2 * ip6_flowlabel.c IPv6 flowlabel manager.
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.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
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>
25 #include <net/net_namespace.h>
29 #include <net/addrconf.h>
30 #include <net/rawv6.h>
31 #include <net/transp_v6.h>
33 #include <asm/uaccess.h>
35 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
36 in old IPv6 RFC. Well, it was reasonable value.
38 #define FL_MAX_LINGER 150 /* Maximal linger timeout */
42 #define FL_MAX_PER_SOCK 32
43 #define FL_MAX_SIZE 4096
44 #define FL_HASH_MASK 255
45 #define FL_HASH(l) (ntohl(l)&FL_HASH_MASK)
47 static atomic_t fl_size = ATOMIC_INIT(0);
48 static struct ip6_flowlabel __rcu *fl_ht[FL_HASH_MASK+1];
50 static void ip6_fl_gc(unsigned long dummy);
51 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc, 0, 0);
53 /* FL hash table lock: it protects only of GC */
55 static DEFINE_SPINLOCK(ip6_fl_lock);
59 static DEFINE_SPINLOCK(ip6_sk_fl_lock);
61 #define for_each_fl_rcu(hash, fl) \
62 for (fl = rcu_dereference_bh(fl_ht[(hash)]); \
64 fl = rcu_dereference_bh(fl->next))
65 #define for_each_fl_continue_rcu(fl) \
66 for (fl = rcu_dereference_bh(fl->next); \
68 fl = rcu_dereference_bh(fl->next))
70 #define for_each_sk_fl_rcu(np, sfl) \
71 for (sfl = rcu_dereference_bh(np->ipv6_fl_list); \
73 sfl = rcu_dereference_bh(sfl->next))
75 static inline struct ip6_flowlabel *__fl_lookup(struct net *net, __be32 label)
77 struct ip6_flowlabel *fl;
79 for_each_fl_rcu(FL_HASH(label), fl) {
80 if (fl->label == label && net_eq(fl->fl_net, net))
86 static struct ip6_flowlabel *fl_lookup(struct net *net, __be32 label)
88 struct ip6_flowlabel *fl;
91 fl = __fl_lookup(net, label);
92 if (fl && !atomic_inc_not_zero(&fl->users))
99 static void fl_free(struct ip6_flowlabel *fl)
102 if (fl->share == IPV6_FL_S_PROCESS)
103 put_pid(fl->owner.pid);
104 release_net(fl->fl_net);
110 static void fl_release(struct ip6_flowlabel *fl)
112 spin_lock_bh(&ip6_fl_lock);
114 fl->lastuse = jiffies;
115 if (atomic_dec_and_test(&fl->users)) {
116 unsigned long ttd = fl->lastuse + fl->linger;
117 if (time_after(ttd, fl->expires))
120 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
121 struct ipv6_txoptions *opt = fl->opt;
125 if (!timer_pending(&ip6_fl_gc_timer) ||
126 time_after(ip6_fl_gc_timer.expires, ttd))
127 mod_timer(&ip6_fl_gc_timer, ttd);
129 spin_unlock_bh(&ip6_fl_lock);
132 static void ip6_fl_gc(unsigned long dummy)
135 unsigned long now = jiffies;
136 unsigned long sched = 0;
138 spin_lock(&ip6_fl_lock);
140 for (i=0; i<=FL_HASH_MASK; i++) {
141 struct ip6_flowlabel *fl;
142 struct ip6_flowlabel __rcu **flp;
145 while ((fl = rcu_dereference_protected(*flp,
146 lockdep_is_held(&ip6_fl_lock))) != NULL) {
147 if (atomic_read(&fl->users) == 0) {
148 unsigned long ttd = fl->lastuse + fl->linger;
149 if (time_after(ttd, fl->expires))
152 if (time_after_eq(now, ttd)) {
155 atomic_dec(&fl_size);
158 if (!sched || time_before(ttd, sched))
164 if (!sched && atomic_read(&fl_size))
165 sched = now + FL_MAX_LINGER;
167 mod_timer(&ip6_fl_gc_timer, sched);
169 spin_unlock(&ip6_fl_lock);
172 static void __net_exit ip6_fl_purge(struct net *net)
176 spin_lock(&ip6_fl_lock);
