blob: a93f6dc51979f84f61e5fc277a451d5fdcdb1032 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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#include <linux/seq_file.h>
23
24#include <net/sock.h>
25
26#include <net/ipv6.h>
27#include <net/ndisc.h>
28#include <net/protocol.h>
29#include <net/ip6_route.h>
30#include <net/addrconf.h>
31#include <net/rawv6.h>
32#include <net/icmp.h>
33#include <net/transp_v6.h>
34
35#include <asm/uaccess.h>
36
37#define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
38 in old IPv6 RFC. Well, it was reasonable value.
39 */
40#define FL_MAX_LINGER 60 /* Maximal linger timeout */
41
42/* FL hash table */
43
44#define FL_MAX_PER_SOCK 32
45#define FL_MAX_SIZE 4096
46#define FL_HASH_MASK 255
47#define FL_HASH(l) (ntohl(l)&FL_HASH_MASK)
48
49static atomic_t fl_size = ATOMIC_INIT(0);
50static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1];
51
52static void ip6_fl_gc(unsigned long dummy);
53static struct timer_list ip6_fl_gc_timer = TIMER_INITIALIZER(ip6_fl_gc, 0, 0);
54
55/* FL hash table lock: it protects only of GC */
56
57static DEFINE_RWLOCK(ip6_fl_lock);
58
59/* Big socket sock */
60
61static DEFINE_RWLOCK(ip6_sk_fl_lock);
62
63
64static __inline__ struct ip6_flowlabel * __fl_lookup(u32 label)
65{
66 struct ip6_flowlabel *fl;
67
68 for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) {
69 if (fl->label == label)
70 return fl;
71 }
72 return NULL;
73}
74
75static struct ip6_flowlabel * fl_lookup(u32 label)
76{
77 struct ip6_flowlabel *fl;
78
79 read_lock_bh(&ip6_fl_lock);
80 fl = __fl_lookup(label);
81 if (fl)
82 atomic_inc(&fl->users);
83 read_unlock_bh(&ip6_fl_lock);
84 return fl;
85}
86
87
88static void fl_free(struct ip6_flowlabel *fl)
89{
90 if (fl)
91 kfree(fl->opt);
92 kfree(fl);
93}
94
95static void fl_release(struct ip6_flowlabel *fl)
96{
97 write_lock_bh(&ip6_fl_lock);
98
99 fl->lastuse = jiffies;
100 if (atomic_dec_and_test(&fl->users)) {
101 unsigned long ttd = fl->lastuse + fl->linger;
102 if (time_after(ttd, fl->expires))
103 fl->expires = ttd;
104 ttd = fl->expires;
105 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
106 struct ipv6_txoptions *opt = fl->opt;
107 fl->opt = NULL;
108 kfree(opt);
109 }
110 if (!timer_pending(&ip6_fl_gc_timer) ||
111 time_after(ip6_fl_gc_timer.expires, ttd))
112 mod_timer(&ip6_fl_gc_timer, ttd);
113 }
114
115 write_unlock_bh(&ip6_fl_lock);
116}
117
118static void ip6_fl_gc(unsigned long dummy)
119{
120 int i;
121 unsigned long now = jiffies;
122 unsigned long sched = 0;
123
124 write_lock(&ip6_fl_lock);
125
126 for (i=0; i<=FL_HASH_MASK; i++) {
127 struct ip6_flowlabel *fl, **flp;
128 flp = &fl_ht[i];
129 while ((fl=*flp) != NULL) {
130 if (atomic_read(&fl->users) == 0) {
131 unsigned long ttd = fl->lastuse + fl->linger;
132 if (time_after(ttd, fl->expires))
133 fl->expires = ttd;
134 ttd = fl->expires;
135 if (time_after_eq(now, ttd)) {
136 *flp = fl->next;
137 fl_free(fl);
138 atomic_dec(&fl_size);
139 continue;
140 }
141 if (!sched || time_before(ttd, sched))
142 sched = ttd;
143 }
144 flp = &fl->next;
145 }
146 }
147 if (!sched && atomic_read(&fl_size))
148 sched = now + FL_MAX_LINGER;
149 if (sched) {
150 ip6_fl_gc_timer.expires = sched;
151 add_timer(&ip6_fl_gc_timer);
152 }
153 write_unlock(&ip6_fl_lock);
154}
155
