2 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
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: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
11 * The filters are packed to hash tables of key nodes
12 * with a set of 32bit key/mask pairs at every node.
13 * Nodes reference next level hash tables etc.
15 * This scheme is the best universal classifier I managed to
16 * invent; it is not super-fast, but it is not slow (provided you
17 * program it correctly), and general enough. And its relative
18 * speed grows as the number of rules becomes larger.
20 * It seems that it represents the best middle point between
21 * speed and manageability both by human and by machine.
23 * It is especially useful for link sharing combined with QoS;
24 * pure RSVP doesn't need such a general approach and can use
25 * much simpler (and faster) schemes, sort of cls_rsvp.c.
27 * JHS: We should remove the CONFIG_NET_CLS_IND from here
28 * eventually when the meta match extension is made available
30 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
33 #include <linux/module.h>
34 #include <linux/slab.h>
35 #include <linux/types.h>
36 #include <linux/kernel.h>
37 #include <linux/string.h>
38 #include <linux/errno.h>
39 #include <linux/percpu.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/skbuff.h>
42 #include <linux/bitmap.h>
43 #include <linux/netdevice.h>
44 #include <linux/hash.h>
45 #include <net/netlink.h>
46 #include <net/act_api.h>
47 #include <net/pkt_cls.h>
48 #include <linux/netdevice.h>
51 struct tc_u_knode __rcu *next;
53 struct tc_u_hnode __rcu *ht_up;
55 #ifdef CONFIG_NET_CLS_IND
59 struct tcf_result res;
60 struct tc_u_hnode __rcu *ht_down;
61 #ifdef CONFIG_CLS_U32_PERF
62 struct tc_u32_pcnt __percpu *pf;
65 #ifdef CONFIG_CLS_U32_MARK
68 u32 __percpu *pcpu_success;
72 struct work_struct work;
75 /* The 'sel' field MUST be the last field in structure to allow for
76 * tc_u32_keys allocated at end of structure.
78 struct tc_u32_sel sel;
82 struct tc_u_hnode __rcu *next;
85 struct tc_u_common *tp_c;
89 /* The 'ht' field MUST be the last field in structure to allow for
90 * more entries allocated at end of structure.
92 struct tc_u_knode __rcu *ht[1];
96 struct tc_u_hnode __rcu *hlist;
100 struct hlist_node hnode;
104 static inline unsigned int u32_hash_fold(__be32 key,
105 const struct tc_u32_sel *sel,
108 unsigned int h = ntohl(key & sel->hmask) >> fshift;
113 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
114 struct tcf_result *res)
117 struct tc_u_knode *knode;
119 } stack[TC_U32_MAXDEPTH];
121 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
122 unsigned int off = skb_network_offset(skb);
123 struct tc_u_knode *n;
127 #ifdef CONFIG_CLS_U32_PERF
133 n = rcu_dereference_bh(ht->ht[sel]);
137 struct tc_u32_key *key = n->sel.keys;
139 #ifdef CONFIG_CLS_U32_PERF
140 __this_cpu_inc(n->pf->rcnt);
144 if (tc_skip_sw(n->flags)) {
145 n = rcu_dereference_bh(n->next);
149 #ifdef CONFIG_CLS_U32_MARK
150 if ((skb->mark & n->mask) != n->val) {
151 n = rcu_dereference_bh(n->next);
154 __this_cpu_inc(*n->pcpu_success);
158 for (i = n->sel.nkeys; i > 0; i--, key++) {
159 int toff = off + key->off + (off2 & key->offmask);
162 if (skb_headroom(skb) + toff > INT_MAX)
165 data = skb_header_pointer(skb, toff, 4, &hdata);
168 if ((*data ^ key->val) & key->mask) {
169 n = rcu_dereference_bh(n->next);
172 #ifdef CONFIG_CLS_U32_PERF
173 __this_cpu_inc(n->pf->kcnts[j]);
178 ht = rcu_dereference_bh(n->ht_down);
181 if (n->sel.flags & TC_U32_TERMINAL) {
184 #ifdef CONFIG_NET_CLS_IND
