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 subsytem 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/tc_ematch/tc_em_meta.h>
70 #include <net/route.h>
71 #include <net/pkt_cls.h>
82 struct tcf_meta_val hdr;
89 struct meta_value lvalue;
90 struct meta_value rvalue;
93 static inline int meta_id(struct meta_value *v)
95 return TCF_META_ID(v->hdr.kind);
98 static inline int meta_type(struct meta_value *v)
100 return TCF_META_TYPE(v->hdr.kind);
103 #define META_COLLECTOR(FUNC) static void meta_##FUNC(struct sk_buff *skb, \
104 struct tcf_pkt_info *info, struct meta_value *v, \
105 struct meta_obj *dst, int *err)
107 /**************************************************************************
108 * System status & misc
109 **************************************************************************/
111 META_COLLECTOR(int_random)
113 get_random_bytes(&dst->value, sizeof(dst->value));
116 static inline unsigned long fixed_loadavg(int load)
118 int rnd_load = load + (FIXED_1/200);
119 int rnd_frac = ((rnd_load & (FIXED_1-1)) * 100) >> FSHIFT;
121 return ((rnd_load >> FSHIFT) * 100) + rnd_frac;
124 META_COLLECTOR(int_loadavg_0)
126 dst->value = fixed_loadavg(avenrun[0]);
129 META_COLLECTOR(int_loadavg_1)
131 dst->value = fixed_loadavg(avenrun[1]);
134 META_COLLECTOR(int_loadavg_2)
136 dst->value = fixed_loadavg(avenrun[2]);
139 /**************************************************************************
140 * Device names & indices
141 **************************************************************************/
143 static inline int int_dev(struct net_device *dev, struct meta_obj *dst)
145 if (unlikely(dev == NULL))
148 dst->value = dev->ifindex;
152 static inline int var_dev(struct net_device *dev, struct meta_obj *dst)
154 if (unlikely(dev == NULL))
157 dst->value = (unsigned long) dev->name;
158 dst->len = strlen(dev->name);
162 META_COLLECTOR(int_dev)
164 *err = int_dev(skb->dev, dst);
167 META_COLLECTOR(var_dev)
169 *err = var_dev(skb->dev, dst);
172 /**************************************************************************
174 **************************************************************************/
176 META_COLLECTOR(int_priority)
178 dst->value = skb->priority;
181 META_COLLECTOR(int_protocol)
183 /* Let userspace take care of the byte ordering */
184 dst->value = skb->protocol;
187 META_COLLECTOR(int_pkttype)
189 dst->value = skb->pkt_type;
192 META_COLLECTOR(int_pktlen)
194 dst->value = skb->len;
197 META_COLLECTOR(int_datalen)
199 dst->value = skb->data_len;
202 META_COLLECTOR(int_maclen)
204 dst->value = skb->mac_len;
207 /**************************************************************************
209 **************************************************************************/
211 META_COLLECTOR(int_mark)
213 dst->value = skb->mark;
216 /**************************************************************************
218 **************************************************************************/
220 META_COLLECTOR(int_tcindex)
222 dst->value = skb->tc_index;
225 /**************************************************************************
227 **************************************************************************/
229 META_COLLECTOR(int_rtclassid)
231 if (unlikely(skb->dst == NULL))
234 #ifdef CONFIG_NET_CLS_ROUTE
