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 tag = vlan_tx_tag_get(skb);
180 if (!tag && __vlan_get_tag(skb, &tag))
188 /**************************************************************************
190 **************************************************************************/
192 META_COLLECTOR(int_priority)
194 dst->value = skb->priority;
197 META_COLLECTOR(int_protocol)
199 /* Let userspace take care of the byte ordering */
200 dst->value = skb->protocol;
203 META_COLLECTOR(int_pkttype)
205 dst->value = skb->pkt_type;
208 META_COLLECTOR(int_pktlen)
210 dst->value = skb->len;
213 META_COLLECTOR(int_datalen)
215 dst->value = skb->data_len;
218 META_COLLECTOR(int_maclen)
220 dst->value = skb->mac_len;
223 META_COLLECTOR(int_rxhash)
225 dst->value = skb_get_rxhash(skb);
228 /**************************************************************************
230 **************************************************************************/
232 META_COLLECTOR(int_mark)
234 dst->value = skb->mark;
237 /**************************************************************************
239 **************************************************************************/
241 META_COLLECTOR(int_tcindex)
243 dst->value = skb->tc_index;
246 /**************************************************************************
248 **************************************************************************/
250 META_COLLECTOR(int_rtclassid)
252 if (unlikely(skb_dst(skb) == NULL))
255 #ifdef CONFIG_IP_ROUTE_CLASSID
256 dst->value = skb_dst(skb)->tclassid;
262 META_COLLECTOR(int_rtiif)
264 if (unlikely(skb_rtable(skb) == NULL))
267 dst->value = skb_rtable(skb)->rt_iif;
270 /**************************************************************************
272 **************************************************************************/
274 #define SKIP_NONLOCAL(skb) \
275 if (unlikely(skb->sk == NULL)) { \
280 META_COLLECTOR(int_sk_family)
283 dst->value = skb->sk->sk_family;
286 META_COLLECTOR(int_sk_state)
289 dst->value = skb->sk->sk_state;
292 META_COLLECTOR(int_sk_reuse)
295 dst->value = skb->sk->sk_reuse;
298 META_COLLECTOR(int_sk_bound_if)
301 /* No error if bound_dev_if is 0, legal userspace check */
302 dst->value = skb->sk->sk_bound_dev_if;
305 META_COLLECTOR(var_sk_bound_if)
309 if (skb->sk->sk_bound_dev_if == 0) {
310 dst->value = (unsigned long) "any";
313 struct net_device *dev;
316 dev = dev_get_by_index_rcu(sock_net(skb->sk),
317 skb->sk->sk_bound_dev_if);
318 *err = var_dev(dev, dst);
323 META_COLLECTOR(int_sk_refcnt)
326 dst->value = atomic_read(&skb->sk->sk_refcnt);
329 META_COLLECTOR(int_sk_rcvbuf)
332 dst->value = skb->sk->sk_rcvbuf;
335 META_COLLECTOR(int_sk_shutdown)
338 dst->value = skb->sk->sk_shutdown;
341 META_COLLECTOR(int_sk_proto)
344 dst->value = skb->sk->sk_protocol;
347 META_COLLECTOR(int_sk_type)
350 dst->value = skb->sk->sk_type;
353 META_COLLECTOR(int_sk_rmem_alloc)
356 dst->value = sk_rmem_alloc_get(skb->sk);
359 META_COLLECTOR(int_sk_wmem_alloc)
362 dst->value = sk_wmem_alloc_get(skb->sk);
365 META_COLLECTOR(int_sk_omem_alloc)
368 dst->value = atomic_read(&skb->sk->sk_omem_alloc);
371 META_COLLECTOR(int_sk_rcv_qlen)
374 dst->value = skb->sk->sk_receive_queue.qlen;
377 META_COLLECTOR(int_sk_snd_qlen)
380 dst->value = skb->sk->sk_write_queue.qlen;
383 META_COLLECTOR(int_sk_wmem_queued)
386 dst->value = skb->sk->sk_wmem_queued;
389 META_COLLECTOR(int_sk_fwd_alloc)
392 dst->value = skb->sk->sk_forward_alloc;
395 META_COLLECTOR(int_sk_sndbuf)
398 dst->value = skb->sk->sk_sndbuf;
401 META_COLLECTOR(int_sk_alloc)
404 dst->value = (__force int) skb->sk->sk_allocation;
407 META_COLLECTOR(int_sk_hash)
