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
;
332 fl
= kzalloc(sizeof(*fl
), GFP_KERNEL
);
336 olen
= optlen
- CMSG_ALIGN(sizeof(*freq
));
343 fl
->opt
= kmalloc(sizeof(*fl
->opt
) + olen
, GFP_KERNEL
);
347 memset(fl
->opt
, 0, sizeof(*fl
->opt
));
348 fl
->opt
->tot_len
= sizeof(*fl
->opt
) + olen
;
350 if (copy_from_user(fl
->opt
+1, optval
+CMSG_ALIGN(sizeof(*freq
)), olen
))
353 msg
.msg_controllen
= olen
;
354 msg
.msg_control
= (void*)(fl
->opt
+1);
357 err
= datagram_send_ctl(net
, &msg
, &flowi
, fl
->opt
, &junk
, &junk
);
361 if (fl
->opt
->opt_flen
)
363 if (fl
->opt
->opt_nflen
== 0) {
369 fl
->fl_net
= hold_net(net
);
370 fl
->expires
= jiffies
;
371 err
= fl6_renew(fl
, freq
->flr_linger
, freq
->flr_expires
);
374 fl
->share
= freq
->flr_share
;
375 addr_type
= ipv6_addr_type(&freq
->flr_dst
);
376 if ((addr_type
&IPV6_ADDR_MAPPED
)
377 || addr_type
== IPV6_ADDR_ANY
) {
381 ipv6_addr_copy(&fl
->dst
, &freq
->flr_dst
);
382 atomic_set(&fl
->users
, 1);
387 case IPV6_FL_S_PROCESS
:
388 fl
->owner
= current
->pid
;
391 fl
->owner
= current
->euid
;
405 static int mem_check(struct sock
*sk
)
407 struct ipv6_pinfo
*np
= inet6_sk(sk
);
408 struct ipv6_fl_socklist
*sfl
;
409 int room
= FL_MAX_SIZE
- atomic_read(&fl_size
);
412 if (room
> FL_MAX_SIZE
- FL_MAX_PER_SOCK
)
415 for (sfl
= np
->ipv6_fl_list
; sfl
; sfl
= sfl
->next
)
419 ((count
>= FL_MAX_PER_SOCK
||
420 (count
> 0 && room
< FL_MAX_SIZE
/2) || room
< FL_MAX_SIZE
/4)
421 && !capable(CAP_NET_ADMIN
)))
427 static int ipv6_hdr_cmp(struct ipv6_opt_hdr
*h1
, struct ipv6_opt_hdr
*h2
)
431 if (h1
== NULL
|| h2
== NULL
)
433 if (h1
->hdrlen
!= h2
->hdrlen
)
435 return memcmp(h1
+1, h2
+1, ((h1
->hdrlen
+1)<<3) - sizeof(*h1
));
438 static int ipv6_opt_cmp(struct ipv6_txoptions
*o1
, struct ipv6_txoptions
*o2
)
442 if (o1
== NULL
|| o2
== NULL
)
444 if (o1
->opt_nflen
!= o2
->opt_nflen
)
446 if (ipv6_hdr_cmp(o1
->hopopt
, o2
->hopopt
))
448 if (ipv6_hdr_cmp(o1
->dst0opt
, o2
->dst0opt
))
450 if (ipv6_hdr_cmp((struct ipv6_opt_hdr
*)o1
->srcrt
, (struct ipv6_opt_hdr
*)o2
->srcrt
))
455 static inline void fl_link(struct ipv6_pinfo
*np
, struct ipv6_fl_socklist
*sfl
,
456 struct ip6_flowlabel
*fl
)
458 write_lock_bh(&ip6_sk_fl_lock
);
460 sfl
->next
= np
->ipv6_fl_list
;
461 np
->ipv6_fl_list
= sfl
;
462 write_unlock_bh(&ip6_sk_fl_lock
);
465 int ipv6_flowlabel_opt(struct sock
*sk
, char __user
*optval
, int optlen
)
468 struct net
*net
= sock_net(sk
);
469 struct ipv6_pinfo
*np
= inet6_sk(sk
);
470 struct in6_flowlabel_req freq
;
471 struct ipv6_fl_socklist
*sfl1
=NULL
;
472 struct ipv6_fl_socklist
*sfl
, **sflp
;
473 struct ip6_flowlabel
*fl
, *fl1
= NULL
;
476 if (optlen
< sizeof(freq
))
479 if (copy_from_user(&freq
, optval
, sizeof(freq
)))
482 switch (freq
.flr_action
) {
