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/idr.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;
88 struct idr handle_idr;
91 /* The 'ht' field MUST be the last field in structure to allow for
92 * more entries allocated at end of structure.
94 struct tc_u_knode __rcu *ht[1];
98 struct tc_u_hnode __rcu *hlist;
99 struct tcf_block *block;
101 struct idr handle_idr;
102 struct hlist_node hnode;
106 static inline unsigned int u32_hash_fold(__be32 key,
107 const struct tc_u32_sel *sel,
110 unsigned int h = ntohl(key & sel->hmask) >> fshift;
115 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
116 struct tcf_result *res)
119 struct tc_u_knode *knode;
121 } stack[TC_U32_MAXDEPTH];
123 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
124 unsigned int off = skb_network_offset(skb);
125 struct tc_u_knode *n;
129 #ifdef CONFIG_CLS_U32_PERF
135 n = rcu_dereference_bh(ht->ht[sel]);
139 struct tc_u32_key *key = n->sel.keys;
141 #ifdef CONFIG_CLS_U32_PERF
142 __this_cpu_inc(n->pf->rcnt);
146 if (tc_skip_sw(n->flags)) {
147 n = rcu_dereference_bh(n->next);
151 #ifdef CONFIG_CLS_U32_MARK
152 if ((skb->mark & n->mask) != n->val) {
153 n = rcu_dereference_bh(n->next);
156 __this_cpu_inc(*n->pcpu_success);
160 for (i = n->sel.nkeys; i > 0; i--, key++) {
161 int toff = off + key->off + (off2 & key->offmask);
164 if (skb_headroom(skb) + toff > INT_MAX)
167 data = skb_header_pointer(skb, toff, 4, &hdata);
170 if ((*data ^ key->val) & key->mask) {
171 n = rcu_dereference_bh(n->next);
174 #ifdef CONFIG_CLS_U32_PERF
175 __this_cpu_inc(n->pf->kcnts[j]);
180 ht = rcu_dereference_bh(n->ht_down);
183 if (n->sel.flags & TC_U32_TERMINAL) {
186 #ifdef CONFIG_NET_CLS_IND
187 if (!tcf_match_indev(skb, n->ifindex)) {
188 n = rcu_dereference_bh(n->next);
192 #ifdef CONFIG_CLS_U32_PERF
193 __this_cpu_inc(n->pf->rhit);
195 r = tcf_exts_exec(skb, &n->exts, res);
197 n = rcu_dereference_bh(n->next);
203 n = rcu_dereference_bh(n->next);
208 if (sdepth >= TC_U32_MAXDEPTH)
210 stack[sdepth].knode = n;
211 stack[sdepth].off = off;
214 ht = rcu_dereference_bh(n->ht_down);
219 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
223 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
226 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
229 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
230 off2 = n->sel.off + 3;
231 if (n->sel.flags & TC_U32_VAROFFSET) {
234 data = skb_header_pointer(skb,
239 off2 += ntohs(n->sel.offmask & *data) >>
244 if (n->sel.flags & TC_U32_EAT) {
255 n = stack[sdepth].knode;
256 ht = rcu_dereference_bh(n->ht_up);
257 off = stack[sdepth].off;
264 net_warn_ratelimited("cls_u32: dead loop\n");
268 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
270 struct tc_u_hnode *ht;
272 for (ht = rtnl_dereference(tp_c->hlist);
274 ht = rtnl_dereference(ht->next))
275 if (ht->handle == handle)
281 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
284 struct tc_u_knode *n = NULL;
286 sel = TC_U32_HASH(handle);
287 if (sel > ht->divisor)
290 for (n = rtnl_dereference(ht->ht[sel]);
292 n = rtnl_dereference(n->next))
293 if (n->handle == handle)
300 static void *u32_get(struct tcf_proto *tp, u32 handle)
302 struct tc_u_hnode *ht;
303 struct tc_u_common *tp_c = tp->data;
305 if (TC_U32_HTID(handle) == TC_U32_ROOT)
306 ht = rtnl_dereference(tp->root);
308 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
313 if (TC_U32_KEY(handle) == 0)
316 return u32_lookup_key(ht, handle);
319 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
321 unsigned long idr_index;
324 /* This is only used inside rtnl lock it is safe to increment
325 * without read _copy_ update semantics
327 err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index,
328 1, 0x7FF, GFP_KERNEL);
331 return (u32)(idr_index | 0x800) << 20;
334 static struct hlist_head *tc_u_common_hash;
336 #define U32_HASH_SHIFT 10
