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: INDEV | | id: VALUE |
31 * | data: | | data: 3 |
32 * +-----------+ +-----------+
34 * ---> meta_ops[INT][INDEV](...) |
38 * +-----------+ +-----------+
39 * | type: INT | | type: INT |
40 * obj | id: INDEV | | 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/config.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/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 META_COLLECTOR(int_indev)
175 *err = int_dev(skb->input_dev, dst);
178 META_COLLECTOR(var_indev)
180 *err = var_dev(skb->input_dev, dst);
183 /**************************************************************************
185 **************************************************************************/
187 META_COLLECTOR(int_priority)
189 dst->value = skb->priority;
192 META_COLLECTOR(int_protocol)
194 /* Let userspace take care of the byte ordering */
195 dst->value = skb->protocol;
198 META_COLLECTOR(int_pkttype)
200 dst->value = skb->pkt_type;
203 META_COLLECTOR(int_pktlen)
205 dst->value = skb->len;
208 META_COLLECTOR(int_datalen)
210 dst->value = skb->data_len;
213 META_COLLECTOR(int_maclen)
215 dst->value = skb->mac_len;
218 /**************************************************************************
220 **************************************************************************/
222 #ifdef CONFIG_NETFILTER
223 META_COLLECTOR(int_nfmark)
225 dst->value = skb->nfmark;
229 /**************************************************************************
231 **************************************************************************/
233 META_COLLECTOR(int_tcindex)
235 dst->value = skb->tc_index;
238 #ifdef CONFIG_NET_CLS_ACT
239 META_COLLECTOR(int_tcverd)
241 dst->value = skb->tc_verd;
245 /**************************************************************************
247 **************************************************************************/
249 #ifdef CONFIG_NET_CLS_ROUTE
250 META_COLLECTOR(int_rtclassid)
252 if (unlikely(skb->dst == NULL))
255 dst->value = skb->dst->tclassid;
259 META_COLLECTOR(int_rtiif)
261 if (unlikely(skb->dst == NULL))
264 dst->value = ((struct rtable*) skb->dst)->fl.iif;
267 /**************************************************************************
269 **************************************************************************/
271 #define SKIP_NONLOCAL(skb) \
272 if (unlikely(skb->sk == NULL)) { \
277 META_COLLECTOR(int_sk_family)
280 dst->value = skb->sk->sk_family;
283 META_COLLECTOR(int_sk_state)
286 dst->value = skb->sk->sk_state;
289 META_COLLECTOR(int_sk_reuse)
292 dst->value = skb->sk->sk_reuse;
295 META_COLLECTOR(int_sk_bound_if)
298 /* No error if bound_dev_if is 0, legal userspace check */
299 dst->value = skb->sk->sk_bound_dev_if;
302 META_COLLECTOR(var_sk_bound_if)
306 if (skb->sk->sk_bound_dev_if == 0) {
307 dst->value = (unsigned long) "any";
310 struct net_device *dev;
312 dev = dev_get_by_index(skb->sk->sk_bound_dev_if);
313 *err = var_dev(dev, dst);
319 META_COLLECTOR(int_sk_refcnt)
322 dst->value = atomic_read(&skb->sk->sk_refcnt);
325 META_COLLECTOR(int_sk_rcvbuf)
328 dst->value = skb->sk->sk_rcvbuf;