410 dst->value = skb->sk->sk_hash;
413 META_COLLECTOR(int_sk_lingertime)
416 dst->value = skb->sk->sk_lingertime / HZ;
419 META_COLLECTOR(int_sk_err_qlen)
422 dst->value = skb->sk->sk_error_queue.qlen;
425 META_COLLECTOR(int_sk_ack_bl)
428 dst->value = skb->sk->sk_ack_backlog;
431 META_COLLECTOR(int_sk_max_ack_bl)
434 dst->value = skb->sk->sk_max_ack_backlog;
437 META_COLLECTOR(int_sk_prio)
440 dst->value = skb->sk->sk_priority;
443 META_COLLECTOR(int_sk_rcvlowat)
446 dst->value = skb->sk->sk_rcvlowat;
449 META_COLLECTOR(int_sk_rcvtimeo)
452 dst->value = skb->sk->sk_rcvtimeo / HZ;
455 META_COLLECTOR(int_sk_sndtimeo)
458 dst->value = skb->sk->sk_sndtimeo / HZ;
461 META_COLLECTOR(int_sk_sendmsg_off)
464 dst->value = skb->sk->sk_sndmsg_off;
467 META_COLLECTOR(int_sk_write_pend)
470 dst->value = skb->sk->sk_write_pending;
473 /**************************************************************************
474 * Meta value collectors assignment table
475 **************************************************************************/
478 void (*get)(struct sk_buff *, struct tcf_pkt_info *,
479 struct meta_value *, struct meta_obj *, int *);
482 #define META_ID(name) TCF_META_ID_##name
483 #define META_FUNC(name) { .get = meta_##name }
485 /* Meta value operations table listing all meta value collectors and
486 * assigns them to a type and meta id. */
487 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX + 1][TCF_META_ID_MAX + 1] = {
488 [TCF_META_TYPE_VAR] = {
489 [META_ID(DEV)] = META_FUNC(var_dev),
490 [META_ID(SK_BOUND_IF)] = META_FUNC(var_sk_bound_if),
492 [TCF_META_TYPE_INT] = {
493 [META_ID(RANDOM)] = META_FUNC(int_random),
494 [META_ID(LOADAVG_0)] = META_FUNC(int_loadavg_0),
495 [META_ID(LOADAVG_1)] = META_FUNC(int_loadavg_1),
496 [META_ID(LOADAVG_2)] = META_FUNC(int_loadavg_2),
497 [META_ID(DEV)] = META_FUNC(int_dev),
498 [META_ID(PRIORITY)] = META_FUNC(int_priority),
499 [META_ID(PROTOCOL)] = META_FUNC(int_protocol),
500 [META_ID(PKTTYPE)] = META_FUNC(int_pkttype),
501 [META_ID(PKTLEN)] = META_FUNC(int_pktlen),
502 [META_ID(DATALEN)] = META_FUNC(int_datalen),
503 [META_ID(MACLEN)] = META_FUNC(int_maclen),
504 [META_ID(NFMARK)] = META_FUNC(int_mark),
505 [META_ID(TCINDEX)] = META_FUNC(int_tcindex),
506 [META_ID(RTCLASSID)] = META_FUNC(int_rtclassid),
507 [META_ID(RTIIF)] = META_FUNC(int_rtiif),
508 [META_ID(SK_FAMILY)] = META_FUNC(int_sk_family),
509 [META_ID(SK_STATE)] = META_FUNC(int_sk_state),
510 [META_ID(SK_REUSE)] = META_FUNC(int_sk_reuse),
511 [META_ID(SK_BOUND_IF)] = META_FUNC(int_sk_bound_if),
512 [META_ID(SK_REFCNT)] = META_FUNC(int_sk_refcnt),
513 [META_ID(SK_RCVBUF)] = META_FUNC(int_sk_rcvbuf),
514 [META_ID(SK_SNDBUF)] = META_FUNC(int_sk_sndbuf),
515 [META_ID(SK_SHUTDOWN)] = META_FUNC(int_sk_shutdown),
516 [META_ID(SK_PROTO)] = META_FUNC(int_sk_proto),
517 [META_ID(SK_TYPE)] = META_FUNC(int_sk_type),
518 [META_ID(SK_RMEM_ALLOC)] = META_FUNC(int_sk_rmem_alloc),
519 [META_ID(SK_WMEM_ALLOC)] = META_FUNC(int_sk_wmem_alloc),
520 [META_ID(SK_OMEM_ALLOC)] = META_FUNC(int_sk_omem_alloc),
521 [META_ID(SK_WMEM_QUEUED)] = META_FUNC(int_sk_wmem_queued),
522 [META_ID(SK_RCV_QLEN)] = META_FUNC(int_sk_rcv_qlen),
523 [META_ID(SK_SND_QLEN)] = META_FUNC(int_sk_snd_qlen),
524 [META_ID(SK_ERR_QLEN)] = META_FUNC(int_sk_err_qlen),
525 [META_ID(SK_FORWARD_ALLOCS)] = META_FUNC(int_sk_fwd_alloc),
526 [META_ID(SK_ALLOCS)] = META_FUNC(int_sk_alloc),
527 [META_ID(SK_HASH)] = META_FUNC(int_sk_hash),
528 [META_ID(SK_LINGERTIME)] = META_FUNC(int_sk_lingertime),
529 [META_ID(SK_ACK_BACKLOG)] = META_FUNC(int_sk_ack_bl),
530 [META_ID(SK_MAX_ACK_BACKLOG)] = META_FUNC(int_sk_max_ack_bl),
531 [META_ID(SK_PRIO)] = META_FUNC(int_sk_prio),
