xref: /openbmc/linux/net/ipv6/ip6_flowlabel.c (revision 534420c6)
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(struct timer_list *unused);
50 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc);
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(struct timer_list *unused)
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 		if (fl->label == label) {
258 			fl->lastuse = jiffies;
259 			atomic_inc(&fl->users);
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 ipcm6_cookie ipc6;
383 
384 		err = -ENOMEM;
385 		fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
386 		if (!fl->opt)
387 			goto done;
388 
389 		memset(fl->opt, 0, sizeof(*fl->opt));
390 		fl->opt->tot_len = sizeof(*fl->opt) + olen;
391 		err = -EFAULT;
392 		if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
393 			goto done;
394 
395 		msg.msg_controllen = olen;
396 		msg.msg_control = (void *)(fl->opt+1);
397 		memset(&flowi6, 0, sizeof(flowi6));
398 
399 		ipc6.opt = fl->opt;
400 		err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, &ipc6);
401 		if (err)
402 			goto done;
403 		err = -EINVAL;
404 		if (fl->opt->opt_flen)
405 			goto done;
406 		if (fl->opt->opt_nflen == 0) {
407 			kfree(fl->opt);
408 			fl->opt = NULL;
409 		}
410 	}
411 
412 	fl->fl_net = net;
413 	fl->expires = jiffies;
414 	err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
415 	if (err)
416 		goto done;
417 	fl->share = freq->flr_share;
418 	addr_type = ipv6_addr_type(&freq->flr_dst);
419 	if ((addr_type & IPV6_ADDR_MAPPED) ||
420 	    addr_type == IPV6_ADDR_ANY) {
421 		err = -EINVAL;
422 		goto done;
423 	}
424 	fl->dst = freq->flr_dst;
425 	atomic_set(&fl->users, 1);
426 	switch (fl->share) {
427 	case IPV6_FL_S_EXCL:
428 	case IPV6_FL_S_ANY:
429 		break;
430 	case IPV6_FL_S_PROCESS:
431 		fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
432 		break;
433 	case IPV6_FL_S_USER:
434 		fl->owner.uid = current_euid();
435 		break;
436 	default:
437 		err = -EINVAL;
438 		goto done;
439 	}
440 	return fl;
441 
442 done:
443 	fl_free(fl);
444 	*err_p = err;
445 	return NULL;
446 }
447 
448 static int mem_check(struct sock *sk)
449 {
450 	struct ipv6_pinfo *np = inet6_sk(sk);
451 	struct ipv6_fl_socklist *sfl;
452 	int room = FL_MAX_SIZE - atomic_read(&fl_size);
453 	int count = 0;
454 
455 	if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
456 		return 0;
457 
458 	rcu_read_lock_bh();
459 	for_each_sk_fl_rcu(np, sfl)
460 		count++;
461 	rcu_read_unlock_bh();
462 
463 	if (room <= 0 ||
464 	    ((count >= FL_MAX_PER_SOCK ||
465 	      (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
466 	     !capable(CAP_NET_ADMIN)))
467 		return -ENOBUFS;
468 
469 	return 0;
470 }
471 
472 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
473 		struct ip6_flowlabel *fl)
474 {
475 	spin_lock_bh(&ip6_sk_fl_lock);
476 	sfl->fl = fl;
477 	sfl->next = np->ipv6_fl_list;
478 	rcu_assign_pointer(np->ipv6_fl_list, sfl);
479 	spin_unlock_bh(&ip6_sk_fl_lock);
480 }
481 
482 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
483 			   int flags)
484 {
485 	struct ipv6_pinfo *np = inet6_sk(sk);
486 	struct ipv6_fl_socklist *sfl;
487 
488 	if (flags & IPV6_FL_F_REMOTE) {
489 		freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
