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 unsigned int in_hw_count;
66 #ifdef CONFIG_CLS_U32_MARK
69 u32 __percpu *pcpu_success;
71 struct rcu_work rwork;
72 /* The 'sel' field MUST be the last field in structure to allow for
73 * tc_u32_keys allocated at end of structure.
75 struct tc_u32_sel sel;
79 struct tc_u_hnode __rcu *next;
84 struct idr handle_idr;
88 /* The 'ht' field MUST be the last field in structure to allow for
89 * more entries allocated at end of structure.
91 struct tc_u_knode __rcu *ht[1];
95 struct tc_u_hnode __rcu *hlist;
98 struct idr handle_idr;
99 struct hlist_node hnode;
103 static inline unsigned int u32_hash_fold(__be32 key,
104 const struct tc_u32_sel *sel,
107 unsigned int h = ntohl(key & sel->hmask) >> fshift;
112 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
113 struct tcf_result *res)
116 struct tc_u_knode *knode;
118 } stack[TC_U32_MAXDEPTH];
120 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
121 unsigned int off = skb_network_offset(skb);
122 struct tc_u_knode *n;
126 #ifdef CONFIG_CLS_U32_PERF
132 n = rcu_dereference_bh(ht->ht[sel]);
136 struct tc_u32_key *key = n->sel.keys;
138 #ifdef CONFIG_CLS_U32_PERF
139 __this_cpu_inc(n->pf->rcnt);
143 if (tc_skip_sw(n->flags)) {
144 n = rcu_dereference_bh(n->next);
148 #ifdef CONFIG_CLS_U32_MARK
149 if ((skb->mark & n->mask) != n->val) {
150 n = rcu_dereference_bh(n->next);
153 __this_cpu_inc(*n->pcpu_success);
157 for (i = n->sel.nkeys; i > 0; i--, key++) {
158 int toff = off + key->off + (off2 & key->offmask);
161 if (skb_headroom(skb) + toff > INT_MAX)
164 data = skb_header_pointer(skb, toff, 4, &hdata);
167 if ((*data ^ key->val) & key->mask) {
168 n = rcu_dereference_bh(n->next);
171 #ifdef CONFIG_CLS_U32_PERF
172 __this_cpu_inc(n->pf->kcnts[j]);
177 ht = rcu_dereference_bh(n->ht_down);
180 if (n->sel.flags & TC_U32_TERMINAL) {
183 #ifdef CONFIG_NET_CLS_IND
184 if (!tcf_match_indev(skb, n->ifindex)) {
185 n = rcu_dereference_bh(n->next);
189 #ifdef CONFIG_CLS_U32_PERF
190 __this_cpu_inc(n->pf->rhit);
192 r = tcf_exts_exec(skb, &n->exts, res);
194 n = rcu_dereference_bh(n->next);
200 n = rcu_dereference_bh(n->next);
205 if (sdepth >= TC_U32_MAXDEPTH)
207 stack[sdepth].knode = n;
208 stack[sdepth].off = off;
211 ht = rcu_dereference_bh(n->ht_down);
216 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
220 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
223 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
226 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
227 off2 = n->sel.off + 3;
228 if (n->sel.flags & TC_U32_VAROFFSET) {
231 data = skb_header_pointer(skb,
236 off2 += ntohs(n->sel.offmask & *data) >>
241 if (n->sel.flags & TC_U32_EAT) {
252 n = stack[sdepth].knode;
253 ht = rcu_dereference_bh(n->ht_up);
254 off = stack[sdepth].off;
261 net_warn_ratelimited("cls_u32: dead loop\n");
265 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
267 struct tc_u_hnode *ht;
269 for (ht = rtnl_dereference(tp_c->hlist);
271 ht = rtnl_dereference(ht->next))
272 if (ht->handle == handle)
278 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
281 struct tc_u_knode *n = NULL;
283 sel = TC_U32_HASH(handle);
284 if (sel > ht->divisor)
287 for (n = rtnl_dereference(ht->ht[sel]);
289 n = rtnl_dereference(n->next))
290 if (n->handle == handle)
297 static void *u32_get(struct tcf_proto *tp, u32 handle)
299 struct tc_u_hnode *ht;
300 struct tc_u_common *tp_c = tp->data;
302 if (TC_U32_HTID(handle) == TC_U32_ROOT)
303 ht = rtnl_dereference(tp->root);
305 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
310 if (TC_U32_KEY(handle) == 0)
313 return u32_lookup_key(ht, handle);
316 /* Protected by rtnl lock */
317 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
319 int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
322 return (id | 0x800U) << 20;
325 static struct hlist_head *tc_u_common_hash;
327 #define U32_HASH_SHIFT 10
328 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
330 static void *tc_u_common_ptr(const struct tcf_proto *tp)
332 struct tcf_block *block = tp->chain->block;
334 /* The block sharing is currently supported only
335 * for classless qdiscs. In that case we use block
336 * for tc_u_common identification. In case the
337 * block is not shared, block->q is a valid pointer
338 * and we can use that. That works for classful qdiscs.