532 [META_ID(SK_RCVLOWAT)] = META_FUNC(int_sk_rcvlowat),
533 [META_ID(SK_RCVTIMEO)] = META_FUNC(int_sk_rcvtimeo),
534 [META_ID(SK_SNDTIMEO)] = META_FUNC(int_sk_sndtimeo),
535 [META_ID(SK_SENDMSG_OFF)] = META_FUNC(int_sk_sendmsg_off),
536 [META_ID(SK_WRITE_PENDING)] = META_FUNC(int_sk_write_pend),
537 [META_ID(VLAN_TAG)] = META_FUNC(int_vlan_tag),
538 [META_ID(RXHASH)] = META_FUNC(int_rxhash),
542 static inline struct meta_ops *meta_ops(struct meta_value *val)
544 return &__meta_ops[meta_type(val)][meta_id(val)];
547 /**************************************************************************
548 * Type specific operations for TCF_META_TYPE_VAR
549 **************************************************************************/
551 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
553 int r = a->len - b->len;
556 r = memcmp((void *) a->value, (void *) b->value, a->len);
561 static int meta_var_change(struct meta_value *dst, struct nlattr *nla)
563 int len = nla_len(nla);
565 dst->val = (unsigned long)kmemdup(nla_data(nla), len, GFP_KERNEL);
572 static void meta_var_destroy(struct meta_value *v)
574 kfree((void *) v->val);
577 static void meta_var_apply_extras(struct meta_value *v,
578 struct meta_obj *dst)
580 int shift = v->hdr.shift;
582 if (shift && shift < dst->len)
586 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
588 if (v->val && v->len)
589 NLA_PUT(skb, tlv, v->len, (void *) v->val);
596 /**************************************************************************
597 * Type specific operations for TCF_META_TYPE_INT
598 **************************************************************************/
600 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
602 /* Let gcc optimize it, the unlikely is not really based on
603 * some numbers but jump free code for mismatches seems
605 if (unlikely(a->value == b->value))
607 else if (a->value < b->value)
613 static int meta_int_change(struct meta_value *dst, struct nlattr *nla)
615 if (nla_len(nla) >= sizeof(unsigned long)) {
616 dst->val = *(unsigned long *) nla_data(nla);
617 dst->len = sizeof(unsigned long);
618 } else if (nla_len(nla) == sizeof(u32)) {
619 dst->val = nla_get_u32(nla);
620 dst->len = sizeof(u32);
627 static void meta_int_apply_extras(struct meta_value *v,
628 struct meta_obj *dst)
631 dst->value >>= v->hdr.shift;
634 dst->value &= v->val;
637 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
639 if (v->len == sizeof(unsigned long))
640 NLA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
641 else if (v->len == sizeof(u32))
642 NLA_PUT_U32(skb, tlv, v->val);
650 /**************************************************************************
651 * Type specific operations table
652 **************************************************************************/
654 struct meta_type_ops {
655 void (*destroy)(struct meta_value *);
656 int (*compare)(struct meta_obj *, struct meta_obj *);
657 int (*change)(struct meta_value *, struct nlattr *);
658 void (*apply_extras)(struct meta_value *, struct meta_obj *);
659 int (*dump)(struct sk_buff *, struct meta_value *, int);
662 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX + 1] = {
663 [TCF_META_TYPE_VAR] = {
664 .destroy = meta_var_destroy,
665 .compare = meta_var_compare,
666 .change = meta_var_change,
667 .apply_extras = meta_var_apply_extras,
668 .dump = meta_var_dump
670 [TCF_META_TYPE_INT] = {
671 .compare = meta_int_compare,
672 .change = meta_int_change,
673 .apply_extras = meta_int_apply_extras,
674 .dump = meta_int_dump
678 static inline struct meta_type_ops *meta_type_ops(struct meta_value *v)
680 return &__meta_type_ops[meta_type(v)];
