xref: /openbmc/linux/net/ipv6/ip6_flowlabel.c (revision 3d9b448b)
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 <asm/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(unsigned long dummy);
50 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc, 0, 0);
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 
98 static void fl_free(struct ip6_flowlabel *fl)
99 {
100 	if (fl) {
101 		if (fl->share == IPV6_FL_S_PROCESS)
102 			put_pid(fl->owner.pid);
103 		release_net(fl->fl_net);
104 		kfree(fl->opt);
105 		kfree_rcu(fl, rcu);
106 	}
107 }
108 
109 static void fl_release(struct ip6_flowlabel *fl)
110 {
111 	spin_lock_bh(&ip6_fl_lock);
112 
113 	fl->lastuse = jiffies;
114 	if (atomic_dec_and_test(&fl->users)) {
115 		unsigned long ttd = fl->lastuse + fl->linger;
116 		if (time_after(ttd, fl->expires))
117 			fl->expires = ttd;
118 		ttd = fl->expires;
119 		if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
120 			struct ipv6_txoptions *opt = fl->opt;
121 			fl->opt = NULL;
122 			kfree(opt);
123 		}
124 		if (!timer_pending(&ip6_fl_gc_timer) ||
125 		    time_after(ip6_fl_gc_timer.expires, ttd))
126 			mod_timer(&ip6_fl_gc_timer, ttd);
127 	}
128 	spin_unlock_bh(&ip6_fl_lock);
129 }
130 
131 static void ip6_fl_gc(unsigned long dummy)
132 {
133 	int i;
134 	unsigned long now = jiffies;
135 	unsigned long sched = 0;
136 
137 	spin_lock(&ip6_fl_lock);
138 
139 	for (i = 0; i <= FL_HASH_MASK; i++) {
140 		struct ip6_flowlabel *fl;
141 		struct ip6_flowlabel __rcu **flp;
142 
143 		flp = &fl_ht[i];
144 		while ((fl = rcu_dereference_protected(*flp,
145 						       lockdep_is_held(&ip6_fl_lock))) != NULL) {
146 			if (atomic_read(&fl->users) == 0) {
147 				unsigned long ttd = fl->lastuse + fl->linger;
148 				if (time_after(ttd, fl->expires))
149 					fl->expires = ttd;
150 				ttd = fl->expires;
151 				if (time_after_eq(now, ttd)) {
152 					*flp = fl->next;
153 					fl_free(fl);
154 					atomic_dec(&fl_size);
155 					continue;
156 				}
157 				if (!sched || time_before(ttd, sched))
158 					sched = ttd;
159 			}
160 			flp = &fl->next;
161 		}
162 	}
163 	if (!sched && atomic_read(&fl_size))
164 		sched = now + FL_MAX_LINGER;
165 	if (sched) {
166 		mod_timer(&ip6_fl_gc_timer, sched);
167 	}
168 	spin_unlock(&ip6_fl_lock);
169 }
170 
171 static void __net_exit ip6_fl_purge(struct net *net)
172 {
173 	int i;
174 
175 	spin_lock_bh(&ip6_fl_lock);
176 	for (i = 0; i <= FL_HASH_MASK; i++) {
177 		struct ip6_flowlabel *fl;
178 		struct ip6_flowlabel __rcu **flp;
179 
180 		flp = &fl_ht[i];
181 		while ((fl = rcu_dereference_protected(*flp,
182 						       lockdep_is_held(&ip6_fl_lock))) != NULL) {
183 			if (net_eq(fl->fl_net, net) &&
184 			    atomic_read(&fl->users) == 0) {
185 				*flp = fl->next;
186 				fl_free(fl);
187 				atomic_dec(&fl_size);
188 				continue;
189 			}
190 			flp = &fl->next;
191 		}
192 	}
193 	spin_unlock_bh(&ip6_fl_lock);
194 }
195 
196 static struct ip6_flowlabel *fl_intern(struct net *net,
197 				       struct ip6_flowlabel *fl, __be32 label)
198 {
199 	struct ip6_flowlabel *lfl;
200 
201 	fl->label = label & IPV6_FLOWLABEL_MASK;
202 
203 	spin_lock_bh(&ip6_fl_lock);
204 	if (label == 0) {
205 		for (;;) {
206 			fl->label = htonl(prandom_u32())&IPV6_FLOWLABEL_MASK;
207 			if (fl->label) {
208 				lfl = __fl_lookup(net, fl->label);
209 				if (lfl == NULL)
210 					break;
211 			}
212 		}
213 	} else {
214 		/*
215 		 * we dropper the ip6_fl_lock, so this entry could reappear
216 		 * and we need to recheck with it.