340 if (tcf_block_shared(block))
346 static struct hlist_head *tc_u_hash(void *key)
348 return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
351 static struct tc_u_common *tc_u_common_find(void *key)
353 struct tc_u_common *tc;
354 hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
361 static int u32_init(struct tcf_proto *tp)
363 struct tc_u_hnode *root_ht;
364 void *key = tc_u_common_ptr(tp);
365 struct tc_u_common *tp_c = tc_u_common_find(key);
367 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
372 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
373 root_ht->prio = tp->prio;
374 root_ht->is_root = true;
375 idr_init(&root_ht->handle_idr);
378 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
384 INIT_HLIST_NODE(&tp_c->hnode);
385 idr_init(&tp_c->handle_idr);
387 hlist_add_head(&tp_c->hnode, tc_u_hash(key));
391 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
392 rcu_assign_pointer(tp_c->hlist, root_ht);
395 rcu_assign_pointer(tp->root, root_ht);
400 static int u32_destroy_key(struct tc_u_knode *n, bool free_pf)
402 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
404 tcf_exts_destroy(&n->exts);
405 tcf_exts_put_net(&n->exts);
406 if (ht && --ht->refcnt == 0)
408 #ifdef CONFIG_CLS_U32_PERF
412 #ifdef CONFIG_CLS_U32_MARK
414 free_percpu(n->pcpu_success);
420 /* u32_delete_key_rcu should be called when free'ing a copied
421 * version of a tc_u_knode obtained from u32_init_knode(). When
422 * copies are obtained from u32_init_knode() the statistics are
423 * shared between the old and new copies to allow readers to
424 * continue to update the statistics during the copy. To support
425 * this the u32_delete_key_rcu variant does not free the percpu
428 static void u32_delete_key_work(struct work_struct *work)
430 struct tc_u_knode *key = container_of(to_rcu_work(work),
434 u32_destroy_key(key, false);
438 /* u32_delete_key_freepf_rcu is the rcu callback variant
439 * that free's the entire structure including the statistics
440 * percpu variables. Only use this if the key is not a copy
441 * returned by u32_init_knode(). See u32_delete_key_rcu()
442 * for the variant that should be used with keys return from
445 static void u32_delete_key_freepf_work(struct work_struct *work)
447 struct tc_u_knode *key = container_of(to_rcu_work(work),
451 u32_destroy_key(key, true);
455 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
457 struct tc_u_common *tp_c = tp->data;
458 struct tc_u_knode __rcu **kp;
459 struct tc_u_knode *pkp;
460 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
463 kp = &ht->ht[TC_U32_HASH(key->handle)];
464 for (pkp = rtnl_dereference(*kp); pkp;
465 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
467 RCU_INIT_POINTER(*kp, key->next);
470 tcf_unbind_filter(tp, &key->res);
471 idr_remove(&ht->handle_idr, key->handle);
472 tcf_exts_get_net(&key->exts);
473 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
482 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
483 struct netlink_ext_ack *extack)
485 struct tcf_block *block = tp->chain->block;
486 struct tc_cls_u32_offload cls_u32 = {};
488 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
489 cls_u32.command = TC_CLSU32_DELETE_HNODE;
490 cls_u32.hnode.divisor = h->divisor;
491 cls_u32.hnode.handle = h->handle;
492 cls_u32.hnode.prio = h->prio;
494 tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
497 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
498 u32 flags, struct netlink_ext_ack *extack)
500 struct tcf_block *block = tp->chain->block;
501 struct tc_cls_u32_offload cls_u32 = {};
502 bool skip_sw = tc_skip_sw(flags);
503 bool offloaded = false;
506 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
