xref: /openbmc/linux/net/ipv6/ip6_flowlabel.c (revision da1d9caf)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	ip6_flowlabel.c		IPv6 flowlabel manager.
4  *
5  *	Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  */
7 
8 #include <linux/capability.h>
9 #include <linux/errno.h>
10 #include <linux/types.h>
11 #include <linux/socket.h>
12 #include <linux/net.h>
13 #include <linux/netdevice.h>
14 #include <linux/in6.h>
15 #include <linux/proc_fs.h>
16 #include <linux/seq_file.h>
17 #include <linux/slab.h>
18 #include <linux/export.h>
19 #include <linux/pid_namespace.h>
20 #include <linux/jump_label_ratelimit.h>
21 
22 #include <net/net_namespace.h>
23 #include <net/sock.h>
24 
25 #include <net/ipv6.h>
26 #include <net/rawv6.h>
27 #include <net/transp_v6.h>
28 
29 #include <linux/uaccess.h>
30 
31 #define FL_MIN_LINGER	6	/* Minimal linger. It is set to 6sec specified
32 				   in old IPv6 RFC. Well, it was reasonable value.
33 				 */
34 #define FL_MAX_LINGER	150	/* Maximal linger timeout */
35 
36 /* FL hash table */
37 
38 #define FL_MAX_PER_SOCK	32
39 #define FL_MAX_SIZE	4096
40 #define FL_HASH_MASK	255
41 #define FL_HASH(l)	(ntohl(l)&FL_HASH_MASK)
42 
43 static atomic_t fl_size = ATOMIC_INIT(0);
44 static struct ip6_flowlabel __rcu *fl_ht[FL_HASH_MASK+1];
45 
46 static void ip6_fl_gc(struct timer_list *unused);
47 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc);
48 
49 /* FL hash table lock: it protects only of GC */
50 
51 static DEFINE_SPINLOCK(ip6_fl_lock);
52 
53 /* Big socket sock */
54 
55 static DEFINE_SPINLOCK(ip6_sk_fl_lock);
56 
57 DEFINE_STATIC_KEY_DEFERRED_FALSE(ipv6_flowlabel_exclusive, HZ);
58 EXPORT_SYMBOL(ipv6_flowlabel_exclusive);
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 bool fl_shared_exclusive(struct ip6_flowlabel *fl)
98 {
99 	return fl->share == IPV6_FL_S_EXCL ||
100 	       fl->share == IPV6_FL_S_PROCESS ||
101 	       fl->share == IPV6_FL_S_USER;
102 }
103 
104 static void fl_free_rcu(struct rcu_head *head)
105 {
106 	struct ip6_flowlabel *fl = container_of(head, struct ip6_flowlabel, rcu);
107 
108 	if (fl->share == IPV6_FL_S_PROCESS)
109 		put_pid(fl->owner.pid);
110 	kfree(fl->opt);
111 	kfree(fl);
112 }
113 
114 
115 static void fl_free(struct ip6_flowlabel *fl)
116 {
117 	if (!fl)
118 		return;
119 
120 	if (fl_shared_exclusive(fl) || fl->opt)
121 		static_branch_slow_dec_deferred(&ipv6_flowlabel_exclusive);
122 
123 	call_rcu(&fl->rcu, fl_free_rcu);
124 }
125 
126 static void fl_release(struct ip6_flowlabel *fl)
127 {
128 	spin_lock_bh(&ip6_fl_lock);
129 
130 	fl->lastuse = jiffies;
131 	if (atomic_dec_and_test(&fl->users)) {
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 (fl->opt && fl->share == IPV6_FL_S_EXCL) {
137 			struct ipv6_txoptions *opt = fl->opt;
138 			fl->opt = NULL;
139 			kfree(opt);
140 		}
141 		if (!timer_pending(&ip6_fl_gc_timer) ||
142 		    time_after(ip6_fl_gc_timer.expires, ttd))
143 			mod_timer(&ip6_fl_gc_timer, ttd);
144 	}
145 	spin_unlock_bh(&ip6_fl_lock);
146 }
147 
148 static void ip6_fl_gc(struct timer_list *unused)
149 {
150 	int i;
151 	unsigned long now = jiffies;
152 	unsigned long sched = 0;
153 
154 	spin_lock(&ip6_fl_lock);
155 
156 	for (i = 0; i <= FL_HASH_MASK; i++) {
157 		struct ip6_flowlabel *fl;
158 		struct ip6_flowlabel __rcu **flp;
