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 * Additionaly, type dependant 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/module.h>
62 #include <linux/types.h>
63 #include <linux/kernel.h>
64 #include <linux/sched.h>
65 #include <linux/string.h>
66 #include <linux/skbuff.h>
67 #include <linux/random.h>
68 #include <linux/if_vlan.h>
69 #include <linux/tc_ematch/tc_em_meta.h>
71 #include <net/route.h>
72 #include <net/pkt_cls.h>
83 struct tcf_meta_val hdr;
90 struct meta_value lvalue;
91 struct meta_value rvalue;
94 static inline int meta_id(struct meta_value *v)
96 return TCF_META_ID(v->hdr.kind);
99 static inline int meta_type(struct meta_value *v)
101 return TCF_META_TYPE(v->hdr.kind);
104 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
105 struct tcf_pkt_info *info, struct meta_value *v, \
106 struct meta_obj *dst, int *err)
108 /**************************************************************************
109 * System status & misc
110 **************************************************************************/
112 META_COLLECTOR(int_random)
114 get_random_bytes(&dst->value, sizeof(dst->value));
117 static inline unsigned long fixed_loadavg(int load)
119 int rnd_load = load + (FIXED_1/200);
120 int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
122 return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
125 META_COLLECTOR(int_loadavg_0)
127 dst->value = fixed_loadavg(avenrun[0]);
130 META_COLLECTOR(int_loadavg_1)
132 dst->value = fixed_loadavg(avenrun[1]);
135 META_COLLECTOR(int_loadavg_2)
137 dst->value = fixed_loadavg(avenrun[2]);
140 /**************************************************************************
141 * Device names & indices
142 **************************************************************************/
144 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
146 if (unlikely(dev == NULL))
149 dst->value = dev->ifindex;
153 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
155 if (unlikely(dev == NULL))
158 dst->value = (unsigned long) dev->name;
159 dst->len = strlen(dev->name);
163 META_COLLECTOR(int_dev)
165 *err = int_dev(skb->dev, dst);
168 META_COLLECTOR(var_dev)
170 *err = var_dev(skb->dev, dst);
173 /**************************************************************************
175 **************************************************************************/
177 META_COLLECTOR(int_vlan_tag)
179 unsigned short uninitialized_var(tag);
180 if (vlan_get_tag(skb, &tag) < 0)
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 /**************************************************************************
225 **************************************************************************/
227 META_COLLECTOR(int_mark)
229 dst->value = skb->mark;
232 /**************************************************************************
234 **************************************************************************/
236 META_COLLECTOR(int_tcindex)
238 dst->value = skb->tc_index;
241 /**************************************************************************
243 **************************************************************************/
245 META_COLLECTOR(int_rtclassid)
247 if (unlikely(skb->dst == NULL))
250 #ifdef CONFIG_NET_CLS_ROUTE
251 dst->value = skb->dst->tclassid;
257 META_COLLECTOR(int_rtiif)
259 if (unlikely(skb->rtable == NULL))
262 dst->value = skb->rtable->fl.iif;
265 /**************************************************************************
267 **************************************************************************/
269 #define SKIP_NONLOCAL(skb) \
270 if (unlikely(skb->sk == NULL)) { \
275 META_COLLECTOR(int_sk_family)
278 dst->value = skb->sk->sk_family;
281 META_COLLECTOR(int_sk_state)
284 dst->value = skb->sk->sk_state;
287 META_COLLECTOR(int_sk_reuse)
290 dst->value = skb->sk->sk_reuse;
293 META_COLLECTOR(int_sk_bound_if)
296 /* No error if bound_dev_if is 0, legal userspace check */
297 dst->value = skb->sk->sk_bound_dev_if;
300 META_COLLECTOR(var_sk_bound_if)
304 if (skb->sk->sk_bound_dev_if == 0) {
305 dst->value = (unsigned long) "any";
308 struct net_device *dev;
310 dev = dev_get_by_index(&init_net, skb->sk->sk_bound_dev_if);
311 *err = var_dev(dev, dst);
317 META_COLLECTOR(int_sk_refcnt)
320 dst->value = atomic_read(&skb->sk->sk_refcnt);
323 META_COLLECTOR(int_sk_rcvbuf)
326 dst->value = skb->sk->sk_rcvbuf;
329 META_COLLECTOR(int_sk_shutdown)
332 dst->value = skb->sk->sk_shutdown;
335 META_COLLECTOR(int_sk_proto)
338 dst->value = skb->sk->sk_protocol;
341 META_COLLECTOR(int_sk_type)
344 dst->value = skb->sk->sk_type;
347 META_COLLECTOR(int_sk_rmem_alloc)
350 dst->value = atomic_read(&skb->sk->sk_rmem_alloc);
353 META_COLLECTOR(int_sk_wmem_alloc)
356 dst->value = atomic_read(&skb->sk->sk_wmem_alloc);
359 META_COLLECTOR(int_sk_omem_alloc)
362 dst->value = atomic_read(&skb->sk->sk_omem_alloc);
365 META_COLLECTOR(int_sk_rcv_qlen)
368 dst->value = skb->sk->sk_receive_queue.qlen;
371 META_COLLECTOR(int_sk_snd_qlen)
374 dst->value = skb->sk->sk_write_queue.qlen;
377 META_COLLECTOR(int_sk_wmem_queued)
380 dst->value = skb->sk->sk_wmem_queued;
383 META_COLLECTOR(int_sk_fwd_alloc)
386 dst->value = skb->sk->sk_forward_alloc;
389 META_COLLECTOR(int_sk_sndbuf)
392 dst->value = skb->sk->sk_sndbuf;
395 META_COLLECTOR(int_sk_alloc)
398 dst->value = skb->sk->sk_allocation;
401 META_COLLECTOR(int_sk_route_caps)
404 dst->value = skb->sk->sk_route_caps;
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 **************************************************************************/