156static int fl_intern(struct ip6_flowlabel *fl, __u32 label)
157{
158 fl->label = label & IPV6_FLOWLABEL_MASK;
159
160 write_lock_bh(&ip6_fl_lock);
161 if (label == 0) {
162 for (;;) {
163 fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK;
164 if (fl->label) {
165 struct ip6_flowlabel *lfl;
166 lfl = __fl_lookup(fl->label);
167 if (lfl == NULL)
168 break;
169 }
170 }
171 }
172
173 fl->lastuse = jiffies;
174 fl->next = fl_ht[FL_HASH(fl->label)];
175 fl_ht[FL_HASH(fl->label)] = fl;
176 atomic_inc(&fl_size);
177 write_unlock_bh(&ip6_fl_lock);
178 return 0;
179}
180
181
182
183/* Socket flowlabel lists */
184
185struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, u32 label)
186{
187 struct ipv6_fl_socklist *sfl;
188 struct ipv6_pinfo *np = inet6_sk(sk);
189
190 label &= IPV6_FLOWLABEL_MASK;
191
192 for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) {
193 struct ip6_flowlabel *fl = sfl->fl;
194 if (fl->label == label) {
195 fl->lastuse = jiffies;
196 atomic_inc(&fl->users);
197 return fl;
198 }
199 }
200 return NULL;
201}
202
203void fl6_free_socklist(struct sock *sk)
204{
205 struct ipv6_pinfo *np = inet6_sk(sk);
206 struct ipv6_fl_socklist *sfl;
207
208 while ((sfl = np->ipv6_fl_list) != NULL) {
209 np->ipv6_fl_list = sfl->next;
210 fl_release(sfl->fl);
211 kfree(sfl);
212 }
213}
214
215/* Service routines */
216
217
218/*
219 It is the only difficult place. flowlabel enforces equal headers
220 before and including routing header, however user may supply options
221 following rthdr.
222 */
223
224struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
225 struct ip6_flowlabel * fl,
226 struct ipv6_txoptions * fopt)
227{
228 struct ipv6_txoptions * fl_opt = fl->opt;
229
230 if (fopt == NULL || fopt->opt_flen == 0)
231 return fl_opt;
232
233 if (fl_opt != NULL) {
234 opt_space->hopopt = fl_opt->hopopt;
235 opt_space->dst0opt = fl_opt->dst0opt;
236 opt_space->srcrt = fl_opt->srcrt;
237 opt_space->opt_nflen = fl_opt->opt_nflen;
238 } else {
239 if (fopt->opt_nflen == 0)
240 return fopt;
241 opt_space->hopopt = NULL;
242 opt_space->dst0opt = NULL;
243 opt_space->srcrt = NULL;
244 opt_space->opt_nflen = 0;
245 }
246 opt_space->dst1opt = fopt->dst1opt;
247 opt_space->auth = fopt->auth;
248 opt_space->opt_flen = fopt->opt_flen;
249 return opt_space;
250}
251
252static unsigned long check_linger(unsigned long ttl)
253{
254 if (ttl < FL_MIN_LINGER)
255 return FL_MIN_LINGER*HZ;
256 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
257 return 0;
258 return ttl*HZ;
259}
260
261static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
262{
263 linger = check_linger(linger);
264 if (!linger)
265 return -EPERM;
266 expires = check_linger(expires);
267 if (!expires)
268 return -EPERM;
269 fl->lastuse = jiffies;
270 if (time_before(fl->linger, linger))
271 fl->linger = linger;
272 if (time_before(expires, fl->linger))
273 expires = fl->linger;
274 if (time_before(fl->expires, fl->lastuse + expires))
275 fl->expires = fl->lastuse + expires;
276 return 0;
277}
278
279static struct ip6_flowlabel *
280fl_create(struct in6_flowlabel_req *freq, char __user *optval, int optlen, int *err_p)
281{
282 struct ip6_flowlabel *fl;
283 int olen;
284 int addr_type;
285 int err;
286
287 err = -ENOMEM;
288 fl = kmalloc(sizeof(*fl), GFP_KERNEL);
