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/err.h>
26 #include <linux/slab.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 */
34 static LIST_HEAD(tcf_proto_base
);
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock
);
39 /* Find classifier type by string name */
41 static const struct tcf_proto_ops
*tcf_proto_lookup_ops(struct nlattr
*kind
)
43 const struct tcf_proto_ops
*t
, *res
= NULL
;
46 read_lock(&cls_mod_lock
);
47 list_for_each_entry(t
, &tcf_proto_base
, head
) {
48 if (nla_strcmp(kind
, t
->kind
) == 0) {
49 if (try_module_get(t
->owner
))
54 read_unlock(&cls_mod_lock
);
59 /* Register(unregister) new classifier type */
61 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
63 struct tcf_proto_ops
*t
;
66 write_lock(&cls_mod_lock
);
67 list_for_each_entry(t
, &tcf_proto_base
, head
)
68 if (!strcmp(ops
->kind
, t
->kind
))
71 list_add_tail(&ops
->head
, &tcf_proto_base
);
74 write_unlock(&cls_mod_lock
);
77 EXPORT_SYMBOL(register_tcf_proto_ops
);
79 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
81 struct tcf_proto_ops
*t
;
84 /* Wait for outstanding call_rcu()s, if any, from a
85 * tcf_proto_ops's destroy() handler.
89 write_lock(&cls_mod_lock
);
90 list_for_each_entry(t
, &tcf_proto_base
, head
) {
97 write_unlock(&cls_mod_lock
);
100 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
102 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
103 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
104 unsigned long fh
, int event
);
106 static void tfilter_notify_chain(struct net
*net
, struct sk_buff
*oskb
,
108 struct tcf_proto __rcu
**chain
, int event
)
110 struct tcf_proto __rcu
**it_chain
;
111 struct tcf_proto
*tp
;
113 for (it_chain
= chain
; (tp
= rtnl_dereference(*it_chain
)) != NULL
;
114 it_chain
= &tp
->next
)
115 tfilter_notify(net
, oskb
, n
, tp
, 0, event
);
118 /* Select new prio value from the range, managed by kernel. */
120 static inline u32
tcf_auto_prio(struct tcf_proto
*tp
)
122 u32 first
= TC_H_MAKE(0xC0000000U
, 0U);
125 first
= tp
->prio
- 1;
130 /* Add/change/delete/get a filter node */
132 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
)
134 struct net
*net
= sock_net(skb
->sk
);
135 struct nlattr
*tca
[TCA_MAX
+ 1];
141 struct net_device
*dev
;
143 struct tcf_proto __rcu
**back
;
144 struct tcf_proto __rcu
**chain
;
145 struct tcf_proto
*tp
;
146 const struct tcf_proto_ops
*tp_ops
;
147 const struct Qdisc_class_ops
*cops
;
153 if ((n
->nlmsg_type
!= RTM_GETTFILTER
) &&
154 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
158 err
= nlmsg_parse(n
, sizeof(*t
), tca
, TCA_MAX
, NULL
);
163 protocol
= TC_H_MIN(t
->tcm_info
);
164 prio
= TC_H_MAJ(t
->tcm_info
);
166 parent
= t
->tcm_parent
;
170 switch (n
->nlmsg_type
) {
172 if (protocol
|| t
->tcm_handle
|| tca
[TCA_KIND
])
176 /* If no priority is provided by the user,
179 if (n
->nlmsg_flags
& NLM_F_CREATE
) {
180 prio
= TC_H_MAKE(0x80000000U
, 0U);
189 /* Find head of filter chain. */
192 dev
= __dev_get_by_index(net
, t
->tcm_ifindex
);
201 q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
));
206 /* Is it classful? */
207 cops
= q
->ops
->cl_ops
;
211 if (cops
->tcf_chain
== NULL
)
214 /* Do we search for filter, attached to class? */
215 if (TC_H_MIN(parent
)) {
216 cl
= cops
->get(q
, parent
);
221 /* And the last stroke */
222 chain
= cops
->tcf_chain(q
, cl
);
226 if (n
->nlmsg_type
== RTM_DELTFILTER
&& prio
== 0) {
227 tfilter_notify_chain(net
, skb
, n
, chain
, RTM_DELTFILTER
);
228 tcf_destroy_chain(chain
);
233 /* Check the chain for existence of proto-tcf with this priority */
235 (tp
= rtnl_dereference(*back
)) != NULL
;
237 if (tp
->prio
>= prio
) {
238 if (tp
->prio
== prio
) {
240 (tp
->protocol
!= protocol
&& protocol
))
249 /* Proto-tcf does not exist, create new one */
251 if (tca
[TCA_KIND
] == NULL
|| !protocol
)
255 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
256 !(n
->nlmsg_flags
& NLM_F_CREATE
))
260 /* Create new proto tcf */
263 tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
);
267 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