337 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
339 static unsigned int tc_u_hash(const struct tcf_proto *tp)
341 return hash_ptr(tp->chain->block, U32_HASH_SHIFT);
344 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
346 struct tc_u_common *tc;
350 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
351 if (tc->block == tp->chain->block)
357 static int u32_init(struct tcf_proto *tp)
359 struct tc_u_hnode *root_ht;
360 struct tc_u_common *tp_c;
363 tp_c = tc_u_common_find(tp);
365 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
370 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
371 root_ht->prio = tp->prio;
372 idr_init(&root_ht->handle_idr);
375 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
380 tp_c->block = tp->chain->block;
381 INIT_HLIST_NODE(&tp_c->hnode);
382 idr_init(&tp_c->handle_idr);
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 int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
401 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
403 tcf_exts_destroy(&n->exts);
404 tcf_exts_put_net(&n->exts);
405 if (ht && --ht->refcnt == 0)
407 #ifdef CONFIG_CLS_U32_PERF
411 #ifdef CONFIG_CLS_U32_MARK
413 free_percpu(n->pcpu_success);
419 /* u32_delete_key_rcu should be called when free'ing a copied
420 * version of a tc_u_knode obtained from u32_init_knode(). When
421 * copies are obtained from u32_init_knode() the statistics are
422 * shared between the old and new copies to allow readers to
423 * continue to update the statistics during the copy. To support
424 * this the u32_delete_key_rcu variant does not free the percpu
427 static void u32_delete_key_work(struct work_struct *work)
429 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
432 u32_destroy_key(key->tp, key, false);
436 static void u32_delete_key_rcu(struct rcu_head *rcu)
438 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
440 INIT_WORK(&key->work, u32_delete_key_work);
441 tcf_queue_work(&key->work);
444 /* u32_delete_key_freepf_rcu is the rcu callback variant
445 * that free's the entire structure including the statistics
446 * percpu variables. Only use this if the key is not a copy
447 * returned by u32_init_knode(). See u32_delete_key_rcu()
448 * for the variant that should be used with keys return from
451 static void u32_delete_key_freepf_work(struct work_struct *work)
453 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
456 u32_destroy_key(key->tp, key, true);
460 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
462 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
464 INIT_WORK(&key->work, u32_delete_key_freepf_work);
465 tcf_queue_work(&key->work);
468 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
470 struct tc_u_knode __rcu **kp;
471 struct tc_u_knode *pkp;
472 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
475 kp = &ht->ht[TC_U32_HASH(key->handle)];
476 for (pkp = rtnl_dereference(*kp); pkp;
477 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
479 RCU_INIT_POINTER(*kp, key->next);
481 tcf_unbind_filter(tp, &key->res);
482 tcf_exts_get_net(&key->exts);
483 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
492 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
493 struct netlink_ext_ack *extack)
495 struct tcf_block *block = tp->chain->block;
496 struct tc_cls_u32_offload cls_u32 = {};
498 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
499 cls_u32.command = TC_CLSU32_DELETE_HNODE;
500 cls_u32.hnode.divisor = h->divisor;
501 cls_u32.hnode.handle = h->handle;
502 cls_u32.hnode.prio = h->prio;
504 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
507 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
508 u32 flags, struct netlink_ext_ack *extack)
510 struct tcf_block *block = tp->chain->block;
511 struct tc_cls_u32_offload cls_u32 = {};
512 bool skip_sw = tc_skip_sw(flags);
513 bool offloaded = false;