289 if (fl == NULL)
290 goto done;
291 memset(fl, 0, sizeof(*fl));
292
293 olen = optlen - CMSG_ALIGN(sizeof(*freq));
294 if (olen > 0) {
295 struct msghdr msg;
296 struct flowi flowi;
297 int junk;
298
299 err = -ENOMEM;
300 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
301 if (fl->opt == NULL)
302 goto done;
303
304 memset(fl->opt, 0, sizeof(*fl->opt));
305 fl->opt->tot_len = sizeof(*fl->opt) + olen;
306 err = -EFAULT;
307 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
308 goto done;
309
310 msg.msg_controllen = olen;
311 msg.msg_control = (void*)(fl->opt+1);
312 flowi.oif = 0;
313
314 err = datagram_send_ctl(&msg, &flowi, fl->opt, &junk);
315 if (err)
316 goto done;
317 err = -EINVAL;
318 if (fl->opt->opt_flen)
319 goto done;
320 if (fl->opt->opt_nflen == 0) {
321 kfree(fl->opt);
322 fl->opt = NULL;
323 }
324 }
325
326 fl->expires = jiffies;
327 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
328 if (err)
329 goto done;
330 fl->share = freq->flr_share;
331 addr_type = ipv6_addr_type(&freq->flr_dst);
332 if ((addr_type&IPV6_ADDR_MAPPED)
333 || addr_type == IPV6_ADDR_ANY)
334 goto done;
335 ipv6_addr_copy(&fl->dst, &freq->flr_dst);
336 atomic_set(&fl->users, 1);
337 switch (fl->share) {
338 case IPV6_FL_S_EXCL:
339 case IPV6_FL_S_ANY:
340 break;
341 case IPV6_FL_S_PROCESS:
342 fl->owner = current->pid;
343 break;
344 case IPV6_FL_S_USER:
345 fl->owner = current->euid;
346 break;
347 default:
348 err = -EINVAL;
349 goto done;
350 }
351 return fl;
352
353done:
354 fl_free(fl);
355 *err_p = err;
356 return NULL;
357}
358
359static int mem_check(struct sock *sk)
360{
361 struct ipv6_pinfo *np = inet6_sk(sk);
362 struct ipv6_fl_socklist *sfl;
363 int room = FL_MAX_SIZE - atomic_read(&fl_size);
364 int count = 0;
365
366 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
367 return 0;
368
369 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next)
370 count++;
371
372 if (room <= 0 ||
373 ((count >= FL_MAX_PER_SOCK ||
374 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4)
375 && !capable(CAP_NET_ADMIN)))
376 return -ENOBUFS;
377
378 return 0;
379}
380
381static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
382{
383 if (h1 == h2)
384 return 0;
385 if (h1 == NULL || h2 == NULL)
386 return 1;
387 if (h1->hdrlen != h2->hdrlen)
388 return 1;
389 return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
390}
391
392static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
393{
394 if (o1 == o2)
395 return 0;
396 if (o1 == NULL || o2 == NULL)
397 return 1;
398 if (o1->opt_nflen != o2->opt_nflen)
399 return 1;
400 if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
401 return 1;
402 if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
403 return 1;
404 if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
405 return 1;
406 return 0;
407}
408
409int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
410{
411 int err;
412 struct ipv6_pinfo *np = inet6_sk(sk);
413 struct in6_flowlabel_req freq;
414 struct ipv6_fl_socklist *sfl1=NULL;
415 struct ipv6_fl_socklist *sfl, **sflp;
416 struct ip6_flowlabel *fl;
417
418 if (optlen < sizeof(freq))
419 return -EINVAL;
420
421 if (copy_from_user(&freq, optval, sizeof(freq)))
422 return -EFAULT;
423
424 switch (freq.flr_action) {
