1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * ip6_flowlabel.c IPv6 flowlabel manager.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
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>
22 #include <net/net_namespace.h>
26 #include <net/rawv6.h>
27 #include <net/transp_v6.h>
29 #include <linux/uaccess.h>
31 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
32 in old IPv6 RFC. Well, it was reasonable value.
34 #define FL_MAX_LINGER 150 /* Maximal linger timeout */
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)
43 static atomic_t fl_size = ATOMIC_INIT(0);
44 static struct ip6_flowlabel __rcu *fl_ht[FL_HASH_MASK+1];
46 static void ip6_fl_gc(struct timer_list *unused);
47 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc);
49 /* FL hash table lock: it protects only of GC */
51 static DEFINE_SPINLOCK(ip6_fl_lock);
55 static DEFINE_SPINLOCK(ip6_sk_fl_lock);
57 DEFINE_STATIC_KEY_DEFERRED_FALSE(ipv6_flowlabel_exclusive, HZ);
58 EXPORT_SYMBOL(ipv6_flowlabel_exclusive);
60 #define for_each_fl_rcu(hash, fl) \
61 for (fl = rcu_dereference_bh(fl_ht[(hash)]); \
63 fl = rcu_dereference_bh(fl->next))
64 #define for_each_fl_continue_rcu(fl) \
65 for (fl = rcu_dereference_bh(fl->next); \
67 fl = rcu_dereference_bh(fl->next))
69 #define for_each_sk_fl_rcu(np, sfl) \
70 for (sfl = rcu_dereference_bh(np->ipv6_fl_list); \
72 sfl = rcu_dereference_bh(sfl->next))
74 static inline struct ip6_flowlabel *__fl_lookup(struct net *net, __be32 label)
76 struct ip6_flowlabel *fl;
78 for_each_fl_rcu(FL_HASH(label), fl) {
79 if (fl->label == label && net_eq(fl->fl_net, net))
85 static struct ip6_flowlabel *fl_lookup(struct net *net, __be32 label)
87 struct ip6_flowlabel *fl;
90 fl = __fl_lookup(net, label);
91 if (fl && !atomic_inc_not_zero(&fl->users))
97 static bool fl_shared_exclusive(struct ip6_flowlabel *fl)
99 return fl->share == IPV6_FL_S_EXCL ||
100 fl->share == IPV6_FL_S_PROCESS ||
101 fl->share == IPV6_FL_S_USER;
104 static void fl_free_rcu(struct rcu_head *head)
106 struct ip6_flowlabel *fl = container_of(head, struct ip6_flowlabel, rcu);
108 if (fl->share == IPV6_FL_S_PROCESS)
109 put_pid(fl->owner.pid);
115 static void fl_free(struct ip6_flowlabel *fl)
120 if (fl_shared_exclusive(fl) || fl->opt)
121 static_branch_slow_dec_deferred(&ipv6_flowlabel_exclusive);
123 call_rcu(&fl->rcu, fl_free_rcu);
126 static void fl_release(struct ip6_flowlabel *fl)
128 spin_lock_bh(&ip6_fl_lock);
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))
136 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
137 struct ipv6_txoptions *opt = fl->opt;
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);
145 spin_unlock_bh(&ip6_fl_lock);
148 static void ip6_fl_gc(struct timer_list *unused)
151 unsigned long now = jiffies;
152 unsigned long sched = 0;
154 spin_lock(&ip6_fl_lock);
156 for (i = 0; i <= FL_HASH_MASK; i++) {
157 struct ip6_flowlabel *fl;
158 struct ip6_flowlabel __rcu **flp;
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))
168 if (time_after_eq(now, ttd)) {
171 atomic_dec(&fl_size);
174 if (!sched || time_before(ttd, sched))
180 if (!sched && atomic_read(&fl_size))
181 sched = now + FL_MAX_LINGER;
183 mod_timer(&ip6_fl_gc_timer, sched);
185 spin_unlock(&ip6_fl_lock);
188 static void __net_exit ip6_fl_purge(struct net *net)
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;
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) {
204 atomic_dec(&fl_size);
210 spin_unlock_bh(&ip6_fl_lock);
213 static struct ip6_flowlabel *fl_intern(struct net *net,
214 struct ip6_flowlabel *fl, __be32 label)
216 struct ip6_flowlabel *lfl;
218 fl->label = label & IPV6_FLOWLABEL_MASK;
220 spin_lock_bh(&ip6_fl_lock);
223 fl->label = htonl(prandom_u32())&IPV6_FLOWLABEL_MASK;
225 lfl = __fl_lookup(net, fl->label);
232 * we dropper the ip6_fl_lock, so this entry could reappear
233 * and we need to recheck with it.