185 if (!tcf_match_indev(skb, n->ifindex)) {
186 n = rcu_dereference_bh(n->next);
190 #ifdef CONFIG_CLS_U32_PERF
191 __this_cpu_inc(n->pf->rhit);
193 r = tcf_exts_exec(skb, &n->exts, res);
195 n = rcu_dereference_bh(n->next);
201 n = rcu_dereference_bh(n->next);
206 if (sdepth >= TC_U32_MAXDEPTH)
208 stack[sdepth].knode = n;
209 stack[sdepth].off = off;
212 ht = rcu_dereference_bh(n->ht_down);
217 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
221 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
224 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
227 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
228 off2 = n->sel.off + 3;
229 if (n->sel.flags & TC_U32_VAROFFSET) {
232 data = skb_header_pointer(skb,
237 off2 += ntohs(n->sel.offmask & *data) >>
242 if (n->sel.flags & TC_U32_EAT) {
253 n = stack[sdepth].knode;
254 ht = rcu_dereference_bh(n->ht_up);
255 off = stack[sdepth].off;
262 net_warn_ratelimited("cls_u32: dead loop\n");
266 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
268 struct tc_u_hnode *ht;
270 for (ht = rtnl_dereference(tp_c->hlist);
272 ht = rtnl_dereference(ht->next))
273 if (ht->handle == handle)
279 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
282 struct tc_u_knode *n = NULL;
284 sel = TC_U32_HASH(handle);
285 if (sel > ht->divisor)
288 for (n = rtnl_dereference(ht->ht[sel]);
290 n = rtnl_dereference(n->next))
291 if (n->handle == handle)
298 static void *u32_get(struct tcf_proto *tp, u32 handle)
300 struct tc_u_hnode *ht;
301 struct tc_u_common *tp_c = tp->data;
303 if (TC_U32_HTID(handle) == TC_U32_ROOT)
304 ht = rtnl_dereference(tp->root);
306 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
311 if (TC_U32_KEY(handle) == 0)
314 return u32_lookup_key(ht, handle);
317 static u32 gen_new_htid(struct tc_u_common *tp_c)
321 /* hgenerator only used inside rtnl lock it is safe to increment
322 * without read _copy_ update semantics
325 if (++tp_c->hgenerator == 0x7FF)
326 tp_c->hgenerator = 1;
327 } while (--i > 0 && u32_lookup_ht(tp_c, (tp_c->hgenerator|0x800)<<20));
329 return i > 0 ? (tp_c->hgenerator|0x800)<<20 : 0;
332 static struct hlist_head *tc_u_common_hash;
334 #define U32_HASH_SHIFT 10
335 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
337 static unsigned int tc_u_hash(const struct tcf_proto *tp)
339 struct net_device *dev = tp->q->dev_queue->dev;
340 u32 qhandle = tp->q->handle;
341 int ifindex = dev->ifindex;
343 return hash_64((u64)ifindex << 32 | qhandle, U32_HASH_SHIFT);
346 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
348 struct tc_u_common *tc;
352 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
359 static int u32_init(struct tcf_proto *tp)
361 struct tc_u_hnode *root_ht;
362 struct tc_u_common *tp_c;
365 tp_c = tc_u_common_find(tp);
367 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
372 root_ht->handle = tp_c ? gen_new_htid(tp_c) : 0x80000000;
373 root_ht->prio = tp->prio;
376 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
382 INIT_HLIST_NODE(&tp_c->hnode);
385 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
389 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
390 rcu_assign_pointer(tp_c->hlist, root_ht);
391 root_ht->tp_c = tp_c;
393 rcu_assign_pointer(tp->root, root_ht);
398 static void __u32_destroy_key(struct tc_u_knode *n)
400 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
402 tcf_exts_destroy(&n->exts);
403 if (ht && --ht->refcnt == 0)
408 static void u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
411 tcf_exts_put_net(&n->exts);