507 cls_u32.command = TC_CLSU32_NEW_HNODE;
508 cls_u32.hnode.divisor = h->divisor;
509 cls_u32.hnode.handle = h->handle;
510 cls_u32.hnode.prio = h->prio;
512 err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
514 u32_clear_hw_hnode(tp, h, NULL);
516 } else if (err > 0) {
520 if (skip_sw && !offloaded)
526 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
527 struct netlink_ext_ack *extack)
529 struct tcf_block *block = tp->chain->block;
530 struct tc_cls_u32_offload cls_u32 = {};
532 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
533 cls_u32.command = TC_CLSU32_DELETE_KNODE;
534 cls_u32.knode.handle = n->handle;
536 tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
537 tcf_block_offload_dec(block, &n->flags);
540 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
541 u32 flags, struct netlink_ext_ack *extack)
543 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
544 struct tcf_block *block = tp->chain->block;
545 struct tc_cls_u32_offload cls_u32 = {};
546 bool skip_sw = tc_skip_sw(flags);
549 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
550 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
551 cls_u32.knode.handle = n->handle;
552 cls_u32.knode.fshift = n->fshift;
553 #ifdef CONFIG_CLS_U32_MARK
554 cls_u32.knode.val = n->val;
555 cls_u32.knode.mask = n->mask;
557 cls_u32.knode.val = 0;
558 cls_u32.knode.mask = 0;
560 cls_u32.knode.sel = &n->sel;
561 cls_u32.knode.res = &n->res;
562 cls_u32.knode.exts = &n->exts;
564 cls_u32.knode.link_handle = ht->handle;
566 err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
568 u32_remove_hw_knode(tp, n, NULL);
570 } else if (err > 0) {
571 n->in_hw_count = err;
572 tcf_block_offload_inc(block, &n->flags);
575 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
581 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
582 struct netlink_ext_ack *extack)
584 struct tc_u_common *tp_c = tp->data;
585 struct tc_u_knode *n;
588 for (h = 0; h <= ht->divisor; h++) {
589 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
590 RCU_INIT_POINTER(ht->ht[h],
591 rtnl_dereference(n->next));
593 tcf_unbind_filter(tp, &n->res);
594 u32_remove_hw_knode(tp, n, extack);
595 idr_remove(&ht->handle_idr, n->handle);
596 if (tcf_exts_get_net(&n->exts))
597 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
599 u32_destroy_key(n, true);
604 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
605 struct netlink_ext_ack *extack)
607 struct tc_u_common *tp_c = tp->data;
608 struct tc_u_hnode __rcu **hn;
609 struct tc_u_hnode *phn;
611 WARN_ON(--ht->refcnt);
613 u32_clear_hnode(tp, ht, extack);
616 for (phn = rtnl_dereference(*hn);
618 hn = &phn->next, phn = rtnl_dereference(*hn)) {
620 u32_clear_hw_hnode(tp, ht, extack);
621 idr_destroy(&ht->handle_idr);
622 idr_remove(&tp_c->handle_idr, ht->handle);
623 RCU_INIT_POINTER(*hn, ht->next);
632 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
633 struct netlink_ext_ack *extack)
635 struct tc_u_common *tp_c = tp->data;
636 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
638 WARN_ON(root_ht == NULL);
640 if (root_ht && --root_ht->refcnt == 1)
641 u32_destroy_hnode(tp, root_ht, extack);
643 if (--tp_c->refcnt == 0) {
644 struct tc_u_hnode *ht;
646 hlist_del(&tp_c->hnode);
648 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
649 u32_clear_hnode(tp, ht, extack);
650 RCU_INIT_POINTER(tp_c->hlist, ht->next);
652 /* u32_destroy_key() will later free ht for us, if it's
653 * still referenced by some knode
655 if (--ht->refcnt == 0)
659 idr_destroy(&tp_c->handle_idr);
666 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
667 bool rtnl_held, struct netlink_ext_ack *extack)
669 struct tc_u_hnode *ht = arg;