331 META_COLLECTOR(int_sk_shutdown)
334 dst->value = skb->sk->sk_shutdown;
337 META_COLLECTOR(int_sk_proto)
340 dst->value = skb->sk->sk_protocol;
343 META_COLLECTOR(int_sk_type)
346 dst->value = skb->sk->sk_type;
349 META_COLLECTOR(int_sk_rmem_alloc)
352 dst->value = atomic_read(&skb->sk->sk_rmem_alloc);
355 META_COLLECTOR(int_sk_wmem_alloc)
358 dst->value = atomic_read(&skb->sk->sk_wmem_alloc);
361 META_COLLECTOR(int_sk_omem_alloc)
364 dst->value = atomic_read(&skb->sk->sk_omem_alloc);
367 META_COLLECTOR(int_sk_rcv_qlen)
370 dst->value = skb->sk->sk_receive_queue.qlen;
373 META_COLLECTOR(int_sk_snd_qlen)
376 dst->value = skb->sk->sk_write_queue.qlen;
379 META_COLLECTOR(int_sk_wmem_queued)
382 dst->value = skb->sk->sk_wmem_queued;
385 META_COLLECTOR(int_sk_fwd_alloc)
388 dst->value = skb->sk->sk_forward_alloc;
391 META_COLLECTOR(int_sk_sndbuf)
394 dst->value = skb->sk->sk_sndbuf;
397 META_COLLECTOR(int_sk_alloc)
400 dst->value = skb->sk->sk_allocation;
403 META_COLLECTOR(int_sk_route_caps)
406 dst->value = skb->sk->sk_route_caps;
409 META_COLLECTOR(int_sk_hashent)
412 dst->value = skb->sk->sk_hashent;
415 META_COLLECTOR(int_sk_lingertime)
418 dst->value = skb->sk->sk_lingertime / HZ;
421 META_COLLECTOR(int_sk_err_qlen)
424 dst->value = skb->sk->sk_error_queue.qlen;
427 META_COLLECTOR(int_sk_ack_bl)
430 dst->value = skb->sk->sk_ack_backlog;
433 META_COLLECTOR(int_sk_max_ack_bl)
436 dst->value = skb->sk->sk_max_ack_backlog;
439 META_COLLECTOR(int_sk_prio)
442 dst->value = skb->sk->sk_priority;
445 META_COLLECTOR(int_sk_rcvlowat)
448 dst->value = skb->sk->sk_rcvlowat;
451 META_COLLECTOR(int_sk_rcvtimeo)
454 dst->value = skb->sk->sk_rcvtimeo / HZ;
457 META_COLLECTOR(int_sk_sndtimeo)
460 dst->value = skb->sk->sk_sndtimeo / HZ;
463 META_COLLECTOR(int_sk_sendmsg_off)
466 dst->value = skb->sk->sk_sndmsg_off;
469 META_COLLECTOR(int_sk_write_pend)
472 dst->value = skb->sk->sk_write_pending;
475 /**************************************************************************
476 * Meta value collectors assignment table
477 **************************************************************************/
481 void (*get)(struct sk_buff *, struct tcf_pkt_info *,
482 struct meta_value *, struct meta_obj *, int *);
485 #define META_ID(name) TCF_META_ID_##name
486 #define META_FUNC(name) { .get = meta_##name }
488 /* Meta value operations table listing all meta value collectors and
489 * assigns them to a type and meta id. */
490 static struct meta_ops __meta_ops[TCF_META_TYPE_MAX+1][TCF_META_ID_MAX+1] = {
491 [TCF_META_TYPE_VAR] = {
492 [META_ID(DEV)] = META_FUNC(var_dev),
493 [META_ID(INDEV)] = META_FUNC(var_indev),
494 [META_ID(SK_BOUND_IF)] = META_FUNC(var_sk_bound_if),
496 [TCF_META_TYPE_INT] = {
497 [META_ID(RANDOM)] = META_FUNC(int_random),
498 [META_ID(LOADAVG_0)] = META_FUNC(int_loadavg_0),
499 [META_ID(LOADAVG_1)] = META_FUNC(int_loadavg_1),
500 [META_ID(LOADAVG_2)] = META_FUNC(int_loadavg_2),
501 [META_ID(DEV)] = META_FUNC(int_dev),
502 [META_ID(INDEV)] = META_FUNC(int_indev),
503 [META_ID(PRIORITY)] = META_FUNC(int_priority),