235 dst->value = skb->dst->tclassid;
241 META_COLLECTOR(int_rtiif)
243 if (unlikely(skb->dst == NULL))
246 dst->value = ((struct rtable*) skb->dst)->fl.iif;
249 /**************************************************************************
251 **************************************************************************/
253 #define SKIP_NONLOCAL(skb) \
254 if (unlikely(skb->sk == NULL)) { \
259 META_COLLECTOR(int_sk_family)
262 dst->value = skb->sk->sk_family;
265 META_COLLECTOR(int_sk_state)
268 dst->value = skb->sk->sk_state;
271 META_COLLECTOR(int_sk_reuse)
274 dst->value = skb->sk->sk_reuse;
277 META_COLLECTOR(int_sk_bound_if)
280 /* No error if bound_dev_if is 0, legal userspace check */
281 dst->value = skb->sk->sk_bound_dev_if;
284 META_COLLECTOR(var_sk_bound_if)
288 if (skb->sk->sk_bound_dev_if == 0) {
289 dst->value = (unsigned long) "any";
292 struct net_device *dev;
294 dev = dev_get_by_index(skb->sk->sk_bound_dev_if);
295 *err = var_dev(dev, dst);
301 META_COLLECTOR(int_sk_refcnt)
304 dst->value = atomic_read(&skb->sk->sk_refcnt);
307 META_COLLECTOR(int_sk_rcvbuf)
310 dst->value = skb->sk->sk_rcvbuf;
313 META_COLLECTOR(int_sk_shutdown)
316 dst->value = skb->sk->sk_shutdown;
319 META_COLLECTOR(int_sk_proto)
322 dst->value = skb->sk->sk_protocol;
325 META_COLLECTOR(int_sk_type)
328 dst->value = skb->sk->sk_type;
331 META_COLLECTOR(int_sk_rmem_alloc)
334 dst->value = atomic_read(&skb->sk->sk_rmem_alloc);
337 META_COLLECTOR(int_sk_wmem_alloc)
340 dst->value = atomic_read(&skb->sk->sk_wmem_alloc);
343 META_COLLECTOR(int_sk_omem_alloc)
346 dst->value = atomic_read(&skb->sk->sk_omem_alloc);
349 META_COLLECTOR(int_sk_rcv_qlen)
352 dst->value = skb->sk->sk_receive_queue.qlen;
355 META_COLLECTOR(int_sk_snd_qlen)
358 dst->value = skb->sk->sk_write_queue.qlen;
361 META_COLLECTOR(int_sk_wmem_queued)
364 dst->value = skb->sk->sk_wmem_queued;
367 META_COLLECTOR(int_sk_fwd_alloc)
370 dst->value = skb->sk->sk_forward_alloc;
373 META_COLLECTOR(int_sk_sndbuf)
376 dst->value = skb->sk->sk_sndbuf;
379 META_COLLECTOR(int_sk_alloc)
382 dst->value = skb->sk->sk_allocation;
385 META_COLLECTOR(int_sk_route_caps)
388 dst->value = skb->sk->sk_route_caps;
391 META_COLLECTOR(int_sk_hash)
394 dst->value = skb->sk->sk_hash;
397 META_COLLECTOR(int_sk_lingertime)
400 dst->value = skb->sk->sk_lingertime / HZ;
403 META_COLLECTOR(int_sk_err_qlen)
406 dst->value = skb->sk->sk_error_queue.qlen;
409 META_COLLECTOR(int_sk_ack_bl)
412 dst->value = skb->sk->sk_ack_backlog;
415 META_COLLECTOR(int_sk_max_ack_bl)
418 dst->value = skb->sk->sk_max_ack_backlog;
421 META_COLLECTOR(int_sk_prio)
424 dst->value = skb->sk->sk_priority;
427 META_COLLECTOR(int_sk_rcvlowat)
430 dst->value = skb->sk->sk_rcvlowat;
433 META_COLLECTOR(int_sk_rcvtimeo)
436 dst->value = skb->sk->sk_rcvtimeo / HZ;
439 META_COLLECTOR(int_sk_sndtimeo)
442 dst->value = skb->sk->sk_sndtimeo / HZ;
445 META_COLLECTOR(int_sk_sendmsg_off)
448 dst->value = skb->sk->sk_sndmsg_off;
451 META_COLLECTOR(int_sk_write_pend)
454 dst->value = skb->sk->sk_write_pending;
457 /**************************************************************************
458 * Meta value collectors assignment table
459 **************************************************************************/
463 void (*get)(struct sk_buff *, struct tcf_pkt_info *,