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_ROUTE_CAPS)] = META_FUNC(int_sk_route_caps),
529 [META_ID(SK_HASH)] = META_FUNC(int_sk_hash),
530 [META_ID(SK_LINGERTIME)] = META_FUNC(int_sk_lingertime),
531 [META_ID(SK_ACK_BACKLOG)] = META_FUNC(int_sk_ack_bl),
532 [META_ID(SK_MAX_ACK_BACKLOG)] = META_FUNC(int_sk_max_ack_bl),
533 [META_ID(SK_PRIO)] = META_FUNC(int_sk_prio),
534 [META_ID(SK_RCVLOWAT)] = META_FUNC(int_sk_rcvlowat),
535 [META_ID(SK_RCVTIMEO)] = META_FUNC(int_sk_rcvtimeo),
536 [META_ID(SK_SNDTIMEO)] = META_FUNC(int_sk_sndtimeo),
537 [META_ID(SK_SENDMSG_OFF)] = META_FUNC(int_sk_sendmsg_off),
538 [META_ID(SK_WRITE_PENDING)] = META_FUNC(int_sk_write_pend),
539 [META_ID(VLAN_TAG)] = META_FUNC(int_vlan_tag),
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);
652 /**************************************************************************
653 * Type specific operations table
654 **************************************************************************/
658 void (*destroy)(struct meta_value *);
659 int (*compare)(struct meta_obj *, struct meta_obj *);
660 int (*change)(struct meta_value *, struct nlattr *);
661 void (*apply_extras)(struct meta_value *, struct meta_obj *);
662 int (*dump)(struct sk_buff *, struct meta_value *, int);
665 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX+1] = {
666 [TCF_META_TYPE_VAR] = {
667 .destroy = meta_var_destroy,
668 .compare = meta_var_compare,
669 .change = meta_var_change,
670 .apply_extras = meta_var_apply_extras,
671 .dump = meta_var_dump
673 [TCF_META_TYPE_INT] = {
674 .compare = meta_int_compare,
675 .change = meta_int_change,
676 .apply_extras = meta_int_apply_extras,
677 .dump = meta_int_dump
681 static inline struct meta_type_ops * meta_type_ops(struct meta_value *v)
683 return &__meta_type_ops[meta_type(v)];
686 /**************************************************************************
688 **************************************************************************/
690 static int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
691 struct meta_value *v, struct meta_obj *dst)
695 if (meta_id(v) == TCF_META_ID_VALUE) {
701 meta_ops(v)->get(skb, info, v, dst, &err);
705 if (meta_type_ops(v)->apply_extras)
706 meta_type_ops(v)->apply_extras(v, dst);
711 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
712 struct tcf_pkt_info *info)
715 struct meta_match *meta = (struct meta_match *) m->data;
716 struct meta_obj l_value, r_value;
718 if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
719 meta_get(skb, info, &meta->rvalue, &r_value) < 0)
722 r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
724 switch (meta->lvalue.hdr.op) {
736 static void meta_delete(struct meta_match *meta)
739 struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
741 if (ops && ops->destroy) {
742 ops->destroy(&meta->lvalue);
743 ops->destroy(&meta->rvalue);
750 static inline int meta_change_data(struct meta_value *dst, struct nlattr *nla)
753 if (nla_len(nla) == 0)
756 return meta_type_ops(dst)->change(dst, nla);
762 static inline int meta_is_supported(struct meta_value *val)
764 return (!meta_id(val) || meta_ops(val)->get);
767 static const struct nla_policy meta_policy[TCA_EM_META_MAX + 1] = {
768 [TCA_EM_META_HDR] = { .len = sizeof(struct tcf_meta_hdr) },
771 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
772 struct tcf_ematch *m)
775 struct nlattr *tb[TCA_EM_META_MAX + 1];
776 struct tcf_meta_hdr *hdr;
777 struct meta_match *meta = NULL;
779 err = nla_parse(tb, TCA_EM_META_MAX, data, len, meta_policy);
784 if (tb[TCA_EM_META_HDR] == NULL)
786 hdr = nla_data(tb[TCA_EM_META_HDR]);
788 if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
789 TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
790 TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
791 TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
794 meta = kzalloc(sizeof(*meta), GFP_KERNEL);
798 memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
799 memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
801 if (!meta_is_supported(&meta->lvalue) ||
802 !meta_is_supported(&meta->rvalue)) {
807 if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE]) < 0 ||
808 meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE]) < 0)
811 m->datalen = sizeof(*meta);
812 m->data = (unsigned long) meta;
821 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
824 meta_delete((struct meta_match *) m->data);
827 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
829 struct meta_match *meta = (struct meta_match *) em->data;
830 struct tcf_meta_hdr hdr;
831 struct meta_type_ops *ops;
833 memset(&hdr, 0, sizeof(hdr));
834 memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
835 memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
837 NLA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
839 ops = meta_type_ops(&meta->lvalue);
840 if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
841 ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
842 goto nla_put_failure;
850 static struct tcf_ematch_ops em_meta_ops = {
852 .change = em_meta_change,
853 .match = em_meta_match,
854 .destroy = em_meta_destroy,
855 .dump = em_meta_dump,
856 .owner = THIS_MODULE,
857 .link = LIST_HEAD_INIT(em_meta_ops.link)
860 static int __init init_em_meta(void)
862 return tcf_em_register(&em_meta_ops);
865 static void __exit exit_em_meta(void)
867 tcf_em_unregister(&em_meta_ops);
870 MODULE_LICENSE("GPL");
872 module_init(init_em_meta);
873 module_exit(exit_em_meta);
875 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);