xref: /linux/net/ipv6/ip6_flowlabel.c (revision 8f7aa3d3c7323f4ca2768a9e74ebbe359c4f8f88)
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(fl_ht[(hash)]);		\
62 	     fl != NULL;					\
63 	     fl = rcu_dereference(fl->next))
64 #define for_each_fl_continue_rcu(fl)				\
65 	for (fl = rcu_dereference(fl->next);			\
66 	     fl != NULL;					\
67 	     fl = rcu_dereference(fl->next))
68 
69 #define for_each_sk_fl_rcu(sk, sfl)				\
70 	for (sfl = rcu_dereference(inet_sk(sk)->ipv6_fl_list);	\
71 	     sfl != NULL;					\
72 	     sfl = rcu_dereference(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();
90 	fl = __fl_lookup(net, label);
91 	if (fl && !atomic_inc_not_zero(&fl->users))
92 		fl = NULL;
93 	rcu_read_unlock();
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 	rcu_read_lock();
221 	spin_lock_bh(&ip6_fl_lock);
222 	if (label == 0) {
223 		for (;;) {
224 			fl->label = htonl(get_random_u32())&IPV6_FLOWLABEL_MASK;
225 			if (fl->label) {
226 				lfl = __fl_lookup(net, fl->label);
227 				if (!lfl)
228 					break;
229 			}
230 		}
231 	} else {
232 		/*
233 		 * we dropper the ip6_fl_lock, so this entry could reappear
234 		 * and we need to recheck with it.
235 		 *
236 		 * OTOH no need to search the active socket first, like it is
237 		 * done in ipv6_flowlabel_opt - sock is locked, so new entry
238 		 * with the same label can only appear on another sock
239 		 */
240 		lfl = __fl_lookup(net, fl->label);
241 		if (lfl) {
242 			atomic_inc(&lfl->users);
243 			spin_unlock_bh(&ip6_fl_lock);
244 			rcu_read_unlock();
245 			return lfl;
246 		}
247 	}
248 
249 	fl->lastuse = jiffies;
250 	fl->next = fl_ht[FL_HASH(fl->label)];
251 	rcu_assign_pointer(fl_ht[FL_HASH(fl->label)], fl);
252 	atomic_inc(&fl_size);
253 	spin_unlock_bh(&ip6_fl_lock);
254 	rcu_read_unlock();
255 	return NULL;
256 }
257 
258 
259 
260 /* Socket flowlabel lists */
261 
262 struct ip6_flowlabel *__fl6_sock_lookup(struct sock *sk, __be32 label)
263 {
264 	struct ipv6_fl_socklist *sfl;
265 
266 	label &= IPV6_FLOWLABEL_MASK;
267 
268 	rcu_read_lock();
269 	for_each_sk_fl_rcu(sk, sfl) {
270 		struct ip6_flowlabel *fl = sfl->fl;
271 
272 		if (fl->label == label && atomic_inc_not_zero(&fl->users)) {
273 			fl->lastuse = jiffies;
274 			rcu_read_unlock();
275 			return fl;
276 		}
277 	}
278 	rcu_read_unlock();
279 	return NULL;
280 }
281 EXPORT_SYMBOL_GPL(__fl6_sock_lookup);
282 
283 void fl6_free_socklist(struct sock *sk)
284 {
285 	struct inet_sock *inet = inet_sk(sk);
286 	struct ipv6_fl_socklist *sfl;
287 
288 	if (!rcu_access_pointer(inet->ipv6_fl_list))
289 		return;
290 
291 	spin_lock_bh(&ip6_sk_fl_lock);
292 	while ((sfl = rcu_dereference_protected(inet->ipv6_fl_list,
293 						lockdep_is_held(&ip6_sk_fl_lock))) != NULL) {
294 		inet->ipv6_fl_list = sfl->next;
295 		spin_unlock_bh(&ip6_sk_fl_lock);
296 
297 		fl_release(sfl->fl);
298 		kfree_rcu(sfl, rcu);
299 
300 		spin_lock_bh(&ip6_sk_fl_lock);
301 	}
302 	spin_unlock_bh(&ip6_sk_fl_lock);
303 }
304 
305 /* Service routines */
306 
307 
308 /*
309    It is the only difficult place. flowlabel enforces equal headers
310    before and including routing header, however user may supply options
311    following rthdr.
