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/if_arp.h> 19 #include <linux/in6.h> 20 #include <linux/route.h> 21 #include <linux/proc_fs.h> 22 #include <linux/seq_file.h> 23 24 #include <net/net_namespace.h> 25 #include <net/sock.h> 26 27 #include <net/ipv6.h> 28 #include <net/ndisc.h> 29 #include <net/protocol.h> 30 #include <net/ip6_route.h> 31 #include <net/addrconf.h> 32 #include <net/rawv6.h> 33 #include <net/icmp.h> 34 #include <net/transp_v6.h> 35 36 #include <asm/uaccess.h> 37 38 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified 39 in old IPv6 RFC. Well, it was reasonable value. 40 */ 41 #define FL_MAX_LINGER 60 /* Maximal linger timeout */ 42 43 /* FL hash table */ 44 45 #define FL_MAX_PER_SOCK 32 46 #define FL_MAX_SIZE 4096 47 #define FL_HASH_MASK 255 48 #define FL_HASH(l) (ntohl(l)&FL_HASH_MASK) 49 50 static atomic_t fl_size = ATOMIC_INIT(0); 51 static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1]; 52 53 static void ip6_fl_gc(unsigned long dummy); 54 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc, 0, 0); 55 56 /* FL hash table lock: it protects only of GC */ 57 58 static DEFINE_RWLOCK(ip6_fl_lock); 59 60 /* Big socket sock */ 61 62 static DEFINE_RWLOCK(ip6_sk_fl_lock); 63 64 65 static __inline__ struct ip6_flowlabel * __fl_lookup(__be32 label) 66 { 67 struct ip6_flowlabel *fl; 68 69 for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) { 70 if (fl->label == label) 71 return fl; 72 } 73 return NULL; 74 } 75 76 static struct ip6_flowlabel * fl_lookup(__be32 label) 77 { 78 struct ip6_flowlabel *fl; 79 80 read_lock_bh(&ip6_fl_lock); 81 fl = __fl_lookup(label); 82 if (fl) 83 atomic_inc(&fl->users); 84 read_unlock_bh(&ip6_fl_lock); 85 return fl; 86 } 87 88 89 static void fl_free(struct ip6_flowlabel *fl) 90 { 91 if (fl) 92 kfree(fl->opt); 93 kfree(fl); 94 } 95 96 static void fl_release(struct ip6_flowlabel *fl) 97 { 98 write_lock_bh(&ip6_fl_lock); 99 100 fl->lastuse = jiffies; 101 if (atomic_dec_and_test(&fl->users)) { 102 unsigned long ttd = fl->lastuse + fl->linger; 103 if (time_after(ttd, fl->expires)) 104 fl->expires = ttd; 105 ttd = fl->expires; 106 if (fl->opt && fl->share == IPV6_FL_S_EXCL) { 107 struct ipv6_txoptions *opt = fl->opt; 108 fl->opt = NULL; 109 kfree(opt); 110 } 111 if (!timer_pending(&ip6_fl_gc_timer) || 112 time_after(ip6_fl_gc_timer.expires, ttd)) 113 mod_timer(&ip6_fl_gc_timer, ttd); 114 } 115 116 write_unlock_bh(&ip6_fl_lock); 117 } 118 119 static void ip6_fl_gc(unsigned long dummy) 120 { 121 int i; 122 unsigned long now = jiffies; 123 unsigned long sched = 0; 124 125 write_lock(&ip6_fl_lock); 126 127 for (i=0; i<=FL_HASH_MASK; i++) { 128 struct ip6_flowlabel *fl, **flp; 129 flp = &fl_ht[i]; 130 while ((fl=*flp) != NULL) { 131 if (atomic_read(&fl->users) == 0) { 132 unsigned long ttd = fl->lastuse + fl->linger; 133 if (time_after(ttd, fl->expires)) 134 fl->expires = ttd; 135 ttd = fl->expires; 136 if (time_after_eq(now, ttd)) { 137 *flp = fl->next; 138 fl_free(fl); 139 atomic_dec(&fl_size); 140 continue; 141 } 142 if (!sched || time_before(ttd, sched)) 143 sched = ttd; 144 } 145 flp = &fl->next; 146 } 147 } 148 if (!sched && atomic_read(&fl_size)) 149 sched = now + FL_MAX_LINGER; 150 if (sched) { 151 ip6_fl_gc_timer.expires = sched; 