670 struct tc_u_common *tp_c = tp->data;
673 if (TC_U32_KEY(ht->handle)) {
674 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
675 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
680 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
684 if (ht->refcnt == 1) {
685 u32_destroy_hnode(tp, ht, extack);
687 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
692 *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
696 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
698 u32 index = htid | 0x800;
699 u32 max = htid | 0xFFF;
701 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
703 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
711 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
712 [TCA_U32_CLASSID] = { .type = NLA_U32 },
713 [TCA_U32_HASH] = { .type = NLA_U32 },
714 [TCA_U32_LINK] = { .type = NLA_U32 },
715 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
716 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
717 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
718 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
719 [TCA_U32_FLAGS] = { .type = NLA_U32 },
722 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
724 struct tc_u_knode *n, struct nlattr **tb,
725 struct nlattr *est, bool ovr,
726 struct netlink_ext_ack *extack)
730 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
734 if (tb[TCA_U32_LINK]) {
735 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
736 struct tc_u_hnode *ht_down = NULL, *ht_old;
738 if (TC_U32_KEY(handle)) {
739 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
744 ht_down = u32_lookup_ht(tp->data, handle);
747 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
750 if (ht_down->is_root) {
751 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
757 ht_old = rtnl_dereference(n->ht_down);
758 rcu_assign_pointer(n->ht_down, ht_down);
763 if (tb[TCA_U32_CLASSID]) {
764 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
765 tcf_bind_filter(tp, &n->res, base);
768 #ifdef CONFIG_NET_CLS_IND
769 if (tb[TCA_U32_INDEV]) {
771 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
780 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
781 struct tc_u_knode *n)
783 struct tc_u_knode __rcu **ins;
784 struct tc_u_knode *pins;
785 struct tc_u_hnode *ht;
787 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
788 ht = rtnl_dereference(tp->root);
790 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
792 ins = &ht->ht[TC_U32_HASH(n->handle)];
794 /* The node must always exist for it to be replaced if this is not the
795 * case then something went very wrong elsewhere.
797 for (pins = rtnl_dereference(*ins); ;
798 ins = &pins->next, pins = rtnl_dereference(*ins))
799 if (pins->handle == n->handle)
802 idr_replace(&ht->handle_idr, n, n->handle);
803 RCU_INIT_POINTER(n->next, pins->next);
804 rcu_assign_pointer(*ins, n);
807 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
808 struct tc_u_knode *n)
810 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
811 struct tc_u32_sel *s = &n->sel;
812 struct tc_u_knode *new;
814 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
820 RCU_INIT_POINTER(new->next, n->next);
821 new->handle = n->handle;
822 RCU_INIT_POINTER(new->ht_up, n->ht_up);
824 #ifdef CONFIG_NET_CLS_IND
825 new->ifindex = n->ifindex;
827 new->fshift = n->fshift;
829 new->flags = n->flags;
830 RCU_INIT_POINTER(new->ht_down, ht);
832 /* bump reference count as long as we hold pointer to structure */
836 #ifdef CONFIG_CLS_U32_PERF
837 /* Statistics may be incremented by readers during update
838 * so we must keep them in tact. When the node is later destroyed
839 * a special destroy call must be made to not free the pf memory.