]);
268 if (tp_ops
== NULL
) {
269 #ifdef CONFIG_MODULES
270 struct nlattr
*kind
= tca
[TCA_KIND
];
274 nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
276 request_module("cls_%s", name
);
278 tp_ops
= tcf_proto_lookup_ops(kind
);
279 /* We dropped the RTNL semaphore in order to
280 * perform the module load. So, even if we
281 * succeeded in loading the module we have to
282 * replay the request. We indicate this using
285 if (tp_ops
!= NULL
) {
286 module_put(tp_ops
->owner
);
295 tp
->protocol
= protocol
;
297 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back
)));
299 tp
->classify
= tp_ops
->classify
;
300 tp
->classid
= parent
;
302 err
= tp_ops
->init(tp
);
304 module_put(tp_ops
->owner
);
311 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
))
314 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
317 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
318 struct tcf_proto
*next
= rtnl_dereference(tp
->next
);
320 RCU_INIT_POINTER(*back
, next
);
322 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
323 tcf_destroy(tp
, true);
329 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
330 !(n
->nlmsg_flags
& NLM_F_CREATE
))
333 switch (n
->nlmsg_type
) {
336 if (n
->nlmsg_flags
& NLM_F_EXCL
) {
338 tcf_destroy(tp
, true);
343 err
= tp
->ops
->delete(tp
, fh
);
345 struct tcf_proto
*next
= rtnl_dereference(tp
->next
);
347 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
348 if (tcf_destroy(tp
, false))
349 RCU_INIT_POINTER(*back
, next
);
353 err
= tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
361 err
= tp
->ops
->change(net
, skb
, tp
, cl
, t
->tcm_handle
, tca
, &fh
,
362 n
->nlmsg_flags
& NLM_F_CREATE
? TCA_ACT_NOREPLACE
: TCA_ACT_REPLACE
);
365 RCU_INIT_POINTER(tp
->next
, rtnl_dereference(*back
));
366 rcu_assign_pointer(*back
, tp
);
368 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
371 tcf_destroy(tp
, true);
378 /* Replay the request. */
383 static int tcf_fill_node(struct net
*net
, struct sk_buff
*skb
,
384 struct tcf_proto
*tp
, unsigned long fh
, u32 portid
,
385 u32 seq
, u16 flags
, int event
)
388 struct nlmsghdr
*nlh
;
389 unsigned char *b
= skb_tail_pointer(skb
);
391 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
394 tcm
= nlmsg_data(nlh
);
395 tcm
->tcm_family
= AF_UNSPEC
;
398 tcm
->tcm_ifindex
= qdisc_dev(tp
->q
)->ifindex
;
399 tcm
->tcm_parent
= tp
->classid
;
400 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
401 if (nla_put_string(skb
, TCA_KIND
, tp
->ops
->kind
))
402 goto nla_put_failure
;
403 tcm
->tcm_handle
= fh
;
404 if (RTM_DELTFILTER
!= event
) {
406 if (tp
->ops
->dump
&& tp
->ops
->dump(net
, tp
, fh
, skb
, tcm
) < 0)
407 goto nla_put_failure
;
409 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
418 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
419 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
420 unsigned long fh
, int event
)
423 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
425 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
429 if (tcf_fill_node(net
, skb
, tp
, fh
, portid
, n
->nlmsg_seq
, 0, event
) <= 0) {
434 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
435 n
->nlmsg_flags
& NLM_F_ECHO
);
438 struct tcf_dump_args
{
441 struct netlink_callback
*cb
;
444 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
,
445 struct tcf_walker
*arg
)
447 struct tcf_dump_args
*a
= (void *)arg
;
448 struct net
*net
= sock_net(a
->skb
->sk
);
450 return tcf_fill_node(net
, a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).portid
,
451 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
454 /* called with RTNL */
455 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
457 struct net
*net
= sock_net(skb
->sk
);
460 struct net_device
*dev
;
462 struct tcf_proto
*tp
, __rcu
**chain
;
463 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
464 unsigned long cl
= 0;
465 const struct Qdisc_class_ops
*cops
;
466 struct tcf_dump_args arg
;
468 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
470 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
