1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
7 * The filters are packed to hash tables of key nodes
8 * with a set of 32bit key/mask pairs at every node.
9 * Nodes reference next level hash tables etc.
11 * This scheme is the best universal classifier I managed to
12 * invent; it is not super-fast, but it is not slow (provided you
13 * program it correctly), and general enough. And its relative
14 * speed grows as the number of rules becomes larger.
16 * It seems that it represents the best middle point between
17 * speed and manageability both by human and by machine.
19 * It is especially useful for link sharing combined with QoS;
20 * pure RSVP doesn't need such a general approach and can use
21 * much simpler (and faster) schemes, sort of cls_rsvp.c.
23 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
26 #include <linux/module.h>
27 #include <linux/slab.h>
28 #include <linux/types.h>
29 #include <linux/kernel.h>
30 #include <linux/string.h>
31 #include <linux/errno.h>
32 #include <linux/percpu.h>
33 #include <linux/rtnetlink.h>
34 #include <linux/skbuff.h>
35 #include <linux/bitmap.h>
36 #include <linux/netdevice.h>
37 #include <linux/hash.h>
38 #include <net/netlink.h>
39 #include <net/act_api.h>
40 #include <net/pkt_cls.h>
41 #include <linux/idr.h>
44 struct tc_u_knode __rcu *next;
46 struct tc_u_hnode __rcu *ht_up;
50 struct tcf_result res;
51 struct tc_u_hnode __rcu *ht_down;
52 #ifdef CONFIG_CLS_U32_PERF
53 struct tc_u32_pcnt __percpu *pf;
56 unsigned int in_hw_count;
57 #ifdef CONFIG_CLS_U32_MARK
60 u32 __percpu *pcpu_success;
62 struct rcu_work rwork;
63 /* The 'sel' field MUST be the last field in structure to allow for
64 * tc_u32_keys allocated at end of structure.
66 struct tc_u32_sel sel;
70 struct tc_u_hnode __rcu *next;
75 struct idr handle_idr;
79 /* The 'ht' field MUST be the last field in structure to allow for
80 * more entries allocated at end of structure.
82 struct tc_u_knode __rcu *ht[];
86 struct tc_u_hnode __rcu *hlist;
89 struct idr handle_idr;
90 struct hlist_node hnode;
94 static inline unsigned int u32_hash_fold(__be32 key,
95 const struct tc_u32_sel *sel,
98 unsigned int h = ntohl(key & sel->hmask) >> fshift;
103 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
104 struct tcf_result *res)
107 struct tc_u_knode *knode;
109 } stack[TC_U32_MAXDEPTH];
111 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
112 unsigned int off = skb_network_offset(skb);
113 struct tc_u_knode *n;
117 #ifdef CONFIG_CLS_U32_PERF
123 n = rcu_dereference_bh(ht->ht[sel]);
127 struct tc_u32_key *key = n->sel.keys;
129 #ifdef CONFIG_CLS_U32_PERF
130 __this_cpu_inc(n->pf->rcnt);
134 if (tc_skip_sw(n->flags)) {
135 n = rcu_dereference_bh(n->next);
139 #ifdef CONFIG_CLS_U32_MARK
140 if ((skb->mark & n->mask) != n->val) {
141 n = rcu_dereference_bh(n->next);
144 __this_cpu_inc(*n->pcpu_success);
148 for (i = n->sel.nkeys; i > 0; i--, key++) {
149 int toff = off + key->off + (off2 & key->offmask);
152 if (skb_headroom(skb) + toff > INT_MAX)
155 data = skb_header_pointer(skb, toff, 4, &hdata);
158 if ((*data ^ key->val) & key->mask) {
159 n = rcu_dereference_bh(n->next);
162 #ifdef CONFIG_CLS_U32_PERF
163 __this_cpu_inc(n->pf->kcnts[j]);
168 ht = rcu_dereference_bh(n->ht_down);
171 if (n->sel.flags & TC_U32_TERMINAL) {
174 if (!tcf_match_indev(skb, n->ifindex)) {
175 n = rcu_dereference_bh(n->next);
178 #ifdef CONFIG_CLS_U32_PERF
179 __this_cpu_inc(n->pf->rhit);
181 r = tcf_exts_exec(skb, &n->exts, res);
183 n = rcu_dereference_bh(n->next);
189 n = rcu_dereference_bh(n->next);
194 if (sdepth >= TC_U32_MAXDEPTH)
196 stack[sdepth].knode = n;
197 stack[sdepth].off = off;
200 ht = rcu_dereference_bh(n->ht_down);
205 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
209 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
212 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
215 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
216 off2 = n->sel.off + 3;
217 if (n->sel.flags & TC_U32_VAROFFSET) {
220 data = skb_header_pointer(skb,
225 off2 += ntohs(n->sel.offmask & *data) >>
230 if (n->sel.flags & TC_U32_EAT) {
241 n = stack[sdepth].knode;
242 ht = rcu_dereference_bh(n->ht_up);