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
239 lfl = __fl_lookup(net, fl->label);
241 atomic_inc(&lfl->users);
242 spin_unlock_bh(&ip6_fl_lock);
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);
257 /* Socket flowlabel lists */
259 struct ip6_flowlabel *__fl6_sock_lookup(struct sock *sk, __be32 label)
261 struct ipv6_fl_socklist *sfl;
262 struct ipv6_pinfo *np = inet6_sk(sk);
264 label &= IPV6_FLOWLABEL_MASK;
267 for_each_sk_fl_rcu(np, sfl) {
268 struct ip6_flowlabel *fl = sfl->fl;
270 if (fl->label == label && atomic_inc_not_zero(&fl->users)) {
271 fl->lastuse = jiffies;
272 rcu_read_unlock_bh();
276 rcu_read_unlock_bh();
279 EXPORT_SYMBOL_GPL(__fl6_sock_lookup);
281 void fl6_free_socklist(struct sock *sk)
283 struct ipv6_pinfo *np = inet6_sk(sk);
284 struct ipv6_fl_socklist *sfl;
286 if (!rcu_access_pointer(np->ipv6_fl_list))
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);
298 spin_lock_bh(&ip6_sk_fl_lock);
300 spin_unlock_bh(&ip6_sk_fl_lock);
303 /* Service routines */
307 It is the only difficult place. flowlabel enforces equal headers
308 before and including routing header, however user may supply options
312 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions *opt_space,
313 struct ip6_flowlabel *fl,
314 struct ipv6_txoptions *fopt)
316 struct ipv6_txoptions *fl_opt = fl->opt;
318 if (!fopt || fopt->opt_flen == 0)
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;
327 if (fopt->opt_nflen == 0)
329 opt_space->hopopt = NULL;
330 opt_space->dst0opt = NULL;
331 opt_space->srcrt = NULL;
332 opt_space->opt_nflen = 0;
334 opt_space->dst1opt = fopt->dst1opt;
335 opt_space->opt_flen = fopt->opt_flen;
336 opt_space->tot_len = fopt->tot_len;
339 EXPORT_SYMBOL_GPL(fl6_merge_options);
341 static unsigned long check_linger(unsigned long ttl)
343 if (ttl < FL_MIN_LINGER)
344 return FL_MIN_LINGER*HZ;
345 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
350 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
352 linger = check_linger(linger);
355 expires = check_linger(expires);
359 spin_lock_bh(&ip6_fl_lock);
360 fl->lastuse = jiffies;
361 if (time_before(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);
372 static struct ip6_flowlabel *
373 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
374 char __user *optval, int optlen, int *err_p)
376 struct ip6_flowlabel *fl = NULL;
381 olen = optlen - CMSG_ALIGN(sizeof(*freq));
383 if (olen > 64 * 1024)
387 fl = kzalloc(sizeof(*fl), GFP_KERNEL);
393 struct flowi6 flowi6;
394 struct ipcm6_cookie ipc6;
397 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
401 memset(fl->opt, 0, sizeof(*fl->opt));
402 fl->opt->tot_len = sizeof(*fl->opt) + olen;
404 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
407 msg.msg_controllen = olen;
408 msg.msg_control = (void *)(fl->opt+1);
409 memset(&flowi6, 0, sizeof(flowi6));
412 err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, &ipc6);
416 if (fl->opt->opt_flen)
418 if (fl->opt->opt_nflen == 0) {
425 fl->expires = jiffies;
426 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
429 fl->share = freq->flr_share;
430 addr_type = ipv6_addr_type(&freq->flr_dst);
431 if ((addr_type & IPV6_ADDR_MAPPED) ||
432 addr_type == IPV6_ADDR_ANY) {
436 fl->dst = freq->flr_dst;
437 atomic_set(&fl->users, 1);
438 if (fl_shared_exclusive(fl) || fl->opt)
439 static_branch_deferred_inc(&ipv6_flowlabel_exclusive);
444 case IPV6_FL_S_PROCESS:
445 fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
448 fl->owner.uid = current_euid();
462 static int mem_check(struct sock *sk)
464 struct ipv6_pinfo *np = inet6_sk(sk);
465 struct ipv6_fl_socklist *sfl;
466 int room = FL_MAX_SIZE - atomic_read(&fl_size);