516 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
517 cls_u32.command = TC_CLSU32_NEW_HNODE;
518 cls_u32.hnode.divisor = h->divisor;
519 cls_u32.hnode.handle = h->handle;
520 cls_u32.hnode.prio = h->prio;
522 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
524 u32_clear_hw_hnode(tp, h, NULL);
526 } else if (err > 0) {
530 if (skip_sw && !offloaded)
536 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
537 struct netlink_ext_ack *extack)
539 struct tcf_block *block = tp->chain->block;
540 struct tc_cls_u32_offload cls_u32 = {};
542 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
543 cls_u32.command = TC_CLSU32_DELETE_KNODE;
544 cls_u32.knode.handle = n->handle;
546 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
547 tcf_block_offload_dec(block, &n->flags);
550 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
551 u32 flags, struct netlink_ext_ack *extack)
553 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
554 struct tcf_block *block = tp->chain->block;
555 struct tc_cls_u32_offload cls_u32 = {};
556 bool skip_sw = tc_skip_sw(flags);
559 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
560 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
561 cls_u32.knode.handle = n->handle;
562 cls_u32.knode.fshift = n->fshift;
563 #ifdef CONFIG_CLS_U32_MARK
564 cls_u32.knode.val = n->val;
565 cls_u32.knode.mask = n->mask;
567 cls_u32.knode.val = 0;
568 cls_u32.knode.mask = 0;
570 cls_u32.knode.sel = &n->sel;
571 cls_u32.knode.exts = &n->exts;
573 cls_u32.knode.link_handle = ht->handle;
575 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
577 u32_remove_hw_knode(tp, n, NULL);
579 } else if (err > 0) {
580 tcf_block_offload_inc(block, &n->flags);
583 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
589 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
590 struct netlink_ext_ack *extack)
592 struct tc_u_knode *n;
595 for (h = 0; h <= ht->divisor; h++) {
596 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
597 RCU_INIT_POINTER(ht->ht[h],
598 rtnl_dereference(n->next));
599 tcf_unbind_filter(tp, &n->res);
600 u32_remove_hw_knode(tp, n, extack);
601 idr_remove_ext(&ht->handle_idr, n->handle);
602 if (tcf_exts_get_net(&n->exts))
603 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
605 u32_destroy_key(n->tp, n, true);
610 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
611 struct netlink_ext_ack *extack)
613 struct tc_u_common *tp_c = tp->data;
614 struct tc_u_hnode __rcu **hn;
615 struct tc_u_hnode *phn;
619 u32_clear_hnode(tp, ht, extack);
622 for (phn = rtnl_dereference(*hn);
624 hn = &phn->next, phn = rtnl_dereference(*hn)) {
626 u32_clear_hw_hnode(tp, ht, extack);
627 idr_destroy(&ht->handle_idr);
628 idr_remove_ext(&tp_c->handle_idr, ht->handle);
629 RCU_INIT_POINTER(*hn, ht->next);
638 static bool ht_empty(struct tc_u_hnode *ht)
642 for (h = 0; h <= ht->divisor; h++)
643 if (rcu_access_pointer(ht->ht[h]))
649 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
651 struct tc_u_common *tp_c = tp->data;
652 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
654 WARN_ON(root_ht == NULL);
656 if (root_ht && --root_ht->refcnt == 0)
657 u32_destroy_hnode(tp, root_ht, extack);
659 if (--tp_c->refcnt == 0) {
660 struct tc_u_hnode *ht;
662 hlist_del(&tp_c->hnode);
664 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
665 u32_clear_hnode(tp, ht, extack);
666 RCU_INIT_POINTER(tp_c->hlist, ht->next);
668 /* u32_destroy_key() will later free ht for us, if it's
669 * still referenced by some knode
671 if (--ht->refcnt == 0)
675 idr_destroy(&tp_c->handle_idr);
682 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
683 struct netlink_ext_ack *extack)
685 struct tc_u_hnode *ht = arg;
686 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
687 struct tc_u_common *tp_c = tp->data;
693 if (TC_U32_KEY(ht->handle)) {
694 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
