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>
24 #include <net/net_namespace.h>
28 #include <net/ndisc.h>
29 #include <net/protocol.h>
30 #include <net/ip6_route.h>
31 #include <net/addrconf.h>
32 #include <net/rawv6.h>
34 #include <net/transp_v6.h>
36 #include <asm/uaccess.h>
38 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
39 in old IPv6 RFC. Well, it was reasonable value.
41 #define FL_MAX_LINGER 60 /* Maximal linger timeout */
45 #define FL_MAX_PER_SOCK 32
46 #define FL_MAX_SIZE 4096
47 #define FL_HASH_MASK 255
48 #define FL_HASH(l) (ntohl(l)&FL_HASH_MASK)
50 static atomic_t fl_size = ATOMIC_INIT(0);
51 static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1];
53 static void ip6_fl_gc(unsigned long dummy);
54 static DEFINE_TIMER(ip6_fl_gc_timer, ip6_fl_gc, 0, 0);
56 /* FL hash table lock: it protects only of GC */
58 static DEFINE_RWLOCK(ip6_fl_lock);
62 static DEFINE_RWLOCK(ip6_sk_fl_lock);
65 static inline struct ip6_flowlabel *__fl_lookup(struct net *net, __be32 label)
67 struct ip6_flowlabel *fl;
69 for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) {
70 if (fl->label == label && fl->fl_net == net)
76 static struct ip6_flowlabel *fl_lookup(struct net *net, __be32 label)
78 struct ip6_flowlabel *fl;
80 read_lock_bh(&ip6_fl_lock);
81 fl = __fl_lookup(net, label);
83 atomic_inc(&fl->users);
84 read_unlock_bh(&ip6_fl_lock);
89 static void fl_free(struct ip6_flowlabel *fl)
92 release_net(fl->fl_net);
98 static void fl_release(struct ip6_flowlabel *fl)
100 write_lock_bh(&ip6_fl_lock);
102 fl->lastuse = jiffies;
103 if (atomic_dec_and_test(&fl->users)) {
104 unsigned long ttd = fl->lastuse + fl->linger;
105 if (time_after(ttd, fl->expires))
108 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
109 struct ipv6_txoptions *opt = fl->opt;
113 if (!timer_pending(&ip6_fl_gc_timer) ||
114 time_after(ip6_fl_gc_timer.expires, ttd))
115 mod_timer(&ip6_fl_gc_timer, ttd);
117 write_unlock_bh(&ip6_fl_lock);
120 static void ip6_fl_gc(unsigned long dummy)
123 unsigned long now = jiffies;
124 unsigned long sched = 0;
126 write_lock(&ip6_fl_lock);
128 for (i=0; i<=FL_HASH_MASK; i++) {
129 struct ip6_flowlabel *fl, **flp;
131 while ((fl=*flp) != NULL) {
132 if (atomic_read(&fl->users) == 0) {
133 unsigned long ttd = fl->lastuse + fl->linger;
134 if (time_after(ttd, fl->expires))
137 if (time_after_eq(now, ttd)) {
140 atomic_dec(&fl_size);
143 if (!sched || time_before(ttd, sched))
149 if (!sched && atomic_read(&fl_size))
150 sched = now + FL_MAX_LINGER;
152 mod_timer(&ip6_fl_gc_timer, sched);
154 write_unlock(&ip6_fl_lock);
157 static void ip6_fl_purge(struct net *net)
161 write_lock(&ip6_fl_lock);
162 for (i = 0; i <= FL_HASH_MASK; i++) {
163 struct ip6_flowlabel *fl, **flp;
165 while ((fl = *flp) != NULL) {
166 if (fl->fl_net == net && atomic_read(&fl->users) == 0) {
169 atomic_dec(&fl_size);
175 write_unlock(&ip6_fl_lock);
178 static struct ip6_flowlabel *fl_intern(struct net *net,
179 struct ip6_flowlabel *fl, __be32 label)
181 struct ip6_flowlabel *lfl;
183 fl->label = label & IPV6_FLOWLABEL_MASK;
185 write_lock_bh(&ip6_fl_lock);
188 fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK;
190 lfl = __fl_lookup(net, fl->label);
197 * we dropper the ip6_fl_lock, so this entry could reappear
198 * and we need to recheck with it.
