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/if_arp.h>
19 #include <linux/in6.h>
20 #include <linux/route.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/slab.h>
25 #include <net/net_namespace.h>
29 #include <net/ndisc.h>
30 #include <net/protocol.h>
31 #include <net/ip6_route.h>
32 #include <net/addrconf.h>
33 #include <net/rawv6.h>
35 #include <net/transp_v6.h>
37 #include <asm/uaccess.h>
39 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
40 in old IPv6 RFC. Well, it was reasonable value.
42 #define FL_MAX_LINGER 60 /* Maximal linger timeout */
46 #define FL_MAX_PER_SOCK 32
47 #define FL_MAX_SIZE 4096
48 #define FL_HASH_MASK 255
49 #define FL_HASH(l) (ntohl(l)&FL_HASH_MASK)
51 static atomic_t fl_size = ATOMIC_INIT(0);
52 static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1];
54 static void ip6_fl_gc(unsigned long dummy);
55 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc, 0, 0);
57 /* FL hash table lock: it protects only of GC */
59 static DEFINE_RWLOCK(ip6_fl_lock);
63 static DEFINE_RWLOCK(ip6_sk_fl_lock);
66 static inline struct ip6_flowlabel *__fl_lookup(struct net *net, __be32 label)
68 struct ip6_flowlabel *fl;
70 for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) {
71 if (fl->label == label && net_eq(fl->fl_net, net))
77 static struct ip6_flowlabel *fl_lookup(struct net *net, __be32 label)
79 struct ip6_flowlabel *fl;
81 read_lock_bh(&ip6_fl_lock);
82 fl = __fl_lookup(net, label);
84 atomic_inc(&fl->users);
85 read_unlock_bh(&ip6_fl_lock);
90 static void fl_free(struct ip6_flowlabel *fl)
93 release_net(fl->fl_net);
99 static void fl_release(struct ip6_flowlabel *fl)
101 write_lock_bh(&ip6_fl_lock);
103 fl->lastuse = jiffies;
104 if (atomic_dec_and_test(&fl->users)) {
105 unsigned long ttd = fl->lastuse + fl->linger;
106 if (time_after(ttd, fl->expires))
109 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
110 struct ipv6_txoptions *opt = fl->opt;
114 if (!timer_pending(&ip6_fl_gc_timer) ||
115 time_after(ip6_fl_gc_timer.expires, ttd))
116 mod_timer(&ip6_fl_gc_timer, ttd);
118 write_unlock_bh(&ip6_fl_lock);
121 static void ip6_fl_gc(unsigned long dummy)
124 unsigned long now = jiffies;
125 unsigned long sched = 0;
127 write_lock(&ip6_fl_lock);
129 for (i=0; i<=FL_HASH_MASK; i++) {
130 struct ip6_flowlabel *fl, **flp;
132 while ((fl=*flp) != NULL) {
133 if (atomic_read(&fl->users) == 0) {
134 unsigned long ttd = fl->lastuse + fl->linger;
135 if (time_after(ttd, fl->expires))
138 if (time_after_eq(now, ttd)) {
141 atomic_dec(&fl_size);
144 if (!sched || time_before(ttd, sched))
150 if (!sched && atomic_read(&fl_size))
151 sched = now + FL_MAX_LINGER;
153 mod_timer(&ip6_fl_gc_timer, sched);
155 write_unlock(&ip6_fl_lock);
158 static void __net_exit ip6_fl_purge(struct net *net)
162 write_lock(&ip6_fl_lock);
163 for (i = 0; i <= FL_HASH_MASK; i++) {
164 struct ip6_flowlabel *fl, **flp;
166 while ((fl = *flp) != NULL) {
167 if (net_eq(fl->fl_net, net) &&
168 atomic_read(&fl->users) == 0) {
171 atomic_dec(&fl_size);
177 write_unlock(&ip6_fl_lock);
180 static struct ip6_flowlabel *fl_intern(struct net *net,
181 struct ip6_flowlabel *fl, __be32 label)
183 struct ip6_flowlabel *lfl;
185 fl->label = label & IPV6_FLOWLABEL_MASK;
187 write_lock_bh(&ip6_fl_lock);
190 fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK;
192 lfl = __fl_lookup(net, fl->label);
199 * we dropper the ip6_fl_lock, so this entry could reappear
200 * and we need to recheck with it.
