2 * net/sched/em_meta.c Metadata ematch
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: Thomas Graf <tgraf@suug.ch>
11 * ==========================================================================
13 * The metadata ematch compares two meta objects where each object
14 * represents either a meta value stored in the kernel or a static
15 * value provided by userspace. The objects are not provided by
16 * userspace itself but rather a definition providing the information
17 * to build them. Every object is of a certain type which must be
18 * equal to the object it is being compared to.
20 * The definition of a objects conists of the type (meta type), a
21 * identifier (meta id) and additional type specific information.
22 * The meta id is either TCF_META_TYPE_VALUE for values provided by
23 * userspace or a index to the meta operations table consisting of
24 * function pointers to type specific meta data collectors returning
25 * the value of the requested meta value.
28 * +-----------+ +-----------+
29 * | type: INT | | type: INT |
30 * def | id: DEV | | id: VALUE |
31 * | data: | | data: 3 |
32 * +-----------+ +-----------+
34 * ---> meta_ops[INT][DEV](...) |
38 * +-----------+ +-----------+
39 * | type: INT | | type: INT |
40 * obj | id: DEV | | id: VALUE |
41 * | data: 2 |<--data got filled out | data: 3 |
42 * +-----------+ +-----------+
44 * --------------> 2 equals 3 <--------------
46 * This is a simplified schema, the complexity varies depending
47 * on the meta type. Obviously, the length of the data must also
48 * be provided for non-numeric types.
50 * Additionally, type dependent modifiers such as shift operators
51 * or mask may be applied to extend the functionaliy. As of now,
52 * the variable length type supports shifting the byte string to
53 * the right, eating up any number of octets and thus supporting
54 * wildcard interface name comparisons such as "ppp%" matching
57 * NOTE: Certain meta values depend on other subsystems and are
58 * only available if that subsystem is enabled in the kernel.
61 #include <linux/slab.h>
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/kernel.h>
65 #include <linux/sched.h>
66 #include <linux/string.h>
67 #include <linux/skbuff.h>
68 #include <linux/random.h>
69 #include <linux/if_vlan.h>
70 #include <linux/tc_ematch/tc_em_meta.h>
72 #include <net/route.h>
73 #include <net/pkt_cls.h>
82 struct tcf_meta_val hdr;
88 struct meta_value lvalue;
89 struct meta_value rvalue;
92 static inline int meta_id(struct meta_value *v)
94 return TCF_META_ID(v->hdr.kind);
97 static inline int meta_type(struct meta_value *v)
99 return TCF_META_TYPE(v->hdr.kind);
102 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
103 struct tcf_pkt_info *info, struct meta_value *v, \
104 struct meta_obj *dst, int *err)
106 /**************************************************************************
107 * System status & misc
108 **************************************************************************/
110 META_COLLECTOR(int_random)
112 get_random_bytes(&dst->value, sizeof(dst->value));
115 static inline unsigned long fixed_loadavg(int load)
117 int rnd_load = load + (FIXED_1/200);
118 int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
120 return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
123 META_COLLECTOR(int_loadavg_0)
125 dst->value = fixed_loadavg(avenrun[0]);
128 META_COLLECTOR(int_loadavg_1)
130 dst->value = fixed_loadavg(avenrun[1]);
133 META_COLLECTOR(int_loadavg_2)
135 dst->value = fixed_loadavg(avenrun[2]);
138 /**************************************************************************
139 * Device names & indices
140 **************************************************************************/
142 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