425 case IPV6_FL_A_PUT:
426 write_lock_bh(&ip6_sk_fl_lock);
427 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
428 if (sfl->fl->label == freq.flr_label) {
429 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
430 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
431 *sflp = sfl->next;
432 write_unlock_bh(&ip6_sk_fl_lock);
433 fl_release(sfl->fl);
434 kfree(sfl);
435 return 0;
436 }
437 }
438 write_unlock_bh(&ip6_sk_fl_lock);
439 return -ESRCH;
440
441 case IPV6_FL_A_RENEW:
442 read_lock_bh(&ip6_sk_fl_lock);
443 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
444 if (sfl->fl->label == freq.flr_label) {
445 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
446 read_unlock_bh(&ip6_sk_fl_lock);
447 return err;
448 }
449 }
450 read_unlock_bh(&ip6_sk_fl_lock);
451
452 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
453 fl = fl_lookup(freq.flr_label);
454 if (fl) {
455 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
456 fl_release(fl);
457 return err;
458 }
459 }
460 return -ESRCH;
461
462 case IPV6_FL_A_GET:
463 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
464 return -EINVAL;
465
466 fl = fl_create(&freq, optval, optlen, &err);
467 if (fl == NULL)
468 return err;
469 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
470
471 if (freq.flr_label) {
472 struct ip6_flowlabel *fl1 = NULL;
473
474 err = -EEXIST;
475 read_lock_bh(&ip6_sk_fl_lock);
476 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
477 if (sfl->fl->label == freq.flr_label) {
478 if (freq.flr_flags&IPV6_FL_F_EXCL) {
479 read_unlock_bh(&ip6_sk_fl_lock);
480 goto done;
481 }
482 fl1 = sfl->fl;
483 atomic_inc(&fl->users);
484 break;
485 }
486 }
487 read_unlock_bh(&ip6_sk_fl_lock);
488
489 if (fl1 == NULL)
490 fl1 = fl_lookup(freq.flr_label);
491 if (fl1) {
492 err = -EEXIST;
493 if (freq.flr_flags&IPV6_FL_F_EXCL)
494 goto release;
495 err = -EPERM;
496 if (fl1->share == IPV6_FL_S_EXCL ||
497 fl1->share != fl->share ||
498 fl1->owner != fl->owner)
499 goto release;
500
501 err = -EINVAL;
502 if (!ipv6_addr_equal(&fl1->dst, &fl->dst) ||
503 ipv6_opt_cmp(fl1->opt, fl->opt))
504 goto release;
505
506 err = -ENOMEM;
507 if (sfl1 == NULL)
508 goto release;
509 if (fl->linger > fl1->linger)
510 fl1->linger = fl->linger;
511 if ((long)(fl->expires - fl1->expires) > 0)
512 fl1->expires = fl->expires;
513 write_lock_bh(&ip6_sk_fl_lock);
514 sfl1->fl = fl1;
515 sfl1->next = np->ipv6_fl_list;
516 np->ipv6_fl_list = sfl1;
517 write_unlock_bh(&ip6_sk_fl_lock);
518 fl_free(fl);
519 return 0;
520
521release:
522 fl_release(fl1);
523 goto done;
524 }
525 }
526 err = -ENOENT;
527 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
528 goto done;
529
530 err = -ENOMEM;
531 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
532 goto done;
533
534 err = fl_intern(fl, freq.flr_label);
535 if (err)
536 goto done;
537
538 /* Do not check for fault */
539 if (!freq.flr_label)
540 copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
541 &fl->label, sizeof(fl->label));
542
543 sfl1->fl = fl;
544 sfl1->next = np->ipv6_fl_list;
545 np->ipv6_fl_list = sfl1;
546 return 0;
547
548 default:
549 return -EINVAL;
550 }
551
552done:
553 fl_free(fl);
554 kfree(sfl1);
555 return err;
556}
557
558#ifdef CONFIG_PROC_FS
559
560struct ip6fl_iter_state {
561 int bucket;
