ipv6_fl_list) != NULL) {
 204                 np->ipv6_fl_list = sfl->next;
 205                 fl_release(sfl->fl);
 206                 kfree(sfl);
 207         }
 208 }
 209
 210 /* Service routines */
 211
 212
 213 /*
 214    It is the only difficult place. flowlabel enforces equal headers
 215    before and including routing header, however user may supply options
 216    following rthdr.
 217  */
 218
 219 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
 220                                          struct ip6_flowlabel * fl,
 221                                          struct ipv6_txoptions * fopt)
 222 {
 223         struct ipv6_txoptions * fl_opt = fl->opt;
 224
 225         if (fopt == NULL || fopt->opt_flen == 0)
 226                 return fl_opt;
 227
 228         if (fl_opt != NULL) {
 229                 opt_space->hopopt = fl_opt->hopopt;
 230                 opt_space->dst0opt = fl_opt->dst0opt;
 231                 opt_space->srcrt = fl_opt->srcrt;
 232                 opt_space->opt_nflen = fl_opt->opt_nflen;
 233         } else {
 234                 if (fopt->opt_nflen == 0)
 235                         return fopt;
 236                 opt_space->hopopt = NULL;
 237                 opt_space->dst0opt = NULL;
 238                 opt_space->srcrt = NULL;
 239                 opt_space->opt_nflen = 0;
 240         }
 241         opt_space->dst1opt = fopt->dst1opt;
 242         opt_space->auth = fopt->auth;
 243         opt_space->opt_flen = fopt->opt_flen;
 244         return opt_space;
 245 }
 246
 247 static __u32 check_linger(__u16 ttl)
 248 {
 249         if (ttl < FL_MIN_LINGER)
 250                 return FL_MIN_LINGER*HZ;
 251         if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
 252                 return 0;
 253         return ttl*HZ;
 254 }
 255
 256 static int fl6_renew(struct ip6_flowlabel *fl, unsigned linger, unsigned expires)
 257 {
 258         linger = check_linger(linger);
 259         if (!linger)
 260                 return -EPERM;
 261         expires = check_linger(expires);
 262         if (!expires)
 263                 return -EPERM;
 264         fl->lastuse = jiffies;
 265         if (fl->linger < linger)
 266                 fl->linger = linger;
 267         if (expires < fl->linger)
 268                 expires = fl->linger;
 269         if ((long)(fl->expires - (fl->lastuse+expires)) < 0)
 270                 fl->expires = fl->lastuse + expires;
 271         return 0;
 272 }
 273
 274 static struct ip6_flowlabel *
 275 fl_create(struct in6_flowlabel_req *freq, char *optval, int optlen, int *err_p)
 276 {
 277         struct ip6_flowlabel *fl;
 278         int olen;
 279         int addr_type;
 280         int err;
 281
 282         err = -ENOMEM;
 283         fl = kmalloc(sizeof(*fl), GFP_KERNEL);
 284         if (fl == NULL)
 285                 goto done;
 286         memset(fl, 0, sizeof(*fl));
 287
 288         olen = optlen - CMSG_ALIGN(sizeof(*freq));
 289         if (olen > 0) {
 290                 struct msghdr msg;
 291                 struct flowi flowi;
 292                 int junk;
 293
 294                 err = -ENOMEM;
 295                 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
 296                 if (fl->opt == NULL)
 297                         goto done;
 298
 299                 memset(fl->opt, 0, sizeof(*fl->opt));
 300                 fl->opt->tot_len = sizeof(*fl->opt) + olen;
 301                 err = -EFAULT;
 302                 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
 303                         goto done;
 304
 305                 msg.msg_controllen = olen;
 306                 msg.msg_control = (void*)(fl->opt+1);
 307                 flowi.oif = 0;
 308
 309                 err = datagram_send_ctl(&msg, &flowi, fl->opt, &junk);
 310                 if (err)
 311                         goto done;
 312                 err = -EINVAL;
 313                 if (fl->opt->opt_flen)
 314                         goto done;
 315                 if (fl->opt->opt_nflen == 0) {
 316                         kfree(fl->opt);
 317                         fl->opt = NULL;
 318                 }
 319         }
 320
 321         fl->expires = jiffies;
 322         err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
 323         if (err)
 324                 goto done;
 325         fl->share = freq->flr_share;
 326         addr_type = ipv6_addr_type(&freq->flr_dst);
 327         if ((addr_type&IPV6_ADDR_MAPPED)
 328             || addr_type == IPV6_ADDR_ANY)
 329                 goto done;
 330         ipv6_addr_copy(&fl->dst, &freq->flr_dst);
 331         atomic_set(&fl->users, 1);
 332         switch (fl->share) {
 333         case IPV6_FL_S_EXCL:
 334         case IPV6_FL_S_ANY:
 335                 break;
 336         case IPV6_FL_S_PROCESS:
 337                 fl->owner = current->pid;
 338                 break;
 339         case IPV6_FL_S_USER:
 340                 fl->owner = current->euid;
 341                 break;
 342         default:
 343                 err = -EINVAL;
 344                 goto done;
 345         }
 346         return fl;
 347
 348 done:
 349         if (fl)
 350                 fl_free(fl);
 351         *err_p = err;
 352         return NULL;
 353 }
 354
 355 static int mem_check(struct sock *sk)
 356 {
 357         struct ipv6_fl_socklist *sfl;
 358         int room = FL_MAX_SIZE - atomic_read(&fl_size);
