2 * net/sched/cls_api.c Packet classifier API.
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>
13 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/skbuff.h>
23 #include <linux/init.h>
24 #include <linux/kmod.h>
25 #include <linux/netlink.h>
26 #include <linux/err.h>
27 #include <net/net_namespace.h>
29 #include <net/netlink.h>
30 #include <net/pkt_sched.h>
31 #include <net/pkt_cls.h>
33 /* The list of all installed classifier types */
35 static struct tcf_proto_ops
*tcf_proto_base __read_mostly
;
37 /* Protects list of registered TC modules. It is pure SMP lock. */
38 static DEFINE_RWLOCK(cls_mod_lock
);
40 /* Find classifier type by string name */
42 static struct tcf_proto_ops
*tcf_proto_lookup_ops(struct nlattr
*kind
)
44 struct tcf_proto_ops
*t
= NULL
;
47 read_lock(&cls_mod_lock
);
48 for (t
= tcf_proto_base
; t
; t
= t
->next
) {
49 if (nla_strcmp(kind
, t
->kind
) == 0) {
50 if (!try_module_get(t
->owner
))
55 read_unlock(&cls_mod_lock
);
60 /* Register(unregister) new classifier type */
62 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
64 struct tcf_proto_ops
*t
, **tp
;
67 write_lock(&cls_mod_lock
);
68 for (tp
= &tcf_proto_base
; (t
= *tp
) != NULL
; tp
= &t
->next
)
69 if (!strcmp(ops
->kind
, t
->kind
))
76 write_unlock(&cls_mod_lock
);
79 EXPORT_SYMBOL(register_tcf_proto_ops
);
81 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
83 struct tcf_proto_ops
*t
, **tp
;
86 write_lock(&cls_mod_lock
);
87 for (tp
= &tcf_proto_base
; (t
=*tp
) != NULL
; tp
= &t
->next
)
96 write_unlock(&cls_mod_lock
);
99 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
101 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
102 struct tcf_proto
*tp
, unsigned long fh
, int event
);
105 /* Select new prio value from the range, managed by kernel. */
107 static inline u32
tcf_auto_prio(struct tcf_proto
*tp
)
109 u32 first
= TC_H_MAKE(0xC0000000U
, 0U);
117 /* Add/change/delete/get a filter node */
119 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
, void *arg
)
121 struct net
*net
= sock_net(skb
->sk
);
122 struct nlattr
*tca
[TCA_MAX
+ 1];
123 spinlock_t
*root_lock
;
129 struct net_device
*dev
;
131 struct tcf_proto
**back
, **chain
;
132 struct tcf_proto
*tp
;
133 struct tcf_proto_ops
*tp_ops
;
134 const struct Qdisc_class_ops
*cops
;
139 if (net
!= &init_net
)
144 protocol
= TC_H_MIN(t
->tcm_info
);
145 prio
= TC_H_MAJ(t
->tcm_info
);
147 parent
= t
->tcm_parent
;
151 /* If no priority is given, user wants we allocated it. */
152 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
154 prio
= TC_H_MAKE(0x80000000U
, 0U);
157 /* Find head of filter chain. */
160 dev
= __dev_get_by_index(&init_net
, t
->tcm_ifindex
);
164 err
= nlmsg_parse(n
, sizeof(*t
), tca
, TCA_MAX
, NULL
);
170 struct netdev_queue
*dev_queue
= netdev_get_tx_queue(dev
, 0);
171 q
= dev_queue
->qdisc_sleeping
;
174 q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
));
179 /* Is it classful? */
180 if ((cops
= q
->ops
->cl_ops
) == NULL
)
183 /* Do we search for filter, attached to class? */
184 if (TC_H_MIN(parent
)) {
185 cl
= cops
->get(q
, parent
);
190 /* And the last stroke */
191 chain
= cops
->tcf_chain(q
, cl
);
196 /* Check the chain for existence of proto-tcf with this priority */
197 for (back
= chain
; (tp
=*back
) != NULL
; back
= &tp
->next
) {
198 if (tp
->prio
>= prio
) {
199 if (tp
->prio
== prio
) {
200 if (!nprio
|| (tp
->protocol
!= protocol
&& protocol
))
208 root_lock
= qdisc_root_sleeping_lock(q
);
211 /* Proto-tcf does not exist, create new one */
213 if (tca
[TCA_KIND
] == NULL
|| !protocol
)
217 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
221 /* Create new proto tcf */