484 write_lock_bh(&ip6_sk_fl_lock
);
485 for (sflp
= &np
->ipv6_fl_list
; (sfl
=*sflp
)!=NULL
; sflp
= &sfl
->next
) {
486 if (sfl
->fl
->label
== freq
.flr_label
) {
487 if (freq
.flr_label
== (np
->flow_label
&IPV6_FLOWLABEL_MASK
))
488 np
->flow_label
&= ~IPV6_FLOWLABEL_MASK
;
490 write_unlock_bh(&ip6_sk_fl_lock
);
496 write_unlock_bh(&ip6_sk_fl_lock
);
499 case IPV6_FL_A_RENEW
:
500 read_lock_bh(&ip6_sk_fl_lock
);
501 for (sfl
= np
->ipv6_fl_list
; sfl
; sfl
= sfl
->next
) {
502 if (sfl
->fl
->label
== freq
.flr_label
) {
503 err
= fl6_renew(sfl
->fl
, freq
.flr_linger
, freq
.flr_expires
);
504 read_unlock_bh(&ip6_sk_fl_lock
);
508 read_unlock_bh(&ip6_sk_fl_lock
);
510 if (freq
.flr_share
== IPV6_FL_S_NONE
&& capable(CAP_NET_ADMIN
)) {
511 fl
= fl_lookup(net
, freq
.flr_label
);
513 err
= fl6_renew(fl
, freq
.flr_linger
, freq
.flr_expires
);
521 if (freq
.flr_label
& ~IPV6_FLOWLABEL_MASK
)
524 fl
= fl_create(net
, &freq
, optval
, optlen
, &err
);
527 sfl1
= kmalloc(sizeof(*sfl1
), GFP_KERNEL
);
529 if (freq
.flr_label
) {
531 read_lock_bh(&ip6_sk_fl_lock
);
532 for (sfl
= np
->ipv6_fl_list
; sfl
; sfl
= sfl
->next
) {
533 if (sfl
->fl
->label
== freq
.flr_label
) {
534 if (freq
.flr_flags
&IPV6_FL_F_EXCL
) {
535 read_unlock_bh(&ip6_sk_fl_lock
);
539 atomic_inc(&fl1
->users
);
543 read_unlock_bh(&ip6_sk_fl_lock
);
546 fl1
= fl_lookup(net
, freq
.flr_label
);
550 if (freq
.flr_flags
&IPV6_FL_F_EXCL
)
553 if (fl1
->share
== IPV6_FL_S_EXCL
||
554 fl1
->share
!= fl
->share
||
555 fl1
->owner
!= fl
->owner
)
559 if (!ipv6_addr_equal(&fl1
->dst
, &fl
->dst
) ||
560 ipv6_opt_cmp(fl1
->opt
, fl
->opt
))
566 if (fl
->linger
> fl1
->linger
)
567 fl1
->linger
= fl
->linger
;
568 if ((long)(fl
->expires
- fl1
->expires
) > 0)
569 fl1
->expires
= fl
->expires
;
570 fl_link(np
, sfl1
, fl1
);
580 if (!(freq
.flr_flags
&IPV6_FL_F_CREATE
))
584 if (sfl1
== NULL
|| (err
= mem_check(sk
)) != 0)
587 fl1
= fl_intern(net
, fl
, freq
.flr_label
);
591 if (!freq
.flr_label
) {
592 if (copy_to_user(&((struct in6_flowlabel_req __user
*) optval
)->flr_label
,
593 &fl
->label
, sizeof(fl
->label
))) {
594 /* Intentionally ignore fault. */
598 fl_link(np
, sfl1
, fl
);
611 #ifdef CONFIG_PROC_FS
613 struct ip6fl_iter_state
{
614 struct seq_net_private p
;
618 #define ip6fl_seq_private(seq) ((struct ip6fl_iter_state *)(seq)->private)
620 static struct ip6_flowlabel
*ip6fl_get_first(struct seq_file
*seq
)
622 struct ip6_flowlabel
*fl
= NULL
;
623 struct ip6fl_iter_state
*state
= ip6fl_seq_private(seq
);
624 struct net
*net
= seq_file_net(seq
);
626 for (state
->bucket
= 0; state
->bucket
<= FL_HASH_MASK
; ++state
->bucket
) {
627 fl
= fl_ht
[state
->bucket
];
629 while (fl
&& fl
->fl_net
!= net
)
637 static struct ip6_flowlabel
*ip6fl_get_next(struct seq_file
*seq
, struct ip6_flowlabel