504 [META_ID(PROTOCOL)] = META_FUNC(int_protocol),
505 [META_ID(PKTTYPE)] = META_FUNC(int_pkttype),
506 [META_ID(PKTLEN)] = META_FUNC(int_pktlen),
507 [META_ID(DATALEN)] = META_FUNC(int_datalen),
508 [META_ID(MACLEN)] = META_FUNC(int_maclen),
509 #ifdef CONFIG_NETFILTER
510 [META_ID(NFMARK)] = META_FUNC(int_nfmark),
512 [META_ID(TCINDEX)] = META_FUNC(int_tcindex),
513 #ifdef CONFIG_NET_CLS_ACT
514 [META_ID(TCVERDICT)] = META_FUNC(int_tcverd),
516 #ifdef CONFIG_NET_CLS_ROUTE
517 [META_ID(RTCLASSID)] = META_FUNC(int_rtclassid),
519 [META_ID(RTIIF)] = META_FUNC(int_rtiif),
520 [META_ID(SK_FAMILY)] = META_FUNC(int_sk_family),
521 [META_ID(SK_STATE)] = META_FUNC(int_sk_state),
522 [META_ID(SK_REUSE)] = META_FUNC(int_sk_reuse),
523 [META_ID(SK_BOUND_IF)] = META_FUNC(int_sk_bound_if),
524 [META_ID(SK_REFCNT)] = META_FUNC(int_sk_refcnt),
525 [META_ID(SK_RCVBUF)] = META_FUNC(int_sk_rcvbuf),
526 [META_ID(SK_SNDBUF)] = META_FUNC(int_sk_sndbuf),
527 [META_ID(SK_SHUTDOWN)] = META_FUNC(int_sk_shutdown),
528 [META_ID(SK_PROTO)] = META_FUNC(int_sk_proto),
529 [META_ID(SK_TYPE)] = META_FUNC(int_sk_type),
530 [META_ID(SK_RMEM_ALLOC)] = META_FUNC(int_sk_rmem_alloc),
531 [META_ID(SK_WMEM_ALLOC)] = META_FUNC(int_sk_wmem_alloc),
532 [META_ID(SK_OMEM_ALLOC)] = META_FUNC(int_sk_omem_alloc),
533 [META_ID(SK_WMEM_QUEUED)] = META_FUNC(int_sk_wmem_queued),
534 [META_ID(SK_RCV_QLEN)] = META_FUNC(int_sk_rcv_qlen),
535 [META_ID(SK_SND_QLEN)] = META_FUNC(int_sk_snd_qlen),
536 [META_ID(SK_ERR_QLEN)] = META_FUNC(int_sk_err_qlen),
537 [META_ID(SK_FORWARD_ALLOCS)] = META_FUNC(int_sk_fwd_alloc),
538 [META_ID(SK_ALLOCS)] = META_FUNC(int_sk_alloc),
539 [META_ID(SK_ROUTE_CAPS)] = META_FUNC(int_sk_route_caps),
540 [META_ID(SK_HASHENT)] = META_FUNC(int_sk_hashent),
541 [META_ID(SK_LINGERTIME)] = META_FUNC(int_sk_lingertime),
542 [META_ID(SK_ACK_BACKLOG)] = META_FUNC(int_sk_ack_bl),
543 [META_ID(SK_MAX_ACK_BACKLOG)] = META_FUNC(int_sk_max_ack_bl),
544 [META_ID(SK_PRIO)] = META_FUNC(int_sk_prio),
545 [META_ID(SK_RCVLOWAT)] = META_FUNC(int_sk_rcvlowat),
546 [META_ID(SK_RCVTIMEO)] = META_FUNC(int_sk_rcvtimeo),
547 [META_ID(SK_SNDTIMEO)] = META_FUNC(int_sk_sndtimeo),
548 [META_ID(SK_SENDMSG_OFF)] = META_FUNC(int_sk_sendmsg_off),
549 [META_ID(SK_WRITE_PENDING)] = META_FUNC(int_sk_write_pend),
553 static inline struct meta_ops * meta_ops(struct meta_value *val)
555 return &__meta_ops[meta_type(val)][meta_id(val)];
558 /**************************************************************************
559 * Type specific operations for TCF_META_TYPE_VAR
560 **************************************************************************/
562 static int meta_var_compare(struct meta_obj *a, struct meta_obj *b)
564 int r = a->len - b->len;
567 r = memcmp((void *) a->value, (void *) b->value, a->len);
572 static int meta_var_change(struct meta_value *dst, struct rtattr *rta)
574 int len = RTA_PAYLOAD(rta);
576 dst->val = (unsigned long) kmalloc(len, GFP_KERNEL);
579 memcpy((void *) dst->val, RTA_DATA(rta), len);
584 static void meta_var_destroy(struct meta_value *v)
587 kfree((void *) v->val);
590 static void meta_var_apply_extras(struct meta_value *v,