412 #ifdef CONFIG_CLS_U32_PERF
416 #ifdef CONFIG_CLS_U32_MARK
418 free_percpu(n->pcpu_success);
420 __u32_destroy_key(n);
423 /* u32_delete_key_rcu should be called when free'ing a copied
424 * version of a tc_u_knode obtained from u32_init_knode(). When
425 * copies are obtained from u32_init_knode() the statistics are
426 * shared between the old and new copies to allow readers to
427 * continue to update the statistics during the copy. To support
428 * this the u32_delete_key_rcu variant does not free the percpu
431 static void u32_delete_key_work(struct work_struct *work)
433 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
436 u32_destroy_key(key->tp, key, false);
440 static void u32_delete_key_rcu(struct rcu_head *rcu)
442 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
444 INIT_WORK(&key->work, u32_delete_key_work);
445 tcf_queue_work(&key->work);
448 /* u32_delete_key_freepf_rcu is the rcu callback variant
449 * that free's the entire structure including the statistics
450 * percpu variables. Only use this if the key is not a copy
451 * returned by u32_init_knode(). See u32_delete_key_rcu()
452 * for the variant that should be used with keys return from
455 static void u32_delete_key_freepf_work(struct work_struct *work)
457 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
460 u32_destroy_key(key->tp, key, true);
464 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
466 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
468 INIT_WORK(&key->work, u32_delete_key_freepf_work);
469 tcf_queue_work(&key->work);
472 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
474 struct tc_u_knode __rcu **kp;
475 struct tc_u_knode *pkp;
476 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
479 kp = &ht->ht[TC_U32_HASH(key->handle)];
480 for (pkp = rtnl_dereference(*kp); pkp;
481 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
483 RCU_INIT_POINTER(*kp, key->next);
485 tcf_unbind_filter(tp, &key->res);
486 tcf_exts_get_net(&key->exts);
487 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
496 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
498 struct net_device *dev = tp->q->dev_queue->dev;
499 struct tc_cls_u32_offload cls_u32 = {};
501 if (!tc_should_offload(dev, 0))
504 tc_cls_common_offload_init(&cls_u32.common, tp);
505 cls_u32.command = TC_CLSU32_DELETE_KNODE;
506 cls_u32.knode.handle = handle;
508 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
511 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
514 struct net_device *dev = tp->q->dev_queue->dev;
515 struct tc_cls_u32_offload cls_u32 = {};
518 if (!tc_should_offload(dev, flags))
519 return tc_skip_sw(flags) ? -EINVAL : 0;
521 tc_cls_common_offload_init(&cls_u32.common, tp);
522 cls_u32.command = TC_CLSU32_NEW_HNODE;
523 cls_u32.hnode.divisor = h->divisor;
524 cls_u32.hnode.handle = h->handle;
525 cls_u32.hnode.prio = h->prio;
527 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
528 if (tc_skip_sw(flags))
534 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
536 struct net_device *dev = tp->q->dev_queue->dev;
537 struct tc_cls_u32_offload cls_u32 = {};
539 if (!tc_should_offload(dev, 0))
542 tc_cls_common_offload_init(&cls_u32.common, tp);
543 cls_u32.command = TC_CLSU32_DELETE_HNODE;
544 cls_u32.hnode.divisor = h->divisor;
545 cls_u32.hnode.handle = h->handle;
546 cls_u32.hnode.prio = h->prio;
548 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
551 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
554 struct net_device *dev = tp->q->dev_queue->dev;