159 
160 		flp = &fl_ht[i];
161 		while ((fl = rcu_dereference_protected(*flp,
162 						       lockdep_is_held(&ip6_fl_lock))) != NULL) {
163 			if (atomic_read(&fl->users) == 0) {
164 				unsigned long ttd = fl->lastuse + fl->linger;
165 				if (time_after(ttd, fl->expires))
166 					fl->expires = ttd;
167 				ttd = fl->expires;
168 				if (time_after_eq(now, ttd)) {
169 					*flp = fl->next;
170 					fl_free(fl);
171 					atomic_dec(&fl_size);
172 					continue;
173 				}
174 				if (!sched || time_before(ttd, sched))
175 					sched = ttd;
176 			}
177 			flp = &fl->next;
178 		}
179 	}
180 	if (!sched && atomic_read(&fl_size))
181 		sched = now + FL_MAX_LINGER;
182 	if (sched) {
183 		mod_timer(&ip6_fl_gc_timer, sched);
184 	}
185 	spin_unlock(&ip6_fl_lock);
186 }
187 
188 static void __net_exit ip6_fl_purge(struct net *net)
189 {
190 	int i;
191 
192 	spin_lock_bh(&ip6_fl_lock);
193 	for (i = 0; i <= FL_HASH_MASK; i++) {
194 		struct ip6_flowlabel *fl;
195 		struct ip6_flowlabel __rcu **flp;
196 
197 		flp = &fl_ht[i];
198 		while ((fl = rcu_dereference_protected(*flp,
199 						       lockdep_is_held(&ip6_fl_lock))) != NULL) {
200 			if (net_eq(fl->fl_net, net) &&
201 			    atomic_read(&fl->users) == 0) {
202 				*flp = fl->next;
203 				fl_free(fl);
204 				atomic_dec(&fl_size);
205 				continue;
206 			}
207 			flp = &fl->next;
208 		}
209 	}
210 	spin_unlock_bh(&ip6_fl_lock);
211 }
212 
213 static struct ip6_flowlabel *fl_intern(struct net *net,
214 				       struct ip6_flowlabel *fl, __be32 label)
215 {
216 	struct ip6_flowlabel *lfl;
217 
218 	fl->label = label & IPV6_FLOWLABEL_MASK;
219 
220 	spin_lock_bh(&ip6_fl_lock);
221 	if (label == 0) {
222 		for (;;) {
223 			fl->label = htonl(prandom_u32())&IPV6_FLOWLABEL_MASK;
224 			if (fl->label) {
225 				lfl = __fl_lookup(net, fl->label);
226 				if (!lfl)
227 					break;
228 			}
229 		}
230 	} else {
231 		/*
232 		 * we dropper the ip6_fl_lock, so this entry could reappear
233 		 * and we need to recheck with it.
234 		 *
235 		 * OTOH no need to search the active socket first, like it is
236 		 * done in ipv6_flowlabel_opt - sock is locked, so new entry
237 		 * with the same label can only appear on another sock
238 		 */
239 		lfl = __fl_lookup(net, fl->label);
240 		if (lfl) {
241 			atomic_inc(&lfl->users);
242 			spin_unlock_bh(&ip6_fl_lock);
243 			return lfl;
244 		}
245 	}
246 
247 	fl->lastuse = jiffies;
248 	fl->next = fl_ht[FL_HASH(fl->label)];
249 	rcu_assign_pointer(fl_ht[FL_HASH(fl->label)], fl);
250 	atomic_inc(&fl_size);
251 	spin_unlock_bh(&ip6_fl_lock);
252 	return NULL;
253 }
254 
255 
256 
257 /* Socket flowlabel lists */
258 
259 struct ip6_flowlabel *__fl6_sock_lookup(struct sock *sk, __be32 label)
260 {
261 	struct ipv6_fl_socklist *sfl;
262 	struct ipv6_pinfo *np = inet6_sk(sk);
263 
264 	label &= IPV6_FLOWLABEL_MASK;
265 
266 	rcu_read_lock_bh();
267 	for_each_sk_fl_rcu(np, sfl) {
268 		struct ip6_flowlabel *fl = sfl->fl;
269 
270 		if (fl->label == label && atomic_inc_not_zero(&fl->users)) {
271 			fl->lastuse = jiffies;
272 			rcu_read_unlock_bh();
273 			return fl;
274 		}
275 	}
276 	rcu_read_unlock_bh();
277 	return NULL;
278 }
279 EXPORT_SYMBOL_GPL(__fl6_sock_lookup);
280 
281 void fl6_free_socklist(struct sock *sk)
282 {
283 	struct ipv6_pinfo *np = inet6_sk(sk);