144 if (unlikely(dev == NULL))
147 dst->value = dev->ifindex;
151 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
153 if (unlikely(dev == NULL))
156 dst->value = (unsigned long) dev->name;
157 dst->len = strlen(dev->name);
161 META_COLLECTOR(int_dev)
163 *err = int_dev(skb->dev, dst);
166 META_COLLECTOR(var_dev)
168 *err = var_dev(skb->dev, dst);
171 /**************************************************************************
173 **************************************************************************/
175 META_COLLECTOR(int_vlan_tag)
179 if (vlan_tx_tag_present(skb))
180 dst->value = vlan_tx_tag_get(skb);
181 else if (!__vlan_get_tag(skb, &tag))
189 /**************************************************************************
191 **************************************************************************/
193 META_COLLECTOR(int_priority)
195 dst->value = skb->priority;
198 META_COLLECTOR(int_protocol)
200 /* Let userspace take care of the byte ordering */
201 dst->value = skb->protocol;
204 META_COLLECTOR(int_pkttype)
206 dst->value = skb->pkt_type;
209 META_COLLECTOR(int_pktlen)
211 dst->value = skb->len;
214 META_COLLECTOR(int_datalen)
216 dst->value = skb->data_len;
219 META_COLLECTOR(int_maclen)
221 dst->value = skb->mac_len;
224 META_COLLECTOR(int_rxhash)
226 dst->value = skb_get_rxhash(skb);
229 /**************************************************************************
231 **************************************************************************/
233 META_COLLECTOR(int_mark)
235 dst->value = skb->mark;
238 /**************************************************************************
240 **************************************************************************/
242 META_COLLECTOR(int_tcindex)
244 dst->value = skb->tc_index;
247 /**************************************************************************
249 **************************************************************************/
251 META_COLLECTOR(int_rtclassid)
253 if (unlikely(skb_dst(skb) == NULL))
256 #ifdef CONFIG_IP_ROUTE_CLASSID
257 dst->value = skb_dst(skb)->tclassid;
263 META_COLLECTOR(int_rtiif)
265 if (unlikely(skb_rtable(skb) == NULL))
268 dst->value = skb_rtable(skb)->rt_iif;
271 /**************************************************************************
273 **************************************************************************/
275 #define SKIP_NONLOCAL(skb) \
276 if (unlikely(skb->sk == NULL)) { \
281 META_COLLECTOR(int_sk_family)
284 dst->value = skb->sk->sk_family;
287 META_COLLECTOR(int_sk_state)
290 dst->value = skb->sk->sk_state;
293 META_COLLECTOR(int_sk_reuse)
296 dst->value = skb->sk->sk_reuse;
299 META_COLLECTOR(int_sk_bound_if)
302 /* No error if bound_dev_if is 0, legal userspace check */
303 dst->value = skb->sk->sk_bound_dev_if;
306 META_COLLECTOR(var_sk_bound_if)
310 if (skb->sk->sk_bound_dev_if == 0) {
311 dst->value = (unsigned long) "any";
314 struct net_device *dev;
317 dev = dev_get_by_index_rcu(sock_net(skb->sk),
318 skb->sk->sk_bound_dev_if);
319 *err = var_dev(dev, dst);
324 META_COLLECTOR(int_sk_refcnt)
327 dst->value = atomic_read(&skb->sk->sk_refcnt);
330 META_COLLECTOR(int_sk_rcvbuf)
333 dst->value = skb->sk->sk_rcvbuf;
336 META_COLLECTOR(int_sk_shutdown)
339 dst->value = skb->sk->sk_shutdown;
342 META_COLLECTOR(int_sk_proto)
345 dst->value = skb->sk->sk_protocol;
348 META_COLLECTOR(int_sk_type)
351 dst->value = skb->sk->sk_type;
354 META_COLLECTOR(int_sk_rmem_alloc)
357 dst->value = sk_rmem_alloc_get(skb->sk);
360 META_COLLECTOR(int_sk_wmem_alloc)
363 dst->value = sk_wmem_alloc_get(skb->sk);
366 META_COLLECTOR(int_sk_omem_alloc)
369 dst->value = atomic_read(&skb->sk->sk_omem_alloc);
372 META_COLLECTOR(int_sk_rcv_qlen)
375 dst->value = skb->sk->sk_receive_queue.qlen;
378 META_COLLECTOR(int_sk_snd_qlen)
381 dst->value = skb->sk->sk_write_queue.qlen;