224 tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
);
228 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
]);
229 if (tp_ops
== NULL
) {
230 #ifdef CONFIG_MODULES
231 struct nlattr
*kind
= tca
[TCA_KIND
];
235 nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
237 request_module("cls_%s", name
);
239 tp_ops
= tcf_proto_lookup_ops(kind
);
240 /* We dropped the RTNL semaphore in order to
241 * perform the module load. So, even if we
242 * succeeded in loading the module we have to
243 * replay the request. We indicate this using
246 if (tp_ops
!= NULL
) {
247 module_put(tp_ops
->owner
);
256 tp
->protocol
= protocol
;
257 tp
->prio
= nprio
? : TC_H_MAJ(tcf_auto_prio(*back
));
259 tp
->classify
= tp_ops
->classify
;
260 tp
->classid
= parent
;
262 err
= tp_ops
->init(tp
);
264 module_put(tp_ops
->owner
);
269 spin_lock_bh(root_lock
);
272 spin_unlock_bh(root_lock
);
274 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
))
277 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
280 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
281 spin_lock_bh(root_lock
);
283 spin_unlock_bh(root_lock
);
285 tfilter_notify(skb
, n
, tp
, fh
, RTM_DELTFILTER
);
292 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
293 !(n
->nlmsg_flags
& NLM_F_CREATE
))
296 switch (n
->nlmsg_type
) {
299 if (n
->nlmsg_flags
& NLM_F_EXCL
)
303 err
= tp
->ops
->delete(tp
, fh
);
305 tfilter_notify(skb
, n
, tp
, fh
, RTM_DELTFILTER
);
308 err
= tfilter_notify(skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
316 err
= tp
->ops
->change(tp
, cl
, t
->tcm_handle
, tca
, &fh
);
318 tfilter_notify(skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
324 /* Replay the request. */
329 static int tcf_fill_node(struct sk_buff
*skb
, struct tcf_proto
*tp
,
330 unsigned long fh
, u32 pid
, u32 seq
, u16 flags
, int event
)
333 struct nlmsghdr
*nlh
;
334 unsigned char *b
= skb_tail_pointer(skb
);
336 nlh
= NLMSG_NEW(skb
, pid
, seq
, event
, sizeof(*tcm
), flags
);
337 tcm
= NLMSG_DATA(nlh
);
338 tcm
->tcm_family
= AF_UNSPEC
;
341 tcm
->tcm_ifindex
= qdisc_dev(tp
->q
)->ifindex
;
342 tcm
->tcm_parent
= tp
->classid
;
343 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
344 NLA_PUT_STRING(skb
, TCA_KIND
, tp
->ops
->kind
);
345 tcm
->tcm_handle
= fh
;
346 if (RTM_DELTFILTER
!= event
) {
348 if (tp
->ops
->dump
&& tp
->ops
->dump(tp
, fh
, skb
, tcm
) < 0)
349 goto nla_put_failure
;
351 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
360 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
361 struct tcf_proto
*tp
, unsigned long fh
, int event
)
364 u32 pid
= oskb
? NETLINK_CB(oskb
).pid
: 0;
366 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
370 if (tcf_fill_node(skb
, tp
, fh
, pid
, n
->nlmsg_seq
, 0, event
) <= 0) {
375 return rtnetlink_send(skb
, &init_net
, pid
, RTNLGRP_TC
,
376 n
->nlmsg_flags
& NLM_F_ECHO
);
379 struct tcf_dump_args
{
382 struct netlink_callback
*cb
;
385 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
,
386 struct tcf_walker
*arg
)
388 struct tcf_dump_args
*a
= (void *)arg
;
390 return tcf_fill_node(a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).pid
,
391 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
394 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
396 struct net
*net
= sock_net(skb
->sk
);
397 struct netdev_queue
*dev_queue
;
400 struct net_device
*dev
;
402 struct tcf_proto
*tp
, **chain
;
403 struct tcmsg
*tcm
= (struct tcmsg
*)NLMSG_DATA(cb
->nlh
);
404 unsigned long cl
= 0;
405 const struct Qdisc_class_ops
*cops
;
406 struct tcf_dump_args arg
;
408 if (net
!= &init_net
)
411 if (cb
->nlh
->nlmsg_len
< NLMSG_LENGTH(sizeof(*tcm
)))
413 if ((dev
= dev_get_by_index(&init_net