284 	struct ipv6_fl_socklist *sfl;
285 
286 	if (!rcu_access_pointer(np->ipv6_fl_list))
287 		return;
288 
289 	spin_lock_bh(&ip6_sk_fl_lock);
290 	while ((sfl = rcu_dereference_protected(np->ipv6_fl_list,
291 						lockdep_is_held(&ip6_sk_fl_lock))) != NULL) {
292 		np->ipv6_fl_list = sfl->next;
293 		spin_unlock_bh(&ip6_sk_fl_lock);
294 
295 		fl_release(sfl->fl);
296 		kfree_rcu(sfl, rcu);
297 
298 		spin_lock_bh(&ip6_sk_fl_lock);
299 	}
300 	spin_unlock_bh(&ip6_sk_fl_lock);
301 }
302 
303 /* Service routines */
304 
305 
306 /*
307    It is the only difficult place. flowlabel enforces equal headers
308    before and including routing header, however user may supply options
309    following rthdr.
310  */
311 
312 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
313 					 struct ip6_flowlabel *fl,
314 					 struct ipv6_txoptions *fopt)
315 {
316 	struct ipv6_txoptions *fl_opt = fl->opt;
317 
318 	if (!fopt || fopt->opt_flen == 0)
319 		return fl_opt;
320 
321 	if (fl_opt) {
322 		opt_space->hopopt = fl_opt->hopopt;
323 		opt_space->dst0opt = fl_opt->dst0opt;
324 		opt_space->srcrt = fl_opt->srcrt;
325 		opt_space->opt_nflen = fl_opt->opt_nflen;
326 	} else {
327 		if (fopt->opt_nflen == 0)
328 			return fopt;
329 		opt_space->hopopt = NULL;
330 		opt_space->dst0opt = NULL;
331 		opt_space->srcrt = NULL;
332 		opt_space->opt_nflen = 0;
333 	}
334 	opt_space->dst1opt = fopt->dst1opt;
335 	opt_space->opt_flen = fopt->opt_flen;
336 	opt_space->tot_len = fopt->tot_len;
337 	return opt_space;
338 }
339 EXPORT_SYMBOL_GPL(fl6_merge_options);
340 
341 static unsigned long check_linger(unsigned long ttl)
342 {
343 	if (ttl < FL_MIN_LINGER)
344 		return FL_MIN_LINGER*HZ;
345 	if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
346 		return 0;
347 	return ttl*HZ;
348 }
349 
350 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
351 {
352 	linger = check_linger(linger);
353 	if (!linger)
354 		return -EPERM;
355 	expires = check_linger(expires);
356 	if (!expires)
357 		return -EPERM;
358 
359 	spin_lock_bh(&ip6_fl_lock);
360 	fl->lastuse = jiffies;
361 	if (time_before(fl->linger, linger))
362 		fl->linger = linger;
363 	if (time_before(expires, fl->linger))
364 		expires = fl->linger;
365 	if (time_before(fl->expires, fl->lastuse + expires))
366 		fl->expires = fl->lastuse + expires;
367 	spin_unlock_bh(&ip6_fl_lock);
368 
369 	return 0;
370 }
371 
372 static struct ip6_flowlabel *
373 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
374 	  sockptr_t optval, int optlen, int *err_p)
375 {
376 	struct ip6_flowlabel *fl = NULL;
377 	int olen;
378 	int addr_type;
379 	int err;
380 
381 	olen = optlen - CMSG_ALIGN(sizeof(*freq));
382 	err = -EINVAL;
383 	if (olen > 64 * 1024)
384 		goto done;
385 
386 	err = -ENOMEM;
387 	fl = kzalloc(sizeof(*fl), GFP_KERNEL);
388 	if (!fl)
389 		goto done;
390 
391 	if (olen > 0) {
392 		struct msghdr msg;
393 		struct flowi6 flowi6;
394 		struct ipcm6_cookie ipc6;
395 
396 		err = -ENOMEM;
397 		fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
398 		if (!fl->opt)
399 			goto done;
400 
401 		memset(fl->opt, 0, sizeof(*fl->opt));
402 		fl->opt->tot_len = sizeof(*fl->opt) + olen;
403 		err = -EFAULT;
404 		if (copy_from_sockptr_offset(fl->opt + 1, optval,
405 				CMSG_ALIGN(sizeof(*freq)), olen))