 359         int count = 0;
 360
 361         if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
 362                 return 0;
 363
 364         for (sfl = sk->net_pinfo.af_inet6.ipv6_fl_list; sfl; sfl = sfl->next)
 365                 count++;
 366
 367         if (room <= 0 ||
 368             ((count >= FL_MAX_PER_SOCK ||
 369              (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4)
 370              && !capable(CAP_NET_ADMIN)))
 371                 return -ENOBUFS;
 372
 373         return 0;
 374 }
 375
 376 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
 377 {
 378         if (h1 == h2)
 379                 return 0;
 380         if (h1 == NULL || h2 == NULL)
 381                 return 1;
 382         if (h1->hdrlen != h2->hdrlen)
 383                 return 1;
 384         return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
 385 }
 386
 387 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
 388 {
 389         if (o1 == o2)
 390                 return 0;
 391         if (o1 == NULL || o2 == NULL)
 392                 return 1;
 393         if (o1->opt_nflen != o2->opt_nflen)
 394                 return 1;
 395         if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
 396                 return 1;
 397         if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
 398                 return 1;
 399         if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
 400                 return 1;
 401         return 0;
 402 }
 403
 404 int ipv6_flowlabel_opt(struct sock *sk, char *optval, int optlen)
 405 {
 406         int err;
 407         struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
 408         struct in6_flowlabel_req freq;
 409         struct ipv6_fl_socklist *sfl1=NULL;
 410         struct ipv6_fl_socklist *sfl, **sflp;
 411         struct ip6_flowlabel *fl;
 412
 413         if (optlen < sizeof(freq))
 414                 return -EINVAL;
 415
 416         if (copy_from_user(&freq, optval, sizeof(freq)))
 417                 return -EFAULT;
 418
 419         switch (freq.flr_action) {
 420         case IPV6_FL_A_PUT:
 421                 write_lock_bh(&ip6_sk_fl_lock);
 422                 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
 423                         if (sfl->fl->label == freq.flr_label) {
 424                                 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
 425                                         np->flow_label &= ~IPV6_FLOWLABEL_MASK;
 426                                 *sflp = sfl->next;
 427                                 write_unlock_bh(&ip6_sk_fl_lock);
 428                                 fl_release(sfl->fl);
 429                                 kfree(sfl);
 430                                 return 0;
 431                         }
 432                 }
 433                 write_unlock_bh(&ip6_sk_fl_lock);
 434                 return -ESRCH;
 435
 436         case IPV6_FL_A_RENEW:
 437                 read_lock_bh(&ip6_sk_fl_lock);
 438                 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
 439                         if (sfl->fl->label == freq.flr_label) {
 440                                 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
 441                                 read_unlock_bh(&ip6_sk_fl_lock);
 442                                 return err;
 443                         }
 444                 }
 445                 read_unlock_bh(&ip6_sk_fl_lock);
 446
 447                 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
 448                         fl = fl_lookup(freq.flr_label);
 449                         if (fl) {
 450                                 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
 451                                 fl_release(fl);
 452                                 return err;
 453                         }
 454                 }
 455                 return -ESRCH;
 456
 457         case IPV6_FL_A_GET:
 458                 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
 459                         return -EINVAL;
 460
 461                 fl = fl_create(&freq, optval, optlen, &err);
 462                 if (fl == NULL)
 463                         return err;
 464                 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
 465
 466                 if (freq.flr_label) {
 467                         struct ip6_flowlabel *fl1 = NULL;
 468
 469                         err = -EEXIST;
 470                         read_lock_bh(&ip6_sk_fl_lock);
 471                         for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
 472                                 if (sfl->fl->label == freq.flr_label) {
 473                                         if (freq.flr_flags&IPV6_FL_F_EXCL) {
 474                                                 read_unlock_bh(&ip6_sk_fl_lock);
 475                                                 goto done;
 476                                         }
 477                                         fl1 = sfl->fl;
 478                                         atomic_inc(&fl1->users);
 479                                         break;
 480                                 }
 481                         }
 482                         read_unlock_bh(&ip6_sk_fl_lock);
 483
 484                         if (fl1 == NULL)
 485                                 fl1 = fl_lookup(freq.flr_label);
 486                         if (fl1) {