, tcm
->tcm_ifindex
)) == NULL
)
416 dev_queue
= netdev_get_tx_queue(dev
, 0);
417 if (!tcm
->tcm_parent
)
418 q
= dev_queue
->qdisc_sleeping
;
420 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
423 if ((cops
= q
->ops
->cl_ops
) == NULL
)
425 if (TC_H_MIN(tcm
->tcm_parent
)) {
426 cl
= cops
->get(q
, tcm
->tcm_parent
);
430 chain
= cops
->tcf_chain(q
, cl
);
436 for (tp
=*chain
, t
=0; tp
; tp
= tp
->next
, t
++) {
437 if (t
< s_t
) continue;
438 if (TC_H_MAJ(tcm
->tcm_info
) &&
439 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
441 if (TC_H_MIN(tcm
->tcm_info
) &&
442 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
445 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
446 if (cb
->args
[1] == 0) {
447 if (tcf_fill_node(skb
, tp
, 0, NETLINK_CB(cb
->skb
).pid
,
448 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
449 RTM_NEWTFILTER
) <= 0)
454 if (tp
->ops
->walk
== NULL
)
456 arg
.w
.fn
= tcf_node_dump
;
460 arg
.w
.skip
= cb
->args
[1]-1;
462 tp
->ops
->walk(tp
, &arg
.w
);
463 cb
->args
[1] = arg
.w
.count
+1;
478 void tcf_exts_destroy(struct tcf_proto
*tp
, struct tcf_exts
*exts
)
480 #ifdef CONFIG_NET_CLS_ACT
482 tcf_action_destroy(exts
->action
, TCA_ACT_UNBIND
);
487 EXPORT_SYMBOL(tcf_exts_destroy
);
489 int tcf_exts_validate(struct tcf_proto
*tp
, struct nlattr
**tb
,
490 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
,
491 const struct tcf_ext_map
*map
)
493 memset(exts
, 0, sizeof(*exts
));
495 #ifdef CONFIG_NET_CLS_ACT
497 struct tc_action
*act
;
499 if (map
->police
&& tb
[map
->police
]) {
500 act
= tcf_action_init_1(tb
[map
->police
], rate_tlv
,
501 "police", TCA_ACT_NOREPLACE
,
506 act
->type
= TCA_OLD_COMPAT
;
508 } else if (map
->action
&& tb
[map
->action
]) {
509 act
= tcf_action_init(tb
[map
->action
], rate_tlv
, NULL
,
510 TCA_ACT_NOREPLACE
, TCA_ACT_BIND
);
518 if ((map
->action
&& tb
[map
->action
]) ||
519 (map
->police
&& tb
[map
->police
]))
525 EXPORT_SYMBOL(tcf_exts_validate
);
527 void tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
528 struct tcf_exts
*src
)
530 #ifdef CONFIG_NET_CLS_ACT
532 struct tc_action
*act
;
535 dst
->action
= src
->action
;
538 tcf_action_destroy(act
, TCA_ACT_UNBIND
);
542 EXPORT_SYMBOL(tcf_exts_change
);
544 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
,
545 const struct tcf_ext_map
*map
)
547 #ifdef CONFIG_NET_CLS_ACT
548 if (map
->action
&& exts
->action
) {
550 * again for backward compatible mode - we want
551 * to work with both old and new modes of entering
552 * tc data even if iproute2 was newer - jhs
556 if (exts
->action
->type
!= TCA_OLD_COMPAT
) {
557 nest
= nla_nest_start(skb
, map
->action
);
559 goto nla_put_failure
;
560 if (tcf_action_dump(skb
, exts
->action
, 0, 0) < 0)
561 goto nla_put_failure
;
562 nla_nest_end(skb
, nest
);
563 } else if (map
->police
) {
564 nest
= nla_nest_start(skb
, map
->police
);
566 goto nla_put_failure
;
567 if (tcf_action_dump_old(skb
, exts
->action
, 0, 0) < 0)
568 goto nla_put_failure
;
569 nla_nest_end(skb
, nest
);
574 nla_put_failure
: __attribute__ ((unused
))
577 EXPORT_SYMBOL(tcf_exts_dump
);
580 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
,
581 const struct tcf_ext_map
*map
)
583 #ifdef CONFIG_NET_CLS_ACT
585 if (tcf_action_copy_stats(skb
, exts
->action
, 1) < 0)
586 goto nla_put_failure
;
589 nla_put_failure
: __attribute__ ((unused
))
592 EXPORT_SYMBOL(tcf_exts_dump_stats
);
594 static int __init
tc_filter_init(void)
596 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
);
597 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
);
598 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
604 subsys_initcall(tc_filter_init
);