312  */
313 
314 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
315 					 struct ip6_flowlabel *fl,
316 					 struct ipv6_txoptions *fopt)
317 {
318 	struct ipv6_txoptions *fl_opt = fl->opt;
319 
320 	if (!fopt || fopt->opt_flen == 0)
321 		return fl_opt;
322 
323 	if (fl_opt) {
324 		opt_space->hopopt = fl_opt->hopopt;
325 		opt_space->dst0opt = fl_opt->dst0opt;
326 		opt_space->srcrt = fl_opt->srcrt;
327 		opt_space->opt_nflen = fl_opt->opt_nflen;
328 	} else {
329 		if (fopt->opt_nflen == 0)
330 			return fopt;
331 		opt_space->hopopt = NULL;
332 		opt_space->dst0opt = NULL;
333 		opt_space->srcrt = NULL;
334 		opt_space->opt_nflen = 0;
335 	}
336 	opt_space->dst1opt = fopt->dst1opt;
337 	opt_space->opt_flen = fopt->opt_flen;
338 	opt_space->tot_len = fopt->tot_len;
339 	return opt_space;
340 }
341 EXPORT_SYMBOL_GPL(fl6_merge_options);
342 
343 static unsigned long check_linger(unsigned long ttl)
344 {
345 	if (ttl < FL_MIN_LINGER)
346 		return FL_MIN_LINGER*HZ;
347 	if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
348 		return 0;
349 	return ttl*HZ;
350 }
351 
352 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
353 {
354 	linger = check_linger(linger);
355 	if (!linger)
356 		return -EPERM;
357 	expires = check_linger(expires);
358 	if (!expires)
359 		return -EPERM;
360 
361 	spin_lock_bh(&ip6_fl_lock);
362 	fl->lastuse = jiffies;
363 	if (time_before(fl->linger, linger))
364 		fl->linger = linger;
365 	if (time_before(expires, fl->linger))
366 		expires = fl->linger;
367 	if (time_before(fl->expires, fl->lastuse + expires))
368 		fl->expires = fl->lastuse + expires;
369 	spin_unlock_bh(&ip6_fl_lock);
370 
371 	return 0;
372 }
373 
374 static struct ip6_flowlabel *
375 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
376 	  sockptr_t optval, int optlen, int *err_p)
377 {
378 	struct ip6_flowlabel *fl = NULL;
379 	int olen;
380 	int addr_type;
381 	int err;
382 
383 	olen = optlen - CMSG_ALIGN(sizeof(*freq));
384 	err = -EINVAL;
385 	if (olen > 64 * 1024)
386 		goto done;
387 
388 	err = -ENOMEM;
389 	fl = kzalloc(sizeof(*fl), GFP_KERNEL);
390 	if (!fl)
391 		goto done;
392 
393 	if (olen > 0) {
394 		struct msghdr msg;
395 		struct flowi6 flowi6;
396 		struct ipcm6_cookie ipc6;
397 
398 		err = -ENOMEM;
399 		fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
400 		if (!fl->opt)
401 			goto done;
402 
403 		memset(fl->opt, 0, sizeof(*fl->opt));
404 		fl->opt->tot_len = sizeof(*fl->opt) + olen;
405 		err = -EFAULT;
406 		if (copy_from_sockptr_offset(fl->opt + 1, optval,
407 				CMSG_ALIGN(sizeof(*freq)), olen))
408 			goto done;
409 
410 		msg.msg_controllen = olen;
411 		msg.msg_control = (void *)(fl->opt+1);
412 		memset(&flowi6, 0, sizeof(flowi6));
413 
414 		ipc6.opt = fl->opt;
415 		err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, &ipc6);
416 		if (err)
417 			goto done;
418 		err = -EINVAL;
419 		if (fl->opt->opt_flen)
420 			goto done;
421 		if (fl->opt->opt_nflen == 0) {
422 			kfree(fl->opt);
423 			fl->opt = NULL;
424 		}
425 	}
426 
427 	fl->fl_net = net;
428 	fl->expires = jiffies;
429 	err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
430 	if (err)