555 struct tc_cls_u32_offload cls_u32 = {};
558 if (!tc_should_offload(dev, flags))
559 return tc_skip_sw(flags) ? -EINVAL : 0;
561 tc_cls_common_offload_init(&cls_u32.common, tp);
562 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
563 cls_u32.knode.handle = n->handle;
564 cls_u32.knode.fshift = n->fshift;
565 #ifdef CONFIG_CLS_U32_MARK
566 cls_u32.knode.val = n->val;
567 cls_u32.knode.mask = n->mask;
569 cls_u32.knode.val = 0;
570 cls_u32.knode.mask = 0;
572 cls_u32.knode.sel = &n->sel;
573 cls_u32.knode.exts = &n->exts;
575 cls_u32.knode.link_handle = n->ht_down->handle;
577 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_CLSU32, &cls_u32);
580 n->flags |= TCA_CLS_FLAGS_IN_HW;
582 if (tc_skip_sw(flags))
588 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
590 struct tc_u_knode *n;
593 for (h = 0; h <= ht->divisor; h++) {
594 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
595 RCU_INIT_POINTER(ht->ht[h],
596 rtnl_dereference(n->next));
597 tcf_unbind_filter(tp, &n->res);
598 u32_remove_hw_knode(tp, n->handle);
599 if (tcf_exts_get_net(&n->exts))
600 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
602 u32_destroy_key(n->tp, n, true);
607 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
609 struct tc_u_common *tp_c = tp->data;
610 struct tc_u_hnode __rcu **hn;
611 struct tc_u_hnode *phn;
615 u32_clear_hnode(tp, ht);
618 for (phn = rtnl_dereference(*hn);
620 hn = &phn->next, phn = rtnl_dereference(*hn)) {
622 u32_clear_hw_hnode(tp, ht);
623 RCU_INIT_POINTER(*hn, ht->next);
632 static bool ht_empty(struct tc_u_hnode *ht)
636 for (h = 0; h <= ht->divisor; h++)
637 if (rcu_access_pointer(ht->ht[h]))
643 static void u32_destroy(struct tcf_proto *tp)
645 struct tc_u_common *tp_c = tp->data;
646 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
648 WARN_ON(root_ht == NULL);
650 if (root_ht && --root_ht->refcnt == 0)
651 u32_destroy_hnode(tp, root_ht);
653 if (--tp_c->refcnt == 0) {
654 struct tc_u_hnode *ht;
656 hlist_del(&tp_c->hnode);
658 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
659 u32_clear_hnode(tp, ht);
660 RCU_INIT_POINTER(tp_c->hlist, ht->next);
662 /* u32_destroy_key() will later free ht for us, if it's
663 * still referenced by some knode
665 if (--ht->refcnt == 0)
675 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last)
677 struct tc_u_hnode *ht = arg;
678 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
679 struct tc_u_common *tp_c = tp->data;
685 if (TC_U32_KEY(ht->handle)) {
686 u32_remove_hw_knode(tp, ht->handle);
687 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
694 if (ht->refcnt == 1) {
696 u32_destroy_hnode(tp, ht);
704 if (root_ht->refcnt > 1) {
708 if (root_ht->refcnt == 1) {
709 if (!ht_empty(root_ht)) {
716 if (tp_c->refcnt > 1) {
721 if (tp_c->refcnt == 1) {
722 struct tc_u_hnode *ht;
724 for (ht = rtnl_dereference(tp_c->hlist);
726 ht = rtnl_dereference(ht->next))
737 #define NR_U32_NODE (1<<12)
738 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
740 struct tc_u_knode *n;
742 unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
745 return handle | 0xFFF;
747 for (n = rtnl_dereference(ht->ht[TC_U32_HASH(handle)]);
749 n = rtnl_dereference(n->next))
750 set_bit(TC_U32_NODE(n->handle), bitmap);
752 i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
753 if (i >= NR_U32_NODE)
754 i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
757 return handle | (i >= NR_U32_NODE ? 0xFFF : i);
760 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
761 [TCA_U32_CLASSID] = { .type = NLA_U32 },