683 /**************************************************************************
685 **************************************************************************/
687 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
688 struct meta_value *v, struct meta_obj *dst)
692 if (meta_id(v) == TCF_META_ID_VALUE) {
698 meta_ops(v)->get(skb, info, v, dst, &err);
702 if (meta_type_ops(v)->apply_extras)
703 meta_type_ops(v)->apply_extras(v, dst);
708 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
709 struct tcf_pkt_info *info)
712 struct meta_match *meta = (struct meta_match *) m->data;
713 struct meta_obj l_value, r_value;
715 if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
716 meta_get(skb, info, &meta->rvalue, &r_value) < 0)
719 r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
721 switch (meta->lvalue.hdr.op) {
733 static void meta_delete(struct meta_match *meta)
736 struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
738 if (ops && ops->destroy) {
739 ops->destroy(&meta->lvalue);
740 ops->destroy(&meta->rvalue);
747 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
750 if (nla_len(nla) == 0)
753 return meta_type_ops(dst)->change(dst, nla);
759 static inline int meta_is_supported(struct meta_value *val)
761 return !meta_id(val) || meta_ops(val)->get;
764 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
765 [TCA_EM_META_HDR] = { .len = sizeof(struct tcf_meta_hdr) },
768 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
769 struct tcf_ematch *m)
772 struct nlattr *tb[TCA_EM_META_MAX + 1];
773 struct tcf_meta_hdr *hdr;
774 struct meta_match *meta = NULL;
776 err = nla_parse(tb, TCA_EM_META_MAX, data, len, meta_policy);
781 if (tb[TCA_EM_META_HDR] == NULL)
783 hdr = nla_data(tb[TCA_EM_META_HDR]);
785 if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
786 TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
787 TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
788 TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
791 meta = kzalloc(sizeof(*meta), GFP_KERNEL);
795 memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
796 memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
798 if (!meta_is_supported(&meta->lvalue) ||
799 !meta_is_supported(&meta->rvalue)) {
804 if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE]) < 0 ||
805 meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE]) < 0)
808 m->datalen = sizeof(*meta);
809 m->data = (unsigned long) meta;
818 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
821 meta_delete((struct meta_match *) m->data);
824 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
826 struct meta_match *meta = (struct meta_match *) em->data;
827 struct tcf_meta_hdr hdr;
828 struct meta_type_ops *ops;
830 memset(&hdr, 0, sizeof(hdr));
831 memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
832 memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
834 NLA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
836 ops = meta_type_ops(&meta->lvalue);
837 if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
838 ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
839 goto nla_put_failure;
847 static struct tcf_ematch_ops em_meta_ops = {
849 .change = em_meta_change,
850 .match = em_meta_match,
851 .destroy = em_meta_destroy,
852 .dump = em_meta_dump,
853 .owner = THIS_MODULE,
854 .link = LIST_HEAD_INIT(em_meta_ops.link)
857 static int __init init_em_meta(void)
859 return tcf_em_register(&em_meta_ops);
862 static void __exit exit_em_meta(void)
864 tcf_em_unregister(&em_meta_ops);
867 MODULE_LICENSE("GPL");
869 module_init(init_em_meta);
870 module_exit(exit_em_meta);
872 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);