695 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
700 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
704 if (ht->refcnt == 1) {
706 u32_destroy_hnode(tp, ht, extack);
708 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
715 if (root_ht->refcnt > 1) {
719 if (root_ht->refcnt == 1) {
720 if (!ht_empty(root_ht)) {
727 if (tp_c->refcnt > 1) {
732 if (tp_c->refcnt == 1) {
733 struct tc_u_hnode *ht;
735 for (ht = rtnl_dereference(tp_c->hlist);
737 ht = rtnl_dereference(ht->next))
748 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
750 unsigned long idr_index;
751 u32 start = htid | 0x800;
752 u32 max = htid | 0xFFF;
755 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
756 start, max + 1, GFP_KERNEL)) {
757 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
758 min + 1, max + 1, GFP_KERNEL))
762 return (u32)idr_index;
765 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
766 [TCA_U32_CLASSID] = { .type = NLA_U32 },
767 [TCA_U32_HASH] = { .type = NLA_U32 },
768 [TCA_U32_LINK] = { .type = NLA_U32 },
769 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
770 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
771 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
772 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
773 [TCA_U32_FLAGS] = { .type = NLA_U32 },
776 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
777 unsigned long base, struct tc_u_hnode *ht,
778 struct tc_u_knode *n, struct nlattr **tb,
779 struct nlattr *est, bool ovr,
780 struct netlink_ext_ack *extack)
784 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack);
788 if (tb[TCA_U32_LINK]) {
789 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
790 struct tc_u_hnode *ht_down = NULL, *ht_old;
792 if (TC_U32_KEY(handle)) {
793 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
798 ht_down = u32_lookup_ht(ht->tp_c, handle);
801 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
807 ht_old = rtnl_dereference(n->ht_down);
808 rcu_assign_pointer(n->ht_down, ht_down);
813 if (tb[TCA_U32_CLASSID]) {
814 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
815 tcf_bind_filter(tp, &n->res, base);
818 #ifdef CONFIG_NET_CLS_IND
819 if (tb[TCA_U32_INDEV]) {
821 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
830 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
831 struct tc_u_knode *n)
833 struct tc_u_knode __rcu **ins;
834 struct tc_u_knode *pins;
835 struct tc_u_hnode *ht;
837 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
838 ht = rtnl_dereference(tp->root);
840 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
842 ins = &ht->ht[TC_U32_HASH(n->handle)];
844 /* The node must always exist for it to be replaced if this is not the
845 * case then something went very wrong elsewhere.
847 for (pins = rtnl_dereference(*ins); ;
848 ins = &pins->next, pins = rtnl_dereference(*ins))
849 if (pins->handle == n->handle)
852 idr_replace_ext(&ht->handle_idr, n, n->handle);
853 RCU_INIT_POINTER(n->next, pins->next);
854 rcu_assign_pointer(*ins, n);
857 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
858 struct tc_u_knode *n)
860 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
861 struct tc_u32_sel *s = &n->sel;
862 struct tc_u_knode *new;
864 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
870 RCU_INIT_POINTER(new->next, n->next);
871 new->handle = n->handle;
872 RCU_INIT_POINTER(new->ht_up, n->ht_up);
874 #ifdef CONFIG_NET_CLS_IND
875 new->ifindex = n->ifindex;
877 new->fshift = n->fshift;
879 new->flags = n->flags;
880 RCU_INIT_POINTER(new->ht_down, ht);
882 /* bump reference count as long as we hold pointer to structure */
886 #ifdef CONFIG_CLS_U32_PERF
887 /* Statistics may be incremented by readers during update
888 * so we must keep them in tact. When the node is later destroyed
889 * a special destroy call must be made to not free the pf memory.