*fl
)
639 struct ip6fl_iter_state
*state
= ip6fl_seq_private(seq
);
640 struct net
*net
= seq_file_net(seq
);
644 while (fl
&& fl
->fl_net
!= net
)
648 if (++state
->bucket
<= FL_HASH_MASK
) {
649 fl
= fl_ht
[state
->bucket
];
657 static struct ip6_flowlabel
*ip6fl_get_idx(struct seq_file
*seq
, loff_t pos
)
659 struct ip6_flowlabel
*fl
= ip6fl_get_first(seq
);
661 while (pos
&& (fl
= ip6fl_get_next(seq
, fl
)) != NULL
)
663 return pos
? NULL
: fl
;
666 static void *ip6fl_seq_start(struct seq_file
*seq
, loff_t
*pos
)
667 __acquires(ip6_fl_lock
)
669 read_lock_bh(&ip6_fl_lock
);
670 return *pos
? ip6fl_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
673 static void *ip6fl_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
675 struct ip6_flowlabel
*fl
;
677 if (v
== SEQ_START_TOKEN
)
678 fl
= ip6fl_get_first(seq
);
680 fl
= ip6fl_get_next(seq
, v
);
685 static void ip6fl_seq_stop(struct seq_file
*seq
, void *v
)
686 __releases(ip6_fl_lock
)
688 read_unlock_bh(&ip6_fl_lock
);
691 static int ip6fl_seq_show(struct seq_file
*seq
, void *v
)
693 if (v
== SEQ_START_TOKEN
)
694 seq_printf(seq
, "%-5s %-1s %-6s %-6s %-6s %-8s %-32s %s\n",
695 "Label", "S", "Owner", "Users", "Linger", "Expires", "Dst", "Opt");
697 struct ip6_flowlabel
*fl
= v
;
699 "%05X %-1d %-6d %-6d %-6ld %-8ld " NIP6_SEQFMT
" %-4d\n",
700 (unsigned)ntohl(fl
->label
),
703 atomic_read(&fl
->users
),
705 (long)(fl
->expires
- jiffies
)/HZ
,
707 fl
->opt
? fl
->opt
->opt_nflen
: 0);
712 static const struct seq_operations ip6fl_seq_ops
= {
713 .start
= ip6fl_seq_start
,
714 .next
= ip6fl_seq_next
,
715 .stop
= ip6fl_seq_stop
,
716 .show
= ip6fl_seq_show
,
719 static int ip6fl_seq_open(struct inode
*inode
, struct file
*file
)
721 return seq_open_net(inode
, file
, &ip6fl_seq_ops
,
722 sizeof(struct ip6fl_iter_state
));
725 static const struct file_operations ip6fl_seq_fops
= {
726 .owner
= THIS_MODULE
,
727 .open
= ip6fl_seq_open
,
730 .release
= seq_release_net
,
733 static int ip6_flowlabel_proc_init(struct net
*net
)
735 if (!proc_net_fops_create(net
, "ip6_flowlabel",
736 S_IRUGO
, &ip6fl_seq_fops
))
741 static void ip6_flowlabel_proc_fini(struct net
*net
)
743 proc_net_remove(net
, "ip6_flowlabel");
746 static inline int ip6_flowlabel_proc_init(struct net
*net
)
750 static inline void ip6_flowlabel_proc_fini(struct net
*net
)
756 static inline void ip6_flowlabel_net_exit(struct net
*net
)
759 ip6_flowlabel_proc_fini(net
);
762 static struct pernet_operations ip6_flowlabel_net_ops
= {
763 .init
= ip6_flowlabel_proc_init
,
764 .exit
= ip6_flowlabel_net_exit
,
767 int ip6_flowlabel_init(void)
769 return register_pernet_subsys(&ip6_flowlabel_net_ops
);
772 void ip6_flowlabel_cleanup(void)
774 del_timer(&ip6_fl_gc_timer
);
775 unregister_pernet_subsys(&ip6_flowlabel_net_ops
);