217 		 *
218 		 * OTOH no need to search the active socket first, like it is
219 		 * done in ipv6_flowlabel_opt - sock is locked, so new entry
220 		 * with the same label can only appear on another sock
221 		 */
222 		lfl = __fl_lookup(net, fl->label);
223 		if (lfl != NULL) {
224 			atomic_inc(&lfl->users);
225 			spin_unlock_bh(&ip6_fl_lock);
226 			return lfl;
227 		}
228 	}
229 
230 	fl->lastuse = jiffies;
231 	fl->next = fl_ht[FL_HASH(fl->label)];
232 	rcu_assign_pointer(fl_ht[FL_HASH(fl->label)], fl);
233 	atomic_inc(&fl_size);
234 	spin_unlock_bh(&ip6_fl_lock);
235 	return NULL;
236 }
237 
238 
239 
240 /* Socket flowlabel lists */
241 
242 struct ip6_flowlabel *fl6_sock_lookup(struct sock *sk, __be32 label)
243 {
244 	struct ipv6_fl_socklist *sfl;
245 	struct ipv6_pinfo *np = inet6_sk(sk);
246 
247 	label &= IPV6_FLOWLABEL_MASK;
248 
249 	rcu_read_lock_bh();
250 	for_each_sk_fl_rcu(np, sfl) {
251 		struct ip6_flowlabel *fl = sfl->fl;
252 		if (fl->label == label) {
253 			fl->lastuse = jiffies;
254 			atomic_inc(&fl->users);
255 			rcu_read_unlock_bh();
256 			return fl;
257 		}
258 	}
259 	rcu_read_unlock_bh();
260 	return NULL;
261 }
262 EXPORT_SYMBOL_GPL(fl6_sock_lookup);
263 
264 void fl6_free_socklist(struct sock *sk)
265 {
266 	struct ipv6_pinfo *np = inet6_sk(sk);
267 	struct ipv6_fl_socklist *sfl;
268 
269 	if (!rcu_access_pointer(np->ipv6_fl_list))
270 		return;
271 
272 	spin_lock_bh(&ip6_sk_fl_lock);
273 	while ((sfl = rcu_dereference_protected(np->ipv6_fl_list,
274 						lockdep_is_held(&ip6_sk_fl_lock))) != NULL) {
275 		np->ipv6_fl_list = sfl->next;
276 		spin_unlock_bh(&ip6_sk_fl_lock);
277 
278 		fl_release(sfl->fl);
279 		kfree_rcu(sfl, rcu);
280 
281 		spin_lock_bh(&ip6_sk_fl_lock);
282 	}
283 	spin_unlock_bh(&ip6_sk_fl_lock);
284 }
285 
286 /* Service routines */
287 
288 
289 /*
290    It is the only difficult place. flowlabel enforces equal headers
291    before and including routing header, however user may supply options
292    following rthdr.
293  */
294 
295 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
296 					 struct ip6_flowlabel *fl,
297 					 struct ipv6_txoptions *fopt)
298 {
299 	struct ipv6_txoptions *fl_opt = fl->opt;
300 
301 	if (fopt == NULL || fopt->opt_flen == 0)
302 		return fl_opt;
303 
304 	if (fl_opt != NULL) {
305 		opt_space->hopopt = fl_opt->hopopt;
306 		opt_space->dst0opt = fl_opt->dst0opt;
307 		opt_space->srcrt = fl_opt->srcrt;
308 		opt_space->opt_nflen = fl_opt->opt_nflen;
309 	} else {
310 		if (fopt->opt_nflen == 0)
311 			return fopt;
312 		opt_space->hopopt = NULL;
313 		opt_space->dst0opt = NULL;
314 		opt_space->srcrt = NULL;
315 		opt_space->opt_nflen = 0;
316 	}
317 	opt_space->dst1opt = fopt->dst1opt;
318 	opt_space->opt_flen = fopt->opt_flen;
319 	return opt_space;
320 }
321 EXPORT_SYMBOL_GPL(fl6_merge_options);
322 
323 static unsigned long check_linger(unsigned long ttl)
324 {
325 	if (ttl < FL_MIN_LINGER)
326 		return FL_MIN_LINGER*HZ;
327 	if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
328 		return 0;
329 	return ttl*HZ;
330 }
331 
332 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
333 {
334 	linger = check_linger(linger);
335 	if (!linger)
336 		return -EPERM;
337 	expires = check_linger(expires);
338 	if (!expires)
339 		return -EPERM;
340 
341 	spin_lock_bh(&ip6_fl_lock);
342 	fl->lastuse = jiffies;
343 	if (time_before(fl->linger, linger))
344 		fl->linger = linger;
345 	if (time_before(expires, fl->linger))
346 		expires = fl->linger;
347 	if (time_before(fl->expires, fl->lastuse + expires))
348 		fl->expires = fl->lastuse + expires;
349 	spin_unlock_bh(&ip6_fl_lock);
350 
351 	return 0;
352 }
353 
354 static struct ip6_flowlabel *