894 #ifdef CONFIG_CLS_U32_MARK
897 /* Similarly success statistics must be moved as pointers */
898 new->pcpu_success = n->pcpu_success;
901 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
903 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
911 static int u32_change(struct net *net, struct sk_buff *in_skb,
912 struct tcf_proto *tp, unsigned long base, u32 handle,
913 struct nlattr **tca, void **arg, bool ovr,
914 struct netlink_ext_ack *extack)
916 struct tc_u_common *tp_c = tp->data;
917 struct tc_u_hnode *ht;
918 struct tc_u_knode *n;
919 struct tc_u32_sel *s;
920 struct nlattr *opt = tca[TCA_OPTIONS];
921 struct nlattr *tb[TCA_U32_MAX + 1];
924 #ifdef CONFIG_CLS_U32_PERF
930 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
937 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack);
941 if (tb[TCA_U32_FLAGS]) {
942 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
943 if (!tc_flags_valid(flags)) {
944 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
951 struct tc_u_knode *new;
953 if (TC_U32_KEY(n->handle) == 0) {
954 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
958 if (n->flags != flags) {
959 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
963 new = u32_init_knode(tp, n);
967 err = u32_set_parms(net, tp, base,
968 rtnl_dereference(n->ht_up), new, tb,
969 tca[TCA_RATE], ovr, extack);
972 u32_destroy_key(tp, new, false);
976 err = u32_replace_hw_knode(tp, new, flags, extack);
978 u32_destroy_key(tp, new, false);
982 if (!tc_in_hw(new->flags))
983 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
985 u32_replace_knode(tp, tp_c, new);
986 tcf_unbind_filter(tp, &n->res);
987 tcf_exts_get_net(&n->exts);
988 call_rcu(&n->rcu, u32_delete_key_rcu);
992 if (tb[TCA_U32_DIVISOR]) {
993 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
995 if (--divisor > 0x100) {
996 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
999 if (TC_U32_KEY(handle)) {
1000 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
1003 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
1007 handle = gen_new_htid(tp->data, ht);
1013 err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL,
1014 handle, handle + 1, GFP_KERNEL);
1022 ht->divisor = divisor;
1023 ht->handle = handle;
1024 ht->prio = tp->prio;
1025 idr_init(&ht->handle_idr);
1028 err = u32_replace_hw_hnode(tp, ht, flags, extack);
1030 idr_remove_ext(&tp_c->handle_idr, handle);
1035 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1036 rcu_assign_pointer(tp_c->hlist, ht);
1042 if (tb[TCA_U32_HASH]) {
1043 htid = nla_get_u32(tb[TCA_U32_HASH]);
1044 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1045 ht = rtnl_dereference(tp->root);
1048 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1050 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1055 ht = rtnl_dereference(tp->root);
1059 if (ht->divisor < TC_U32_HASH(htid)) {
1060 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1065 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1066 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1069 handle = htid | TC_U32_NODE(handle);
1070 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL,
1076 handle = gen_new_kid(ht, htid);
1078 if (tb[TCA_U32_SEL] == NULL) {
1079 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1084 s = nla_data(tb[TCA_U32_SEL]);
1086 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1092 #ifdef CONFIG_CLS_U32_PERF
1093 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1094 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1101 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1102 RCU_INIT_POINTER(n->ht_up, ht);
1104 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1108 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1112 #ifdef CONFIG_CLS_U32_MARK
1113 n->pcpu_success = alloc_percpu(u32);
1114 if (!n->pcpu_success) {
1119 if (tb[TCA_U32_MARK]) {
1120 struct tc_u32_mark *mark;
1122 mark = nla_data(tb[TCA_U32_MARK]);
1124 n->mask = mark->mask;
1128 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr,
1131 struct tc_u_knode __rcu **ins;
1132 struct tc_u_knode *pins;
1134 err = u32_replace_hw_knode(tp, n, flags, extack);
1138 if (!tc_in_hw(n->flags))
1139 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1141 ins = &ht->ht[TC_U32_HASH(handle)];
1142 for (pins = rtnl_dereference(*ins); pins;
1143 ins = &pins->next, pins = rtnl_dereference(*ins))