384 META_COLLECTOR(int_sk_wmem_queued)
387 dst->value = skb->sk->sk_wmem_queued;
390 META_COLLECTOR(int_sk_fwd_alloc)
393 dst->value = skb->sk->sk_forward_alloc;
396 META_COLLECTOR(int_sk_sndbuf)
399 dst->value = skb->sk->sk_sndbuf;
402 META_COLLECTOR(int_sk_alloc)
405 dst->value = (__force int) skb->sk->sk_allocation;
408 META_COLLECTOR(int_sk_hash)
411 dst->value = skb->sk->sk_hash;
414 META_COLLECTOR(int_sk_lingertime)
417 dst->value = skb->sk->sk_lingertime / HZ;
420 META_COLLECTOR(int_sk_err_qlen)
423 dst->value = skb->sk->sk_error_queue.qlen;
426 META_COLLECTOR(int_sk_ack_bl)
429 dst->value = skb->sk->sk_ack_backlog;
432 META_COLLECTOR(int_sk_max_ack_bl)
435 dst->value = skb->sk->sk_max_ack_backlog;
438 META_COLLECTOR(int_sk_prio)
441 dst->value = skb->sk->sk_priority;
444 META_COLLECTOR(int_sk_rcvlowat)
447 dst->value = skb->sk->sk_rcvlowat;
450 META_COLLECTOR(int_sk_rcvtimeo)
453 dst->value = skb->sk->sk_rcvtimeo / HZ;
456 META_COLLECTOR(int_sk_sndtimeo)
459 dst->value = skb->sk->sk_sndtimeo / HZ;
462 META_COLLECTOR(int_sk_sendmsg_off)
465 dst->value = skb->sk->sk_sndmsg_off;
468 META_COLLECTOR(int_sk_write_pend)
471 dst->value = skb->sk->sk_write_pending;
474 /**************************************************************************
475 * Meta value collectors assignment table
476 **************************************************************************/
479 void (*get)(struct sk_buff *, struct tcf_pkt_info *,
480 struct meta_value *, struct meta_obj *, int *);
483 #define META_ID(name) TCF_META_ID_##name
484 #define META_FUNC(name) { .get = meta_##name }
486 /* Meta value operations table listing all meta value collectors and
487 * assigns them to a type and meta id. */
488 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX + 1][TCF_META_ID_MAX + 1] = {
489 [TCF_META_TYPE_VAR] = {
490 [META_ID(DEV)] = META_FUNC(var_dev),
491 [META_ID(SK_BOUND_IF)] = META_FUNC(var_sk_bound_if),
493 [TCF_META_TYPE_INT] = {
494 [META_ID(RANDOM)] = META_FUNC(int_random),
495 [META_ID(LOADAVG_0)] = META_FUNC(int_loadavg_0),
496 [META_ID(LOADAVG_1)] = META_FUNC(int_loadavg_1),
497 [META_ID(LOADAVG_2)] = META_FUNC(int_loadavg_2),
498 [META_ID(DEV)] = META_FUNC(int_dev),
499 [META_ID(PRIORITY)] = META_FUNC(int_priority),
500 [META_ID(PROTOCOL)] = META_FUNC(int_protocol),
501 [META_ID(PKTTYPE)] = META_FUNC(int_pkttype),
502 [META_ID(PKTLEN)] = META_FUNC(int_pktlen),
503 [META_ID(DATALEN)] = META_FUNC(int_datalen),
504 [META_ID(MACLEN)] = META_FUNC(int_maclen),
505 [META_ID(NFMARK)] = META_FUNC(int_mark),
506 [META_ID(TCINDEX)] = META_FUNC(int_tcindex),
507 [META_ID(RTCLASSID)] = META_FUNC(int_rtclassid),
508 [META_ID(RTIIF)] = META_FUNC(int_rtiif),
509 [META_ID(SK_FAMILY)] = META_FUNC(int_sk_family),
510 [META_ID(SK_STATE)] = META_FUNC(int_sk_state),
511 [META_ID(SK_REUSE)] = META_FUNC(int_sk_reuse),
512 [META_ID(SK_BOUND_IF)] = META_FUNC(int_sk_bound_if),
513 [META_ID(SK_REFCNT)] = META_FUNC(int_sk_refcnt),
514 [META_ID(SK_RCVBUF)] = META_FUNC(int_sk_rcvbuf),
515 [META_ID(SK_SNDBUF)] = META_FUNC(int_sk_sndbuf),
516 [META_ID(SK_SHUTDOWN)] = META_FUNC(int_sk_shutdown),
517 [META_ID(SK_PROTO)] = META_FUNC(int_sk_proto),
518 [META_ID(SK_TYPE)] = META_FUNC(int_sk_type),
519 [META_ID(SK_RMEM_ALLOC)] = META_FUNC(int_sk_rmem_alloc),
520 [META_ID(SK_WMEM_ALLOC)] = META_FUNC(int_sk_wmem_alloc),
521 [META_ID(SK_OMEM_ALLOC)] = META_FUNC(int_sk_omem_alloc),
522 [META_ID(SK_WMEM_QUEUED)] = META_FUNC(int_sk_wmem_queued),
523 [META_ID(SK_RCV_QLEN)] = META_FUNC(int_sk_rcv_qlen),
524 [META_ID(SK_SND_QLEN)] = META_FUNC(int_sk_snd_qlen),
525 [META_ID(SK_ERR_QLEN)] = META_FUNC(int_sk_err_qlen),
526 [META_ID(SK_FORWARD_ALLOCS)] = META_FUNC(int_sk_fwd_alloc),
527 [META_ID(SK_ALLOCS)] = META_FUNC(int_sk_alloc),