431 		goto done;
432 	fl->share = freq->flr_share;
433 	addr_type = ipv6_addr_type(&freq->flr_dst);
434 	if ((addr_type & IPV6_ADDR_MAPPED) ||
435 	    addr_type == IPV6_ADDR_ANY) {
436 		err = -EINVAL;
437 		goto done;
438 	}
439 	fl->dst = freq->flr_dst;
440 	atomic_set(&fl->users, 1);
441 	switch (fl->share) {
442 	case IPV6_FL_S_EXCL:
443 	case IPV6_FL_S_ANY:
444 		break;
445 	case IPV6_FL_S_PROCESS:
446 		fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
447 		break;
448 	case IPV6_FL_S_USER:
449 		fl->owner.uid = current_euid();
450 		break;
451 	default:
452 		err = -EINVAL;
453 		goto done;
454 	}
455 	if (fl_shared_exclusive(fl) || fl->opt) {
456 		WRITE_ONCE(sock_net(sk)->ipv6.flowlabel_has_excl, 1);
457 		static_branch_deferred_inc(&ipv6_flowlabel_exclusive);
458 	}
459 	return fl;
460 
461 done:
462 	if (fl) {
463 		kfree(fl->opt);
464 		kfree(fl);
465 	}
466 	*err_p = err;
467 	return NULL;
468 }
469 
470 static int mem_check(struct sock *sk)
471 {
472 	int room = FL_MAX_SIZE - atomic_read(&fl_size);
473 	struct ipv6_fl_socklist *sfl;
474 	int count = 0;
475 
476 	if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
477 		return 0;
478 
479 	rcu_read_lock();
480 	for_each_sk_fl_rcu(sk, sfl)
481 		count++;
482 	rcu_read_unlock();
483 
484 	if (room <= 0 ||
485 	    ((count >= FL_MAX_PER_SOCK ||
486 	      (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
487 	     !capable(CAP_NET_ADMIN)))
488 		return -ENOBUFS;
489 
490 	return 0;
491 }
492 
493 static inline void fl_link(struct sock *sk, struct ipv6_fl_socklist *sfl,
494 			   struct ip6_flowlabel *fl)
495 {
496 	struct inet_sock *inet = inet_sk(sk);
497 
498 	spin_lock_bh(&ip6_sk_fl_lock);
499 	sfl->fl = fl;
500 	sfl->next = inet->ipv6_fl_list;
501 	rcu_assign_pointer(inet->ipv6_fl_list, sfl);
502 	spin_unlock_bh(&ip6_sk_fl_lock);
503 }
504 
505 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
506 			   int flags)
507 {
508 	struct ipv6_pinfo *np = inet6_sk(sk);
509 	struct ipv6_fl_socklist *sfl;
510 
511 	if (flags & IPV6_FL_F_REMOTE) {
512 		freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
513 		return 0;
514 	}
515 
516 	if (inet6_test_bit(REPFLOW, sk)) {
517 		freq->flr_label = np->flow_label;
518 		return 0;
519 	}
520 
521 	rcu_read_lock();
522 
523 	for_each_sk_fl_rcu(sk, sfl) {
524 		if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
525 			spin_lock_bh(&ip6_fl_lock);
526 			freq->flr_label = sfl->fl->label;
527 			freq->flr_dst = sfl->fl->dst;
528 			freq->flr_share = sfl->fl->share;
529 			freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
530 			freq->flr_linger = sfl->fl->linger / HZ;
531 
532 			spin_unlock_bh(&ip6_fl_lock);
533 			rcu_read_unlock();
534 			return 0;
535 		}
536 	}
537 	rcu_read_unlock();
538 
539 	return -ENOENT;
540 }
541 
542 #define socklist_dereference(__sflp) \
543 	rcu_dereference_protected(__sflp, lockdep_is_held(&ip6_sk_fl_lock))
544 
545 static int ipv6_flowlabel_put(struct sock *sk, struct in6_flowlabel_req *freq)
546 {
547 	struct ipv6_pinfo *np = inet6_sk(sk);
548 	struct ipv6_fl_socklist __rcu **sflp;
549 	struct ipv6_fl_socklist *sfl;
550 