152 add_timer(&ip6_fl_gc_timer); 153 } 154 write_unlock(&ip6_fl_lock); 155 } 156 157 static int fl_intern(struct ip6_flowlabel *fl, __be32 label) 158 { 159 fl->label = label & IPV6_FLOWLABEL_MASK; 160 161 write_lock_bh(&ip6_fl_lock); 162 if (label == 0) { 163 for (;;) { 164 fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK; 165 if (fl->label) { 166 struct ip6_flowlabel *lfl; 167 lfl = __fl_lookup(fl->label); 168 if (lfl == NULL) 169 break; 170 } 171 } 172 } 173 174 fl->lastuse = jiffies; 175 fl->next = fl_ht[FL_HASH(fl->label)]; 176 fl_ht[FL_HASH(fl->label)] = fl; 177 atomic_inc(&fl_size); 178 write_unlock_bh(&ip6_fl_lock); 179 return 0; 180 } 181 182 183 184 /* Socket flowlabel lists */ 185 186 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, __be32 label) 187 { 188 struct ipv6_fl_socklist *sfl; 189 struct ipv6_pinfo *np = inet6_sk(sk); 190 191 label &= IPV6_FLOWLABEL_MASK; 192 193 for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) { 194 struct ip6_flowlabel *fl = sfl->fl; 195 if (fl->label == label) { 196 fl->lastuse = jiffies; 197 atomic_inc(&fl->users); 198 return fl; 199 } 200 } 201 return NULL; 202 } 203 204 EXPORT_SYMBOL_GPL(fl6_sock_lookup); 205 206 void fl6_free_socklist(struct sock *sk) 207 { 208 struct ipv6_pinfo *np = inet6_sk(sk); 209 struct ipv6_fl_socklist *sfl; 210 211 while ((sfl = np->ipv6_fl_list) != NULL) { 212 np->ipv6_fl_list = sfl->next; 213 fl_release(sfl->fl); 214 kfree(sfl); 215 } 216 } 217 218 /* Service routines */ 219 220 221 /* 222 It is the only difficult place. flowlabel enforces equal headers 223 before and including routing header, however user may supply options 224 following rthdr. 225 */ 226 227 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space, 228 struct ip6_flowlabel * fl, 229 struct ipv6_txoptions * fopt) 230 { 231 struct ipv6_txoptions * fl_opt = fl->opt; 232 233 if (fopt == NULL || fopt->opt_flen == 0) 234 return fl_opt; 235 236 if (fl_opt != NULL) { 237 opt_space->hopopt = fl_opt->hopopt; 238 opt_space->dst0opt = fl_opt->dst0opt; 239 opt_space->srcrt = fl_opt->srcrt; 240 opt_space->opt_nflen = fl_opt->opt_nflen; 241 } else { 242 if (fopt->opt_nflen == 0) 243 return fopt; 244 opt_space->hopopt = NULL; 245 opt_space->dst0opt = NULL; 246 opt_space->srcrt = NULL; 247 opt_space->opt_nflen = 0; 248 } 249 opt_space->dst1opt = fopt->dst1opt; 250 opt_space->opt_flen = fopt->opt_flen; 251 return opt_space; 252 } 253 254 static unsigned long check_linger(unsigned long ttl) 255 { 256 if (ttl < FL_MIN_LINGER) 257 return FL_MIN_LINGER*HZ; 258 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN)) 259 return 0; 260 return ttl*HZ; 261 } 262 263 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires) 264 { 265 linger = check_linger(linger); 266 if (!linger) 267 return -EPERM; 268 expires = check_linger(expires); 269 if (!expires) 270 return -EPERM; 271 fl->lastuse = jiffies; 272 if (time_before(fl->linger, linger)) 273 fl->linger = linger; 274 if (time_before(expires, fl->linger)) 275 expires = fl->linger; 276 if (time_before(fl->expires, fl->lastuse + expires)) 277 fl->expires = fl->lastuse + expires; 278 return 0; 279 } 280 281 static struct ip6_flowlabel * 282 fl_create(struct in6_flowlabel_req *freq, char __user *optval, int optlen, int *err_p) 283 { 284 struct ip6_flowlabel *fl; 285 int olen; 286 int addr_type; 287 int err; 288 289 err = -ENOMEM; 290 fl = kzalloc(sizeof(*fl), GFP_KERNEL); 291 if (fl == NULL) 292 goto done; 293 294 olen = optlen - CMSG_ALIGN(sizeof(*freq)); 295 if (olen > 0) { 296 struct msghdr msg; 297 struct flowi flowi; 298 int junk; 299 300 err = -ENOMEM; 301 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL); 302 if (fl->opt == NULL) 303 goto done; 304 305 memset(fl->opt, 0, sizeof(*fl->opt)); 306 fl->opt->tot_len = sizeof(*fl->opt) + olen; 307 err = -EFAULT; 308 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen)) 309 goto done; 310 311 msg.msg_controllen = olen; 312 msg.msg_control = (void*)(fl->opt+1); 313 flowi.oif = 0; 314 315 err = datagram_send_ctl(&msg, &flowi, fl->opt, &junk, &junk); 316 if (err) 317 goto done; 318 err = -EINVAL; 319 if (fl->opt->opt_flen) 320 goto done; 321 if (fl->opt->opt_nflen == 0) { 322 kfree(fl->opt); 323 fl->opt = NULL; 324 } 325 } 326 327 fl->expires = jiffies; 328 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires); 329 if (err) 330 goto done; 331 fl->share = freq->flr_share; 332 addr_type = ipv6_addr_type(&freq->flr_dst); 333 if ((addr_type&IPV6_ADDR_MAPPED) 334 || addr_type == IPV6_ADDR_ANY) { 335 err = -EINVAL; 336 goto done; 337 } 338 ipv6_addr_copy(&fl->dst, &freq->flr_dst); 339 atomic_set(&fl->users, 1); 340 switch (fl->share) { 341 case IPV6_FL_S_EXCL: 342 case IPV6_FL_S_ANY: 343 break; 344 case IPV6_FL_S_PROCESS: 345 fl->owner = current->pid; 346 break; 347 case IPV6_FL_S_USER: 348 fl->owner = current->euid; 349 break; 350 default: 351 err = -EINVAL; 352 goto done; 353 } 354 return fl; 355 356 done: 357 fl_free(fl); 358 *err_p = err; 359 return NULL; 360 } 361 362 static int mem_check(struct sock *sk) 363 { 364 struct ipv6_pinfo *np = inet6_sk(sk); 365 struct ipv6_fl_socklist *sfl; 366 int room = FL_MAX_SIZE - atomic_read(&fl_size); 367 int count = 0; 368 369 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK) 370 return 0; 371 372 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) 373 count++; 374 375 if (room <= 0 || 376 ((count >= FL_MAX_PER_SOCK || 377 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) 378 && !capable(CAP_NET_ADMIN))) 379 return -ENOBUFS; 380 381 return 0; 382 } 383 384 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2) 385 { 386 if (h1 == h2) 387 return 0; 388 if (h1 == NULL || h2 == NULL) 389 return 1; 390 if (h1->hdrlen != h2->hdrlen) 391 return 1; 392 return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1)); 393 } 394 395 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2) 396 { 397 if (o1 == o2) 398 return 0; 399 if (o1 == NULL || o2 == NULL) 400 return 1; 401 if (o1->opt_nflen != o2->opt_nflen) 402 return 1; 403 if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt)) 404 return 1; 405 if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt)) 406 return 1; 407 if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt)) 408 return 1; 409 return 0; 410 } 411 412 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen) 413 { 414 int err; 415 struct ipv6_pinfo *np = inet6_sk(sk); 416 struct in6_flowlabel_req freq; 417 struct ipv6_fl_socklist *sfl1=NULL; 418 struct ipv6_fl_socklist *sfl, **sflp; 419 