762 [TCA_U32_HASH] = { .type = NLA_U32 },
763 [TCA_U32_LINK] = { .type = NLA_U32 },
764 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
765 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
766 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
767 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
768 [TCA_U32_FLAGS] = { .type = NLA_U32 },
771 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
772 unsigned long base, struct tc_u_hnode *ht,
773 struct tc_u_knode *n, struct nlattr **tb,
774 struct nlattr *est, bool ovr)
778 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr);
782 if (tb[TCA_U32_LINK]) {
783 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
784 struct tc_u_hnode *ht_down = NULL, *ht_old;
786 if (TC_U32_KEY(handle))
790 ht_down = u32_lookup_ht(ht->tp_c, handle);
797 ht_old = rtnl_dereference(n->ht_down);
798 rcu_assign_pointer(n->ht_down, ht_down);
803 if (tb[TCA_U32_CLASSID]) {
804 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
805 tcf_bind_filter(tp, &n->res, base);
808 #ifdef CONFIG_NET_CLS_IND
809 if (tb[TCA_U32_INDEV]) {
811 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
820 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
821 struct tc_u_knode *n)
823 struct tc_u_knode __rcu **ins;
824 struct tc_u_knode *pins;
825 struct tc_u_hnode *ht;
827 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
828 ht = rtnl_dereference(tp->root);
830 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
832 ins = &ht->ht[TC_U32_HASH(n->handle)];
834 /* The node must always exist for it to be replaced if this is not the
835 * case then something went very wrong elsewhere.
837 for (pins = rtnl_dereference(*ins); ;
838 ins = &pins->next, pins = rtnl_dereference(*ins))
839 if (pins->handle == n->handle)
842 RCU_INIT_POINTER(n->next, pins->next);
843 rcu_assign_pointer(*ins, n);
846 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
847 struct tc_u_knode *n)
849 struct tc_u_knode *new;
850 struct tc_u32_sel *s = &n->sel;
852 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
858 RCU_INIT_POINTER(new->next, n->next);
859 new->handle = n->handle;
860 RCU_INIT_POINTER(new->ht_up, n->ht_up);
862 #ifdef CONFIG_NET_CLS_IND
863 new->ifindex = n->ifindex;
865 new->fshift = n->fshift;
867 new->flags = n->flags;
868 RCU_INIT_POINTER(new->ht_down, n->ht_down);
870 /* bump reference count as long as we hold pointer to structure */
872 new->ht_down->refcnt++;
874 #ifdef CONFIG_CLS_U32_PERF
875 /* Statistics may be incremented by readers during update
876 * so we must keep them in tact. When the node is later destroyed
877 * a special destroy call must be made to not free the pf memory.
882 #ifdef CONFIG_CLS_U32_MARK
885 /* Similarly success statistics must be moved as pointers */
886 new->pcpu_success = n->pcpu_success;
889 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
891 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
899 static int u32_change(struct net *net, struct sk_buff *in_skb,
900 struct tcf_proto *tp, unsigned long base, u32 handle,
901 struct nlattr **tca, void **arg, bool ovr)
903 struct tc_u_common *tp_c = tp->data;
904 struct tc_u_hnode *ht;
905 struct tc_u_knode *n;
906 struct tc_u32_sel *s;
907 struct nlattr *opt = tca[TCA_OPTIONS];
908 struct nlattr *tb[TCA_U32_MAX + 1];
912 #ifdef CONFIG_CLS_U32_PERF
917 return handle ? -EINVAL : 0;
919 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL);
923 if (tb[TCA_U32_FLAGS]) {
924 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
925 if (!tc_flags_valid(flags))