177 for (i = 0; i <= FL_HASH_MASK; i++) {
178 struct ip6_flowlabel *fl;
179 struct ip6_flowlabel __rcu **flp;
182 while ((fl = rcu_dereference_protected(*flp,
183 lockdep_is_held(&ip6_fl_lock))) != NULL) {
184 if (net_eq(fl->fl_net, net) &&
185 atomic_read(&fl->users) == 0) {
188 atomic_dec(&fl_size);
194 spin_unlock(&ip6_fl_lock);
197 static struct ip6_flowlabel *fl_intern(struct net *net,
198 struct ip6_flowlabel *fl, __be32 label)
200 struct ip6_flowlabel *lfl;
202 fl->label = label & IPV6_FLOWLABEL_MASK;
204 spin_lock_bh(&ip6_fl_lock);
207 fl->label = htonl(prandom_u32())&IPV6_FLOWLABEL_MASK;
209 lfl = __fl_lookup(net, fl->label);
216 * we dropper the ip6_fl_lock, so this entry could reappear
217 * and we need to recheck with it.
219 * OTOH no need to search the active socket first, like it is
220 * done in ipv6_flowlabel_opt - sock is locked, so new entry
221 * with the same label can only appear on another sock
223 lfl = __fl_lookup(net, fl->label);
225 atomic_inc(&lfl->users);
226 spin_unlock_bh(&ip6_fl_lock);
231 fl->lastuse = jiffies;
232 fl->next = fl_ht[FL_HASH(fl->label)];
233 rcu_assign_pointer(fl_ht[FL_HASH(fl->label)], fl);
234 atomic_inc(&fl_size);
235 spin_unlock_bh(&ip6_fl_lock);
241 /* Socket flowlabel lists */
243 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, __be32 label)
245 struct ipv6_fl_socklist *sfl;
246 struct ipv6_pinfo *np = inet6_sk(sk);
248 label &= IPV6_FLOWLABEL_MASK;
251 for_each_sk_fl_rcu(np, sfl) {
252 struct ip6_flowlabel *fl = sfl->fl;
253 if (fl->label == label) {
254 fl->lastuse = jiffies;
255 atomic_inc(&fl->users);
256 rcu_read_unlock_bh();
260 rcu_read_unlock_bh();
264 EXPORT_SYMBOL_GPL(fl6_sock_lookup);
266 void fl6_free_socklist(struct sock *sk)
268 struct ipv6_pinfo *np = inet6_sk(sk);
269 struct ipv6_fl_socklist *sfl;
271 if (!rcu_access_pointer(np->ipv6_fl_list))
274 spin_lock_bh(&ip6_sk_fl_lock);
275 while ((sfl = rcu_dereference_protected(np->ipv6_fl_list,
276 lockdep_is_held(&ip6_sk_fl_lock))) != NULL) {
277 np->ipv6_fl_list = sfl->next;
278 spin_unlock_bh(&ip6_sk_fl_lock);
283 spin_lock_bh(&ip6_sk_fl_lock);
285 spin_unlock_bh(&ip6_sk_fl_lock);
288 /* Service routines */
292 It is the only difficult place. flowlabel enforces equal headers
293 before and including routing header, however user may supply options
297 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
298 struct ip6_flowlabel * fl,
299 struct ipv6_txoptions * fopt)
301 struct ipv6_txoptions * fl_opt = fl->opt;
303 if (fopt == NULL || fopt->opt_flen == 0)
306 if (fl_opt != NULL) {
307 opt_space->hopopt = fl_opt->hopopt;
308 opt_space->dst0opt = fl_opt->dst0opt;
309 opt_space->srcrt = fl_opt->srcrt;
310 opt_space->opt_nflen = fl_opt->opt_nflen;
312 if (fopt->opt_nflen == 0)
314 opt_space->hopopt = NULL;
315 opt_space->dst0opt = NULL;
316 opt_space->srcrt = NULL;
317 opt_space->opt_nflen = 0;
319 opt_space->dst1opt = fopt->dst1opt;
320 opt_space->opt_flen = fopt->opt_flen;
323 EXPORT_SYMBOL_GPL(fl6_merge_options);
325 static unsigned long check_linger(unsigned long ttl)
327 if (ttl < FL_MIN_LINGER)
328 return FL_MIN_LINGER*HZ;
329 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
334 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
336 linger = check_linger(linger);