355 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
356 	  char __user *optval, int optlen, int *err_p)
357 {
358 	struct ip6_flowlabel *fl = NULL;
359 	int olen;
360 	int addr_type;
361 	int err;
362 
363 	olen = optlen - CMSG_ALIGN(sizeof(*freq));
364 	err = -EINVAL;
365 	if (olen > 64 * 1024)
366 		goto done;
367 
368 	err = -ENOMEM;
369 	fl = kzalloc(sizeof(*fl), GFP_KERNEL);
370 	if (fl == NULL)
371 		goto done;
372 
373 	if (olen > 0) {
374 		struct msghdr msg;
375 		struct flowi6 flowi6;
376 		int junk;
377 
378 		err = -ENOMEM;
379 		fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
380 		if (fl->opt == NULL)
381 			goto done;
382 
383 		memset(fl->opt, 0, sizeof(*fl->opt));
384 		fl->opt->tot_len = sizeof(*fl->opt) + olen;
385 		err = -EFAULT;
386 		if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
387 			goto done;
388 
389 		msg.msg_controllen = olen;
390 		msg.msg_control = (void *)(fl->opt+1);
391 		memset(&flowi6, 0, sizeof(flowi6));
392 
393 		err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, fl->opt,
394 					    &junk, &junk, &junk);
395 		if (err)
396 			goto done;
397 		err = -EINVAL;
398 		if (fl->opt->opt_flen)
399 			goto done;
400 		if (fl->opt->opt_nflen == 0) {
401 			kfree(fl->opt);
402 			fl->opt = NULL;
403 		}
404 	}
405 
406 	fl->fl_net = hold_net(net);
407 	fl->expires = jiffies;
408 	err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
409 	if (err)
410 		goto done;
411 	fl->share = freq->flr_share;
412 	addr_type = ipv6_addr_type(&freq->flr_dst);
413 	if ((addr_type & IPV6_ADDR_MAPPED) ||
414 	    addr_type == IPV6_ADDR_ANY) {
415 		err = -EINVAL;
416 		goto done;
417 	}
418 	fl->dst = freq->flr_dst;
419 	atomic_set(&fl->users, 1);
420 	switch (fl->share) {
421 	case IPV6_FL_S_EXCL:
422 	case IPV6_FL_S_ANY:
423 		break;
424 	case IPV6_FL_S_PROCESS:
425 		fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
426 		break;
427 	case IPV6_FL_S_USER:
428 		fl->owner.uid = current_euid();
429 		break;
430 	default:
431 		err = -EINVAL;
432 		goto done;
433 	}
434 	return fl;
435 
436 done:
437 	fl_free(fl);
438 	*err_p = err;
439 	return NULL;
440 }
441 
442 static int mem_check(struct sock *sk)
443 {
444 	struct ipv6_pinfo *np = inet6_sk(sk);
445 	struct ipv6_fl_socklist *sfl;
446 	int room = FL_MAX_SIZE - atomic_read(&fl_size);
447 	int count = 0;
448 
449 	if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
450 		return 0;
451 
452 	rcu_read_lock_bh();
453 	for_each_sk_fl_rcu(np, sfl)
454 		count++;
455 	rcu_read_unlock_bh();
456 
457 	if (room <= 0 ||
458 	    ((count >= FL_MAX_PER_SOCK ||
459 	      (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
460 	     !capable(CAP_NET_ADMIN)))
461 		return -ENOBUFS;
462 
463 	return 0;
464 }
465 
466 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
467 		struct ip6_flowlabel *fl)
468 {
469 	spin_lock_bh(&ip6_sk_fl_lock);
470 	sfl->fl = fl;
471 	sfl->next = np->ipv6_fl_list;
472 	rcu_assign_pointer(np->ipv6_fl_list, sfl);
473 	spin_unlock_bh(&ip6_sk_fl_lock);
474 }
475 
476 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
477 			   int flags)
478 {
479 	struct ipv6_pinfo *np = inet6_sk(sk);
480 	struct ipv6_fl_socklist *sfl;
481 
482 	if (flags & IPV6_FL_F_REMOTE) {
483 		freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
484 		return 0;
485 	}
486 
487 	if (np->repflow) {
488 		freq->flr_label = np->flow_label;
489 		return 0;
490 	}
491 
492 	rcu_read_lock_bh();
493 
494 	for_each_sk_fl_rcu(np, sfl) {
495 		if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
496 			spin_lock_bh(&ip6_fl_lock);
497 			freq->flr_label = sfl->fl->label;
498 			freq->flr_dst = sfl->fl->dst;
499 			freq->flr_share = sfl->fl->share;
