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 
__fl_lookup(u32 label)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 
fl_lookup(u32 label)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 
fl_free(struct ip6_flowlabel * fl)86 static void fl_free(struct ip6_flowlabel *fl)
87 {
88 	if (fl->opt)
89 		kfree(fl->opt);
90 	kfree(fl);
91 }
92 
fl_release(struct ip6_flowlabel * fl)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 
ip6_fl_gc(unsigned long dummy)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 
fl_intern(struct ip6_flowlabel * fl,__u32 label)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 
fl6_sock_lookup(struct sock * sk,u32 label)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 
fl6_free_socklist(struct sock * sk)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 
fl6_merge_options(struct ipv6_txoptions * opt_space,struct ip6_flowlabel * fl,struct ipv6_txoptions * fopt)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 
check_linger(__u16 ttl)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 
fl6_renew(struct ip6_flowlabel * fl,unsigned linger,unsigned expires)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 *
fl_create(struct in6_flowlabel_req * freq,char * optval,int optlen,int * err_p)275 fl_create(struct in6_flowlabel_req *freq, char *optval, int optlen, int *err_p)
276 {
277 	struct ip6_flowlabel *fl = NULL;
278 	int olen;
279 	int addr_type;
280 	int err;
281 
282 	olen = optlen - CMSG_ALIGN(sizeof(*freq));
283 	err = -EINVAL;
284 	if (olen > 64 * 1024)
285 		goto done;
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 	if (olen > 0) {
294 		struct msghdr msg;
295 		struct flowi flowi;
296 		int junk;
297 
298 		err = -ENOMEM;
299 		fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
300 		if (fl->opt == NULL)
301 			goto done;
302 
303 		memset(fl->opt, 0, sizeof(*fl->opt));
304 		fl->opt->tot_len = sizeof(*fl->opt) + olen;
305 		err = -EFAULT;
306 		if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
307 			goto done;
308 
309 		msg.msg_controllen = olen;
310 		msg.msg_control = (void*)(fl->opt+1);
311 		flowi.oif = 0;
312 
313 		err = datagram_send_ctl(&msg, &flowi, fl->opt, &junk);
314 		if (err)
315 			goto done;
316 		err = -EINVAL;
317 		if (fl->opt->opt_flen)
318 			goto done;
319 		if (fl->opt->opt_nflen == 0) {
320 			kfree(fl->opt);
321 			fl->opt = NULL;
322 		}
323 	}
324 
325 	fl->expires = jiffies;
326 	err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
327 	if (err)
328 		goto done;
329 	fl->share = freq->flr_share;
330 	addr_type = ipv6_addr_type(&freq->flr_dst);
331 	if ((addr_type&IPV6_ADDR_MAPPED)
332 	    || addr_type == IPV6_ADDR_ANY)
333 		goto done;
334 	ipv6_addr_copy(&fl->dst, &freq->flr_dst);
335 	atomic_set(&fl->users, 1);
336 	switch (fl->share) {
337 	case IPV6_FL_S_EXCL:
338 	case IPV6_FL_S_ANY:
339 		break;
340 	case IPV6_FL_S_PROCESS:
341 		fl->owner = current->pid;
342 		break;
343 	case IPV6_FL_S_USER:
344 		fl->owner = current->euid;
345 		break;
346 	default:
347 		err = -EINVAL;
348 		goto done;
349 	}
350 	return fl;
351 
352 done:
353 	if (fl)
354 		fl_free(fl);
355 	*err_p = err;
356 	return NULL;
357 }
358 
mem_check(struct sock * sk)359 static int mem_check(struct sock *sk)
360 {
361 	struct ipv6_fl_socklist *sfl;
362 	int room = FL_MAX_SIZE - atomic_read(&fl_size);
363 	int count = 0;
364 
365 	if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
366 		return 0;
367 
368 	for (sfl = sk->net_pinfo.af_inet6.ipv6_fl_list; sfl; sfl = sfl->next)
369 		count++;
370 
371 	if (room <= 0 ||