469 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
473 for_each_sk_fl_rcu(np, sfl)
475 rcu_read_unlock_bh();
478 ((count >= FL_MAX_PER_SOCK ||
479 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
480 !capable(CAP_NET_ADMIN)))
486 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
487 struct ip6_flowlabel *fl)
489 spin_lock_bh(&ip6_sk_fl_lock);
491 sfl->next = np->ipv6_fl_list;
492 rcu_assign_pointer(np->ipv6_fl_list, sfl);
493 spin_unlock_bh(&ip6_sk_fl_lock);
496 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
499 struct ipv6_pinfo *np = inet6_sk(sk);
500 struct ipv6_fl_socklist *sfl;
502 if (flags & IPV6_FL_F_REMOTE) {
503 freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
508 freq->flr_label = np->flow_label;
514 for_each_sk_fl_rcu(np, sfl) {
515 if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
516 spin_lock_bh(&ip6_fl_lock);
517 freq->flr_label = sfl->fl->label;
518 freq->flr_dst = sfl->fl->dst;
519 freq->flr_share = sfl->fl->share;
520 freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
521 freq->flr_linger = sfl->fl->linger / HZ;
523 spin_unlock_bh(&ip6_fl_lock);
524 rcu_read_unlock_bh();
528 rcu_read_unlock_bh();
533 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
535 int uninitialized_var(err);
536 struct net *net = sock_net(sk);
537 struct ipv6_pinfo *np = inet6_sk(sk);
538 struct in6_flowlabel_req freq;
539 struct ipv6_fl_socklist *sfl1 = NULL;
540 struct ipv6_fl_socklist *sfl;
541 struct ipv6_fl_socklist __rcu **sflp;
542 struct ip6_flowlabel *fl, *fl1 = NULL;
545 if (optlen < sizeof(freq))
548 if (copy_from_user(&freq, optval, sizeof(freq)))
551 switch (freq.flr_action) {
553 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
554 if (sk->sk_protocol != IPPROTO_TCP)
562 spin_lock_bh(&ip6_sk_fl_lock);
563 for (sflp = &np->ipv6_fl_list;
564 (sfl = rcu_dereference_protected(*sflp,
565 lockdep_is_held(&ip6_sk_fl_lock))) != NULL;
567 if (sfl->fl->label == freq.flr_label) {
568 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
569 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
571 spin_unlock_bh(&ip6_sk_fl_lock);
577 spin_unlock_bh(&ip6_sk_fl_lock);
580 case IPV6_FL_A_RENEW:
582 for_each_sk_fl_rcu(np, sfl) {
583 if (sfl->fl->label == freq.flr_label) {
584 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
585 rcu_read_unlock_bh();
589 rcu_read_unlock_bh();
591 if (freq.flr_share == IPV6_FL_S_NONE &&
592 ns_capable(net->user_ns, CAP_NET_ADMIN)) {
593 fl = fl_lookup(net, freq.flr_label);
595 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
603 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
604 struct net *net = sock_net(sk);
605 if (net->ipv6.sysctl.flowlabel_consistency) {
606 net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
610 if (sk->sk_protocol != IPPROTO_TCP)
617 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
620 if (net->ipv6.sysctl.flowlabel_state_ranges &&
621 (freq.flr_label & IPV6_FLOWLABEL_STATELESS_FLAG))
624 fl = fl_create(net, sk, &freq, optval, optlen, &err);
627 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
629 if (freq.flr_label) {
632 for_each_sk_fl_rcu(np, sfl) {
633 if (sfl->fl->label == freq.flr_label) {
634 if (freq.flr_flags&IPV6_FL_F_EXCL) {
635 rcu_read_unlock_bh();
639 if (!atomic_inc_not_zero(&fl1->users))
644 rcu_read_unlock_bh();
647 fl1 = fl_lookup(net, freq.flr_label);
651 if (freq.flr_flags&IPV6_FL_F_EXCL)
654 if (fl1->share == IPV6_FL_S_EXCL ||
655 fl1->share != fl->share ||
656 ((fl1->share == IPV6_FL_S_PROCESS) &&
657 (fl1->owner.pid != fl->owner.pid)) ||
658 ((fl1->share == IPV6_FL_S_USER) &&
659 !uid_eq(fl1->owner.uid, fl->owner.uid)))
665 if (fl->linger > fl1->linger)
666 fl1->linger = fl->linger;