406 			goto done;
407 
408 		msg.msg_controllen = olen;
409 		msg.msg_control = (void *)(fl->opt+1);
410 		memset(&flowi6, 0, sizeof(flowi6));
411 
412 		ipc6.opt = fl->opt;
413 		err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, &ipc6);
414 		if (err)
415 			goto done;
416 		err = -EINVAL;
417 		if (fl->opt->opt_flen)
418 			goto done;
419 		if (fl->opt->opt_nflen == 0) {
420 			kfree(fl->opt);
421 			fl->opt = NULL;
422 		}
423 	}
424 
425 	fl->fl_net = net;
426 	fl->expires = jiffies;
427 	err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
428 	if (err)
429 		goto done;
430 	fl->share = freq->flr_share;
431 	addr_type = ipv6_addr_type(&freq->flr_dst);
432 	if ((addr_type & IPV6_ADDR_MAPPED) ||
433 	    addr_type == IPV6_ADDR_ANY) {
434 		err = -EINVAL;
435 		goto done;
436 	}
437 	fl->dst = freq->flr_dst;
438 	atomic_set(&fl->users, 1);
439 	switch (fl->share) {
440 	case IPV6_FL_S_EXCL:
441 	case IPV6_FL_S_ANY:
442 		break;
443 	case IPV6_FL_S_PROCESS:
444 		fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
445 		break;
446 	case IPV6_FL_S_USER:
447 		fl->owner.uid = current_euid();
448 		break;
449 	default:
450 		err = -EINVAL;
451 		goto done;
452 	}
453 	if (fl_shared_exclusive(fl) || fl->opt) {
454 		WRITE_ONCE(sock_net(sk)->ipv6.flowlabel_has_excl, 1);
455 		static_branch_deferred_inc(&ipv6_flowlabel_exclusive);
456 	}
457 	return fl;
458 
459 done:
460 	if (fl) {
461 		kfree(fl->opt);
462 		kfree(fl);
463 	}
464 	*err_p = err;
465 	return NULL;
466 }
467 
468 static int mem_check(struct sock *sk)
469 {
470 	struct ipv6_pinfo *np = inet6_sk(sk);
471 	struct ipv6_fl_socklist *sfl;
472 	int room = FL_MAX_SIZE - atomic_read(&fl_size);
473 	int count = 0;
474 
475 	if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
476 		return 0;
477 
478 	rcu_read_lock_bh();
479 	for_each_sk_fl_rcu(np, sfl)
480 		count++;
481 	rcu_read_unlock_bh();
482 
483 	if (room <= 0 ||
484 	    ((count >= FL_MAX_PER_SOCK ||
485 	      (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
486 	     !capable(CAP_NET_ADMIN)))
487 		return -ENOBUFS;
488 
489 	return 0;
490 }
491 
492 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
493 		struct ip6_flowlabel *fl)
494 {
495 	spin_lock_bh(&ip6_sk_fl_lock);
496 	sfl->fl = fl;
497 	sfl->next = np->ipv6_fl_list;
498 	rcu_assign_pointer(np->ipv6_fl_list, sfl);
499 	spin_unlock_bh(&ip6_sk_fl_lock);
500 }
501 
502 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
503 			   int flags)
504 {
505 	struct ipv6_pinfo *np = inet6_sk(sk);
506 	struct ipv6_fl_socklist *sfl;
507 
508 	if (flags & IPV6_FL_F_REMOTE) {
509 		freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
510 		return 0;
511 	}
512 
513 	if (np->repflow) {
514 		freq->flr_label = np->flow_label;
515 		return 0;
516 	}
517 
518 	rcu_read_lock_bh();
519 
520 	for_each_sk_fl_rcu(np, sfl) {
521 		if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
522 			spin_lock_bh(&ip6_fl_lock);
523 			freq->flr_label = sfl->fl->label;
524 			freq->flr_dst = sfl->fl->dst;
525 			freq->flr_share = sfl->fl->share;
526 			freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
527 			freq->flr_linger = sfl->fl->linger / HZ;
528 
529 			spin_unlock_bh(&ip6_fl_lock);
530 			rcu_read_unlock_bh();
531 			return 0;
532 		}
533 	}
534 	rcu_read_unlock_bh();