551 	if (freq->flr_flags & IPV6_FL_F_REFLECT) {
552 		if (sk->sk_protocol != IPPROTO_TCP)
553 			return -ENOPROTOOPT;
554 		if (!inet6_test_bit(REPFLOW, sk))
555 			return -ESRCH;
556 		np->flow_label = 0;
557 		inet6_clear_bit(REPFLOW, sk);
558 		return 0;
559 	}
560 
561 	spin_lock_bh(&ip6_sk_fl_lock);
562 	for (sflp = &inet_sk(sk)->ipv6_fl_list;
563 	     (sfl = socklist_dereference(*sflp)) != NULL;
564 	     sflp = &sfl->next) {
565 		if (sfl->fl->label == freq->flr_label)
566 			goto found;
567 	}
568 	spin_unlock_bh(&ip6_sk_fl_lock);
569 	return -ESRCH;
570 found:
571 	if (freq->flr_label == (np->flow_label & IPV6_FLOWLABEL_MASK))
572 		np->flow_label &= ~IPV6_FLOWLABEL_MASK;
573 	*sflp = sfl->next;
574 	spin_unlock_bh(&ip6_sk_fl_lock);
575 	fl_release(sfl->fl);
576 	kfree_rcu(sfl, rcu);
577 	return 0;
578 }
579 
580 static int ipv6_flowlabel_renew(struct sock *sk, struct in6_flowlabel_req *freq)
581 {
582 	struct net *net = sock_net(sk);
583 	struct ipv6_fl_socklist *sfl;
584 	int err;
585 
586 	rcu_read_lock();
587 	for_each_sk_fl_rcu(sk, sfl) {
588 		if (sfl->fl->label == freq->flr_label) {
589 			err = fl6_renew(sfl->fl, freq->flr_linger,
590 					freq->flr_expires);
591 			rcu_read_unlock();
592 			return err;
593 		}
594 	}
595 	rcu_read_unlock();
596 
597 	if (freq->flr_share == IPV6_FL_S_NONE &&
598 	    ns_capable(net->user_ns, CAP_NET_ADMIN)) {
599 		struct ip6_flowlabel *fl = fl_lookup(net, freq->flr_label);
600 
601 		if (fl) {
602 			err = fl6_renew(fl, freq->flr_linger,
603 					freq->flr_expires);
604 			fl_release(fl);
605 			return err;
606 		}
607 	}
608 	return -ESRCH;
609 }
610 
611 static int ipv6_flowlabel_get(struct sock *sk, struct in6_flowlabel_req *freq,
612 		sockptr_t optval, int optlen)
613 {
614 	struct ipv6_fl_socklist *sfl, *sfl1 = NULL;
615 	struct ip6_flowlabel *fl, *fl1 = NULL;
616 	struct net *net = sock_net(sk);
617 	int err;
618 
619 	if (freq->flr_flags & IPV6_FL_F_REFLECT) {
620 		if (net->ipv6.sysctl.flowlabel_consistency) {
621 			net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
622 			return -EPERM;
623 		}
624 
625 		if (sk->sk_protocol != IPPROTO_TCP)
626 			return -ENOPROTOOPT;
627 		inet6_set_bit(REPFLOW, sk);
628 		return 0;
629 	}
630 
631 	if (freq->flr_label & ~IPV6_FLOWLABEL_MASK)
632 		return -EINVAL;
633 	if (net->ipv6.sysctl.flowlabel_state_ranges &&
634 	    (freq->flr_label & IPV6_FLOWLABEL_STATELESS_FLAG))
635 		return -ERANGE;
636 
637 	fl = fl_create(net, sk, freq, optval, optlen, &err);
638 	if (!fl)
639 		return err;
640 
641 	sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
642 
643 	if (freq->flr_label) {
644 		err = -EEXIST;
645 		rcu_read_lock();
646 		for_each_sk_fl_rcu(sk, sfl) {
647 			if (sfl->fl->label == freq->flr_label) {
648 				if (freq->flr_flags & IPV6_FL_F_EXCL) {
649 					rcu_read_unlock();
650 					goto done;
651 				}
652 				fl1 = sfl->fl;
653 				if (!atomic_inc_not_zero(&fl1->users))
654 					fl1 = NULL;
655 				break;
656 			}
657 		}
658 		rcu_read_unlock();
659 
660 		if (!fl1)
661 			fl1 = fl_lookup(net, freq->flr_label);
662 		if (fl1) {
663 recheck:
664 			err = -EEXIST;
665 			if (freq->flr_flags&IPV6_FL_F_EXCL)
