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>
26 #include <net/ndisc.h>
27 #include <net/protocol.h>
28 #include <net/ip6_route.h>
29 #include <net/addrconf.h>
30 #include <net/rawv6.h>
32 #include <net/transp_v6.h>
34 #include <asm/uaccess.h>
36 #define FL_MIN_LINGER 6 /* Minimal linger. It is set to 6sec specified
37 in old IPv6 RFC. Well, it was reasonable value.
39 #define FL_MAX_LINGER 60 /* Maximal linger timeout */
43 #define FL_MAX_PER_SOCK 32
44 #define FL_MAX_SIZE 4096
45 #define FL_HASH_MASK 255
46 #define FL_HASH(l) (ntohl(l)&FL_HASH_MASK)
48 static atomic_t fl_size = ATOMIC_INIT(0);
49 static struct ip6_flowlabel *fl_ht[FL_HASH_MASK+1];
51 static struct timer_list ip6_fl_gc_timer;
53 /* FL hash table lock: it protects only of GC */
55 static rwlock_t ip6_fl_lock = RW_LOCK_UNLOCKED;
59 static rwlock_t ip6_sk_fl_lock = RW_LOCK_UNLOCKED;
62 static __inline__ struct ip6_flowlabel * __fl_lookup(u32 label)
64 struct ip6_flowlabel *fl;
66 for (fl=fl_ht[FL_HASH(label)]; fl; fl = fl->next) {
67 if (fl->label == label)
73 static struct ip6_flowlabel * fl_lookup(u32 label)
75 struct ip6_flowlabel *fl;
77 read_lock_bh(&ip6_fl_lock);
78 fl = __fl_lookup(label);
80 atomic_inc(&fl->users);
81 read_unlock_bh(&ip6_fl_lock);
86 static void fl_free(struct ip6_flowlabel *fl)
93 static void fl_release(struct ip6_flowlabel *fl)
95 fl->lastuse = jiffies;
96 if (atomic_dec_and_test(&fl->users)) {
97 unsigned long ttd = fl->lastuse + fl->linger;
98 if ((long)(ttd - fl->expires) > 0)
101 if (fl->opt && fl->share == IPV6_FL_S_EXCL) {
102 struct ipv6_txoptions *opt = fl->opt;
106 if (!del_timer(&ip6_fl_gc_timer) ||
107 (long)(ip6_fl_gc_timer.expires - ttd) > 0)
108 ip6_fl_gc_timer.expires = ttd;
109 add_timer(&ip6_fl_gc_timer);
113 static void ip6_fl_gc(unsigned long dummy)
116 unsigned long now = jiffies;
117 unsigned long sched = 0;
119 write_lock(&ip6_fl_lock);
121 for (i=0; i<=FL_HASH_MASK; i++) {
122 struct ip6_flowlabel *fl, **flp;
124 while ((fl=*flp) != NULL) {
125 if (atomic_read(&fl->users) == 0) {
126 unsigned long ttd = fl->lastuse + fl->linger;
127 if ((long)(ttd - fl->expires) > 0)
130 if ((long)(now - ttd) >= 0) {
133 atomic_dec(&fl_size);
136 if (!sched || (long)(ttd - sched) < 0)
142 if (!sched && atomic_read(&fl_size))
143 sched = now + FL_MAX_LINGER;
145 ip6_fl_gc_timer.expires = sched;
146 add_timer(&ip6_fl_gc_timer);
148 write_unlock(&ip6_fl_lock);
151 static int fl_intern(struct ip6_flowlabel *fl, __u32 label)
153 fl->label = label & IPV6_FLOWLABEL_MASK;
155 write_lock_bh(&ip6_fl_lock);
158 fl->label = htonl(net_random())&IPV6_FLOWLABEL_MASK;
160 struct ip6_flowlabel *lfl;
161 lfl = __fl_lookup(fl->label);
168 fl->lastuse = jiffies;
169 fl->next = fl_ht[FL_HASH(fl->label)];
170 fl_ht[FL_HASH(fl->label)] = fl;
171 atomic_inc(&fl_size);
172 write_unlock_bh(&ip6_fl_lock);
178 /* Socket flowlabel lists */
180 struct ip6_flowlabel * fl6_sock_lookup(struct sock *sk, u32 label)
182 struct ipv6_fl_socklist *sfl;
183 struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
185 label &= IPV6_FLOWLABEL_MASK;
187 for (sfl=np->ipv6_fl_list; sfl; sfl = sfl->next) {
188 struct ip6_flowlabel *fl = sfl->fl;
189 if (fl->label == label) {
190 fl->lastuse = jiffies;
191 atomic_inc(&fl->users);
198 void fl6_free_socklist(struct sock *sk)
200 struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
201 struct ipv6_fl_socklist *sfl;
203 while ((sfl = np->ipv6_fl_list) != NULL) {
204 np->ipv6_fl_list = sfl->next;
210 /* Service routines */