535 
536 	return -ENOENT;
537 }
538 
539 #define socklist_dereference(__sflp) \
540 	rcu_dereference_protected(__sflp, lockdep_is_held(&ip6_sk_fl_lock))
541 
542 static int ipv6_flowlabel_put(struct sock *sk, struct in6_flowlabel_req *freq)
543 {
544 	struct ipv6_pinfo *np = inet6_sk(sk);
545 	struct ipv6_fl_socklist __rcu **sflp;
546 	struct ipv6_fl_socklist *sfl;
547 
548 	if (freq->flr_flags & IPV6_FL_F_REFLECT) {
549 		if (sk->sk_protocol != IPPROTO_TCP)
550 			return -ENOPROTOOPT;
551 		if (!np->repflow)
552 			return -ESRCH;
553 		np->flow_label = 0;
554 		np->repflow = 0;
555 		return 0;
556 	}
557 
558 	spin_lock_bh(&ip6_sk_fl_lock);
559 	for (sflp = &np->ipv6_fl_list;
560 	     (sfl = socklist_dereference(*sflp)) != NULL;
561 	     sflp = &sfl->next) {
562 		if (sfl->fl->label == freq->flr_label)
563 			goto found;
564 	}
565 	spin_unlock_bh(&ip6_sk_fl_lock);
566 	return -ESRCH;
567 found:
568 	if (freq->flr_label == (np->flow_label & IPV6_FLOWLABEL_MASK))
569 		np->flow_label &= ~IPV6_FLOWLABEL_MASK;
570 	*sflp = sfl->next;
571 	spin_unlock_bh(&ip6_sk_fl_lock);
572 	fl_release(sfl->fl);
573 	kfree_rcu(sfl, rcu);
574 	return 0;
575 }
576 
577 static int ipv6_flowlabel_renew(struct sock *sk, struct in6_flowlabel_req *freq)
578 {
579 	struct ipv6_pinfo *np = inet6_sk(sk);
580 	struct net *net = sock_net(sk);
581 	struct ipv6_fl_socklist *sfl;
582 	int err;
583 
584 	rcu_read_lock_bh();
585 	for_each_sk_fl_rcu(np, sfl) {
586 		if (sfl->fl->label == freq->flr_label) {
587 			err = fl6_renew(sfl->fl, freq->flr_linger,
588 					freq->flr_expires);
589 			rcu_read_unlock_bh();
590 			return err;
591 		}
592 	}
593 	rcu_read_unlock_bh();
594 
595 	if (freq->flr_share == IPV6_FL_S_NONE &&
596 	    ns_capable(net->user_ns, CAP_NET_ADMIN)) {
597 		struct ip6_flowlabel *fl = fl_lookup(net, freq->flr_label);
598 
599 		if (fl) {
600 			err = fl6_renew(fl, freq->flr_linger,
601 					freq->flr_expires);
602 			fl_release(fl);
603 			return err;
604 		}
605 	}
606 	return -ESRCH;
607 }
608 
609 static int ipv6_flowlabel_get(struct sock *sk, struct in6_flowlabel_req *freq,
610 		sockptr_t optval, int optlen)
611 {
612 	struct ipv6_fl_socklist *sfl, *sfl1 = NULL;
613 	struct ip6_flowlabel *fl, *fl1 = NULL;
614 	struct ipv6_pinfo *np = inet6_sk(sk);
615 	struct net *net = sock_net(sk);
616 	int err;
617 
618 	if (freq->flr_flags & IPV6_FL_F_REFLECT) {
619 		if (net->ipv6.sysctl.flowlabel_consistency) {
620 			net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
621 			return -EPERM;
622 		}
623 
624 		if (sk->sk_protocol != IPPROTO_TCP)
625 			return -ENOPROTOOPT;
626 		np->repflow = 1;
627 		return 0;
628 	}
629 
630 	if (freq->flr_label & ~IPV6_FLOWLABEL_MASK)
631 		return -EINVAL;
632 	if (net->ipv6.sysctl.flowlabel_state_ranges &&
633 	    (freq->flr_label & IPV6_FLOWLABEL_STATELESS_FLAG))
634 		return -ERANGE;
635 
636 	fl = fl_create(net, sk, freq, optval, optlen, &err);
637 	if (!fl)
638 		return err;
639 
640 	sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
641 
642 	if (freq->flr_label) {
643 		err = -EEXIST;
644 		rcu_read_lock_bh();
645 		for_each_sk_fl_rcu(np, sfl) {
646 			if (sfl->fl->label == freq->flr_label) {
647 				if (freq->flr_flags & IPV6_FL_F_EXCL) {
648 					rcu_read_unlock_bh();
649 					goto done;
650 				}
651 				fl1 = sfl->fl;