666 				goto release;
667 			err = -EPERM;
668 			if (fl1->share == IPV6_FL_S_EXCL ||
669 			    fl1->share != fl->share ||
670 			    ((fl1->share == IPV6_FL_S_PROCESS) &&
671 			     (fl1->owner.pid != fl->owner.pid)) ||
672 			    ((fl1->share == IPV6_FL_S_USER) &&
673 			     !uid_eq(fl1->owner.uid, fl->owner.uid)))
674 				goto release;
675 
676 			err = -ENOMEM;
677 			if (!sfl1)
678 				goto release;
679 			if (fl->linger > fl1->linger)
680 				fl1->linger = fl->linger;
681 			if ((long)(fl->expires - fl1->expires) > 0)
682 				fl1->expires = fl->expires;
683 			fl_link(sk, sfl1, fl1);
684 			fl_free(fl);
685 			return 0;
686 
687 release:
688 			fl_release(fl1);
689 			goto done;
690 		}
691 	}
692 	err = -ENOENT;
693 	if (!(freq->flr_flags & IPV6_FL_F_CREATE))
694 		goto done;
695 
696 	err = -ENOMEM;
697 	if (!sfl1)
698 		goto done;
699 
700 	err = mem_check(sk);
701 	if (err != 0)
702 		goto done;
703 
704 	fl1 = fl_intern(net, fl, freq->flr_label);
705 	if (fl1)
706 		goto recheck;
707 
708 	if (!freq->flr_label) {
709 		size_t offset = offsetof(struct in6_flowlabel_req, flr_label);
710 
711 		if (copy_to_sockptr_offset(optval, offset, &fl->label,
712 				sizeof(fl->label))) {
713 			/* Intentionally ignore fault. */
714 		}
715 	}
716 
717 	fl_link(sk, sfl1, fl);
718 	return 0;
719 done:
720 	fl_free(fl);
721 	kfree(sfl1);
722 	return err;
723 }
724 
725 int ipv6_flowlabel_opt(struct sock *sk, sockptr_t optval, int optlen)
726 {
727 	struct in6_flowlabel_req freq;
728 
729 	if (optlen < sizeof(freq))
730 		return -EINVAL;
731 	if (copy_from_sockptr(&freq, optval, sizeof(freq)))
732 		return -EFAULT;
733 
734 	switch (freq.flr_action) {
735 	case IPV6_FL_A_PUT:
736 		return ipv6_flowlabel_put(sk, &freq);
737 	case IPV6_FL_A_RENEW:
738 		return ipv6_flowlabel_renew(sk, &freq);
739 	case IPV6_FL_A_GET:
740 		return ipv6_flowlabel_get(sk, &freq, optval, optlen);
741 	default:
742 		return -EINVAL;
743 	}
744 }
745 
746 #ifdef CONFIG_PROC_FS
747 
748 struct ip6fl_iter_state {
749 	struct seq_net_private p;
750 	struct pid_namespace *pid_ns;
751 	int bucket;
752 };
753 
754 #define ip6fl_seq_private(seq)	((struct ip6fl_iter_state *)(seq)->private)
755 
756 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
757 {
758 	struct ip6_flowlabel *fl = NULL;
759 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
760 	struct net *net = seq_file_net(seq);
761 
762 	for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
763 		for_each_fl_rcu(state->bucket, fl) {
764 			if (net_eq(fl->fl_net, net))
765 				goto out;
766 		}
767 	}
768 	fl = NULL;
769 out:
770 	return fl;
771 }
772 
773 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
774 {
775 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
776 	struct net *net = seq_file_net(seq);
777 
778 	for_each_fl_continue_rcu(fl) {
779 		if (net_eq(fl->fl_net, net))
780 			goto out;
781 	}
782 
783 try_again:
784 	if (++state->bucket <= FL_HASH_MASK) {
785 		for_each_fl_rcu(state->bucket, fl) {
786 			if (net_eq(fl->fl_net, net))
787 				goto out;
788 		}
789 		goto try_again;
790 	}
791 	fl = NULL;
792 
793 out:
794 	return fl;
795 }
796 