339 expires = check_linger(expires);
343 spin_lock_bh(&ip6_fl_lock);
344 fl->lastuse = jiffies;
345 if (time_before(fl->linger, linger))
347 if (time_before(expires, fl->linger))
348 expires = fl->linger;
349 if (time_before(fl->expires, fl->lastuse + expires))
350 fl->expires = fl->lastuse + expires;
351 spin_unlock_bh(&ip6_fl_lock);
356 static struct ip6_flowlabel *
357 fl_create(struct net *net, struct sock *sk, struct in6_flowlabel_req *freq,
358 char __user *optval, int optlen, int *err_p)
360 struct ip6_flowlabel *fl = NULL;
365 olen = optlen - CMSG_ALIGN(sizeof(*freq));
367 if (olen > 64 * 1024)
371 fl = kzalloc(sizeof(*fl), GFP_KERNEL);
377 struct flowi6 flowi6;
381 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
385 memset(fl->opt, 0, sizeof(*fl->opt));
386 fl->opt->tot_len = sizeof(*fl->opt) + olen;
388 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
391 msg.msg_controllen = olen;
392 msg.msg_control = (void*)(fl->opt+1);
393 memset(&flowi6, 0, sizeof(flowi6));
395 err = ip6_datagram_send_ctl(net, sk, &msg, &flowi6, fl->opt,
396 &junk, &junk, &junk);
400 if (fl->opt->opt_flen)
402 if (fl->opt->opt_nflen == 0) {
408 fl->fl_net = hold_net(net);
409 fl->expires = jiffies;
410 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
413 fl->share = freq->flr_share;
414 addr_type = ipv6_addr_type(&freq->flr_dst);
415 if ((addr_type & IPV6_ADDR_MAPPED) ||
416 addr_type == IPV6_ADDR_ANY) {
420 fl->dst = freq->flr_dst;
421 atomic_set(&fl->users, 1);
426 case IPV6_FL_S_PROCESS:
427 fl->owner.pid = get_task_pid(current, PIDTYPE_PID);
430 fl->owner.uid = current_euid();
444 static int mem_check(struct sock *sk)
446 struct ipv6_pinfo *np = inet6_sk(sk);
447 struct ipv6_fl_socklist *sfl;
448 int room = FL_MAX_SIZE - atomic_read(&fl_size);
451 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
455 for_each_sk_fl_rcu(np, sfl)
457 rcu_read_unlock_bh();
460 ((count >= FL_MAX_PER_SOCK ||
461 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
462 !capable(CAP_NET_ADMIN)))
468 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
469 struct ip6_flowlabel *fl)
471 spin_lock_bh(&ip6_sk_fl_lock);
473 sfl->next = np->ipv6_fl_list;
474 rcu_assign_pointer(np->ipv6_fl_list, sfl);
475 spin_unlock_bh(&ip6_sk_fl_lock);
478 int ipv6_flowlabel_opt_get(struct sock *sk, struct in6_flowlabel_req *freq,
481 struct ipv6_pinfo *np = inet6_sk(sk);
482 struct ipv6_fl_socklist *sfl;
484 if (flags & IPV6_FL_F_REMOTE) {
485 freq->flr_label = np->rcv_flowinfo & IPV6_FLOWLABEL_MASK;
490 freq->flr_label = np->flow_label;
496 for_each_sk_fl_rcu(np, sfl) {
497 if (sfl->fl->label == (np->flow_label & IPV6_FLOWLABEL_MASK)) {
498 spin_lock_bh(&ip6_fl_lock);
499 freq->flr_label = sfl->fl->label;
500 freq->flr_dst = sfl->fl->dst;
501 freq->flr_share = sfl->fl->share;
502 freq->flr_expires = (sfl->fl->expires - jiffies) / HZ;
503 freq->flr_linger = sfl->fl->linger / HZ;
505 spin_unlock_bh(&ip6_fl_lock);
506 rcu_read_unlock_bh();
510 rcu_read_unlock_bh();
515 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
517 int uninitialized_var(err);
518 struct net *net = sock_net(sk);
519 struct ipv6_pinfo *np = inet6_sk(sk);
520 struct in6_flowlabel_req freq;
521 struct ipv6_fl_socklist *sfl1=NULL;