500 			freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
501 			freq->flr_linger = sfl->fl->linger / HZ;
502 
503 			spin_unlock_bh(&ip6_fl_lock);
504 			rcu_read_unlock_bh();
505 			return 0;
506 		}
507 	}
508 	rcu_read_unlock_bh();
509 
510 	return -ENOENT;
511 }
512 
513 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
514 {
515 	int uninitialized_var(err);
516 	struct net *net = sock_net(sk);
517 	struct ipv6_pinfo *np = inet6_sk(sk);
518 	struct in6_flowlabel_req freq;
519 	struct ipv6_fl_socklist *sfl1 = NULL;
520 	struct ipv6_fl_socklist *sfl;
521 	struct ipv6_fl_socklist __rcu **sflp;
522 	struct ip6_flowlabel *fl, *fl1 = NULL;
523 
524 
525 	if (optlen < sizeof(freq))
526 		return -EINVAL;
527 
528 	if (copy_from_user(&freq, optval, sizeof(freq)))
529 		return -EFAULT;
530 
531 	switch (freq.flr_action) {
532 	case IPV6_FL_A_PUT:
533 		if (freq.flr_flags & IPV6_FL_F_REFLECT) {
534 			if (sk->sk_protocol != IPPROTO_TCP)
535 				return -ENOPROTOOPT;
536 			if (!np->repflow)
537 				return -ESRCH;
538 			np->flow_label = 0;
539 			np->repflow = 0;
540 			return 0;
541 		}
542 		spin_lock_bh(&ip6_sk_fl_lock);
543 		for (sflp = &np->ipv6_fl_list;
544 		     (sfl = rcu_dereference(*sflp)) != NULL;
545 		     sflp = &sfl->next) {
546 			if (sfl->fl->label == freq.flr_label) {
547 				if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
548 					np->flow_label &= ~IPV6_FLOWLABEL_MASK;
549 				*sflp = rcu_dereference(sfl->next);
550 				spin_unlock_bh(&ip6_sk_fl_lock);
551 				fl_release(sfl->fl);
552 				kfree_rcu(sfl, rcu);
553 				return 0;
554 			}
555 		}
556 		spin_unlock_bh(&ip6_sk_fl_lock);
557 		return -ESRCH;
558 
559 	case IPV6_FL_A_RENEW:
560 		rcu_read_lock_bh();
561 		for_each_sk_fl_rcu(np, sfl) {
562 			if (sfl->fl->label == freq.flr_label) {
563 				err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
564 				rcu_read_unlock_bh();
565 				return err;
566 			}
567 		}
568 		rcu_read_unlock_bh();
569 
570 		if (freq.flr_share == IPV6_FL_S_NONE &&
571 		    ns_capable(net->user_ns, CAP_NET_ADMIN)) {
572 			fl = fl_lookup(net, freq.flr_label);
573 			if (fl) {
574 				err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
575 				fl_release(fl);
576 				return err;
577 			}
578 		}
579 		return -ESRCH;
580 
581 	case IPV6_FL_A_GET:
582 		if (freq.flr_flags & IPV6_FL_F_REFLECT) {
583 			struct net *net = sock_net(sk);
584 			if (net->ipv6.sysctl.flowlabel_consistency) {
585 				net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
586 				return -EPERM;
587 			}
588 
589 			if (sk->sk_protocol != IPPROTO_TCP)
590 				return -ENOPROTOOPT;
591 
592 			np->repflow = 1;
593 			return 0;
594 		}
595 
596 		if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
597 			return -EINVAL;
598 
599 		fl = fl_create(net, sk, &freq, optval, optlen, &err);
600 		if (fl == NULL)
601 			return err;
602 		sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
603 
604 		if (freq.flr_label) {
605 			err = -EEXIST;
606 			rcu_read_lock_bh();
607 			for_each_sk_fl_rcu(np, sfl) {
608 				if (sfl->fl->label == freq.flr_label) {
609 					if (freq.flr_flags&IPV6_FL_F_EXCL) {
610 						rcu_read_unlock_bh();
611 						goto done;
612 					}
613 					fl1 = sfl->fl;
614 					atomic_inc(&fl1->users);
615 					break;
616 				}
617 			}
618 			rcu_read_unlock_bh();
619 
620 			if (fl1 == NULL)
621 				fl1 = fl_lookup(net, freq.flr_label);
622 			if (fl1) {
623 recheck:
624 				err = -EEXIST;
625 				if (freq.flr_flags&IPV6_FL_F_EXCL)
626 					goto release;
627 				err = -EPERM;
628 				if (fl1->share == IPV6_FL_S_EXCL ||
629 				    fl1->share != fl->share ||
630 				    ((fl1->share == IPV6_FL_S_PROCESS) &&
631 				     (fl1->owner.pid == fl->owner.pid)) ||