struct ip6_flowlabel *fl; 420 421 if (optlen < sizeof(freq)) 422 return -EINVAL; 423 424 if (copy_from_user(&freq, optval, sizeof(freq))) 425 return -EFAULT; 426 427 switch (freq.flr_action) { 428 case IPV6_FL_A_PUT: 429 write_lock_bh(&ip6_sk_fl_lock); 430 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) { 431 if (sfl->fl->label == freq.flr_label) { 432 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK)) 433 np->flow_label &= ~IPV6_FLOWLABEL_MASK; 434 *sflp = sfl->next; 435 write_unlock_bh(&ip6_sk_fl_lock); 436 fl_release(sfl->fl); 437 kfree(sfl); 438 return 0; 439 } 440 } 441 write_unlock_bh(&ip6_sk_fl_lock); 442 return -ESRCH; 443 444 case IPV6_FL_A_RENEW: 445 read_lock_bh(&ip6_sk_fl_lock); 446 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) { 447 if (sfl->fl->label == freq.flr_label) { 448 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires); 449 read_unlock_bh(&ip6_sk_fl_lock); 450 return err; 451 } 452 } 453 read_unlock_bh(&ip6_sk_fl_lock); 454 455 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) { 456 fl = fl_lookup(freq.flr_label); 457 if (fl) { 458 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires); 459 fl_release(fl); 460 return err; 461 } 462 } 463 return -ESRCH; 464 465 case IPV6_FL_A_GET: 466 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK) 467 return -EINVAL; 468 469 fl = fl_create(&freq, optval, optlen, &err); 470 if (fl == NULL) 471 return err; 472 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL); 473 474 if (freq.flr_label) { 475 struct ip6_flowlabel *fl1 = NULL; 476 477 err = -EEXIST; 478 read_lock_bh(&ip6_sk_fl_lock); 479 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) { 480 if (sfl->fl->label == freq.flr_label) { 481 if (freq.flr_flags&IPV6_FL_F_EXCL) { 482 read_unlock_bh(&ip6_sk_fl_lock); 483 goto done; 484 } 485 fl1 = sfl->fl; 486 atomic_inc(&fl1->users); 487 break; 488 } 489 } 490 read_unlock_bh(&ip6_sk_fl_lock); 491 492 if (fl1 == NULL) 493 fl1 = fl_lookup(freq.flr_label); 494 if (fl1) { 495 err = -EEXIST; 496 if (freq.flr_flags&IPV6_FL_F_EXCL) 497 goto release; 498 err = -EPERM; 499 if (fl1->share == IPV6_FL_S_EXCL || 500 fl1->share != fl->share || 501 fl1->owner != fl->owner) 502 goto release; 503 504 err = -EINVAL; 505 if (!ipv6_addr_equal(&fl1->dst, &fl->dst) || 506 ipv6_opt_cmp(fl1->opt, fl->opt)) 507 goto release; 508 509 err = -ENOMEM; 510 if (sfl1 == NULL) 511 goto release; 512 if (fl->linger > fl1->linger) 513 fl1->linger = fl->linger; 514 if ((long)(fl->expires - fl1->expires) > 0) 515 fl1->expires = fl->expires; 516 write_lock_bh(&ip6_sk_fl_lock); 517 sfl1->fl = fl1; 518 sfl1->next = np->ipv6_fl_list; 519 np->ipv6_fl_list = sfl1; 520 write_unlock_bh(&ip6_sk_fl_lock); 521 fl_free(fl); 522 return 0; 523 524 release: 525 fl_release(fl1); 526 goto done; 527 } 528 } 529 err = -ENOENT; 530 if (!