200 * OTOH no need to search the active socket first, like it is
201 * done in ipv6_flowlabel_opt - sock is locked, so new entry
202 * with the same label can only appear on another sock
204 lfl = __fl_lookup(net, fl->label);
206 atomic_inc(&lfl->users);
207 write_unlock_bh(&ip6_fl_lock);
212 fl->lastuse = jiffies;
213 fl->next = fl_ht[FL_HASH(fl->label)];
214 fl_ht[FL_HASH(fl->label)] = fl;
215 atomic_inc(&fl_size);
216 write_unlock_bh(&ip6_fl_lock);
222 /* Socket flowlabel lists */
224 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, __be32 label)
226 struct ipv6_fl_socklist *sfl;
227 struct ipv6_pinfo *np = inet6_sk(sk);
229 label &= IPV6_FLOWLABEL_MASK;
231 read_lock_bh(&ip6_sk_fl_lock);
232 for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) {
233 struct ip6_flowlabel *fl = sfl->fl;
234 if (fl->label == label) {
235 fl->lastuse = jiffies;
236 atomic_inc(&fl->users);
237 read_unlock_bh(&ip6_sk_fl_lock);
241 read_unlock_bh(&ip6_sk_fl_lock);
245 EXPORT_SYMBOL_GPL(fl6_sock_lookup);
247 void fl6_free_socklist(struct sock *sk)
249 struct ipv6_pinfo *np = inet6_sk(sk);
250 struct ipv6_fl_socklist *sfl;
252 while ((sfl = np->ipv6_fl_list) != NULL) {
253 np->ipv6_fl_list = sfl->next;
259 /* Service routines */
263 It is the only difficult place. flowlabel enforces equal headers
264 before and including routing header, however user may supply options
268 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
269 struct ip6_flowlabel * fl,
270 struct ipv6_txoptions * fopt)
272 struct ipv6_txoptions * fl_opt = fl->opt;
274 if (fopt == NULL || fopt->opt_flen == 0)
277 if (fl_opt != NULL) {
278 opt_space->hopopt = fl_opt->hopopt;
279 opt_space->dst0opt = fl_opt->dst0opt;
280 opt_space->srcrt = fl_opt->srcrt;
281 opt_space->opt_nflen = fl_opt->opt_nflen;
283 if (fopt->opt_nflen == 0)
285 opt_space->hopopt = NULL;
286 opt_space->dst0opt = NULL;
287 opt_space->srcrt = NULL;
288 opt_space->opt_nflen = 0;
290 opt_space->dst1opt = fopt->dst1opt;
291 opt_space->opt_flen = fopt->opt_flen;
295 static unsigned long check_linger(unsigned long ttl)
297 if (ttl < FL_MIN_LINGER)
298 return FL_MIN_LINGER*HZ;
299 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
304 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
306 linger = check_linger(linger);
309 expires = check_linger(expires);
312 fl->lastuse = jiffies;
313 if (time_before(fl->linger, linger))
315 if (time_before(expires, fl->linger))
316 expires = fl->linger;
317 if (time_before(fl->expires, fl->lastuse + expires))
318 fl->expires = fl->lastuse + expires;
322 static struct ip6_flowlabel *
323 fl_create(struct net *net, struct in6_flowlabel_req *freq, char __user *optval,
324 int optlen, int *err_p)
326 struct ip6_flowlabel *fl = NULL;
331 olen = optlen - CMSG_ALIGN(sizeof(*freq));
333 if (olen > 64 * 1024)
337 fl = kzalloc(sizeof(*fl), GFP_KERNEL);
347 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
351 memset(fl->opt, 0, sizeof(*fl->opt));
352 fl->opt->tot_len = sizeof(*fl->opt) + olen;
354 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
357 msg.msg_controllen = olen;
358 msg.msg_control = (void*)(fl->opt+1);
361 err = datagram_send_ctl(net, &msg, &flowi, fl->opt, &junk, &junk);
365 if (fl->opt->opt_flen)
367 if (fl->opt->opt_nflen == 0) {
373 fl->fl_net = hold_net(net);
374 fl->expires = jiffies;
375 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
378 fl->share = freq->flr_share;
379 addr_type = ipv6_addr_type(&freq->flr_dst);
380 if ((addr_type&IPV6_ADDR_MAPPED)