214 It is the only difficult place. flowlabel enforces equal headers
215 before and including routing header, however user may supply options
219 struct ipv6_txoptions *fl6_merge_options(struct ipv6_txoptions * opt_space,
220 struct ip6_flowlabel * fl,
221 struct ipv6_txoptions * fopt)
223 struct ipv6_txoptions * fl_opt = fl->opt;
225 if (fopt == NULL || fopt->opt_flen == 0)
228 if (fl_opt != NULL) {
229 opt_space->hopopt = fl_opt->hopopt;
230 opt_space->dst0opt = fl_opt->dst0opt;
231 opt_space->srcrt = fl_opt->srcrt;
232 opt_space->opt_nflen = fl_opt->opt_nflen;
234 if (fopt->opt_nflen == 0)
236 opt_space->hopopt = NULL;
237 opt_space->dst0opt = NULL;
238 opt_space->srcrt = NULL;
239 opt_space->opt_nflen = 0;
241 opt_space->dst1opt = fopt->dst1opt;
242 opt_space->auth = fopt->auth;
243 opt_space->opt_flen = fopt->opt_flen;
247 static __u32 check_linger(__u16 ttl)
249 if (ttl < FL_MIN_LINGER)
250 return FL_MIN_LINGER*HZ;
251 if (ttl > FL_MAX_LINGER && !capable(CAP_NET_ADMIN))
256 static int fl6_renew(struct ip6_flowlabel *fl, unsigned linger, unsigned expires)
258 linger = check_linger(linger);
261 expires = check_linger(expires);
264 fl->lastuse = jiffies;
265 if (fl->linger < linger)
267 if (expires < fl->linger)
268 expires = fl->linger;
269 if ((long)(fl->expires - (fl->lastuse+expires)) < 0)
270 fl->expires = fl->lastuse + expires;
274 static struct ip6_flowlabel *
275 fl_create(struct in6_flowlabel_req *freq, char *optval, int optlen, int *err_p)
277 struct ip6_flowlabel *fl;
283 fl = kmalloc(sizeof(*fl), GFP_KERNEL);
286 memset(fl, 0, sizeof(*fl));
288 olen = optlen - CMSG_ALIGN(sizeof(*freq));
295 fl->opt = kmalloc(sizeof(*fl->opt) + olen, GFP_KERNEL);
299 memset(fl->opt, 0, sizeof(*fl->opt));
300 fl->opt->tot_len = sizeof(*fl->opt) + olen;
302 if (copy_from_user(fl->opt+1, optval+CMSG_ALIGN(sizeof(*freq)), olen))
305 msg.msg_controllen = olen;
306 msg.msg_control = (void*)(fl->opt+1);
309 err = datagram_send_ctl(&msg, &flowi, fl->opt, &junk);
313 if (fl->opt->opt_flen)
315 if (fl->opt->opt_nflen == 0) {
321 fl->expires = jiffies;
322 err = fl6_renew(fl, freq->flr_linger, freq->flr_expires);
325 fl->share = freq->flr_share;
326 addr_type = ipv6_addr_type(&freq->flr_dst);
327 if ((addr_type&IPV6_ADDR_MAPPED)
328 || addr_type == IPV6_ADDR_ANY)
330 ipv6_addr_copy(&fl->dst, &freq->flr_dst);
331 atomic_set(&fl->users, 1);
336 case IPV6_FL_S_PROCESS:
337 fl->owner = current->pid;
340 fl->owner = current->euid;
355 static int mem_check(struct sock *sk)
357 struct ipv6_fl_socklist *sfl;
358 int room = FL_MAX_SIZE - atomic_read(&fl_size);
361 if (room > FL_MAX_SIZE - FL_MAX_PER_SOCK)
364 for (sfl = sk->net_pinfo.af_inet6.ipv6_fl_list; sfl; sfl = sfl->next)
368 ((count >= FL_MAX_PER_SOCK ||
369 (count > 0 && room < FL_MAX_SIZE/2) || room < FL_MAX_SIZE/4)
370 && !capable(CAP_NET_ADMIN)))
376 static int ipv6_hdr_cmp(struct ipv6_opt_hdr *h1, struct ipv6_opt_hdr *h2)
380 if (h1 == NULL || h2 == NULL)
382 if (h1->hdrlen != h2->hdrlen)
384 return memcmp(h1+1, h2+1, ((h1->hdrlen+1)<<3) - sizeof(*h1));
387 static int ipv6_opt_cmp(struct ipv6_txoptions *o1, struct ipv6_txoptions *o2)
391 if (o1 == NULL || o2 == NULL)
393 if (o1->opt_nflen != o2->opt_nflen)
395 if (ipv6_hdr_cmp(o1->hopopt, o2->hopopt))
397 if (ipv6_hdr_cmp(o1->dst0opt, o2->dst0opt))
399 if (ipv6_hdr_cmp((struct ipv6_opt_hdr *)o1->srcrt, (struct ipv6_opt_hdr *)o2->srcrt))
404 int ipv6_flowlabel_opt(struct sock *sk, char *optval, int optlen)
407 struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