202 * OTOH no need to search the active socket first, like it is
203 * done in ipv6_flowlabel_opt - sock is locked, so new entry
204 * with the same label can only appear on another sock
206 lfl = __fl_lookup(net, fl->label);
208 atomic_inc(&lfl->users);
209 write_unlock_bh(&ip6_fl_lock);
214 fl->lastuse = jiffies;
215 fl->next = fl_ht[FL_HASH(fl->label)];
216 fl_ht[FL_HASH(fl->label)] = fl;
217 atomic_inc(&fl_size);
218 write_unlock_bh(&ip6_fl_lock);
224 /* Socket flowlabel lists */
226 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, __be32 label)
228 struct ipv6_fl_socklist *sfl;
229 struct ipv6_pinfo *np = inet6_sk(sk);
231 label &= IPV6_FLOWLABEL_MASK;
233 read_lock_bh(&ip6_sk_fl_lock);
234 for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) {
235 struct ip6_flowlabel *fl = sfl->fl;
236 if (fl->label == label) {
237 fl->lastuse = jiffies;
238 atomic_inc(&fl->users);
239 read_unlock_bh(&ip6_sk_fl_lock);
243 read_unlock_bh(&ip6_sk_fl_lock);
247 EXPORT_SYMBOL_GPL(fl6_sock_lookup);
249 void fl6_free_socklist(struct sock *sk)
251 struct ipv6_pinfo *np = inet6_sk(sk);
252 struct ipv6_fl_socklist *sfl;
254 while ((sfl = np->ipv6_fl_list) != NULL) {
255 np->ipv6_fl_list = sfl->next;
261 /* Service routines */
265 It is the only difficult place. flowlabel enforces equal headers
266 before and including routing header, however user may supply options
270 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
271 struct ip6_flowlabel * fl,
272 struct ipv6_txoptions * fopt)
274 struct ipv6_txoptions * fl_opt = fl->opt;
276 if (fopt == NULL || fopt->opt_flen == 0)
279 if (fl_opt != NULL) {
280 opt_space->hopopt = fl_opt->hopopt;
281 opt_space->dst0opt = fl_opt->dst0opt;
282 opt_space->srcrt = fl_opt->srcrt;
283 opt_space->opt_nflen = fl_opt->opt_nflen;
285 if (fopt->opt_nflen == 0)
287 opt_space->hopopt = NULL;
288 opt_space->dst0opt = NULL;
289 opt_space->srcrt = NULL;
290 opt_space->opt_nflen = 0;
292 opt_space->dst1opt = fopt->dst1opt;
293 opt_space->opt_flen = fopt->opt_flen;
297 static unsigned long check_linger(unsigned long ttl)
299 if (ttl < FL_MIN_LINGER)
300 return FL_MIN_LINGER*HZ;
301 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
306 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
308 linger = check_linger(linger);
311 expires = check_linger(expires);
314 fl->lastuse = jiffies;
315 if (time_before(fl->linger, linger))
317 if (time_before(expires, fl->linger))
318 expires = fl->linger;
319 if (time_before(fl->expires, fl->lastuse + expires))
320 fl->expires = fl->lastuse + expires;
324 static struct ip6_flowlabel *
325 fl_create(struct net *net, struct in6_flowlabel_req *freq, char __user *optval,
326 int optlen, int *err_p)
328 struct ip6_flowlabel *fl = NULL;
333 olen = optlen - CMSG_ALIGN(sizeof(*freq));
335 if (olen > 64 * 1024)
339 fl = kzalloc(sizeof(*fl), GFP_KERNEL);
349 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
353 memset(fl->opt, 0, sizeof(*fl->opt));
354 fl->opt->tot_len = sizeof(*fl->opt) + olen;
356 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
359 msg.msg_controllen = olen;
360 msg.msg_control = (void*)(fl->opt+1);
363 err = datagram_send_ctl(net, &msg, &flowi, fl->opt, &junk, &junk);
367 if (fl->opt->opt_flen)
369 if (fl->opt->opt_nflen == 0) {
375 fl->fl_net = hold_net(net);
376 fl->expires = jiffies;
377 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
380 fl->share = freq->flr_share;
381 addr_type = ipv6_addr_type(&freq->flr_dst);
382 if ((addr_type & IPV6_ADDR_MAPPED) ||