381 || addr_type == IPV6_ADDR_ANY) {
385 ipv6_addr_copy(&fl->dst, &freq->flr_dst);
386 atomic_set(&fl->users, 1);
391 case IPV6_FL_S_PROCESS:
392 fl->owner = current->pid;
395 fl->owner = current_euid();
409 static int mem_check(struct sock *sk)
411 struct ipv6_pinfo *np = inet6_sk(sk);
412 struct ipv6_fl_socklist *sfl;
413 int room = FL_MAX_SIZE - atomic_read(&fl_size);
416 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
419 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next)
423 ((count >= FL_MAX_PER_SOCK ||
424 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4)
425 && !capable(CAP_NET_ADMIN)))
431 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
435 if (h1 == NULL || h2 == NULL)
437 if (h1->hdrlen != h2->hdrlen)
439 return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
442 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
446 if (o1 == NULL || o2 == NULL)
448 if (o1->opt_nflen != o2->opt_nflen)
450 if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
452 if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
454 if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
459 static inline void fl_link(struct ipv6_pinfo *np, struct ipv6_fl_socklist *sfl,
460 struct ip6_flowlabel *fl)
462 write_lock_bh(&ip6_sk_fl_lock);
464 sfl->next = np->ipv6_fl_list;
465 np->ipv6_fl_list = sfl;
466 write_unlock_bh(&ip6_sk_fl_lock);
469 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
471 int uninitialized_var(err);
472 struct net *net = sock_net(sk);
473 struct ipv6_pinfo *np = inet6_sk(sk);
474 struct in6_flowlabel_req freq;
475 struct ipv6_fl_socklist *sfl1=NULL;
476 struct ipv6_fl_socklist *sfl, **sflp;
477 struct ip6_flowlabel *fl, *fl1 = NULL;
480 if (optlen < sizeof(freq))
483 if (copy_from_user(&freq, optval, sizeof(freq)))
486 switch (freq.flr_action) {
488 write_lock_bh(&ip6_sk_fl_lock);
489 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
490 if (sfl->fl->label == freq.flr_label) {
491 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
492 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
494 write_unlock_bh(&ip6_sk_fl_lock);
500 write_unlock_bh(&ip6_sk_fl_lock);
503 case IPV6_FL_A_RENEW:
504 read_lock_bh(&ip6_sk_fl_lock);
505 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
506 if (sfl->fl->label == freq.flr_label) {
507 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
508 read_unlock_bh(&ip6_sk_fl_lock);
512 read_unlock_bh(&ip6_sk_fl_lock);
514 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
515 fl = fl_lookup(net, freq.flr_label);
517 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
525 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
528 fl = fl_create(net, &freq, optval, optlen, &err);
531 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
533 if (freq.flr_label) {
535 read_lock_bh(&ip6_sk_fl_lock);
536 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
537 if (sfl->fl->label == freq.flr_label) {
538 if (freq.flr_flags&IPV6_FL_F_EXCL) {
539 read_unlock_bh(&ip6_sk_fl_lock);
543 atomic_inc(&fl1->users);
547 read_unlock_bh(&ip6_sk_fl_lock);
550 fl1 = fl_lookup(net, freq.flr_label);
554 if (freq.flr_flags&IPV6_FL_F_EXCL)
557 if (fl1->share == IPV6_FL_S_EXCL ||
558 fl1->share != fl->share ||
559 fl1->owner != fl->owner)
563 if (!ipv6_addr_equal(&fl1->dst, &fl->dst) ||
564 ipv6_opt_cmp(fl1->opt, fl->opt))
570 if (fl->linger > fl1->linger)
571 fl1->linger = fl->linger;
572 if ((long)(fl->expires - fl1->expires) > 0)
573 fl1->expires = fl->expires;
574 fl_link(np, sfl1, fl1);