528 [META_ID(SK_HASH)] = META_FUNC(int_sk_hash),
529 [META_ID(SK_LINGERTIME)] = META_FUNC(int_sk_lingertime),
530 [META_ID(SK_ACK_BACKLOG)] = META_FUNC(int_sk_ack_bl),
531 [META_ID(SK_MAX_ACK_BACKLOG)] = META_FUNC(int_sk_max_ack_bl),
532 [META_ID(SK_PRIO)] = META_FUNC(int_sk_prio),
533 [META_ID(SK_RCVLOWAT)] = META_FUNC(int_sk_rcvlowat),
534 [META_ID(SK_RCVTIMEO)] = META_FUNC(int_sk_rcvtimeo),
535 [META_ID(SK_SNDTIMEO)] = META_FUNC(int_sk_sndtimeo),
536 [META_ID(SK_SENDMSG_OFF)] = META_FUNC(int_sk_sendmsg_off),
537 [META_ID(SK_WRITE_PENDING)] = META_FUNC(int_sk_write_pend),
538 [META_ID(VLAN_TAG)] = META_FUNC(int_vlan_tag),
539 [META_ID(RXHASH)] = META_FUNC(int_rxhash),
543 static inline struct meta_ops *meta_ops(struct meta_value *val)
545 return &__meta_ops[meta_type(val)][meta_id(val)];
548 /**************************************************************************
549 * Type specific operations for TCF_META_TYPE_VAR
550 **************************************************************************/
552 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
554 int r = a->len - b->len;
557 r = memcmp((void *) a->value, (void *) b->value, a->len);
562 static int meta_var_change(struct meta_value *dst, struct nlattr *nla)
564 int len = nla_len(nla);
566 dst->val = (unsigned long)kmemdup(nla_data(nla), len, GFP_KERNEL);
573 static void meta_var_destroy(struct meta_value *v)
575 kfree((void *) v->val);
578 static void meta_var_apply_extras(struct meta_value *v,
579 struct meta_obj *dst)
581 int shift = v->hdr.shift;
583 if (shift && shift < dst->len)
587 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
589 if (v->val && v->len)
590 NLA_PUT(skb, tlv, v->len, (void *) v->val);
597 /**************************************************************************
598 * Type specific operations for TCF_META_TYPE_INT
599 **************************************************************************/
601 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
603 /* Let gcc optimize it, the unlikely is not really based on
604 * some numbers but jump free code for mismatches seems
606 if (unlikely(a->value == b->value))
608 else if (a->value < b->value)
614 static int meta_int_change(struct meta_value *dst, struct nlattr *nla)
616 if (nla_len(nla) >= sizeof(unsigned long)) {
617 dst->val = *(unsigned long *) nla_data(nla);
618 dst->len = sizeof(unsigned long);
619 } else if (nla_len(nla) == sizeof(u32)) {
620 dst->val = nla_get_u32(nla);
621 dst->len = sizeof(u32);
628 static void meta_int_apply_extras(struct meta_value *v,
629 struct meta_obj *dst)
632 dst->value >>= v->hdr.shift;
635 dst->value &= v->val;
638 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
640 if (v->len == sizeof(unsigned long))
641 NLA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
642 else if (v->len == sizeof(u32))
643 NLA_PUT_U32(skb, tlv, v->val);
651 /**************************************************************************
652 * Type specific operations table
653 **************************************************************************/
655 struct meta_type_ops {
656 void (*destroy)(struct meta_value *);
657 int (*compare)(struct meta_obj *, struct meta_obj *);
658 int (*change)(struct meta_value *, struct nlattr *);
659 void (*apply_extras)(struct meta_value *, struct meta_obj *);
660 int (*dump)(struct sk_buff *, struct meta_value *, int);
663 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX + 1] = {
664 [TCF_META_TYPE_VAR] = {
665 .destroy = meta_var_destroy,
666 .compare = meta_var_compare,
667 .change = meta_var_change,
668 .apply_extras = meta_var_apply_extras,
669 .dump = meta_var_dump
671 [TCF_META_TYPE_INT] = {
672 .compare = meta_int_compare,
673 .change = meta_int_change,
674 .apply_extras = meta_int_apply_extras,
675 .dump = meta_int_dump