464 struct meta_value *, struct meta_obj *, int *);
467 #define META_ID(name) TCF_META_ID_##name
468 #define META_FUNC(name) { .get = meta_##name }
470 /* Meta value operations table listing all meta value collectors and
471 * assigns them to a type and meta id. */
472 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX+1][TCF_META_ID_MAX+1] = {
473 [TCF_META_TYPE_VAR] = {
474 [META_ID(DEV)] = META_FUNC(var_dev),
475 [META_ID(SK_BOUND_IF)] = META_FUNC(var_sk_bound_if),
477 [TCF_META_TYPE_INT] = {
478 [META_ID(RANDOM)] = META_FUNC(int_random),
479 [META_ID(LOADAVG_0)] = META_FUNC(int_loadavg_0),
480 [META_ID(LOADAVG_1)] = META_FUNC(int_loadavg_1),
481 [META_ID(LOADAVG_2)] = META_FUNC(int_loadavg_2),
482 [META_ID(DEV)] = META_FUNC(int_dev),
483 [META_ID(PRIORITY)] = META_FUNC(int_priority),
484 [META_ID(PROTOCOL)] = META_FUNC(int_protocol),
485 [META_ID(PKTTYPE)] = META_FUNC(int_pkttype),
486 [META_ID(PKTLEN)] = META_FUNC(int_pktlen),
487 [META_ID(DATALEN)] = META_FUNC(int_datalen),
488 [META_ID(MACLEN)] = META_FUNC(int_maclen),
489 [META_ID(NFMARK)] = META_FUNC(int_mark),
490 [META_ID(TCINDEX)] = META_FUNC(int_tcindex),
491 [META_ID(RTCLASSID)] = META_FUNC(int_rtclassid),
492 [META_ID(RTIIF)] = META_FUNC(int_rtiif),
493 [META_ID(SK_FAMILY)] = META_FUNC(int_sk_family),
494 [META_ID(SK_STATE)] = META_FUNC(int_sk_state),
495 [META_ID(SK_REUSE)] = META_FUNC(int_sk_reuse),
496 [META_ID(SK_BOUND_IF)] = META_FUNC(int_sk_bound_if),
497 [META_ID(SK_REFCNT)] = META_FUNC(int_sk_refcnt),
498 [META_ID(SK_RCVBUF)] = META_FUNC(int_sk_rcvbuf),
499 [META_ID(SK_SNDBUF)] = META_FUNC(int_sk_sndbuf),
500 [META_ID(SK_SHUTDOWN)] = META_FUNC(int_sk_shutdown),
501 [META_ID(SK_PROTO)] = META_FUNC(int_sk_proto),
502 [META_ID(SK_TYPE)] = META_FUNC(int_sk_type),
503 [META_ID(SK_RMEM_ALLOC)] = META_FUNC(int_sk_rmem_alloc),
504 [META_ID(SK_WMEM_ALLOC)] = META_FUNC(int_sk_wmem_alloc),
505 [META_ID(SK_OMEM_ALLOC)] = META_FUNC(int_sk_omem_alloc),
506 [META_ID(SK_WMEM_QUEUED)] = META_FUNC(int_sk_wmem_queued),
507 [META_ID(SK_RCV_QLEN)] = META_FUNC(int_sk_rcv_qlen),
508 [META_ID(SK_SND_QLEN)] = META_FUNC(int_sk_snd_qlen),
509 [META_ID(SK_ERR_QLEN)] = META_FUNC(int_sk_err_qlen),
510 [META_ID(SK_FORWARD_ALLOCS)] = META_FUNC(int_sk_fwd_alloc),
511 [META_ID(SK_ALLOCS)] = META_FUNC(int_sk_alloc),
512 [META_ID(SK_ROUTE_CAPS)] = META_FUNC(int_sk_route_caps),
513 [META_ID(SK_HASH)] = META_FUNC(int_sk_hash),
514 [META_ID(SK_LINGERTIME)] = META_FUNC(int_sk_lingertime),
515 [META_ID(SK_ACK_BACKLOG)] = META_FUNC(int_sk_ack_bl),
516 [META_ID(SK_MAX_ACK_BACKLOG)] = META_FUNC(int_sk_max_ack_bl),
517 [META_ID(SK_PRIO)] = META_FUNC(int_sk_prio),
518 [META_ID(SK_RCVLOWAT)] = META_FUNC(int_sk_rcvlowat),
519 [META_ID(SK_RCVTIMEO)] = META_FUNC(int_sk_rcvtimeo),
520 [META_ID(SK_SNDTIMEO)] = META_FUNC(int_sk_sndtimeo),
521 [META_ID(SK_SENDMSG_OFF)] = META_FUNC(int_sk_sendmsg_off),
522 [META_ID(SK_WRITE_PENDING)] = META_FUNC(int_sk_write_pend),
526 static inline struct meta_ops * meta_ops(struct meta_value *val)
528 return &__meta_ops[meta_type(val)][meta_id(val)];
531 /**************************************************************************
532 * Type specific operations for TCF_META_TYPE_VAR
533 **************************************************************************/
