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/config.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>
27 #include <net/ndisc.h>
28 #include <net/protocol.h>
29 #include <net/ip6_route.h>
30 #include <net/addrconf.h>
31 #include <net/rawv6.h>
33 #include <net/transp_v6.h>
35 #include <asm/uaccess.h>
37 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
38 in old IPv6 RFC. Well, it was reasonable value.
40 #define FL_MAX_LINGER 60 /* Maximal linger timeout */
44 #define FL_MAX_PER_SOCK 32
45 #define FL_MAX_SIZE 4096
46 #define FL_HASH_MASK 255
47 #define FL_HASH(l) (ntohl(l)&FL_HASH_MASK)
49 static atomic_t fl_size = ATOMIC_INIT(0);
50 static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1];
52 static void ip6_fl_gc(unsigned long dummy);
53 static struct timer_list ip6_fl_gc_timer = TIMER_INITIALIZER(ip6_fl_gc, 0, 0);
55 /* FL hash table lock: it protects only of GC */
57 static DEFINE_RWLOCK(ip6_fl_lock);
61 static DEFINE_RWLOCK(ip6_sk_fl_lock);
64 static __inline__ struct ip6_flowlabel * __fl_lookup(u32 label)
66 struct ip6_flowlabel *fl;
68 for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) {
69 if (fl->label == label)
75 static struct ip6_flowlabel * fl_lookup(u32 label)
77 struct ip6_flowlabel *fl;
79 read_lock_bh(&ip6_fl_lock);
80 fl = __fl_lookup(label);
82 atomic_inc(&fl->users);
83 read_unlock_bh(&ip6_fl_lock);
88 static void fl_free(struct ip6_flowlabel *fl)
95 static void fl_release(struct ip6_flowlabel *fl)
97 write_lock_bh(&ip6_fl_lock);
99 fl->lastuse = jiffies;
100 if (atomic_dec_and_test(&fl->users)) {
101 unsigned long ttd = fl->lastuse + fl->linger;
102 if (time_after(ttd, fl->expires))
105 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
106 struct ipv6_txoptions *opt = fl->opt;
110 if (!timer_pending(&ip6_fl_gc_timer) ||
111 time_after(ip6_fl_gc_timer.expires, ttd))
112 mod_timer(&ip6_fl_gc_timer, ttd);
115 write_unlock_bh(&ip6_fl_lock);
118 static void ip6_fl_gc(unsigned long dummy)
121 unsigned long now = jiffies;
122 unsigned long sched = 0;
124 write_lock(&ip6_fl_lock);
126 for (i=0; i<=FL_HASH_MASK; i++) {
127 struct ip6_flowlabel *fl, **flp;
129 while ((fl=*flp) != NULL) {
130 if (atomic_read(&fl->users) == 0) {
131 unsigned long ttd = fl->lastuse + fl->linger;
132 if (time_after(ttd, fl->expires))
135 if (time_after_eq(now, ttd)) {
138 atomic_dec(&fl_size);
141 if (!sched || time_before(ttd, sched))
147 if (!sched && atomic_read(&fl_size))
148 sched = now + FL_MAX_LINGER;
150 ip6_fl_gc_timer.expires = sched;
151 add_timer(&ip6_fl_gc_timer);
153 write_unlock(&ip6_fl_lock);
156 static int fl_intern(struct ip6_flowlabel *fl, __u32 label)
158 fl->label = label & IPV6_FLOWLABEL_MASK;
160 write_lock_bh(&ip6_fl_lock);
163 fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK;
165 struct ip6_flowlabel *lfl;
166 lfl = __fl_lookup(fl->label);
173 fl->lastuse = jiffies;
174 fl->next = fl_ht[FL_HASH(fl->label)];
175 fl_ht[FL_HASH(fl->label)] = fl;
176 atomic_inc(&fl_size);
177 write_unlock_bh(&ip6_fl_lock);
183 /* Socket flowlabel lists */
185 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, u32 label)
187 struct ipv6_fl_socklist *sfl;
188 struct ipv6_pinfo *np = inet6_sk(sk);
190 label &= IPV6_FLOWLABEL_MASK;
192 for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) {
193 struct ip6_flowlabel *fl = sfl->fl;
194 if (fl->label == label) {
195 fl->lastuse = jiffies;
196 atomic_inc(&fl->users);
203 void fl6_free_socklist(struct sock *sk)
205 struct ipv6_pinfo *np = inet6_sk(sk);
206 struct ipv6_fl_socklist *sfl;
208 while ((sfl = np->ipv6_fl_list) != NULL) {
209 np->ipv6_fl_list = sfl->next;
215 /* Service routines */
219 It is the only difficult place. flowlabel enforces equal headers
220 before and including routing header, however user may supply options
224 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
225 struct ip6_flowlabel * fl,
226 struct ipv6_txoptions * fopt)
228 struct ipv6_txoptions * fl_opt = fl->opt;
230 if (fopt == NULL || fopt->opt_flen == 0)
233 if (fl_opt != NULL) {
234 opt_space->hopopt = fl_opt->hopopt;
235 opt_space->dst0opt = fl_opt->dst0opt;