797 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
798 {
799 	struct ip6_flowlabel *fl = ip6fl_get_first(seq);
800 	if (fl)
801 		while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
802 			--pos;
803 	return pos ? NULL : fl;
804 }
805 
806 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
807 	__acquires(RCU)
808 {
809 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
810 
811 	state->pid_ns = proc_pid_ns(file_inode(seq->file)->i_sb);
812 
813 	rcu_read_lock();
814 	return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
815 }
816 
817 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
818 {
819 	struct ip6_flowlabel *fl;
820 
821 	if (v == SEQ_START_TOKEN)
822 		fl = ip6fl_get_first(seq);
823 	else
824 		fl = ip6fl_get_next(seq, v);
825 	++*pos;
826 	return fl;
827 }
828 
829 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
830 	__releases(RCU)
831 {
832 	rcu_read_unlock();
833 }
834 
835 static int ip6fl_seq_show(struct seq_file *seq, void *v)
836 {
837 	struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
838 	if (v == SEQ_START_TOKEN) {
839 		seq_puts(seq, "Label S Owner  Users  Linger Expires  Dst                              Opt\n");
840 	} else {
841 		struct ip6_flowlabel *fl = v;
842 		seq_printf(seq,
843 			   "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
844 			   (unsigned int)ntohl(fl->label),
845 			   fl->share,
846 			   ((fl->share == IPV6_FL_S_PROCESS) ?
847 			    pid_nr_ns(fl->owner.pid, state->pid_ns) :
848 			    ((fl->share == IPV6_FL_S_USER) ?
849 			     from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
850 			     0)),
851 			   atomic_read(&fl->users),
852 			   fl->linger/HZ,
853 			   (long)(fl->expires - jiffies)/HZ,
854 			   &fl->dst,
855 			   fl->opt ? fl->opt->opt_nflen : 0);
856 	}
857 	return 0;
858 }
859 
860 static const struct seq_operations ip6fl_seq_ops = {
861 	.start	=	ip6fl_seq_start,
862 	.next	=	ip6fl_seq_next,
863 	.stop	=	ip6fl_seq_stop,
864 	.show	=	ip6fl_seq_show,
865 };
866 
867 static int __net_init ip6_flowlabel_proc_init(struct net *net)
868 {
869 	if (!proc_create_net("ip6_flowlabel", 0444, net->proc_net,
870 			&ip6fl_seq_ops, sizeof(struct ip6fl_iter_state)))
871 		return -ENOMEM;
872 	return 0;
873 }
874 
875 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
876 {
877 	remove_proc_entry("ip6_flowlabel", net->proc_net);
878 }
879 #else
880 static inline int ip6_flowlabel_proc_init(struct net *net)
881 {
882 	return 0;
883 }
884 static inline void ip6_flowlabel_proc_fini(struct net *net)
885 {
886 }
887 #endif
888 
889 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
890 {
891 	ip6_fl_purge(net);
892 	ip6_flowlabel_proc_fini(net);
893 }
894 
895 static struct pernet_operations ip6_flowlabel_net_ops = {
896 	.init = ip6_flowlabel_proc_init,
897 	.exit = ip6_flowlabel_net_exit,
898 };
899 
900 int ip6_flowlabel_init(void)
901 {
902 	return register_pernet_subsys(&ip6_flowlabel_net_ops);
903 }
904 
905 void ip6_flowlabel_cleanup(void)
906 {
907 	static_key_deferred_flush(&ipv6_flowlabel_exclusive);
908 	timer_delete(&ip6_fl_gc_timer);
909 	unregister_pernet_subsys(&ip6_flowlabel_net_ops);
910 }
911