652 				if (!atomic_inc_not_zero(&fl1->users))
653 					fl1 = NULL;
654 				break;
655 			}
656 		}
657 		rcu_read_unlock_bh();
658 
659 		if (!fl1)
660 			fl1 = fl_lookup(net, freq->flr_label);
661 		if (fl1) {
662 recheck:
663 			err = -EEXIST;
664 			if (freq->flr_flags&IPV6_FL_F_EXCL)
665 				goto release;
666 			err = -EPERM;
667 			if (fl1->share == IPV6_FL_S_EXCL ||
668 			    fl1->share != fl->share ||
669 			    ((fl1->share == IPV6_FL_S_PROCESS) &&
670 			     (fl1->owner.pid != fl->owner.pid)) ||
671 			    ((fl1->share == IPV6_FL_S_USER) &&
672 			     !uid_eq(fl1->owner.uid, fl->owner.uid)))
673 				goto release;
674 
675 			err = -ENOMEM;
676 			if (!sfl1)
677 				goto release;
678 			if (fl->linger > fl1->linger)
679 				fl1->linger = fl->linger;
680 			if ((long)(fl->expires - fl1->expires) > 0)
681 				fl1->expires = fl->expires;
682 			fl_link(np, sfl1, fl1);
683 			fl_free(fl);
684 			return 0;
685 
686 release:
687 			fl_release(fl1);
688 			goto done;
689 		}
690 	}
691 	err = -ENOENT;
692 	if (!(freq->flr_flags & IPV6_FL_F_CREATE))
693 		goto done;
694 
695 	err = -ENOMEM;
696 	if (!sfl1)
697 		goto done;
698 
699 	err = mem_check(sk);
700 	if (err != 0)
701 		goto done;
702 
703 	fl1 = fl_intern(net, fl, freq->flr_label);
704 	if (fl1)
705 		goto recheck;
706 
707 	if (!freq->flr_label) {
708 		size_t offset = offsetof(struct in6_flowlabel_req, flr_label);
709 
710 		if (copy_to_sockptr_offset(optval, offset, &fl->label,
711 				sizeof(fl->label))) {
712 			/* Intentionally ignore fault. */
713 		}
714 	}
715 
716 	fl_link(np, sfl1, fl);
717 	return 0;
718 done:
719 	fl_free(fl);
720 	kfree(sfl1);
721 	return err;
722 }
723 
724 int ipv6_flowlabel_opt(struct sock *sk, sockptr_t optval, int optlen)
725 {
726 	struct in6_flowlabel_req freq;
727 
728 	if (optlen < sizeof(freq))
729 		return -EINVAL;
730 	if (copy_from_sockptr(&freq, optval, sizeof(freq)))
731 		return -EFAULT;
732 
733 	switch (freq.flr_action) {
734 	case IPV6_FL_A_PUT:
735 		return ipv6_flowlabel_put(sk, &freq);
736 	case IPV6_FL_A_RENEW:
737 		return ipv6_flowlabel_renew(sk, &freq);
738 	case IPV6_FL_A_GET:
739 		return ipv6_flowlabel_get(sk, &freq, optval, optlen);
740 	default:
741 		return -EINVAL;
742 	}
743 }
744 
745 #ifdef CONFIG_PROC_FS
746 
747 struct ip6fl_iter_state {
748 	struct seq_net_private p;
749 	struct pid_namespace *pid_ns;
750 	int bucket;
751 };
752 
753 #define ip6fl_seq_private(seq)	((struct ip6fl_iter_state *)(seq)->private)
754 
755 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
756 {
757 	struct ip6_flowlabel *fl = NULL;
758 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
759 	struct net *net = seq_file_net(seq);
760 
761 	for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
762 		for_each_fl_rcu(state->bucket, fl) {
763 			if (net_eq(fl->fl_net, net))
764 				goto out;
765 		}
766 	}
767 	fl = NULL;
768 out:
769 	return fl;
770 }
771 
772 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
773 {
774 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
775 	struct net *net = seq_file_net(seq);
776 
777 	for_each_fl_continue_rcu(fl) {
778 		if (net_eq(fl->fl_net, net))
779 			goto out;
780 	}
781 
782 try_again:
783 	if (++state->bucket <= FL_HASH_MASK) {
784 		for_each_fl_rcu(state->bucket, fl) {