236 opt_space->srcrt = fl_opt->srcrt;
237 opt_space->opt_nflen = fl_opt->opt_nflen;
239 if (fopt->opt_nflen == 0)
241 opt_space->hopopt = NULL;
242 opt_space->dst0opt = NULL;
243 opt_space->srcrt = NULL;
244 opt_space->opt_nflen = 0;
246 opt_space->dst1opt = fopt->dst1opt;
247 opt_space->opt_flen = fopt->opt_flen;
251 static unsigned long check_linger(unsigned long ttl)
253 if (ttl < FL_MIN_LINGER)
254 return FL_MIN_LINGER*HZ;
255 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
260 static int fl6_renew(struct ip6_flowlabel *fl, unsigned long linger, unsigned long expires)
262 linger = check_linger(linger);
265 expires = check_linger(expires);
268 fl->lastuse = jiffies;
269 if (time_before(fl->linger, linger))
271 if (time_before(expires, fl->linger))
272 expires = fl->linger;
273 if (time_before(fl->expires, fl->lastuse + expires))
274 fl->expires = fl->lastuse + expires;
278 static struct ip6_flowlabel *
279 fl_create(struct in6_flowlabel_req *freq, char __user *optval, int optlen, int *err_p)
281 struct ip6_flowlabel *fl;
287 fl = kmalloc(sizeof(*fl), GFP_KERNEL);
290 memset(fl, 0, sizeof(*fl));
292 olen = optlen - CMSG_ALIGN(sizeof(*freq));
299 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
303 memset(fl->opt, 0, sizeof(*fl->opt));
304 fl->opt->tot_len = sizeof(*fl->opt) + olen;
306 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
309 msg.msg_controllen = olen;
310 msg.msg_control = (void*)(fl->opt+1);
313 err = datagram_send_ctl(&msg, &flowi, fl->opt, &junk);
317 if (fl->opt->opt_flen)
319 if (fl->opt->opt_nflen == 0) {
325 fl->expires = jiffies;
326 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
329 fl->share = freq->flr_share;
330 addr_type = ipv6_addr_type(&freq->flr_dst);
331 if ((addr_type&IPV6_ADDR_MAPPED)
332 || addr_type == IPV6_ADDR_ANY)
334 ipv6_addr_copy(&fl->dst, &freq->flr_dst);
335 atomic_set(&fl->users, 1);
340 case IPV6_FL_S_PROCESS:
341 fl->owner = current->pid;
344 fl->owner = current->euid;
358 static int mem_check(struct sock *sk)
360 struct ipv6_pinfo *np = inet6_sk(sk);
361 struct ipv6_fl_socklist *sfl;
362 int room = FL_MAX_SIZE - atomic_read(&fl_size);
365 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
368 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next)
372 ((count >= FL_MAX_PER_SOCK ||
373 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4)
374 && !capable(CAP_NET_ADMIN)))
380 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
384 if (h1 == NULL || h2 == NULL)
386 if (h1->hdrlen != h2->hdrlen)
388 return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
391 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
395 if (o1 == NULL || o2 == NULL)
397 if (o1->opt_nflen != o2->opt_nflen)
399 if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
401 if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
403 if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
408 int ipv6_flowlabel_opt(struct sock *sk, char __user *optval, int optlen)
411 struct ipv6_pinfo *np = inet6_sk(sk);
412 struct in6_flowlabel_req freq;
413 struct ipv6_fl_socklist *sfl1=NULL;
414 struct ipv6_fl_socklist *sfl, **sflp;
415 struct ip6_flowlabel *fl;
417 if (optlen < sizeof(freq))
420 if (copy_from_user(&freq, optval, sizeof(freq)))
423 switch (freq.flr_action) {
425 write_lock_bh(&ip6_sk_fl_lock);
426 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
427 if (sfl->fl->label == freq.flr_label) {
428 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
429 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
431 write_unlock_bh(&ip6_sk_fl_lock);
437 write_unlock_bh(&ip6_sk_fl_lock);
440 case IPV6_FL_A_RENEW:
441 read_lock_bh(&ip6_sk_fl_lock);
442 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
443 if (sfl->fl->label == freq.flr_label) {
444 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
445 read_unlock_bh(&ip6_sk_fl_lock);
449 read_unlock_bh(&ip6_sk_fl_lock);
451 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
452 fl = fl_lookup(freq.flr_label);
454 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