372 	    ((count >= FL_MAX_PER_SOCK ||
373 	     (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4)
374 	     && !capable(CAP_NET_ADMIN)))
375 		return -ENOBUFS;
376 
377 	return 0;
378 }
379 
ipv6_hdr_cmp(struct ipv6_opt_hdr * h1,struct ipv6_opt_hdr * h2)380 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
381 {
382 	if (h1 == h2)
383 		return 0;
384 	if (h1 == NULL || h2 == NULL)
385 		return 1;
386 	if (h1->hdrlen != h2->hdrlen)
387 		return 1;
388 	return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
389 }
390 
ipv6_opt_cmp(struct ipv6_txoptions * o1,struct ipv6_txoptions * o2)391 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
392 {
393 	if (o1 == o2)
394 		return 0;
395 	if (o1 == NULL || o2 == NULL)
396 		return 1;
397 	if (o1->opt_nflen != o2->opt_nflen)
398 		return 1;
399 	if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
400 		return 1;
401 	if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
402 		return 1;
403 	if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
404 		return 1;
405 	return 0;
406 }
407 
ipv6_flowlabel_opt(struct sock * sk,char * optval,int optlen)408 int ipv6_flowlabel_opt(struct sock *sk, char *optval, int optlen)
409 {
410 	int err;
411 	struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
412 	struct in6_flowlabel_req freq;
413 	struct ipv6_fl_socklist *sfl1=NULL;
414 	struct ipv6_fl_socklist *sfl, **sflp;
415 	struct ip6_flowlabel *fl;
416 
417 	if (optlen < sizeof(freq))
418 		return -EINVAL;
419 
420 	if (copy_from_user(&freq, optval, sizeof(freq)))
421 		return -EFAULT;
422 
423 	switch (freq.flr_action) {
424 	case IPV6_FL_A_PUT:
425 		write_lock_bh(&ip6_sk_fl_lock);
426 		for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
427 			if (sfl->fl->label == freq.flr_label) {
428 				if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
429 					np->flow_label &= ~IPV6_FLOWLABEL_MASK;
430 				*sflp = sfl->next;
431 				write_unlock_bh(&ip6_sk_fl_lock);
432 				fl_release(sfl->fl);
433 				kfree(sfl);
434 				return 0;
435 			}
436 		}
437 		write_unlock_bh(&ip6_sk_fl_lock);
438 		return -ESRCH;
439 
440 	case IPV6_FL_A_RENEW:
441 		read_lock_bh(&ip6_sk_fl_lock);
442 		for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
443 			if (sfl->fl->label == freq.flr_label) {
444 				err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
445 				read_unlock_bh(&ip6_sk_fl_lock);
446 				return err;
447 			}
448 		}
449 		read_unlock_bh(&ip6_sk_fl_lock);
450 
451 		if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
452 			fl = fl_lookup(freq.flr_label);
453 			if (fl) {
454 				err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
455 				fl_release(fl);
456 				return err;
457 			}
458 		}
459 		return -ESRCH;
460 
461 	case IPV6_FL_A_GET:
462 		if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
463 			return -EINVAL;
464 
465 		fl = fl_create(&freq, optval, optlen, &err);
466 		if (fl == NULL)
467 			return err;
468 		sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
469 
470 		if (freq.flr_label) {
471 			struct ip6_flowlabel *fl1 = NULL;
472 
473 			err = -EEXIST;
474 			read_lock_bh(&ip6_sk_fl_lock);
475 			for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
476 				if (sfl->fl->label == freq.flr_label) {
477 					if (freq.flr_flags&IPV6_FL_F_EXCL) {
478 						read_unlock_bh(&ip6_sk_fl_lock);
479 						goto done;
480 					}
481 					fl1 = sfl->fl;
482 					atomic_inc(&fl1->users);
483 					break;
484 				}
485 			}
486 			read_unlock_bh(&ip6_sk_fl_lock);
487 
488 			if (fl1 == NULL)
489 				fl1 = fl_lookup(freq.flr_label);
490 			if (fl1) {