591 struct meta_obj *dst)
593 int shift = v->hdr.shift;
595 if (shift && shift < dst->len)
599 static int meta_var_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
601 if (v->val && v->len)
602 RTA_PUT(skb, tlv, v->len, (void *) v->val);
609 /**************************************************************************
610 * Type specific operations for TCF_META_TYPE_INT
611 **************************************************************************/
613 static int meta_int_compare(struct meta_obj *a, struct meta_obj *b)
615 /* Let gcc optimize it, the unlikely is not really based on
616 * some numbers but jump free code for mismatches seems
618 if (unlikely(a->value == b->value))
620 else if (a->value < b->value)
626 static int meta_int_change(struct meta_value *dst, struct rtattr *rta)
628 if (RTA_PAYLOAD(rta) >= sizeof(unsigned long)) {
629 dst->val = *(unsigned long *) RTA_DATA(rta);
630 dst->len = sizeof(unsigned long);
631 } else if (RTA_PAYLOAD(rta) == sizeof(u32)) {
632 dst->val = *(u32 *) RTA_DATA(rta);
633 dst->len = sizeof(u32);
640 static void meta_int_apply_extras(struct meta_value *v,
641 struct meta_obj *dst)
644 dst->value >>= v->hdr.shift;
647 dst->value &= v->val;
650 static int meta_int_dump(struct sk_buff *skb, struct meta_value *v, int tlv)
652 if (v->len == sizeof(unsigned long))
653 RTA_PUT(skb, tlv, sizeof(unsigned long), &v->val);
654 else if (v->len == sizeof(u32)) {
656 RTA_PUT(skb, tlv, sizeof(d), &d);
665 /**************************************************************************
666 * Type specific operations table
667 **************************************************************************/
671 void (*destroy)(struct meta_value *);
672 int (*compare)(struct meta_obj *, struct meta_obj *);
673 int (*change)(struct meta_value *, struct rtattr *);
674 void (*apply_extras)(struct meta_value *, struct meta_obj *);
675 int (*dump)(struct sk_buff *, struct meta_value *, int);
678 static struct meta_type_ops __meta_type_ops[TCF_META_TYPE_MAX+1] = {
679 [TCF_META_TYPE_VAR] = {
680 .destroy = meta_var_destroy,
681 .compare = meta_var_compare,
682 .change = meta_var_change,
683 .apply_extras = meta_var_apply_extras,
684 .dump = meta_var_dump
686 [TCF_META_TYPE_INT] = {
687 .compare = meta_int_compare,
688 .change = meta_int_change,
689 .apply_extras = meta_int_apply_extras,
690 .dump = meta_int_dump
694 static inline struct meta_type_ops * meta_type_ops(struct meta_value *v)
696 return &__meta_type_ops[meta_type(v)];
699 /**************************************************************************
701 **************************************************************************/
703 static inline int meta_get(struct sk_buff *skb, struct tcf_pkt_info *info,
704 struct meta_value *v, struct meta_obj *dst)
708 if (meta_id(v) == TCF_META_ID_VALUE) {
714 meta_ops(v)->get(skb, info, v, dst, &err);
718 if (meta_type_ops(v)->apply_extras)
719 meta_type_ops(v)->apply_extras(v, dst);
724 static int em_meta_match(struct sk_buff *skb, struct tcf_ematch *m,
725 struct tcf_pkt_info *info)