462 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
465 fl = fl_create(&freq, optval, optlen, &err);
468 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
470 if (freq.flr_label) {
471 struct ip6_flowlabel *fl1 = NULL;
474 read_lock_bh(&ip6_sk_fl_lock);
475 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
476 if (sfl->fl->label == freq.flr_label) {
477 if (freq.flr_flags&IPV6_FL_F_EXCL) {
478 read_unlock_bh(&ip6_sk_fl_lock);
482 atomic_inc(&fl->users);
486 read_unlock_bh(&ip6_sk_fl_lock);
489 fl1 = fl_lookup(freq.flr_label);
492 if (freq.flr_flags&IPV6_FL_F_EXCL)
495 if (fl1->share == IPV6_FL_S_EXCL ||
496 fl1->share != fl->share ||
497 fl1->owner != fl->owner)
501 if (!ipv6_addr_equal(&fl1->dst, &fl->dst) ||
502 ipv6_opt_cmp(fl1->opt, fl->opt))
508 if (fl->linger > fl1->linger)
509 fl1->linger = fl->linger;
510 if ((long)(fl->expires - fl1->expires) > 0)
511 fl1->expires = fl->expires;
512 write_lock_bh(&ip6_sk_fl_lock);
514 sfl1->next = np->ipv6_fl_list;
515 np->ipv6_fl_list = sfl1;
516 write_unlock_bh(&ip6_sk_fl_lock);
526 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
530 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
533 err = fl_intern(fl, freq.flr_label);
537 if (!freq.flr_label) {
538 if (copy_to_user(&((struct in6_flowlabel_req __user *) optval)->flr_label,
539 &fl->label, sizeof(fl->label))) {
540 /* Intentionally ignore fault. */
545 sfl1->next = np->ipv6_fl_list;
546 np->ipv6_fl_list = sfl1;
559 #ifdef CONFIG_PROC_FS
561 struct ip6fl_iter_state {
565 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
567 static struct ip6_flowlabel *ip6fl_get_first(struct seq_file *seq)
569 struct ip6_flowlabel *fl = NULL;
570 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
572 for (state->bucket = 0; state->bucket <= FL_HASH_MASK; ++state->bucket) {
573 if (fl_ht[state->bucket]) {
574 fl = fl_ht[state->bucket];
581 static struct ip6_flowlabel *ip6fl_get_next(struct seq_file *seq, struct ip6_flowlabel *fl)
583 struct ip6fl_iter_state *state = ip6fl_seq_private(seq);
587 if (++state->bucket <= FL_HASH_MASK)
588 fl = fl_ht[state->bucket];
593 static struct ip6_flowlabel *ip6fl_get_idx(struct seq_file *seq, loff_t pos)
595 struct ip6_flowlabel *fl = ip6fl_get_first(seq);
597 while (pos && (fl = ip6fl_get_next(seq, fl)) != NULL)
599 return pos ? NULL : fl;
602 static void *ip6fl_seq_start(struct seq_file *seq, loff_t *pos)
604 read_lock_bh(&ip6_fl_lock);
605 return *pos ? ip6fl_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
608 static void *ip6fl_seq_next(struct seq_file *seq, void *v, loff_t *pos)
610 struct ip6_flowlabel *fl;
612 if (v == SEQ_START_TOKEN)
613 fl = ip6fl_get_first(seq);
615 fl = ip6fl_get_next(seq, v);
620 static void ip6fl_seq_stop(struct seq_file *seq, void *v)
622 read_unlock_bh(&ip6_fl_lock);
625 static void ip6fl_fl_seq_show(struct seq_file *seq, struct ip6_flowlabel *fl)
629 "%05X %-1d %-6d %-6d %-6ld %-8ld "
630 "%02x%02x%02x%02x%02x%02x%02x%02x "
632 (unsigned)ntohl(fl->label),
635 atomic_read(&fl->users),
637 (long)(fl->expires - jiffies)/HZ,
639 fl->opt ? fl->opt->opt_nflen : 0);
644 static int ip6fl_seq_show(struct seq_file *seq, void *v)
646 if (v == SEQ_START_TOKEN)
647 seq_puts(seq, "Label S Owner Users Linger Expires "
650 ip6fl_fl_seq_show(seq, v);
654 static struct seq_operations ip6fl_seq_ops = {
655 .start = ip6fl_seq_start,
656 .next = ip6fl_seq_next,
657 .stop = ip6fl_seq_stop,
658 .show = ip6fl_seq_show,
661 static int ip6fl_seq_open(struct inode *inode, struct file *file)
663 struct seq_file *seq;
665 struct ip6fl_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
670 rc = seq_open(file, &ip6fl_seq_ops);
674 seq = file->private_data;
676 memset(s, 0, sizeof(*s));
684 static struct file_operations ip6fl_seq_fops = {
685 .owner = THIS_MODULE,
686 .open = ip6fl_seq_open,
689 .release = seq_release_private,
694 void ip6_flowlabel_init(void)
696 #ifdef CONFIG_PROC_FS
697 proc_net_fops_create("ip6_flowlabel", S_IRUGO, &ip6fl_seq_fops);
701 void ip6_flowlabel_cleanup(void)
703 del_timer(&ip6_fl_gc_timer);
704 #ifdef CONFIG_PROC_FS
705 proc_net_remove("ip6_flowlabel");