844 #ifdef CONFIG_CLS_U32_MARK
847 /* Similarly success statistics must be moved as pointers */
848 new->pcpu_success = n->pcpu_success;
850 memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
852 if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
860 static int u32_change(struct net *net, struct sk_buff *in_skb,
861 struct tcf_proto *tp, unsigned long base, u32 handle,
862 struct nlattr **tca, void **arg, bool ovr, bool rtnl_held,
863 struct netlink_ext_ack *extack)
865 struct tc_u_common *tp_c = tp->data;
866 struct tc_u_hnode *ht;
867 struct tc_u_knode *n;
868 struct tc_u32_sel *s;
869 struct nlattr *opt = tca[TCA_OPTIONS];
870 struct nlattr *tb[TCA_U32_MAX + 1];
874 #ifdef CONFIG_CLS_U32_PERF
880 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
887 err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
892 if (tb[TCA_U32_FLAGS]) {
893 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
894 if (!tc_flags_valid(flags)) {
895 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
902 struct tc_u_knode *new;
904 if (TC_U32_KEY(n->handle) == 0) {
905 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
909 if ((n->flags ^ flags) &
910 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
911 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
915 new = u32_init_knode(net, tp, n);
919 err = u32_set_parms(net, tp, base, new, tb,
920 tca[TCA_RATE], ovr, extack);
923 u32_destroy_key(new, false);
927 err = u32_replace_hw_knode(tp, new, flags, extack);
929 u32_destroy_key(new, false);
933 if (!tc_in_hw(new->flags))
934 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
936 u32_replace_knode(tp, tp_c, new);
937 tcf_unbind_filter(tp, &n->res);
938 tcf_exts_get_net(&n->exts);
939 tcf_queue_work(&n->rwork, u32_delete_key_work);
943 if (tb[TCA_U32_DIVISOR]) {
944 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
946 if (!is_power_of_2(divisor)) {
947 NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
950 if (divisor-- > 0x100) {
951 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
954 if (TC_U32_KEY(handle)) {
955 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
958 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
962 handle = gen_new_htid(tp->data, ht);
968 err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
976 ht->divisor = divisor;
979 idr_init(&ht->handle_idr);
982 err = u32_replace_hw_hnode(tp, ht, flags, extack);
984 idr_remove(&tp_c->handle_idr, handle);
989 RCU_INIT_POINTER(ht->next, tp_c->hlist);
990 rcu_assign_pointer(tp_c->hlist, ht);
996 if (tb[TCA_U32_HASH]) {
997 htid = nla_get_u32(tb[TCA_U32_HASH]);
998 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
999 ht = rtnl_dereference(tp->root);
1002 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1004 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1009 ht = rtnl_dereference(tp->root);
1013 if (ht->divisor < TC_U32_HASH(htid)) {
1014 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1019 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1020 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1023 handle = htid | TC_U32_NODE(handle);
1024 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1029 handle = gen_new_kid(ht, htid);
1031 if (tb[TCA_U32_SEL] == NULL) {
1032 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1037 s = nla_data(tb[TCA_U32_SEL]);
1038 sel_size = struct_size(s, keys, s->nkeys);
1039 if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1044 n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1050 #ifdef CONFIG_CLS_U32_PERF
1051 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1052 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1059 memcpy(&n->sel, s, sel_size);
1060 RCU_INIT_POINTER(n->ht_up, ht);
1062 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1065 err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1069 #ifdef CONFIG_CLS_U32_MARK