383 addr_type == IPV6_ADDR_ANY) {
387 ipv6_addr_copy(&fl->dst, &freq->flr_dst);
388 atomic_set(&fl->users, 1);
393 case IPV6_FL_S_PROCESS:
394 fl->owner = current->pid;
397 fl->owner = current_euid();
411 static int mem_check(struct sock *sk)
413 struct ipv6_pinfo *np = inet6_sk(sk);
414 struct ipv6_fl_socklist *sfl;
415 int room = FL_MAX_SIZE - atomic_read(&fl_size);
418 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
421 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next)
425 ((count >= FL_MAX_PER_SOCK ||
426 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4) &&
427 !capable(CAP_NET_ADMIN)))
433 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
437 if (h1 == NULL || h2 == NULL)
439 if (h1->hdrlen != h2->hdrlen)
441 return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
444 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
448 if (o1 == NULL || o2 == NULL)
450 if (o1->opt_nflen != o2->opt_nflen)
452 if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
454 if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
456 if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
461 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
462 struct ip6_flowlabel *fl)
464 write_lock_bh(&ip6_sk_fl_lock);
466 sfl->next = np->ipv6_fl_list;
467 np->ipv6_fl_list = sfl;
468 write_unlock_bh(&ip6_sk_fl_lock);
471 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
473 int uninitialized_var(err);
474 struct net *net = sock_net(sk);
475 struct ipv6_pinfo *np = inet6_sk(sk);
476 struct in6_flowlabel_req freq;
477 struct ipv6_fl_socklist *sfl1=NULL;
478 struct ipv6_fl_socklist *sfl, **sflp;
479 struct ip6_flowlabel *fl, *fl1 = NULL;
482 if (optlen < sizeof(freq))
485 if (copy_from_user(&freq, optval, sizeof(freq)))
488 switch (freq.flr_action) {
490 write_lock_bh(&ip6_sk_fl_lock);
491 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
492 if (sfl->fl->label == freq.flr_label) {
493 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
494 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
496 write_unlock_bh(&ip6_sk_fl_lock);
502 write_unlock_bh(&ip6_sk_fl_lock);
505 case IPV6_FL_A_RENEW:
506 read_lock_bh(&ip6_sk_fl_lock);
507 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
508 if (sfl->fl->label == freq.flr_label) {
509 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
510 read_unlock_bh(&ip6_sk_fl_lock);
514 read_unlock_bh(&ip6_sk_fl_lock);
516 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
517 fl = fl_lookup(net, freq.flr_label);
519 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
527 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
530 fl = fl_create(net, &freq, optval, optlen, &err);
533 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
535 if (freq.flr_label) {
537 read_lock_bh(&ip6_sk_fl_lock);
538 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
539 if (sfl->fl->label == freq.flr_label) {
540 if (freq.flr_flags&IPV6_FL_F_EXCL) {
541 read_unlock_bh(&ip6_sk_fl_lock);
545 atomic_inc(&fl1->users);
549 read_unlock_bh(&ip6_sk_fl_lock);
552 fl1 = fl_lookup(net, freq.flr_label);
556 if (freq.flr_flags&IPV6_FL_F_EXCL)
559 if (fl1->share == IPV6_FL_S_EXCL ||
560 fl1->share != fl->share ||
561 fl1->owner != fl->owner)
565 if (!ipv6_addr_equal(&fl1->dst, &fl->dst) ||
566 ipv6_opt_cmp(fl1->opt, fl->opt))
572 if (fl->linger > fl1->linger)
573 fl1->linger = fl->linger;
574 if ((long)(fl->expires - fl1->expires) > 0)
575 fl1->expires = fl->expires;
576 fl_link(np, sfl1, fl1);
586 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