584 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
588 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
591 fl1 = fl_intern(net, fl, freq.flr_label);
595 if (!freq.flr_label) {
596 if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
597 &fl->label, sizeof(fl->label))) {
598 /* Intentionally ignore fault. */
602 fl_link(np, sfl1, fl);
615 #ifdef CONFIG_PROC_FS
617 struct ip6fl_iter_state {
618 struct seq_net_private p;
622 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
624 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
626 struct ip6_flowlabel *fl = NULL;
627 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
628 struct net *net = seq_file_net(seq);
630 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
631 fl = fl_ht[state->bucket];
633 while (fl && fl->fl_net != net)
641 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
643 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
644 struct net *net = seq_file_net(seq);
648 while (fl && fl->fl_net != net)
652 if (++state->bucket <= FL_HASH_MASK) {
653 fl = fl_ht[state->bucket];
661 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
663 struct ip6_flowlabel *fl = ip6fl_get_first(seq);
665 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
667 return pos ? NULL : fl;
670 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
671 __acquires(ip6_fl_lock)
673 read_lock_bh(&ip6_fl_lock);
674 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
677 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
679 struct ip6_flowlabel *fl;
681 if (v == SEQ_START_TOKEN)
682 fl = ip6fl_get_first(seq);
684 fl = ip6fl_get_next(seq, v);
689 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
690 __releases(ip6_fl_lock)
692 read_unlock_bh(&ip6_fl_lock);
695 static int ip6fl_seq_show(struct seq_file *seq, void *v)
697 if (v == SEQ_START_TOKEN)
698 seq_printf(seq, "%-5s %-1s %-6s %-6s %-6s %-8s %-32s %s\n",
699 "Label", "S", "Owner", "Users", "Linger", "Expires", "Dst", "Opt");
701 struct ip6_flowlabel *fl = v;
703 "%05X %-1d %-6d %-6d %-6ld %-8ld %pi6 %-4d\n",
704 (unsigned)ntohl(fl->label),
707 atomic_read(&fl->users),
709 (long)(fl->expires - jiffies)/HZ,
711 fl->opt ? fl->opt->opt_nflen : 0);
716 static const struct seq_operations ip6fl_seq_ops = {
717 .start = ip6fl_seq_start,
718 .next = ip6fl_seq_next,
719 .stop = ip6fl_seq_stop,
720 .show = ip6fl_seq_show,
723 static int ip6fl_seq_open(struct inode *inode, struct file *file)
725 return seq_open_net(inode, file, &ip6fl_seq_ops,
726 sizeof(struct ip6fl_iter_state));
729 static const struct file_operations ip6fl_seq_fops = {
730 .owner = THIS_MODULE,
731 .open = ip6fl_seq_open,
734 .release = seq_release_net,
737 static int ip6_flowlabel_proc_init(struct net *net)
739 if (!proc_net_fops_create(net, "ip6_flowlabel",
740 S_IRUGO, &ip6fl_seq_fops))
745 static void ip6_flowlabel_proc_fini(struct net *net)
747 proc_net_remove(net, "ip6_flowlabel");
750 static inline int ip6_flowlabel_proc_init(struct net *net)
754 static inline void ip6_flowlabel_proc_fini(struct net *net)
760 static inline void ip6_flowlabel_net_exit(struct net *net)
763 ip6_flowlabel_proc_fini(net);
766 static struct pernet_operations ip6_flowlabel_net_ops = {
767 .init = ip6_flowlabel_proc_init,
768 .exit = ip6_flowlabel_net_exit,
771 int ip6_flowlabel_init(void)
773 return register_pernet_subsys(&ip6_flowlabel_net_ops);
776 void ip6_flowlabel_cleanup(void)
778 del_timer(&ip6_fl_gc_timer);
779 unregister_pernet_subsys(&ip6_flowlabel_net_ops);