(freq.flr_flags&IPV6_FL_F_CREATE)) 531 goto done; 532 533 err = -ENOMEM; 534 if (sfl1 == NULL || (err = mem_check(sk)) != 0) 535 goto done; 536 537 err = fl_intern(fl, freq.flr_label); 538 if (err) 539 goto done; 540 541 if (!freq.flr_label) { 542 if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label, 543 &fl->label, sizeof(fl->label))) { 544 /* Intentionally ignore fault. */ 545 } 546 } 547 548 sfl1->fl = fl; 549 sfl1->next = np->ipv6_fl_list; 550 np->ipv6_fl_list = sfl1; 551 return 0; 552 553 default: 554 return -EINVAL; 555 } 556 557 done: 558 fl_free(fl); 559 kfree(sfl1); 560 return err; 561 } 562 563 #ifdef CONFIG_PROC_FS 564 565 struct ip6fl_iter_state { 566 int bucket; 567 }; 568 569 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private) 570 571 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq) 572 { 573 struct ip6_flowlabel *fl = NULL; 574 struct ip6fl_iter_state *state = ip6fl_seq_private(seq); 575 576 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) { 577 if (fl_ht[state->bucket]) { 578 fl = fl_ht[state->bucket]; 579 break; 580 } 581 } 582 return fl; 583 } 584 585 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl) 586 { 587 struct ip6fl_iter_state *state = ip6fl_seq_private(seq); 588 589 fl = fl->next; 590 while (!fl) { 591 if (++state->bucket <= FL_HASH_MASK) 592 fl = fl_ht[state->bucket]; 593 else 594 break; 595 } 596 return fl; 597 } 598 599 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos) 600 { 601 struct ip6_flowlabel *fl = ip6fl_get_first(seq); 602 if (fl) 603 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL) 604 --pos; 605 return pos ? NULL : fl; 606 } 607 608 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos) 609 { 610 read_lock_bh(&ip6_fl_lock); 611 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; 612 } 613 614 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos) 615 { 616 struct ip6_flowlabel *fl; 617 618 if (v == SEQ_START_TOKEN) 619 fl = ip6fl_get_first(seq); 620 else 621 fl = ip6fl_get_next(seq, v); 622 ++*pos; 623 return fl; 624 } 625 626 static void ip6fl_seq_stop(struct seq_file *seq, void *v) 627 { 628 read_unlock_bh(&ip6_fl_lock); 629 } 630 631 static int ip6fl_seq_show(struct seq_file *seq, void *v) 632 { 633 if (v == SEQ_START_TOKEN) 634 seq_printf(seq, "%-5s %-1s %-6s %-6s %-6s %-8s %-32s %s\n", 635 "Label", "S", "Owner", "Users", "Linger", "Expires", "Dst", "Opt"); 636 else { 637 struct ip6_flowlabel *fl = v; 638 seq_printf(seq, 639 "%05X %-1d %-6d %-6d %-6ld %-8ld " NIP6_SEQFMT " %-4d\n", 640 (unsigned)ntohl(fl->label), 641 fl->share, 642 (unsigned)fl->owner, 643 atomic_read(&fl->users), 644 fl->linger/HZ, 645 (long)(fl->expires - jiffies)/HZ, 646 NIP6(fl->dst), 647 fl->opt ? fl->opt->opt_nflen : 0); 648 } 649 return 0; 650 } 651 652 static const struct seq_operations ip6fl_seq_ops = { 653 .start = ip6fl_seq_start, 654 .next = ip6fl_seq_next, 655 .stop = ip6fl_seq_stop, 656 .show = ip6fl_seq_show, 657 }; 658 659 static int ip6fl_seq_open(struct inode *inode, struct file *file) 660 { 661 return seq_open_private(file, &ip6fl_seq_ops, 662 sizeof(struct ip6fl_iter_state)); 663 } 664 665 static const struct file_operations ip6fl_seq_fops = { 666 .owner = THIS_MODULE, 667 .open = ip6fl_seq_open, 668 .read = seq_read, 669 .llseek = seq_lseek, 670 .release = seq_release_private, 671 }; 672 #endif 673 674 675 void ip6_flowlabel_init(void) 676 { 677 #ifdef CONFIG_PROC_FS 678 proc_net_fops_create(&init_net, "ip6_flowlabel", S_IRUGO, &ip6fl_seq_fops); 679 #endif 680 } 681 682 void ip6_flowlabel_cleanup(void) 683 { 684 del_timer(&ip6_fl_gc_timer); 685 #ifdef CONFIG_PROC_FS 686 proc_net_remove(&init_net, "ip6_flowlabel"); 687 #endif 688 } 689