667 if ((long)(fl->expires - fl1->expires) > 0)
668 fl1->expires = fl->expires;
669 fl_link(np, sfl1, fl1);
679 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
690 fl1 = fl_intern(net, fl, freq.flr_label);
694 if (!freq.flr_label) {
695 if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
696 &fl->label, sizeof(fl->label))) {
697 /* Intentionally ignore fault. */
701 fl_link(np, sfl1, fl);
714 #ifdef CONFIG_PROC_FS
716 struct ip6fl_iter_state {
717 struct seq_net_private p;
718 struct pid_namespace *pid_ns;
722 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
724 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
726 struct ip6_flowlabel *fl = NULL;
727 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
728 struct net *net = seq_file_net(seq);
730 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
731 for_each_fl_rcu(state->bucket, fl) {
732 if (net_eq(fl->fl_net, net))
741 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
743 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
744 struct net *net = seq_file_net(seq);
746 for_each_fl_continue_rcu(fl) {
747 if (net_eq(fl->fl_net, net))
752 if (++state->bucket <= FL_HASH_MASK) {
753 for_each_fl_rcu(state->bucket, fl) {
754 if (net_eq(fl->fl_net, net))
765 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
767 struct ip6_flowlabel *fl = ip6fl_get_first(seq);
769 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
771 return pos ? NULL : fl;
774 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
777 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
779 state->pid_ns = proc_pid_ns(file_inode(seq->file));
782 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
785 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
787 struct ip6_flowlabel *fl;
789 if (v == SEQ_START_TOKEN)
790 fl = ip6fl_get_first(seq);
792 fl = ip6fl_get_next(seq, v);
797 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
800 rcu_read_unlock_bh();
803 static int ip6fl_seq_show(struct seq_file *seq, void *v)
805 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
806 if (v == SEQ_START_TOKEN) {
807 seq_puts(seq, "Label S Owner Users Linger Expires Dst Opt\n");
809 struct ip6_flowlabel *fl = v;
811 "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
812 (unsigned int)ntohl(fl->label),
814 ((fl->share == IPV6_FL_S_PROCESS) ?
815 pid_nr_ns(fl->owner.pid, state->pid_ns) :
816 ((fl->share == IPV6_FL_S_USER) ?
817 from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
819 atomic_read(&fl->users),
821 (long)(fl->expires - jiffies)/HZ,
823 fl->opt ? fl->opt->opt_nflen : 0);
828 static const struct seq_operations ip6fl_seq_ops = {
829 .start = ip6fl_seq_start,
830 .next = ip6fl_seq_next,
831 .stop = ip6fl_seq_stop,
832 .show = ip6fl_seq_show,
835 static int __net_init ip6_flowlabel_proc_init(struct net *net)
837 if (!proc_create_net("ip6_flowlabel", 0444, net->proc_net,
838 &ip6fl_seq_ops, sizeof(struct ip6fl_iter_state)))
843 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
845 remove_proc_entry("ip6_flowlabel", net->proc_net);
848 static inline int ip6_flowlabel_proc_init(struct net *net)
852 static inline void ip6_flowlabel_proc_fini(struct net *net)
857 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
860 ip6_flowlabel_proc_fini(net);
863 static struct pernet_operations ip6_flowlabel_net_ops = {
864 .init = ip6_flowlabel_proc_init,
865 .exit = ip6_flowlabel_net_exit,
868 int ip6_flowlabel_init(void)
870 return register_pernet_subsys(&ip6_flowlabel_net_ops);
873 void ip6_flowlabel_cleanup(void)
875 static_key_deferred_flush(&ipv6_flowlabel_exclusive);
876 del_timer(&ip6_fl_gc_timer);
877 unregister_pernet_subsys(&ip6_flowlabel_net_ops);