490 		return 0;
491 	}
492 
493 	if (np->repflow) {
494 		freq->flr_label = np->flow_label;
495 		return 0;
496 	}
497 
498 	rcu_read_lock_bh();
499 
500 	for_each_sk_fl_rcu(np, sfl) {
501 		if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
502 			spin_lock_bh(&ip6_fl_lock);
503 			freq->flr_label = sfl->fl->label;
504 			freq->flr_dst = sfl->fl->dst;
505 			freq->flr_share = sfl->fl->share;
506 			freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
507 			freq->flr_linger = sfl->fl->linger / HZ;
508 
509 			spin_unlock_bh(&ip6_fl_lock);
510 			rcu_read_unlock_bh();
511 			return 0;
512 		}
513 	}
514 	rcu_read_unlock_bh();
515 
516 	return -ENOENT;
517 }
518 
519 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
520 {
521 	int uninitialized_var(err);
522 	struct net *net = sock_net(sk);
523 	struct ipv6_pinfo *np = inet6_sk(sk);
524 	struct in6_flowlabel_req freq;
525 	struct ipv6_fl_socklist *sfl1 = NULL;
526 	struct ipv6_fl_socklist *sfl;
527 	struct ipv6_fl_socklist __rcu **sflp;
528 	struct ip6_flowlabel *fl, *fl1 = NULL;
529 
530 
531 	if (optlen < sizeof(freq))
532 		return -EINVAL;
533 
534 	if (copy_from_user(&freq, optval, sizeof(freq)))
535 		return -EFAULT;
536 
537 	switch (freq.flr_action) {
538 	case IPV6_FL_A_PUT:
539 		if (freq.flr_flags & IPV6_FL_F_REFLECT) {
540 			if (sk->sk_protocol != IPPROTO_TCP)
541 				return -ENOPROTOOPT;
542 			if (!np->repflow)
543 				return -ESRCH;
544 			np->flow_label = 0;
545 			np->repflow = 0;
546 			return 0;
547 		}
548 		spin_lock_bh(&ip6_sk_fl_lock);
549 		for (sflp = &np->ipv6_fl_list;
550 		     (sfl = rcu_dereference_protected(*sflp,
551 						      lockdep_is_held(&ip6_sk_fl_lock))) != NULL;
552 		     sflp = &sfl->next) {
553 			if (sfl->fl->label == freq.flr_label) {
554 				if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
555 					np->flow_label &= ~IPV6_FLOWLABEL_MASK;
556 				*sflp = sfl->next;
557 				spin_unlock_bh(&ip6_sk_fl_lock);
558 				fl_release(sfl->fl);
559 				kfree_rcu(sfl, rcu);
560 				return 0;
561 			}
562 		}
563 		spin_unlock_bh(&ip6_sk_fl_lock);
564 		return -ESRCH;
565 
566 	case IPV6_FL_A_RENEW:
567 		rcu_read_lock_bh();
568 		for_each_sk_fl_rcu(np, sfl) {
569 			if (sfl->fl->label == freq.flr_label) {
570 				err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
571 				rcu_read_unlock_bh();
572 				return err;
573 			}
574 		}
575 		rcu_read_unlock_bh();
576 
577 		if (freq.flr_share == IPV6_FL_S_NONE &&
578 		    ns_capable(net->user_ns, CAP_NET_ADMIN)) {
579 			fl = fl_lookup(net, freq.flr_label);
580 			if (fl) {
581 				err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
582 				fl_release(fl);
583 				return err;
584 			}
585 		}
586 		return -ESRCH;
587 
588 	case IPV6_FL_A_GET:
589 		if (freq.flr_flags & IPV6_FL_F_REFLECT) {
590 			struct net *net = sock_net(sk);
591 			if (net->ipv6.sysctl.flowlabel_consistency) {
592 				net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
593 				return -EPERM;
594 			}
595 
596 			if (sk->sk_protocol != IPPROTO_TCP)
597 				return -ENOPROTOOPT;
598 
599 			np->repflow = 1;
600 			return 0;
601 		}
602 
603 		if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
604 			return -EINVAL;
605 
606 		if (net->ipv6.sysctl.flowlabel_state_ranges &&