474 if (!tcm
->tcm_parent
)
477 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
480 cops
= q
->ops
->cl_ops
;
483 if (cops
->tcf_chain
== NULL
)
485 if (TC_H_MIN(tcm
->tcm_parent
)) {
486 cl
= cops
->get(q
, tcm
->tcm_parent
);
490 chain
= cops
->tcf_chain(q
, cl
);
496 for (tp
= rtnl_dereference(*chain
), t
= 0;
497 tp
; tp
= rtnl_dereference(tp
->next
), t
++) {
500 if (TC_H_MAJ(tcm
->tcm_info
) &&
501 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
503 if (TC_H_MIN(tcm
->tcm_info
) &&
504 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
507 memset(&cb
->args
[1], 0,
508 sizeof(cb
->args
)-sizeof(cb
->args
[0]));
509 if (cb
->args
[1] == 0) {
510 if (tcf_fill_node(net
, skb
, tp
, 0,
511 NETLINK_CB(cb
->skb
).portid
,
512 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
513 RTM_NEWTFILTER
) <= 0)
518 if (tp
->ops
->walk
== NULL
)
520 arg
.w
.fn
= tcf_node_dump
;
524 arg
.w
.skip
= cb
->args
[1] - 1;
526 tp
->ops
->walk(tp
, &arg
.w
);
527 cb
->args
[1] = arg
.w
.count
+ 1;
541 void tcf_exts_destroy(struct tcf_exts
*exts
)
543 #ifdef CONFIG_NET_CLS_ACT
546 tcf_exts_to_list(exts
, &actions
);
547 tcf_action_destroy(&actions
, TCA_ACT_UNBIND
);
548 kfree(exts
->actions
);
549 exts
->nr_actions
= 0;
552 EXPORT_SYMBOL(tcf_exts_destroy
);
554 int tcf_exts_validate(struct net
*net
, struct tcf_proto
*tp
, struct nlattr
**tb
,
555 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
, bool ovr
)
557 #ifdef CONFIG_NET_CLS_ACT
559 struct tc_action
*act
;
561 if (exts
->police
&& tb
[exts
->police
]) {
562 act
= tcf_action_init_1(net
, tb
[exts
->police
], rate_tlv
,
568 act
->type
= exts
->type
= TCA_OLD_COMPAT
;
569 exts
->actions
[0] = act
;
570 exts
->nr_actions
= 1;
571 } else if (exts
->action
&& tb
[exts
->action
]) {
575 err
= tcf_action_init(net
, tb
[exts
->action
], rate_tlv
,
577 TCA_ACT_BIND
, &actions
);
580 list_for_each_entry(act
, &actions
, list
)
581 exts
->actions
[i
++] = act
;
582 exts
->nr_actions
= i
;
586 if ((exts
->action
&& tb
[exts
->action
]) ||
587 (exts
->police
&& tb
[exts
->police
]))
593 EXPORT_SYMBOL(tcf_exts_validate
);
595 void tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
596 struct tcf_exts
*src
)
598 #ifdef CONFIG_NET_CLS_ACT
599 struct tcf_exts old
= *dst
;
602 dst
->nr_actions
= src
->nr_actions
;
603 dst
->actions
= src
->actions
;
604 dst
->type
= src
->type
;
607 tcf_exts_destroy(&old
);
610 EXPORT_SYMBOL(tcf_exts_change
);
612 #ifdef CONFIG_NET_CLS_ACT
613 static struct tc_action
*tcf_exts_first_act(struct tcf_exts
*exts
)
615 if (exts
->nr_actions
== 0)
618 return exts
->actions
[0];
622 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
)
624 #ifdef CONFIG_NET_CLS_ACT
627 if (exts
->action
&& exts
->nr_actions
) {
629 * again for backward compatible mode - we want
630 * to work with both old and new modes of entering
631 * tc data even if iproute2 was newer - jhs
633 if (exts
->type
!= TCA_OLD_COMPAT
) {
636 nest
= nla_nest_start(skb
, exts
->action
);
638 goto nla_put_failure
;
640 tcf_exts_to_list(exts
, &actions
);
641 if (tcf_action_dump(skb
, &actions
, 0, 0) < 0)
642 goto nla_put_failure
;
643 nla_nest_end(skb
, nest
);
644 } else if (exts
->police
) {
645 struct tc_action
*act
= tcf_exts_first_act(exts
);
646 nest
= nla_nest_start(skb
, exts
->police
);
647 if (nest
== NULL
|| !act
)
648 goto nla_put_failure
;
649 if (tcf_action_dump_old(skb
, act
, 0, 0) < 0)
650 goto nla_put_failure
;
651 nla_nest_end(skb
, nest
);
657 nla_nest_cancel(skb
, nest
);
663 EXPORT_SYMBOL(tcf_exts_dump
);
666 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
)
668 #ifdef CONFIG_NET_CLS_ACT
669 struct tc_action
*a
= tcf_exts_first_act(exts
);
670 if (a
!= NULL
&& tcf_action_copy_stats(skb
, a
, 1) < 0)
675 EXPORT_SYMBOL(tcf_exts_dump_stats
);
677 static int __init
tc_filter_init(void)
679 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
680 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
681 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
682 tc_dump_tfilter
, NULL
);
687 subsys_initcall(tc_filter_init
);