562};
563
564#define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
565
566static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
567{
568 struct ip6_flowlabel *fl = NULL;
569 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
570
571 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
572 if (fl_ht[state->bucket]) {
573 fl = fl_ht[state->bucket];
574 break;
575 }
576 }
577 return fl;
578}
579
580static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
581{
582 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
583
584 fl = fl->next;
585 while (!fl) {
586 if (++state->bucket <= FL_HASH_MASK)
587 fl = fl_ht[state->bucket];
588 }
589 return fl;
590}
591
592static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
593{
594 struct ip6_flowlabel *fl = ip6fl_get_first(seq);
595 if (fl)
596 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
597 --pos;
598 return pos ? NULL : fl;
599}
600
601static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
602{
603 read_lock_bh(&ip6_fl_lock);
604 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
605}
606
607static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
608{
609 struct ip6_flowlabel *fl;
610
611 if (v == SEQ_START_TOKEN)
612 fl = ip6fl_get_first(seq);
613 else
614 fl = ip6fl_get_next(seq, v);
615 ++*pos;
616 return fl;
617}
618
619static void ip6fl_seq_stop(struct seq_file *seq, void *v)
620{
621 read_unlock_bh(&ip6_fl_lock);
622}
623
624static void ip6fl_fl_seq_show(struct seq_file *seq, struct ip6_flowlabel *fl)
625{
626 while(fl) {
627 seq_printf(seq,
628 "%05X %-1d %-6d %-6d %-6ld %-8ld "
629 "%02x%02x%02x%02x%02x%02x%02x%02x "
630 "%-4d\n",
631 (unsigned)ntohl(fl->label),
632 fl->share,
633 (unsigned)fl->owner,
634 atomic_read(&fl->users),
635 fl->linger/HZ,
636 (long)(fl->expires - jiffies)/HZ,
637 NIP6(fl->dst),
638 fl->opt ? fl->opt->opt_nflen : 0);
639 fl = fl->next;
640 }
641}
642
643static int ip6fl_seq_show(struct seq_file *seq, void *v)
644{
645 if (v == SEQ_START_TOKEN)
646 seq_puts(seq, "Label S Owner Users Linger Expires "
647 "Dst Opt\n");
648 else
649 ip6fl_fl_seq_show(seq, v);
650 return 0;
651}
652
653static struct seq_operations ip6fl_seq_ops = {
654 .start = ip6fl_seq_start,
655 .next = ip6fl_seq_next,
656 .stop = ip6fl_seq_stop,
657 .show = ip6fl_seq_show,
658};
659
660static int ip6fl_seq_open(struct inode *inode, struct file *file)
661{
662 struct seq_file *seq;
663 int rc = -ENOMEM;
664 struct ip6fl_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
665
666 if (!s)
667 goto out;
668
669 rc = seq_open(file, &ip6fl_seq_ops);
670 if (rc)
671 goto out_kfree;
672
673 seq = file->private_data;
674 seq->private = s;
675 memset(s, 0, sizeof(*s));
676out:
677 return rc;
678out_kfree:
679 kfree(s);
680 goto out;
681}
682
683static struct file_operations ip6fl_seq_fops = {
684 .owner = THIS_MODULE,
685 .open = ip6fl_seq_open,
686 .read = seq_read,
687 .llseek = seq_lseek,
688 .release = seq_release_private,
689};
690#endif
691
692
693void ip6_flowlabel_init(void)
694{
695#ifdef CONFIG_PROC_FS
696 proc_net_fops_create("ip6_flowlabel", S_IRUGO, &ip6fl_seq_fops);
697#endif
698}
699
700void ip6_flowlabel_cleanup(void)
701{
702 del_timer(&ip6_fl_gc_timer);
703#ifdef CONFIG_PROC_FS
704 proc_net_remove("ip6_flowlabel");
705#endif
706}