535 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
537 int r = a->len - b->len;
540 r = memcmp((void *) a->value, (void *) b->value, a->len);
545 static int meta_var_change(struct meta_value *dst, struct rtattr *rta)
547 int len = RTA_PAYLOAD(rta);
549 dst->val = (unsigned long)kmemdup(RTA_DATA(rta), len, GFP_KERNEL);
556 static void meta_var_destroy(struct meta_value *v)
558 kfree((void *) v->val);
561 static void meta_var_apply_extras(struct meta_value *v,
562 struct meta_obj *dst)
564 int shift = v->hdr.shift;
566 if (shift && shift < dst->len)
570 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
572 if (v->val && v->len)
573 RTA_PUT(skb, tlv, v->len, (void *) v->val);
580 /**************************************************************************
581 * Type specific operations for TCF_META_TYPE_INT
582 **************************************************************************/
584 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
586 /* Let gcc optimize it, the unlikely is not really based on
587 * some numbers but jump free code for mismatches seems
589 if (unlikely(a->value == b->value))
591 else if (a->value < b->value)
597 static int meta_int_change(struct meta_value *dst, struct rtattr *rta)
599 if (RTA_PAYLOAD(rta) >= sizeof(unsigned long)) {
600 dst->val = *(unsigned long *) RTA_DATA(rta);
601 dst->len = sizeof(unsigned long);
602 } else if (RTA_PAYLOAD(rta) == sizeof(u32)) {
603 dst->val = *(u32 *) RTA_DATA(rta);
604 dst->len = sizeof(u32);
611 static void meta_int_apply_extras(struct meta_value *v,
612 struct meta_obj *dst)
615 dst->value >>= v->hdr.shift;
618 dst->value &= v->val;
621 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
623 if (v->len == sizeof(unsigned long))
624 RTA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
625 else if (v->len == sizeof(u32)) {
627 RTA_PUT(skb, tlv, sizeof(d), &d);
636 /**************************************************************************
637 * Type specific operations table
638 **************************************************************************/
642 void (*destroy)(struct meta_value *);
643 int (*compare)(struct meta_obj *, struct meta_obj *);
644 int (*change)(struct meta_value *, struct rtattr *);
645 void (*apply_extras)(struct meta_value *, struct meta_obj *);
646 int (*dump)(struct sk_buff *, struct meta_value *, int);
649 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX+1] = {
650 [TCF_META_TYPE_VAR] = {
651 .destroy = meta_var_destroy,
652 .compare = meta_var_compare,
653 .change = meta_var_change,
654 .apply_extras = meta_var_apply_extras,
655 .dump = meta_var_dump
657 [TCF_META_TYPE_INT] = {
658 .compare = meta_int_compare,
659 .change = meta_int_change,
660 .apply_extras = meta_int_apply_extras,
661 .dump = meta_int_dump
665 static inline struct meta_type_ops * meta_type_ops(struct meta_value *v)
667 return &__meta_type_ops[meta_type(v)];
670 /**************************************************************************
672 **************************************************************************/
674 static inline int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
675 struct meta_value *v, struct meta_obj *dst)
679 if (meta_id(v) == TCF_META_ID_VALUE) {
685 meta_ops(v)->get(skb, info, v, dst, &err);