590 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
593 fl1 = fl_intern(net, fl, freq.flr_label);
597 if (!freq.flr_label) {
598 if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
599 &fl->label, sizeof(fl->label))) {
600 /* Intentionally ignore fault. */
604 fl_link(np, sfl1, fl);
617 #ifdef CONFIG_PROC_FS
619 struct ip6fl_iter_state {
620 struct seq_net_private p;
624 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
626 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
628 struct ip6_flowlabel *fl = NULL;
629 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
630 struct net *net = seq_file_net(seq);
632 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
633 fl = fl_ht[state->bucket];
635 while (fl && !net_eq(fl->fl_net, net))
643 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
645 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
646 struct net *net = seq_file_net(seq);
650 while (fl && !net_eq(fl->fl_net, net))
654 if (++state->bucket <= FL_HASH_MASK) {
655 fl = fl_ht[state->bucket];
663 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
665 struct ip6_flowlabel *fl = ip6fl_get_first(seq);
667 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
669 return pos ? NULL : fl;
672 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
673 __acquires(ip6_fl_lock)
675 read_lock_bh(&ip6_fl_lock);
676 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
679 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
681 struct ip6_flowlabel *fl;
683 if (v == SEQ_START_TOKEN)
684 fl = ip6fl_get_first(seq);
686 fl = ip6fl_get_next(seq, v);
691 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
692 __releases(ip6_fl_lock)
694 read_unlock_bh(&ip6_fl_lock);
697 static int ip6fl_seq_show(struct seq_file *seq, void *v)
699 if (v == SEQ_START_TOKEN)
700 seq_printf(seq, "%-5s %-1s %-6s %-6s %-6s %-8s %-32s %s\n",
701 "Label", "S", "Owner", "Users", "Linger", "Expires", "Dst", "Opt");
703 struct ip6_flowlabel *fl = v;
705 "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
706 (unsigned)ntohl(fl->label),
709 atomic_read(&fl->users),
711 (long)(fl->expires - jiffies)/HZ,
713 fl->opt ? fl->opt->opt_nflen : 0);
718 static const struct seq_operations ip6fl_seq_ops = {
719 .start = ip6fl_seq_start,
720 .next = ip6fl_seq_next,
721 .stop = ip6fl_seq_stop,
722 .show = ip6fl_seq_show,
725 static int ip6fl_seq_open(struct inode *inode, struct file *file)
727 return seq_open_net(inode, file, &ip6fl_seq_ops,
728 sizeof(struct ip6fl_iter_state));
731 static const struct file_operations ip6fl_seq_fops = {
732 .owner = THIS_MODULE,
733 .open = ip6fl_seq_open,
736 .release = seq_release_net,
739 static int __net_init ip6_flowlabel_proc_init(struct net *net)
741 if (!proc_net_fops_create(net, "ip6_flowlabel",
742 S_IRUGO, &ip6fl_seq_fops))
747 static void __net_exit ip6_flowlabel_proc_fini(struct net *net)
749 proc_net_remove(net, "ip6_flowlabel");
752 static inline int ip6_flowlabel_proc_init(struct net *net)
756 static inline void ip6_flowlabel_proc_fini(struct net *net)
761 static void __net_exit ip6_flowlabel_net_exit(struct net *net)
764 ip6_flowlabel_proc_fini(net);
767 static struct pernet_operations ip6_flowlabel_net_ops = {
768 .init = ip6_flowlabel_proc_init,
769 .exit = ip6_flowlabel_net_exit,
772 int ip6_flowlabel_init(void)
774 return register_pernet_subsys(&ip6_flowlabel_net_ops);
777 void ip6_flowlabel_cleanup(void)
779 del_timer(&ip6_fl_gc_timer);
780 unregister_pernet_subsys(&ip6_flowlabel_net_ops);