679 static inline struct meta_type_ops *meta_type_ops(struct meta_value *v)
681 return &__meta_type_ops[meta_type(v)];
684 /**************************************************************************
686 **************************************************************************/
688 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
689 struct meta_value *v, struct meta_obj *dst)
693 if (meta_id(v) == TCF_META_ID_VALUE) {
699 meta_ops(v)->get(skb, info, v, dst, &err);
703 if (meta_type_ops(v)->apply_extras)
704 meta_type_ops(v)->apply_extras(v, dst);
709 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
710 struct tcf_pkt_info *info)
713 struct meta_match *meta = (struct meta_match *) m->data;
714 struct meta_obj l_value, r_value;
716 if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
717 meta_get(skb, info, &meta->rvalue, &r_value) < 0)
720 r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
722 switch (meta->lvalue.hdr.op) {
734 static void meta_delete(struct meta_match *meta)
737 struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
739 if (ops && ops->destroy) {
740 ops->destroy(&meta->lvalue);
741 ops->destroy(&meta->rvalue);
748 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
751 if (nla_len(nla) == 0)
754 return meta_type_ops(dst)->change(dst, nla);
760 static inline int meta_is_supported(struct meta_value *val)
762 return !meta_id(val) || meta_ops(val)->get;
765 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
766 [TCA_EM_META_HDR] = { .len = sizeof(struct tcf_meta_hdr) },
769 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
770 struct tcf_ematch *m)
773 struct nlattr *tb[TCA_EM_META_MAX + 1];
774 struct tcf_meta_hdr *hdr;
775 struct meta_match *meta = NULL;
777 err = nla_parse(tb, TCA_EM_META_MAX, data, len, meta_policy);
782 if (tb[TCA_EM_META_HDR] == NULL)
784 hdr = nla_data(tb[TCA_EM_META_HDR]);
786 if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
787 TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
788 TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
789 TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
792 meta = kzalloc(sizeof(*meta), GFP_KERNEL);
796 memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
797 memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
799 if (!meta_is_supported(&meta->lvalue) ||
800 !meta_is_supported(&meta->rvalue)) {
805 if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE]) < 0 ||
806 meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE]) < 0)
809 m->datalen = sizeof(*meta);
810 m->data = (unsigned long) meta;
819 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
822 meta_delete((struct meta_match *) m->data);
825 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
827 struct meta_match *meta = (struct meta_match *) em->data;
828 struct tcf_meta_hdr hdr;
829 struct meta_type_ops *ops;
831 memset(&hdr, 0, sizeof(hdr));
832 memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
833 memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
835 NLA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
837 ops = meta_type_ops(&meta->lvalue);
838 if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
839 ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
840 goto nla_put_failure;
848 static struct tcf_ematch_ops em_meta_ops = {
850 .change = em_meta_change,
851 .match = em_meta_match,
852 .destroy = em_meta_destroy,
853 .dump = em_meta_dump,
854 .owner = THIS_MODULE,
855 .link = LIST_HEAD_INIT(em_meta_ops.link)
858 static int __init init_em_meta(void)
860 return tcf_em_register(&em_meta_ops);
863 static void __exit exit_em_meta(void)
865 tcf_em_unregister(&em_meta_ops);
868 MODULE_LICENSE("GPL");
870 module_init(init_em_meta);
871 module_exit(exit_em_meta);
873 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);