491 				err = -EEXIST;
492 				if (freq.flr_flags&IPV6_FL_F_EXCL)
493 					goto release;
494 				err = -EPERM;
495 				if (fl1->share == IPV6_FL_S_EXCL ||
496 				    fl1->share != fl->share ||
497 				    fl1->owner != fl->owner)
498 					goto release;
499 
500 				err = -EINVAL;
501 				if (ipv6_addr_cmp(&fl1->dst, &fl->dst) ||
502 				    ipv6_opt_cmp(fl1->opt, fl->opt))
503 					goto release;
504 
505 				err = -ENOMEM;
506 				if (sfl1 == NULL)
507 					goto release;
508 				if (fl->linger > fl1->linger)
509 					fl1->linger = fl->linger;
510 				if ((long)(fl->expires - fl1->expires) > 0)
511 					fl1->expires = fl->expires;
512 				write_lock_bh(&ip6_sk_fl_lock);
513 				sfl1->fl = fl1;
514 				sfl1->next = np->ipv6_fl_list;
515 				np->ipv6_fl_list = sfl1;
516 				write_unlock_bh(&ip6_sk_fl_lock);
517 				fl_free(fl);
518 				return 0;
519 
520 release:
521 				fl_release(fl1);
522 				goto done;
523 			}
524 		}
525 		err = -ENOENT;
526 		if (!(freq.flr_flags&IPV6_FL_F_CREATE))
527 			goto done;
528 
529 		err = -ENOMEM;
530 		if (sfl1 == NULL || (err = mem_check(sk)) != 0)
531 			goto done;
532 
533 		err = fl_intern(fl, freq.flr_label);
534 		if (err)
535 			goto done;
536 
537 		/* Do not check for fault */
538 		if (!freq.flr_label)
539 			copy_to_user(optval + ((u8*)&freq.flr_label - (u8*)&freq), &fl->label, sizeof(fl->label));
540 
541 		sfl1->fl = fl;
542 		sfl1->next = np->ipv6_fl_list;
543 		np->ipv6_fl_list = sfl1;
544 		return 0;
545 
546 	default:
547 		return -EINVAL;
548 	}
549 
550 done:
551 	if (fl)
552 		fl_free(fl);
553 	if (sfl1)
554 		kfree(sfl1);
555 	return err;
556 }
557 
558 #ifdef CONFIG_PROC_FS
559 
560 
ip6_fl_read_proc(char * buffer,char ** start,off_t offset,int length,int * eof,void * data)561 static int ip6_fl_read_proc(char *buffer, char **start, off_t offset,
562 			    int length, int *eof, void *data)
563 {
564 	off_t pos=0;
565 	off_t begin=0;
566 	int len=0;
567 	int i, k;
568 	struct ip6_flowlabel *fl;
569 
570 	len+= sprintf(buffer,"Label S Owner  Users  Linger Expires  "
571 		      "Dst                              Opt\n");
572 
573 	read_lock_bh(&ip6_fl_lock);
574 	for (i=0; i<=FL_HASH_MASK; i++) {
575 		for (fl = fl_ht[i]; fl; fl = fl->next) {
576 			len+=sprintf(buffer+len,"%05X %-1d %-6d %-6d %-6d %-8ld ",
577 				     (unsigned)ntohl(fl->label),
578 				     fl->share,
579 				     (unsigned)fl->owner,
580 				     atomic_read(&fl->users),
581 				     fl->linger/HZ,
582 				     (long)(fl->expires - jiffies)/HZ);
583 
584 			for (k=0; k<16; k++)
585 				len+=sprintf(buffer+len, "%02x", fl->dst.s6_addr[k]);
586 			buffer[len++]=' ';
587 			len+=sprintf(buffer+len, "%-4d", fl->opt ? fl->opt->opt_nflen : 0);
588 			buffer[len++]='\n';
589 
590 			pos=begin+len;
591 			if(pos<offset) {
592 				len=0;
593 				begin=pos;
594 			}
595 			if(pos>offset+length)
596 				goto done;
597 		}
598 	}
599 	*eof = 1;
600 
601 done:
602 	read_unlock_bh(&ip6_fl_lock);
603 	*start=buffer+(offset-begin);
604 	len-=(offset-begin);
605 	if(len>length)
606 		len=length;
607 	if(len<0)
608 		len=0;
609 	return len;
610 }
611 #endif
612 
613 
ip6_flowlabel_init()614 void ip6_flowlabel_init()
615 {
616 	init_timer(&ip6_fl_gc_timer);
617 	ip6_fl_gc_timer.function = ip6_fl_gc;
618 #ifdef CONFIG_PROC_FS
619 	create_proc_read_entry("net/ip6_flowlabel", 0, 0, ip6_fl_read_proc, NULL);
620 #endif
621 }
622 
ip6_flowlabel_cleanup()623 void ip6_flowlabel_cleanup()
624 {
625 	del_timer(&ip6_fl_gc_timer);
626 #ifdef CONFIG_PROC_FS
627 	remove_proc_entry("net/ip6_flowlabel", 0);
628 #endif
629 }
630