522 struct ipv6_fl_socklist *sfl;
523 struct ipv6_fl_socklist __rcu **sflp;
524 struct ip6_flowlabel *fl, *fl1 = NULL;
527 if (optlen < sizeof(freq))
530 if (copy_from_user(&freq, optval, sizeof(freq)))
533 switch (freq.flr_action) {
535 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
536 if (sk->sk_protocol != IPPROTO_TCP)
544 spin_lock_bh(&ip6_sk_fl_lock);
545 for (sflp = &np->ipv6_fl_list;
546 (sfl = rcu_dereference(*sflp))!=NULL;
548 if (sfl->fl->label == freq.flr_label) {
549 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
550 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
551 *sflp = rcu_dereference(sfl->next);
552 spin_unlock_bh(&ip6_sk_fl_lock);
558 spin_unlock_bh(&ip6_sk_fl_lock);
561 case IPV6_FL_A_RENEW:
563 for_each_sk_fl_rcu(np, sfl) {
564 if (sfl->fl->label == freq.flr_label) {
565 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
566 rcu_read_unlock_bh();
570 rcu_read_unlock_bh();
572 if (freq.flr_share == IPV6_FL_S_NONE &&
573 ns_capable(net->user_ns, CAP_NET_ADMIN)) {
574 fl = fl_lookup(net, freq.flr_label);
576 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
584 if (freq.flr_flags & IPV6_FL_F_REFLECT) {
585 struct net *net = sock_net(sk);
586 if (net->ipv6.sysctl.flowlabel_consistency) {
587 net_info_ratelimited("Can not set IPV6_FL_F_REFLECT if flowlabel_consistency sysctl is enable\n");
591 if (sk->sk_protocol != IPPROTO_TCP)
598 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
601 fl = fl_create(net, sk, &freq, optval, optlen, &err);
604 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
606 if (freq.flr_label) {
609 for_each_sk_fl_rcu(np, sfl) {
610 if (sfl->fl->label == freq.flr_label) {
611 if (freq.flr_flags&IPV6_FL_F_EXCL) {
612 rcu_read_unlock_bh();
616 atomic_inc(&fl1->users);
620 rcu_read_unlock_bh();
623 fl1 = fl_lookup(net, freq.flr_label);
627 if (freq.flr_flags&IPV6_FL_F_EXCL)
630 if (fl1->share == IPV6_FL_S_EXCL ||
631 fl1->share != fl->share ||
632 ((fl1->share == IPV6_FL_S_PROCESS) &&
633 (fl1->owner.pid == fl->owner.pid)) ||
634 ((fl1->share == IPV6_FL_S_USER) &&
635 uid_eq(fl1->owner.uid, fl->owner.uid)))
641 if (fl->linger > fl1->linger)
642 fl1->linger = fl->linger;
643 if ((long)(fl->expires - fl1->expires) > 0)
644 fl1->expires = fl->expires;
645 fl_link(np, sfl1, fl1);
655 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
659 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
662 fl1 = fl_intern(net, fl, freq.flr_label);
666 if (!freq.flr_label) {
667 if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
668 &fl->label, sizeof(fl->label))) {
669 /* Intentionally ignore fault. */
673 fl_link(np, sfl1, fl);
686 #ifdef CONFIG_PROC_FS
688 struct ip6fl_iter_state {
689 struct seq_net_private p;
690 struct pid_namespace *pid_ns;
694 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
696 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
698 struct ip6_flowlabel *fl = NULL;
699 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
700 struct net *net = seq_file_net(seq);
702 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
703 for_each_fl_rcu(state->bucket, fl) {
704 if (net_eq(fl->fl_net, net))
713 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
715 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