1070 n->pcpu_success = alloc_percpu(u32);
1071 if (!n->pcpu_success) {
1076 if (tb[TCA_U32_MARK]) {
1077 struct tc_u32_mark *mark;
1079 mark = nla_data(tb[TCA_U32_MARK]);
1081 n->mask = mark->mask;
1085 err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1088 struct tc_u_knode __rcu **ins;
1089 struct tc_u_knode *pins;
1091 err = u32_replace_hw_knode(tp, n, flags, extack);
1095 if (!tc_in_hw(n->flags))
1096 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1098 ins = &ht->ht[TC_U32_HASH(handle)];
1099 for (pins = rtnl_dereference(*ins); pins;
1100 ins = &pins->next, pins = rtnl_dereference(*ins))
1101 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1104 RCU_INIT_POINTER(n->next, pins);
1105 rcu_assign_pointer(*ins, n);
1112 #ifdef CONFIG_CLS_U32_MARK
1113 free_percpu(n->pcpu_success);
1117 tcf_exts_destroy(&n->exts);
1118 #ifdef CONFIG_CLS_U32_PERF
1124 idr_remove(&ht->handle_idr, handle);
1128 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1131 struct tc_u_common *tp_c = tp->data;
1132 struct tc_u_hnode *ht;
1133 struct tc_u_knode *n;
1139 for (ht = rtnl_dereference(tp_c->hlist);
1141 ht = rtnl_dereference(ht->next)) {
1142 if (ht->prio != tp->prio)
1144 if (arg->count >= arg->skip) {
1145 if (arg->fn(tp, ht, arg) < 0) {
1151 for (h = 0; h <= ht->divisor; h++) {
1152 for (n = rtnl_dereference(ht->ht[h]);
1154 n = rtnl_dereference(n->next)) {
1155 if (arg->count < arg->skip) {
1159 if (arg->fn(tp, n, arg) < 0) {
1169 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1170 bool add, tc_setup_cb_t *cb, void *cb_priv,
1171 struct netlink_ext_ack *extack)
1173 struct tc_cls_u32_offload cls_u32 = {};
1176 tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1177 cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1178 cls_u32.hnode.divisor = ht->divisor;
1179 cls_u32.hnode.handle = ht->handle;
1180 cls_u32.hnode.prio = ht->prio;
1182 err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1183 if (err && add && tc_skip_sw(ht->flags))
1189 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1190 bool add, tc_setup_cb_t *cb, void *cb_priv,
1191 struct netlink_ext_ack *extack)
1193 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1194 struct tcf_block *block = tp->chain->block;
1195 struct tc_cls_u32_offload cls_u32 = {};
1198 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1199 cls_u32.command = add ?
1200 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1201 cls_u32.knode.handle = n->handle;
1204 cls_u32.knode.fshift = n->fshift;
1205 #ifdef CONFIG_CLS_U32_MARK
1206 cls_u32.knode.val = n->val;
1207 cls_u32.knode.mask = n->mask;
1209 cls_u32.knode.val = 0;
1210 cls_u32.knode.mask = 0;
1212 cls_u32.knode.sel = &n->sel;
1213 cls_u32.knode.res = &n->res;
1214 cls_u32.knode.exts = &n->exts;
1216 cls_u32.knode.link_handle = ht->handle;
1219 err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1221 if (add && tc_skip_sw(n->flags))
1226 tc_cls_offload_cnt_update(block, &n->in_hw_count, &n->flags, add);
1231 static int u32_reoffload(struct tcf_proto *tp, bool add, tc_setup_cb_t *cb,
1232 void *cb_priv, struct netlink_ext_ack *extack)
1234 struct tc_u_common *tp_c = tp->data;
1235 struct tc_u_hnode *ht;
1236 struct tc_u_knode *n;
1240 for (ht = rtnl_dereference(tp_c->hlist);
1242 ht = rtnl_dereference(ht->next)) {
1243 if (ht->prio != tp->prio)
1246 /* When adding filters to a new dev, try to offload the
1247 * hashtable first. When removing, do the filters before the
1250 if (add && !tc_skip_hw(ht->flags)) {
1251 err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1257 for (h = 0; h <= ht->divisor; h++) {
1258 for (n = rtnl_dereference(ht->ht[h]);
1260 n = rtnl_dereference(n->next)) {
1261 if (tc_skip_hw(n->flags))