931 struct tc_u_knode *new;
933 if (TC_U32_KEY(n->handle) == 0)
936 if ((n->flags ^ flags) &
937 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW))
940 new = u32_init_knode(tp, n);
944 err = u32_set_parms(net, tp, base,
945 rtnl_dereference(n->ht_up), new, tb,
949 __u32_destroy_key(new);
953 err = u32_replace_hw_knode(tp, new, flags);
955 __u32_destroy_key(new);
959 if (!tc_in_hw(new->flags))
960 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
962 u32_replace_knode(tp, tp_c, new);
963 tcf_unbind_filter(tp, &n->res);
964 tcf_exts_get_net(&n->exts);
965 call_rcu(&n->rcu, u32_delete_key_rcu);
969 if (tb[TCA_U32_DIVISOR]) {
970 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
972 if (--divisor > 0x100)
974 if (TC_U32_KEY(handle))
977 handle = gen_new_htid(tp->data);
981 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
986 ht->divisor = divisor;
990 err = u32_replace_hw_hnode(tp, ht, flags);
996 RCU_INIT_POINTER(ht->next, tp_c->hlist);
997 rcu_assign_pointer(tp_c->hlist, ht);
1003 if (tb[TCA_U32_HASH]) {
1004 htid = nla_get_u32(tb[TCA_U32_HASH]);
1005 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1006 ht = rtnl_dereference(tp->root);
1009 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1014 ht = rtnl_dereference(tp->root);
1018 if (ht->divisor < TC_U32_HASH(htid))
1022 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
1024 handle = htid | TC_U32_NODE(handle);
1026 handle = gen_new_kid(ht, htid);
1028 if (tb[TCA_U32_SEL] == NULL)
1031 s = nla_data(tb[TCA_U32_SEL]);
1032 sel_size = sizeof(*s) + sizeof(*s->keys) * s->nkeys;
1033 if (nla_len(tb[TCA_U32_SEL]) < sel_size)
1036 n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1040 #ifdef CONFIG_CLS_U32_PERF
1041 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1042 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1049 memcpy(&n->sel, s, sel_size);
1050 RCU_INIT_POINTER(n->ht_up, ht);
1052 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1056 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1060 #ifdef CONFIG_CLS_U32_MARK
1061 n->pcpu_success = alloc_percpu(u32);
1062 if (!n->pcpu_success) {
1067 if (tb[TCA_U32_MARK]) {
1068 struct tc_u32_mark *mark;
1070 mark = nla_data(tb[TCA_U32_MARK]);
1072 n->mask = mark->mask;
1076 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
1078 struct tc_u_knode __rcu **ins;
1079 struct tc_u_knode *pins;
1081 err = u32_replace_hw_knode(tp, n, flags);
1085 if (!tc_in_hw(n->flags))
1086 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1088 ins = &ht->ht[TC_U32_HASH(handle)];
1089 for (pins = rtnl_dereference(*ins); pins;
1090 ins = &pins->next, pins = rtnl_dereference(*ins))
1091 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1094 RCU_INIT_POINTER(n->next, pins);
1095 rcu_assign_pointer(*ins, n);
1101 #ifdef CONFIG_CLS_U32_MARK
1102 free_percpu(n->pcpu_success);
1106 tcf_exts_destroy(&n->exts);
1107 #ifdef CONFIG_CLS_U32_PERF
1114 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1116 struct tc_u_common *tp_c = tp->data;
1117 struct tc_u_hnode *ht;
1118 struct tc_u_knode *n;
1124 for (ht = rtnl_dereference(tp_c->hlist);
1126 ht = rtnl_dereference(ht->next)) {
1127 if (ht->prio != tp->prio)
1129 if (arg->count >= arg->skip) {
1130 if (arg->fn(tp, ht, arg) < 0) {
1136 for (h = 0; h <= ht->divisor; h++) {
1137 for (n = rtnl_dereference(ht->ht[h]);
1139 n = rtnl_dereference(n->next)) {
1140 if (arg->count < arg->skip) {
1144 if (arg->fn(tp, n, arg) < 0) {
1154 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1156 struct tc_u_knode *n = fh;