716 struct net *net = seq_file_net(seq);
718 for_each_fl_continue_rcu(fl) {
719 if (net_eq(fl->fl_net, net))
724 if (++state->bucket <= FL_HASH_MASK) {
725 for_each_fl_rcu(state->bucket, fl) {
726 if (net_eq(fl->fl_net, net))
737 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
739 struct ip6_flowlabel *fl = ip6fl_get_first(seq);
741 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
743 return pos ? NULL : fl;
746 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
750 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
753 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
755 struct ip6_flowlabel *fl;
757 if (v == SEQ_START_TOKEN)
758 fl = ip6fl_get_first(seq);
760 fl = ip6fl_get_next(seq, v);
765 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
768 rcu_read_unlock_bh();
771 static int ip6fl_seq_show(struct seq_file *seq, void *v)
773 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
774 if (v == SEQ_START_TOKEN)
775 seq_printf(seq, "%-5s %-1s %-6s %-6s %-6s %-8s %-32s %s\n",
776 "Label", "S", "Owner", "Users", "Linger", "Expires", "Dst", "Opt");
778 struct ip6_flowlabel *fl = v;
780 "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
781 (unsigned int)ntohl(fl->label),
783 ((fl->share == IPV6_FL_S_PROCESS) ?
784 pid_nr_ns(fl->owner.pid, state->pid_ns) :
785 ((fl->share == IPV6_FL_S_USER) ?
786 from_kuid_munged(seq_user_ns(seq), fl->owner.uid) :
788 atomic_read(&fl->users),
790 (long)(fl->expires - jiffies)/HZ,
792 fl->opt ? fl->opt->opt_nflen : 0);
797 static const struct seq_operations ip6fl_seq_ops = {
798 .start = ip6fl_seq_start,
799 .next = ip6fl_seq_next,
800 .stop = ip6fl_seq_stop,
801 .show = ip6fl_seq_show,
804 static int ip6fl_seq_open(struct inode *inode, struct file *file)
806 struct seq_file *seq;
807 struct ip6fl_iter_state *state;
810 err = seq_open_net(inode, file, &ip6fl_seq_ops,
811 sizeof(struct ip6fl_iter_state));
814 seq = file->private_data;
815 state = ip6fl_seq_private(seq);
817 state->pid_ns = get_pid_ns(task_active_pid_ns(current));
823 static int ip6fl_seq_release(struct inode *inode, struct file *file)
825 struct seq_file *seq = file->private_data;
826 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
827 put_pid_ns(state->pid_ns);
828 return seq_release_net(inode, file);
831 static const struct file_operations ip6fl_seq_fops = {
832 .owner = THIS_MODULE,
833 .open = ip6fl_seq_open,
836 .release = ip6fl_seq_release,
839 static int __net_init ip6_flowlabel_proc_init(struct net *net)
841 if (!proc_create("ip6_flowlabel", S_IRUGO, net->proc_net,
847 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
849 remove_proc_entry("ip6_flowlabel", net->proc_net);
852 static inline int ip6_flowlabel_proc_init(struct net *net)
856 static inline void ip6_flowlabel_proc_fini(struct net *net)
861 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
864 ip6_flowlabel_proc_fini(net);
867 static struct pernet_operations ip6_flowlabel_net_ops = {
868 .init = ip6_flowlabel_proc_init,
869 .exit = ip6_flowlabel_net_exit,
872 int ip6_flowlabel_init(void)
874 return register_pernet_subsys(&ip6_flowlabel_net_ops);
877 void ip6_flowlabel_cleanup(void)
879 del_timer(&ip6_fl_gc_timer);
880 unregister_pernet_subsys(&ip6_flowlabel_net_ops);