607 		    (freq.flr_label & IPV6_FLOWLABEL_STATELESS_FLAG))
608 			return -ERANGE;
609 
610 		fl = fl_create(net, sk, &freq, optval, optlen, &err);
611 		if (!fl)
612 			return err;
613 		sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
614 
615 		if (freq.flr_label) {
616 			err = -EEXIST;
617 			rcu_read_lock_bh();
618 			for_each_sk_fl_rcu(np, sfl) {
619 				if (sfl->fl->label == freq.flr_label) {
620 					if (freq.flr_flags&IPV6_FL_F_EXCL) {
621 						rcu_read_unlock_bh();
622 						goto done;
623 					}
624 					fl1 = sfl->fl;
625 					atomic_inc(&fl1->users);
626 					break;
627 				}
628 			}
629 			rcu_read_unlock_bh();
630 
631 			if (!fl1)
632 				fl1 = fl_lookup(net, freq.flr_label);
633 			if (fl1) {
634 recheck:
635 				err = -EEXIST;
636 				if (freq.flr_flags&IPV6_FL_F_EXCL)
637 					goto release;
638 				err = -EPERM;
639 				if (fl1->share == IPV6_FL_S_EXCL ||
640 				    fl1->share != fl->share ||
641 				    ((fl1->share == IPV6_FL_S_PROCESS) &&
642 				     (fl1->owner.pid != fl->owner.pid)) ||
643 				    ((fl1->share == IPV6_FL_S_USER) &&
644 				     !uid_eq(fl1->owner.uid, fl->owner.uid)))
645 					goto release;
646 
647 				err = -ENOMEM;
648 				if (!sfl1)
649 					goto release;
650 				if (fl->linger > fl1->linger)
651 					fl1->linger = fl->linger;
652 				if ((long)(fl->expires - fl1->expires) > 0)
653 					fl1->expires = fl->expires;
654 				fl_link(np, sfl1, fl1);
655 				fl_free(fl);
656 				return 0;
657 
658 release:
659 				fl_release(fl1);
660 				goto done;
661 			}
662 		}
663 		err = -ENOENT;
664 		if (!(freq.flr_flags&IPV6_FL_F_CREATE))
665 			goto done;
666 
667 		err = -ENOMEM;
668 		if (!sfl1)
669 			goto done;
670 
671 		err = mem_check(sk);
672 		if (err != 0)
673 			goto done;
674 
675 		fl1 = fl_intern(net, fl, freq.flr_label);
676 		if (fl1)
677 			goto recheck;
678 
679 		if (!freq.flr_label) {
680 			if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
681 					 &fl->label, sizeof(fl->label))) {
682 				/* Intentionally ignore fault. */
683 			}
684 		}
685 
686 		fl_link(np, sfl1, fl);
687 		return 0;
688 
689 	default:
690 		return -EINVAL;
691 	}
692 
693 done:
694 	fl_free(fl);
695 	kfree(sfl1);
696 	return err;
697 }
698 
699 #ifdef CONFIG_PROC_FS
700 
701 struct ip6fl_iter_state {
702 	struct seq_net_private p;
703 	struct pid_namespace *pid_ns;
704 	int bucket;
705 };
706 
707 #define ip6fl_seq_private(seq)	((struct ip6fl_iter_state *)(seq)->private)
708 
709 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
710 {
711 	struct ip6_flowlabel *fl = NULL;
712 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
713 	struct net *net = seq_file_net(seq);
714 
715 	for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
716 		for_each_fl_rcu(state->bucket, fl) {
717 			if (net_eq(fl->fl_net, net))
718 				goto out;
719 		}
720 	}
721 	fl = NULL;
722 out:
723 	return fl;
724 }
725 
726 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
727 {
728 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
729 	struct net *net = seq_file_net(seq);
730 
731 	for_each_fl_continue_rcu(fl) {
732 		if (net_eq(fl->fl_net, net))
733 			goto out;
734 	}
735 
736 try_again:
737 	if (++state->bucket <= FL_HASH_MASK) {
738 		for_each_fl_rcu(state->bucket, fl) {