632 				    ((fl1->share == IPV6_FL_S_USER) &&
633 				     uid_eq(fl1->owner.uid, fl->owner.uid)))
634 					goto release;
635 
636 				err = -ENOMEM;
637 				if (sfl1 == NULL)
638 					goto release;
639 				if (fl->linger > fl1->linger)
640 					fl1->linger = fl->linger;
641 				if ((long)(fl->expires - fl1->expires) > 0)
642 					fl1->expires = fl->expires;
643 				fl_link(np, sfl1, fl1);
644 				fl_free(fl);
645 				return 0;
646 
647 release:
648 				fl_release(fl1);
649 				goto done;
650 			}
651 		}
652 		err = -ENOENT;
653 		if (!(freq.flr_flags&IPV6_FL_F_CREATE))
654 			goto done;
655 
656 		err = -ENOMEM;
657 		if (sfl1 == NULL)
658 			goto done;
659 
660 		err = mem_check(sk);
661 		if (err != 0)
662 			goto done;
663 
664 		fl1 = fl_intern(net, fl, freq.flr_label);
665 		if (fl1 != NULL)
666 			goto recheck;
667 
668 		if (!freq.flr_label) {
669 			if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
670 					 &fl->label, sizeof(fl->label))) {
671 				/* Intentionally ignore fault. */
672 			}
673 		}
674 
675 		fl_link(np, sfl1, fl);
676 		return 0;
677 
678 	default:
679 		return -EINVAL;
680 	}
681 
682 done:
683 	fl_free(fl);
684 	kfree(sfl1);
685 	return err;
686 }
687 
688 #ifdef CONFIG_PROC_FS
689 
690 struct ip6fl_iter_state {
691 	struct seq_net_private p;
692 	struct pid_namespace *pid_ns;
693 	int bucket;
694 };
695 
696 #define ip6fl_seq_private(seq)	((struct ip6fl_iter_state *)(seq)->private)
697 
698 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
699 {
700 	struct ip6_flowlabel *fl = NULL;
701 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
702 	struct net *net = seq_file_net(seq);
703 
704 	for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
705 		for_each_fl_rcu(state->bucket, fl) {
706 			if (net_eq(fl->fl_net, net))
707 				goto out;
708 		}
709 	}
710 	fl = NULL;
711 out:
712 	return fl;
713 }
714 
715 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
716 {
717 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
718 	struct net *net = seq_file_net(seq);
719 
720 	for_each_fl_continue_rcu(fl) {
721 		if (net_eq(fl->fl_net, net))
722 			goto out;
723 	}
724 
725 try_again:
726 	if (++state->bucket <= FL_HASH_MASK) {
727 		for_each_fl_rcu(state->bucket, fl) {
728 			if (net_eq(fl->fl_net, net))
729 				goto out;
730 		}
731 		goto try_again;
732 	}
733 	fl = NULL;
734 
735 out:
736 	return fl;
737 }
738 
739 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
740 {
741 	struct ip6_flowlabel *fl = ip6fl_get_first(seq);
742 	if (fl)
743 		while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
744 			--pos;
745 	return pos ? NULL : fl;
746 }
747 
748 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
749 	__acquires(RCU)
750 {
751 	rcu_read_lock_bh();
752 	return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
753 }
754 
755 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
756 {
757 	struct ip6_flowlabel *fl;
758 
759 	if (v == SEQ_START_TOKEN)
760 		fl = ip6fl_get_first(seq);
761 	else
762 		fl = ip6fl_get_next(seq, v);
763 	++*pos;
764 	return fl;
765 }
766 
767 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
768 	__releases(RCU)
769 {
770 	rcu_read_unlock_bh();
771 }
772 
773 static int ip6fl_seq_show(struct seq_file *seq, void *v)
774 {
775 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
776 	if (v == SEQ_START_TOKEN) {
777 		seq_puts(seq, "Label S Owner  Users  Linger Expires  Dst                              Opt\n");
778 	} else {
779 		struct ip6_flowlabel *fl = v;
780 		seq_printf(seq,
781 			   "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
782 			   (unsigned int)ntohl(fl->label),
783 			   fl->share,
784 			   ((fl->share == IPV6_FL_S_PROCESS) ?