689 if (meta_type_ops(v)->apply_extras)
690 meta_type_ops(v)->apply_extras(v, dst);
695 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
696 struct tcf_pkt_info *info)
699 struct meta_match *meta = (struct meta_match *) m->data;
700 struct meta_obj l_value, r_value;
702 if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
703 meta_get(skb, info, &meta->rvalue, &r_value) < 0)
706 r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
708 switch (meta->lvalue.hdr.op) {
720 static inline void meta_delete(struct meta_match *meta)
722 struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
724 if (ops && ops->destroy) {
725 ops->destroy(&meta->lvalue);
726 ops->destroy(&meta->rvalue);
732 static inline int meta_change_data(struct meta_value *dst, struct rtattr *rta)
735 if (RTA_PAYLOAD(rta) == 0)
738 return meta_type_ops(dst)->change(dst, rta);
744 static inline int meta_is_supported(struct meta_value *val)
746 return (!meta_id(val) || meta_ops(val)->get);
749 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
750 struct tcf_ematch *m)
753 struct rtattr *tb[TCA_EM_META_MAX];
754 struct tcf_meta_hdr *hdr;
755 struct meta_match *meta = NULL;
757 if (rtattr_parse(tb, TCA_EM_META_MAX, data, len) < 0)
760 if (tb[TCA_EM_META_HDR-1] == NULL ||
761 RTA_PAYLOAD(tb[TCA_EM_META_HDR-1]) < sizeof(*hdr))
763 hdr = RTA_DATA(tb[TCA_EM_META_HDR-1]);
765 if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
766 TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
767 TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
768 TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
771 meta = kzalloc(sizeof(*meta), GFP_KERNEL);
775 memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
776 memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
778 if (!meta_is_supported(&meta->lvalue) ||
779 !meta_is_supported(&meta->rvalue)) {
784 if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE-1]) < 0 ||
785 meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE-1]) < 0)
788 m->datalen = sizeof(*meta);
789 m->data = (unsigned long) meta;
798 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
801 meta_delete((struct meta_match *) m->data);
804 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
806 struct meta_match *meta = (struct meta_match *) em->data;
807 struct tcf_meta_hdr hdr;
808 struct meta_type_ops *ops;
810 memset(&hdr, 0, sizeof(hdr));
811 memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
812 memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
814 RTA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
816 ops = meta_type_ops(&meta->lvalue);
817 if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
818 ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
827 static struct tcf_ematch_ops em_meta_ops = {
829 .change = em_meta_change,
830 .match = em_meta_match,
831 .destroy = em_meta_destroy,
832 .dump = em_meta_dump,
833 .owner = THIS_MODULE,
834 .link = LIST_HEAD_INIT(em_meta_ops.link)
837 static int __init init_em_meta(void)
839 return tcf_em_register(&em_meta_ops);
842 static void __exit exit_em_meta(void)
844 tcf_em_unregister(&em_meta_ops);
847 MODULE_LICENSE("GPL");
849 module_init(init_em_meta);
850 module_exit(exit_em_meta);
852 MODULE_ALIAS_TCF_EMATCH(TCF_EM_META);