739 			if (net_eq(fl->fl_net, net))
740 				goto out;
741 		}
742 		goto try_again;
743 	}
744 	fl = NULL;
745 
746 out:
747 	return fl;
748 }
749 
750 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
751 {
752 	struct ip6_flowlabel *fl = ip6fl_get_first(seq);
753 	if (fl)
754 		while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
755 			--pos;
756 	return pos ? NULL : fl;
757 }
758 
759 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
760 	__acquires(RCU)
761 {
762 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
763 
764 	state->pid_ns = proc_pid_ns(file_inode(seq->file));
765 
766 	rcu_read_lock_bh();
767 	return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
768 }
769 
770 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
771 {
772 	struct ip6_flowlabel *fl;
773 
774 	if (v == SEQ_START_TOKEN)
775 		fl = ip6fl_get_first(seq);
776 	else
777 		fl = ip6fl_get_next(seq, v);
778 	++*pos;
779 	return fl;
780 }
781 
782 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
783 	__releases(RCU)
784 {
785 	rcu_read_unlock_bh();
786 }
787 
788 static int ip6fl_seq_show(struct seq_file *seq, void *v)
789 {
790 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
791 	if (v == SEQ_START_TOKEN) {
792 		seq_puts(seq, "Label S Owner  Users  Linger Expires  Dst                              Opt\n");
793 	} else {
794 		struct ip6_flowlabel *fl = v;
795 		seq_printf(seq,
796 			   "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
797 			   (unsigned int)ntohl(fl->label),
798 			   fl->share,
799 			   ((fl->share == IPV6_FL_S_PROCESS) ?
800 			    pid_nr_ns(fl->owner.pid, state->pid_ns) :
801 			    ((fl->share == IPV6_FL_S_USER) ?
802 			     from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
803 			     0)),
804 			   atomic_read(&fl->users),
805 			   fl->linger/HZ,
806 			   (long)(fl->expires - jiffies)/HZ,
807 			   &fl->dst,
808 			   fl->opt ? fl->opt->opt_nflen : 0);
809 	}
810 	return 0;
811 }
812 
813 static const struct seq_operations ip6fl_seq_ops = {
814 	.start	=	ip6fl_seq_start,
815 	.next	=	ip6fl_seq_next,
816 	.stop	=	ip6fl_seq_stop,
817 	.show	=	ip6fl_seq_show,
818 };
819 
820 static int __net_init ip6_flowlabel_proc_init(struct net *net)
821 {
822 	if (!proc_create_net("ip6_flowlabel", 0444, net->proc_net,
823 			&ip6fl_seq_ops, sizeof(struct ip6fl_iter_state)))
824 		return -ENOMEM;
825 	return 0;
826 }
827 
828 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
829 {
830 	remove_proc_entry("ip6_flowlabel", net->proc_net);
831 }
832 #else
833 static inline int ip6_flowlabel_proc_init(struct net *net)
834 {
835 	return 0;
836 }
837 static inline void ip6_flowlabel_proc_fini(struct net *net)
838 {
839 }
840 #endif
841 
842 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
843 {
844 	ip6_fl_purge(net);
845 	ip6_flowlabel_proc_fini(net);
846 }
847 
848 static struct pernet_operations ip6_flowlabel_net_ops = {
849 	.init = ip6_flowlabel_proc_init,
850 	.exit = ip6_flowlabel_net_exit,
851 };
852 
853 int ip6_flowlabel_init(void)
854 {
855 	return register_pernet_subsys(&ip6_flowlabel_net_ops);
856 }
857 
858 void ip6_flowlabel_cleanup(void)
859 {
860 	del_timer(&ip6_fl_gc_timer);
861 	unregister_pernet_subsys(&ip6_flowlabel_net_ops);
862 }
863