785 			    pid_nr_ns(fl->owner.pid, state->pid_ns) :
786 			    ((fl->share == IPV6_FL_S_USER) ?
787 			     from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
788 			     0)),
789 			   atomic_read(&fl->users),
790 			   fl->linger/HZ,
791 			   (long)(fl->expires - jiffies)/HZ,
792 			   &fl->dst,
793 			   fl->opt ? fl->opt->opt_nflen : 0);
794 	}
795 	return 0;
796 }
797 
798 static const struct seq_operations ip6fl_seq_ops = {
799 	.start	=	ip6fl_seq_start,
800 	.next	=	ip6fl_seq_next,
801 	.stop	=	ip6fl_seq_stop,
802 	.show	=	ip6fl_seq_show,
803 };
804 
805 static int ip6fl_seq_open(struct inode *inode, struct file *file)
806 {
807 	struct seq_file *seq;
808 	struct ip6fl_iter_state *state;
809 	int err;
810 
811 	err = seq_open_net(inode, file, &ip6fl_seq_ops,
812 			   sizeof(struct ip6fl_iter_state));
813 
814 	if (!err) {
815 		seq = file->private_data;
816 		state = ip6fl_seq_private(seq);
817 		rcu_read_lock();
818 		state->pid_ns = get_pid_ns(task_active_pid_ns(current));
819 		rcu_read_unlock();
820 	}
821 	return err;
822 }
823 
824 static int ip6fl_seq_release(struct inode *inode, struct file *file)
825 {
826 	struct seq_file *seq = file->private_data;
827 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
828 	put_pid_ns(state->pid_ns);
829 	return seq_release_net(inode, file);
830 }
831 
832 static const struct file_operations ip6fl_seq_fops = {
833 	.owner		=	THIS_MODULE,
834 	.open		=	ip6fl_seq_open,
835 	.read		=	seq_read,
836 	.llseek		=	seq_lseek,
837 	.release	=	ip6fl_seq_release,
838 };
839 
840 static int __net_init ip6_flowlabel_proc_init(struct net *net)
841 {
842 	if (!proc_create("ip6_flowlabel", S_IRUGO, net->proc_net,
843 			 &ip6fl_seq_fops))
844 		return -ENOMEM;
845 	return 0;
846 }
847 
848 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
849 {
850 	remove_proc_entry("ip6_flowlabel", net->proc_net);
851 }
852 #else
853 static inline int ip6_flowlabel_proc_init(struct net *net)
854 {
855 	return 0;
856 }
857 static inline void ip6_flowlabel_proc_fini(struct net *net)
858 {
859 }
860 #endif
861 
862 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
863 {
864 	ip6_fl_purge(net);
865 	ip6_flowlabel_proc_fini(net);
866 }
867 
868 static struct pernet_operations ip6_flowlabel_net_ops = {
869 	.init = ip6_flowlabel_proc_init,
870 	.exit = ip6_flowlabel_net_exit,
871 };
872 
873 int ip6_flowlabel_init(void)
874 {
875 	return register_pernet_subsys(&ip6_flowlabel_net_ops);
876 }
877 
878 void ip6_flowlabel_cleanup(void)
879 {
880 	del_timer(&ip6_fl_gc_timer);
881 	unregister_pernet_subsys(&ip6_flowlabel_net_ops);
882 }
883