1158 if (n && n->res.classid == classid)
1162 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1163 struct sk_buff *skb, struct tcmsg *t)
1165 struct tc_u_knode *n = fh;
1166 struct tc_u_hnode *ht_up, *ht_down;
1167 struct nlattr *nest;
1172 t->tcm_handle = n->handle;
1174 nest = nla_nest_start(skb, TCA_OPTIONS);
1176 goto nla_put_failure;
1178 if (TC_U32_KEY(n->handle) == 0) {
1179 struct tc_u_hnode *ht = fh;
1180 u32 divisor = ht->divisor + 1;
1182 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1183 goto nla_put_failure;
1185 #ifdef CONFIG_CLS_U32_PERF
1186 struct tc_u32_pcnt *gpf;
1190 if (nla_put(skb, TCA_U32_SEL,
1191 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1193 goto nla_put_failure;
1195 ht_up = rtnl_dereference(n->ht_up);
1197 u32 htid = n->handle & 0xFFFFF000;
1198 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1199 goto nla_put_failure;
1201 if (n->res.classid &&
1202 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1203 goto nla_put_failure;
1205 ht_down = rtnl_dereference(n->ht_down);
1207 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1208 goto nla_put_failure;
1210 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1211 goto nla_put_failure;
1213 #ifdef CONFIG_CLS_U32_MARK
1214 if ((n->val || n->mask)) {
1215 struct tc_u32_mark mark = {.val = n->val,
1220 for_each_possible_cpu(cpum) {
1221 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1223 mark.success += cnt;
1226 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1227 goto nla_put_failure;
1231 if (tcf_exts_dump(skb, &n->exts) < 0)
1232 goto nla_put_failure;
1234 #ifdef CONFIG_NET_CLS_IND
1236 struct net_device *dev;
1237 dev = __dev_get_by_index(net, n->ifindex);
1238 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1239 goto nla_put_failure;
1242 #ifdef CONFIG_CLS_U32_PERF
1243 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1244 n->sel.nkeys * sizeof(u64),
1247 goto nla_put_failure;
1249 for_each_possible_cpu(cpu) {
1251 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1253 gpf->rcnt += pf->rcnt;
1254 gpf->rhit += pf->rhit;
1255 for (i = 0; i < n->sel.nkeys; i++)
1256 gpf->kcnts[i] += pf->kcnts[i];
1259 if (nla_put_64bit(skb, TCA_U32_PCNT,
1260 sizeof(struct tc_u32_pcnt) +
1261 n->sel.nkeys * sizeof(u64),
1262 gpf, TCA_U32_PAD)) {
1264 goto nla_put_failure;
1270 nla_nest_end(skb, nest);
1272 if (TC_U32_KEY(n->handle))
1273 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1274 goto nla_put_failure;
1278 nla_nest_cancel(skb, nest);
1282 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1284 .classify = u32_classify,
1286 .destroy = u32_destroy,
1288 .change = u32_change,
1289 .delete = u32_delete,
1292 .bind_class = u32_bind_class,
1293 .owner = THIS_MODULE,
1296 static int __init init_u32(void)
1300 pr_info("u32 classifier\n");
1301 #ifdef CONFIG_CLS_U32_PERF
1302 pr_info(" Performance counters on\n");
1304 #ifdef CONFIG_NET_CLS_IND
1305 pr_info(" input device check on\n");
1307 #ifdef CONFIG_NET_CLS_ACT
1308 pr_info(" Actions configured\n");
1310 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1311 sizeof(struct hlist_head),
1313 if (!tc_u_common_hash)
1316 for (i = 0; i < U32_HASH_SIZE; i++)
1317 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1319 ret = register_tcf_proto_ops(&cls_u32_ops);
1321 kvfree(tc_u_common_hash);
1325 static void __exit exit_u32(void)
1327 unregister_tcf_proto_ops(&cls_u32_ops);
1328 kvfree(tc_u_common_hash);
1331 module_init(init_u32)
1332 module_exit(exit_u32)
1333 MODULE_LICENSE("GPL");