785 			if (net_eq(fl->fl_net, net))
786 				goto out;
787 		}
788 		goto try_again;
789 	}
790 	fl = NULL;
791 
792 out:
793 	return fl;
794 }
795 
796 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
797 {
798 	struct ip6_flowlabel *fl = ip6fl_get_first(seq);
799 	if (fl)
800 		while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
801 			--pos;
802 	return pos ? NULL : fl;
803 }
804 
805 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
806 	__acquires(RCU)
807 {
808 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
809 
810 	state->pid_ns = proc_pid_ns(file_inode(seq->file)->i_sb);
811 
812 	rcu_read_lock_bh();
813 	return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
814 }
815 
816 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
817 {
818 	struct ip6_flowlabel *fl;
819 
820 	if (v == SEQ_START_TOKEN)
821 		fl = ip6fl_get_first(seq);
822 	else
823 		fl = ip6fl_get_next(seq, v);
824 	++*pos;
825 	return fl;
826 }
827 
828 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
829 	__releases(RCU)
830 {
831 	rcu_read_unlock_bh();
832 }
833 
834 static int ip6fl_seq_show(struct seq_file *seq, void *v)
835 {
836 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
837 	if (v == SEQ_START_TOKEN) {
838 		seq_puts(seq, "Label S Owner  Users  Linger Expires  Dst                              Opt\n");
839 	} else {
840 		struct ip6_flowlabel *fl = v;
841 		seq_printf(seq,
842 			   "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
843 			   (unsigned int)ntohl(fl->label),
844 			   fl->share,
845 			   ((fl->share == IPV6_FL_S_PROCESS) ?
846 			    pid_nr_ns(fl->owner.pid, state->pid_ns) :
847 			    ((fl->share == IPV6_FL_S_USER) ?
848 			     from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
849 			     0)),
850 			   atomic_read(&fl->users),
851 			   fl->linger/HZ,
852 			   (long)(fl->expires - jiffies)/HZ,
853 			   &fl->dst,
854 			   fl->opt ? fl->opt->opt_nflen : 0);
855 	}
856 	return 0;
857 }
858 
859 static const struct seq_operations ip6fl_seq_ops = {
860 	.start	=	ip6fl_seq_start,
861 	.next	=	ip6fl_seq_next,
862 	.stop	=	ip6fl_seq_stop,
863 	.show	=	ip6fl_seq_show,
864 };
865 
866 static int __net_init ip6_flowlabel_proc_init(struct net *net)
867 {
868 	if (!proc_create_net("ip6_flowlabel", 0444, net->proc_net,
869 			&ip6fl_seq_ops, sizeof(struct ip6fl_iter_state)))
870 		return -ENOMEM;
871 	return 0;
872 }
873 
874 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
875 {
876 	remove_proc_entry("ip6_flowlabel", net->proc_net);
877 }
878 #else
879 static inline int ip6_flowlabel_proc_init(struct net *net)
880 {
881 	return 0;
882 }
883 static inline void ip6_flowlabel_proc_fini(struct net *net)
884 {
885 }
886 #endif
887 
888 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
889 {
890 	ip6_fl_purge(net);
891 	ip6_flowlabel_proc_fini(net);
892 }
893 
894 static struct pernet_operations ip6_flowlabel_net_ops = {
895 	.init = ip6_flowlabel_proc_init,
896 	.exit = ip6_flowlabel_net_exit,
897 };
898 
899 int ip6_flowlabel_init(void)
900 {
901 	return register_pernet_subsys(&ip6_flowlabel_net_ops);
902 }
903 
904 void ip6_flowlabel_cleanup(void)
905 {
906 	static_key_deferred_flush(&ipv6_flowlabel_exclusive);
907 	del_timer(&ip6_fl_gc_timer);
908 	unregister_pernet_subsys(&ip6_flowlabel_net_ops);
909 }
910