408 struct in6_flowlabel_req freq;
409 struct ipv6_fl_socklist *sfl1=NULL;
410 struct ipv6_fl_socklist *sfl, **sflp;
411 struct ip6_flowlabel *fl;
413 if (optlen < sizeof(freq))
416 if (copy_from_user(&freq, optval, sizeof(freq)))
419 switch (freq.flr_action) {
421 write_lock_bh(&ip6_sk_fl_lock);
422 for (sflp = &np->ipv6_fl_list; (sfl=*sflp)!=NULL; sflp = &sfl->next) {
423 if (sfl->fl->label == freq.flr_label) {
424 if (freq.flr_label == (np->flow_label&IPV6_FLOWLABEL_MASK))
425 np->flow_label &= ~IPV6_FLOWLABEL_MASK;
427 write_unlock_bh(&ip6_sk_fl_lock);
433 write_unlock_bh(&ip6_sk_fl_lock);
436 case IPV6_FL_A_RENEW:
437 read_lock_bh(&ip6_sk_fl_lock);
438 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
439 if (sfl->fl->label == freq.flr_label) {
440 err = fl6_renew(sfl->fl, freq.flr_linger, freq.flr_expires);
441 read_unlock_bh(&ip6_sk_fl_lock);
445 read_unlock_bh(&ip6_sk_fl_lock);
447 if (freq.flr_share == IPV6_FL_S_NONE && capable(CAP_NET_ADMIN)) {
448 fl = fl_lookup(freq.flr_label);
450 err = fl6_renew(fl, freq.flr_linger, freq.flr_expires);
458 if (freq.flr_label & ~IPV6_FLOWLABEL_MASK)
461 fl = fl_create(&freq, optval, optlen, &err);
464 sfl1 = kmalloc(sizeof(*sfl1), GFP_KERNEL);
466 if (freq.flr_label) {
467 struct ip6_flowlabel *fl1 = NULL;
470 read_lock_bh(&ip6_sk_fl_lock);
471 for (sfl = np->ipv6_fl_list; sfl; sfl = sfl->next) {
472 if (sfl->fl->label == freq.flr_label) {
473 if (freq.flr_flags&IPV6_FL_F_EXCL) {
474 read_unlock_bh(&ip6_sk_fl_lock);
478 atomic_inc(&fl1->users);
482 read_unlock_bh(&ip6_sk_fl_lock);
485 fl1 = fl_lookup(freq.flr_label);
488 if (freq.flr_flags&IPV6_FL_F_EXCL)
491 if (fl1->share == IPV6_FL_S_EXCL ||
492 fl1->share != fl->share ||
493 fl1->owner != fl->owner)
497 if (ipv6_addr_cmp(&fl1->dst, &fl->dst) ||
498 ipv6_opt_cmp(fl1->opt, fl->opt))
504 if (fl->linger > fl1->linger)
505 fl1->linger = fl->linger;
506 if ((long)(fl->expires - fl1->expires) > 0)
507 fl1->expires = fl->expires;
508 write_lock_bh(&ip6_sk_fl_lock);
510 sfl1->next = np->ipv6_fl_list;
511 np->ipv6_fl_list = sfl1;
512 write_unlock_bh(&ip6_sk_fl_lock);
522 if (!(freq.flr_flags&IPV6_FL_F_CREATE))
526 if (sfl1 == NULL || (err = mem_check(sk)) != 0)
529 err = fl_intern(fl, freq.flr_label);
533 /* Do not check for fault */
535 copy_to_user(optval + ((u8*)&freq.flr_label - (u8*)&freq), &fl->label, sizeof(fl->label));
538 sfl1->next = np->ipv6_fl_list;
539 np->ipv6_fl_list = sfl1;
554 #ifdef CONFIG_PROC_FS
557 static int ip6_fl_read_proc(char *buffer, char **start, off_t offset,
558 int length, int *eof, void *data)
564 struct ip6_flowlabel *fl;
566 len+= sprintf(buffer,"Label S Owner Users Linger Expires "
569 read_lock_bh(&ip6_fl_lock);
570 for (i=0; i<=FL_HASH_MASK; i++) {
571 for (fl = fl_ht[i]; fl; fl = fl->next) {
572 len+=sprintf(buffer+len,"%05X %-1d %-6d %-6d %-6d %-8ld ",
573 (unsigned)ntohl(fl->label),
576 atomic_read(&fl->users),
578 (long)(fl->expires - jiffies)/HZ);
581 len+=sprintf(buffer+len, "%02x", fl->dst.s6_addr[k]);
583 len+=sprintf(buffer+len, "%-4d", fl->opt ? fl->opt->opt_nflen : 0);
591 if(pos>offset+length)
598 read_unlock_bh(&ip6_fl_lock);
599 *start=buffer+(offset-begin);
610 void ip6_flowlabel_init()
612 init_timer(&ip6_fl_gc_timer);
613 ip6_fl_gc_timer.function = ip6_fl_gc;
614 #ifdef CONFIG_PROC_FS
615 create_proc_read_entry("net/ip6_flowlabel", 0, 0, ip6_fl_read_proc, NULL);
619 void ip6_flowlabel_cleanup()
621 del_timer(&ip6_fl_gc_timer);
622 #ifdef CONFIG_PROC_FS
623 remove_proc_entry("net/ip6_flowlabel", 0);