1264 err = u32_reoffload_knode(tp, n, add, cb,
1271 if (!add && !tc_skip_hw(ht->flags))
1272 u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1278 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1280 struct tc_u_knode *n = fh;
1282 if (n && n->res.classid == classid)
1286 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1287 struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1289 struct tc_u_knode *n = fh;
1290 struct tc_u_hnode *ht_up, *ht_down;
1291 struct nlattr *nest;
1296 t->tcm_handle = n->handle;
1298 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1300 goto nla_put_failure;
1302 if (TC_U32_KEY(n->handle) == 0) {
1303 struct tc_u_hnode *ht = fh;
1304 u32 divisor = ht->divisor + 1;
1306 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1307 goto nla_put_failure;
1309 #ifdef CONFIG_CLS_U32_PERF
1310 struct tc_u32_pcnt *gpf;
1314 if (nla_put(skb, TCA_U32_SEL,
1315 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1317 goto nla_put_failure;
1319 ht_up = rtnl_dereference(n->ht_up);
1321 u32 htid = n->handle & 0xFFFFF000;
1322 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1323 goto nla_put_failure;
1325 if (n->res.classid &&
1326 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1327 goto nla_put_failure;
1329 ht_down = rtnl_dereference(n->ht_down);
1331 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1332 goto nla_put_failure;
1334 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1335 goto nla_put_failure;
1337 #ifdef CONFIG_CLS_U32_MARK
1338 if ((n->val || n->mask)) {
1339 struct tc_u32_mark mark = {.val = n->val,
1344 for_each_possible_cpu(cpum) {
1345 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1347 mark.success += cnt;
1350 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1351 goto nla_put_failure;
1355 if (tcf_exts_dump(skb, &n->exts) < 0)
1356 goto nla_put_failure;
1358 #ifdef CONFIG_NET_CLS_IND
1360 struct net_device *dev;
1361 dev = __dev_get_by_index(net, n->ifindex);
1362 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1363 goto nla_put_failure;
1366 #ifdef CONFIG_CLS_U32_PERF
1367 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1368 n->sel.nkeys * sizeof(u64),
1371 goto nla_put_failure;
1373 for_each_possible_cpu(cpu) {
1375 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1377 gpf->rcnt += pf->rcnt;
1378 gpf->rhit += pf->rhit;
1379 for (i = 0; i < n->sel.nkeys; i++)
1380 gpf->kcnts[i] += pf->kcnts[i];
1383 if (nla_put_64bit(skb, TCA_U32_PCNT,
1384 sizeof(struct tc_u32_pcnt) +
1385 n->sel.nkeys * sizeof(u64),
1386 gpf, TCA_U32_PAD)) {
1388 goto nla_put_failure;
1394 nla_nest_end(skb, nest);
1396 if (TC_U32_KEY(n->handle))
1397 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1398 goto nla_put_failure;
1402 nla_nest_cancel(skb, nest);
1406 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1408 .classify = u32_classify,
1410 .destroy = u32_destroy,
1412 .change = u32_change,
1413 .delete = u32_delete,
1415 .reoffload = u32_reoffload,
1417 .bind_class = u32_bind_class,
1418 .owner = THIS_MODULE,
1421 static int __init init_u32(void)
1425 pr_info("u32 classifier\n");
1426 #ifdef CONFIG_CLS_U32_PERF
1427 pr_info(" Performance counters on\n");
1429 #ifdef CONFIG_NET_CLS_IND
1430 pr_info(" input device check on\n");
1432 #ifdef CONFIG_NET_CLS_ACT
1433 pr_info(" Actions configured\n");
1435 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1436 sizeof(struct hlist_head),
1438 if (!tc_u_common_hash)
1441 for (i = 0; i < U32_HASH_SIZE; i++)
1442 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1444 ret = register_tcf_proto_ops(&cls_u32_ops);
1446 kvfree(tc_u_common_hash);
1450 static void __exit exit_u32(void)
1452 unregister_tcf_proto_ops(&cls_u32_ops);
1453 kvfree(tc_u_common_hash);
1456 module_init(init_u32)
1457 module_exit(exit_u32)
1458 MODULE_LICENSE("GPL");