728 struct meta_match *meta = (struct meta_match *) m->data;
729 struct meta_obj l_value, r_value;
731 if (meta_get(skb, info, &meta->lvalue, &l_value) < 0 ||
732 meta_get(skb, info, &meta->rvalue, &r_value) < 0)
735 r = meta_type_ops(&meta->lvalue)->compare(&l_value, &r_value);
737 switch (meta->lvalue.hdr.op) {
749 static inline void meta_delete(struct meta_match *meta)
751 struct meta_type_ops *ops = meta_type_ops(&meta->lvalue);
753 if (ops && ops->destroy) {
754 ops->destroy(&meta->lvalue);
755 ops->destroy(&meta->rvalue);
761 static inline int meta_change_data(struct meta_value *dst, struct rtattr *rta)
764 if (RTA_PAYLOAD(rta) == 0)
767 return meta_type_ops(dst)->change(dst, rta);
773 static inline int meta_is_supported(struct meta_value *val)
775 return (!meta_id(val) || meta_ops(val)->get);
778 static int em_meta_change(struct tcf_proto *tp, void *data, int len,
779 struct tcf_ematch *m)
782 struct rtattr *tb[TCA_EM_META_MAX];
783 struct tcf_meta_hdr *hdr;
784 struct meta_match *meta = NULL;
786 if (rtattr_parse(tb, TCA_EM_META_MAX, data, len) < 0)
789 if (tb[TCA_EM_META_HDR-1] == NULL ||
790 RTA_PAYLOAD(tb[TCA_EM_META_HDR-1]) < sizeof(*hdr))
792 hdr = RTA_DATA(tb[TCA_EM_META_HDR-1]);
794 if (TCF_META_TYPE(hdr->left.kind) != TCF_META_TYPE(hdr->right.kind) ||
795 TCF_META_TYPE(hdr->left.kind) > TCF_META_TYPE_MAX ||
796 TCF_META_ID(hdr->left.kind) > TCF_META_ID_MAX ||
797 TCF_META_ID(hdr->right.kind) > TCF_META_ID_MAX)
800 meta = kmalloc(sizeof(*meta), GFP_KERNEL);
803 memset(meta, 0, sizeof(*meta));
805 memcpy(&meta->lvalue.hdr, &hdr->left, sizeof(hdr->left));
806 memcpy(&meta->rvalue.hdr, &hdr->right, sizeof(hdr->right));
808 if (!meta_is_supported(&meta->lvalue) ||
809 !meta_is_supported(&meta->rvalue)) {
814 if (meta_change_data(&meta->lvalue, tb[TCA_EM_META_LVALUE-1]) < 0 ||
815 meta_change_data(&meta->rvalue, tb[TCA_EM_META_RVALUE-1]) < 0)
818 m->datalen = sizeof(*meta);
819 m->data = (unsigned long) meta;
828 static void em_meta_destroy(struct tcf_proto *tp, struct tcf_ematch *m)
831 meta_delete((struct meta_match *) m->data);
834 static int em_meta_dump(struct sk_buff *skb, struct tcf_ematch *em)
836 struct meta_match *meta = (struct meta_match *) em->data;
837 struct tcf_meta_hdr hdr;
838 struct meta_type_ops *ops;
840 memset(&hdr, 0, sizeof(hdr));
841 memcpy(&hdr.left, &meta->lvalue.hdr, sizeof(hdr.left));
842 memcpy(&hdr.right, &meta->rvalue.hdr, sizeof(hdr.right));
844 RTA_PUT(skb, TCA_EM_META_HDR, sizeof(hdr), &hdr);
846 ops = meta_type_ops(&meta->lvalue);
847 if (ops->dump(skb, &meta->lvalue, TCA_EM_META_LVALUE) < 0 ||
848 ops->dump(skb, &meta->rvalue, TCA_EM_META_RVALUE) < 0)
857 static struct tcf_ematch_ops em_meta_ops = {
859 .change = em_meta_change,
860 .match = em_meta_match,
861 .destroy = em_meta_destroy,
862 .dump = em_meta_dump,
863 .owner = THIS_MODULE,
864 .link = LIST_HEAD_INIT(em_meta_ops.link)
867 static int __init init_em_meta(void)
869 return tcf_em_register(&em_meta_ops);
872 static void __exit exit_em_meta(void)
874 tcf_em_unregister(&em_meta_ops);
877 MODULE_LICENSE("GPL");
879 module_init(init_em_meta);
880 module_exit(exit_em_meta);