1144 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1147 RCU_INIT_POINTER(n->next, pins);
1148 rcu_assign_pointer(*ins, n);
1154 #ifdef CONFIG_CLS_U32_MARK
1155 free_percpu(n->pcpu_success);
1159 tcf_exts_destroy(&n->exts);
1160 #ifdef CONFIG_CLS_U32_PERF
1166 idr_remove_ext(&ht->handle_idr, handle);
1170 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1172 struct tc_u_common *tp_c = tp->data;
1173 struct tc_u_hnode *ht;
1174 struct tc_u_knode *n;
1180 for (ht = rtnl_dereference(tp_c->hlist);
1182 ht = rtnl_dereference(ht->next)) {
1183 if (ht->prio != tp->prio)
1185 if (arg->count >= arg->skip) {
1186 if (arg->fn(tp, ht, arg) < 0) {
1192 for (h = 0; h <= ht->divisor; h++) {
1193 for (n = rtnl_dereference(ht->ht[h]);
1195 n = rtnl_dereference(n->next)) {
1196 if (arg->count < arg->skip) {
1200 if (arg->fn(tp, n, arg) < 0) {
1210 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1212 struct tc_u_knode *n = fh;
1214 if (n && n->res.classid == classid)
1218 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1219 struct sk_buff *skb, struct tcmsg *t)
1221 struct tc_u_knode *n = fh;
1222 struct tc_u_hnode *ht_up, *ht_down;
1223 struct nlattr *nest;
1228 t->tcm_handle = n->handle;
1230 nest = nla_nest_start(skb, TCA_OPTIONS);
1232 goto nla_put_failure;
1234 if (TC_U32_KEY(n->handle) == 0) {
1235 struct tc_u_hnode *ht = fh;
1236 u32 divisor = ht->divisor + 1;
1238 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1239 goto nla_put_failure;
1241 #ifdef CONFIG_CLS_U32_PERF
1242 struct tc_u32_pcnt *gpf;
1246 if (nla_put(skb, TCA_U32_SEL,
1247 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1249 goto nla_put_failure;
1251 ht_up = rtnl_dereference(n->ht_up);
1253 u32 htid = n->handle & 0xFFFFF000;
1254 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1255 goto nla_put_failure;
1257 if (n->res.classid &&
1258 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1259 goto nla_put_failure;
1261 ht_down = rtnl_dereference(n->ht_down);
1263 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1264 goto nla_put_failure;
1266 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1267 goto nla_put_failure;
1269 #ifdef CONFIG_CLS_U32_MARK
1270 if ((n->val || n->mask)) {
1271 struct tc_u32_mark mark = {.val = n->val,
1276 for_each_possible_cpu(cpum) {
1277 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1279 mark.success += cnt;
1282 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1283 goto nla_put_failure;
1287 if (tcf_exts_dump(skb, &n->exts) < 0)
1288 goto nla_put_failure;
1290 #ifdef CONFIG_NET_CLS_IND
1292 struct net_device *dev;
1293 dev = __dev_get_by_index(net, n->ifindex);
1294 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1295 goto nla_put_failure;
1298 #ifdef CONFIG_CLS_U32_PERF
1299 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1300 n->sel.nkeys * sizeof(u64),
1303 goto nla_put_failure;
1305 for_each_possible_cpu(cpu) {
1307 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1309 gpf->rcnt += pf->rcnt;
1310 gpf->rhit += pf->rhit;
1311 for (i = 0; i < n->sel.nkeys; i++)
1312 gpf->kcnts[i] += pf->kcnts[i];
1315 if (nla_put_64bit(skb, TCA_U32_PCNT,
1316 sizeof(struct tc_u32_pcnt) +
1317 n->sel.nkeys * sizeof(u64),
1318 gpf, TCA_U32_PAD)) {
1320 goto nla_put_failure;
1326 nla_nest_end(skb, nest);
1328 if (TC_U32_KEY(n->handle))
1329 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1330 goto nla_put_failure;
1334 nla_nest_cancel(skb, nest);
1338 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1340 .classify = u32_classify,
1342 .destroy = u32_destroy,
1344 .change = u32_change,
1345 .delete = u32_delete,
1348 .bind_class = u32_bind_class,
1349 .owner = THIS_MODULE,
1352 static int __init init_u32(void)
1356 pr_info("u32 classifier\n");
1357 #ifdef CONFIG_CLS_U32_PERF
1358 pr_info(" Performance counters on\n");
1360 #ifdef CONFIG_NET_CLS_IND
1361 pr_info(" input device check on\n");
1363 #ifdef CONFIG_NET_CLS_ACT
1364 pr_info(" Actions configured\n");
1366 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1367 sizeof(struct hlist_head),
1369 if (!tc_u_common_hash)
1372 for (i = 0; i < U32_HASH_SIZE; i++)
1373 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1375 ret = register_tcf_proto_ops(&cls_u32_ops);
1377 kvfree(tc_u_common_hash);
1381 static void __exit exit_u32(void)
1383 unregister_tcf_proto_ops(&cls_u32_ops);
1384 kvfree(tc_u_common_hash);
1387 module_init(init_u32)
1388 module_exit(exit_u32)
1389 MODULE_LICENSE("GPL");