 487                                 err = -EEXIST;
 488                                 if (freq.flr_flags&IPV6_FL_F_EXCL)
 489                                         goto release;
 490                                 err = -EPERM;
 491                                 if (fl1->share == IPV6_FL_S_EXCL ||
 492                                     fl1->share != fl->share ||
 493                                     fl1->owner != fl->owner)
 494                                         goto release;
 495
 496                                 err = -EINVAL;
 497                                 if (ipv6_addr_cmp(&fl1->dst, &fl->dst) ||
 498                                     ipv6_opt_cmp(fl1->opt, fl->opt))
 499                                         goto release;
 500
 501                                 err = -ENOMEM;
 502                                 if (sfl1 == NULL)
 503                                         goto release;
 504                                 if (fl->linger > fl1->linger)
 505                                         fl1->linger = fl->linger;
 506                                 if ((long)(fl->expires - fl1->expires) > 0)
 507                                         fl1->expires = fl->expires;
 508                                 write_lock_bh(&ip6_sk_fl_lock);
 509                                 sfl1->fl = fl1;
 510                                 sfl1->next = np->ipv6_fl_list;
 511                                 np->ipv6_fl_list = sfl1;
 512                                 write_unlock_bh(&ip6_sk_fl_lock);
 513                                 fl_free(fl);
 514                                 return 0;
 515
 516 release:
 517                                 fl_release(fl1);
 518                                 goto done;
 519                         }
 520                 }
 521                 err = -ENOENT;
 522                 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
 523                         goto done;
 524
 525                 err = -ENOMEM;
 526                 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
 527                         goto done;
 528
 529                 err = fl_intern(fl, freq.flr_label);
 530                 if (err)
 531                         goto done;
 532
 533                 /* Do not check for fault */
 534                 if (!freq.flr_label)
 535                         copy_to_user(optval + ((u8*)&freq.flr_label - (u8*)&freq), &fl->label, sizeof(fl->label));
 536
 537                 sfl1->fl = fl;
 538                 sfl1->next = np->ipv6_fl_list;
 539                 np->ipv6_fl_list = sfl1;
 540                 return 0;
 541
 542         default:
 543                 return -EINVAL;
 544         }
 545
 546 done:
 547         if (fl)
 548                 fl_free(fl);
 549         if (sfl1)
 550                 kfree(sfl1);
 551         return err;
 552 }
 553
 554 #ifdef CONFIG_PROC_FS
 555
 556
 557 static int ip6_fl_read_proc(char *buffer, char **start, off_t offset,
 558                             int length, int *eof, void *data)
 559 {
 560         off_t pos=0;
 561         off_t begin=0;
 562         int len=0;
 563         int i, k;
 564         struct ip6_flowlabel *fl;
 565
 566         len+= sprintf(buffer,"Label S Owner  Users  Linger Expires  "
 567                       "Dst                              Opt\n");
 568
 569         read_lock_bh(&ip6_fl_lock);
 570         for (i=0; i<=FL_HASH_MASK; i++) {
 571                 for (fl = fl_ht[i]; fl; fl = fl->next) {
 572                         len+=sprintf(buffer+len,"%05X %-1d %-6d %-6d %-6d %-8ld ",
 573                                      (unsigned)ntohl(fl->label),
 574                                      fl->share,
 575                                      (unsigned)fl->owner,
 576                                      atomic_read(&fl->users),
 577                                      fl->linger/HZ,
 578                                      (long)(fl->expires - jiffies)/HZ);
 579
 580                         for (k=0; k<16; k++)
 581                                 len+=sprintf(buffer+len, "%02x", fl->dst.s6_addr[k]);
 582                         buffer[len++]=' ';
 583                         len+=sprintf(buffer+len, "%-4d", fl->opt ? fl->opt->opt_nflen : 0);
 584                         buffer[len++]='\n';
 585
 586                         pos=begin+len;
 587                         if(pos<offset) {
 588                                 len=0;
 589                                 begin=pos;
 590                         }
 591                         if(pos>offset+length)
 592                                 goto done;
 593                 }
 594         }
 595         *eof = 1;
 596
 597 done:
 598         read_unlock_bh(&ip6_fl_lock);
 599         *start=buffer+(offset-begin);
 600         len-=(offset-begin);
 601         if(len>length)
 602                 len=length;
 603         if(len<0)
 604                 len=0;
 605         return len;
 606 }
 607 #endif
 608
 609
 610 void ip6_flowlabel_init()
 611 {
 612         init_timer(&ip6_fl_gc_timer);
 613         ip6_fl_gc_timer.function = ip6_fl_gc;
 614 #ifdef CONFIG_PROC_FS
 615         create_proc_read_entry("net/ip6_flowlabel", 0, 0, ip6_fl_read_proc, NULL);
 616 #endif
 617 }
 618
 619 void ip6_flowlabel_cleanup()
 620 {
 621         del_timer(&ip6_fl_gc_timer);
 622 #ifdef CONFIG_PROC_FS
 623         remove_proc_entry("net/ip6_flowlabel", 0);
 624 #endif
 625 }