243 off = stack[sdepth].off;
250 net_warn_ratelimited("cls_u32: dead loop\n");
254 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
256 struct tc_u_hnode *ht;
258 for (ht = rtnl_dereference(tp_c->hlist);
260 ht = rtnl_dereference(ht->next))
261 if (ht->handle == handle)
267 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
270 struct tc_u_knode *n = NULL;
272 sel = TC_U32_HASH(handle);
273 if (sel > ht->divisor)
276 for (n = rtnl_dereference(ht->ht[sel]);
278 n = rtnl_dereference(n->next))
279 if (n->handle == handle)
286 static void *u32_get(struct tcf_proto *tp, u32 handle)
288 struct tc_u_hnode *ht;
289 struct tc_u_common *tp_c = tp->data;
291 if (TC_U32_HTID(handle) == TC_U32_ROOT)
292 ht = rtnl_dereference(tp->root);
294 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
299 if (TC_U32_KEY(handle) == 0)
302 return u32_lookup_key(ht, handle);
305 /* Protected by rtnl lock */
306 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
308 int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
311 return (id | 0x800U) << 20;
314 static struct hlist_head *tc_u_common_hash;
316 #define U32_HASH_SHIFT 10
317 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
319 static void *tc_u_common_ptr(const struct tcf_proto *tp)
321 struct tcf_block *block = tp->chain->block;
323 /* The block sharing is currently supported only
324 * for classless qdiscs. In that case we use block
325 * for tc_u_common identification. In case the
326 * block is not shared, block->q is a valid pointer
327 * and we can use that. That works for classful qdiscs.
329 if (tcf_block_shared(block))
335 static struct hlist_head *tc_u_hash(void *key)
337 return tc_u_common_hash + hash_ptr(key, U32_HASH_SHIFT);
340 static struct tc_u_common *tc_u_common_find(void *key)
342 struct tc_u_common *tc;
343 hlist_for_each_entry(tc, tc_u_hash(key), hnode) {
350 static int u32_init(struct tcf_proto *tp)
352 struct tc_u_hnode *root_ht;
353 void *key = tc_u_common_ptr(tp);
354 struct tc_u_common *tp_c = tc_u_common_find(key);
356 root_ht = kzalloc(struct_size(root_ht, ht, 1), GFP_KERNEL);
361 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
362 root_ht->prio = tp->prio;
363 root_ht->is_root = true;
364 idr_init(&root_ht->handle_idr);
367 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
373 INIT_HLIST_NODE(&tp_c->hnode);
374 idr_init(&tp_c->handle_idr);
376 hlist_add_head(&tp_c->hnode, tc_u_hash(key));
380 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
381 rcu_assign_pointer(tp_c->hlist, root_ht);
384 rcu_assign_pointer(tp->root, root_ht);
389 static void __u32_destroy_key(struct tc_u_knode *n)
391 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
393 tcf_exts_destroy(&n->exts);
394 if (ht && --ht->refcnt == 0)
399 static void u32_destroy_key(struct tc_u_knode *n, bool free_pf)
401 tcf_exts_put_net(&n->exts);
402 #ifdef CONFIG_CLS_U32_PERF
406 #ifdef CONFIG_CLS_U32_MARK
408 free_percpu(n->pcpu_success);
410 __u32_destroy_key(n);
413 /* u32_delete_key_rcu should be called when free'ing a copied
414 * version of a tc_u_knode obtained from u32_init_knode(). When
415 * copies are obtained from u32_init_knode() the statistics are
416 * shared between the old and new copies to allow readers to
417 * continue to update the statistics during the copy. To support
418 * this the u32_delete_key_rcu variant does not free the percpu
421 static void u32_delete_key_work(struct work_struct *work)
423 struct tc_u_knode *key = container_of(to_rcu_work(work),
427 u32_destroy_key(key, false);
431 /* u32_delete_key_freepf_rcu is the rcu callback variant
432 * that free's the entire structure including the statistics
433 * percpu variables. Only use this if the key is not a copy
434 * returned by u32_init_knode(). See u32_delete_key_rcu()
435 * for the variant that should be used with keys return from
438 static void u32_delete_key_freepf_work(struct work_struct *work)
440 struct tc_u_knode *key = container_of(to_rcu_work(work),
444 u32_destroy_key(key, true);
448 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
450 struct tc_u_common *tp_c = tp->data;
451 struct tc_u_knode __rcu **kp;
452 struct tc_u_knode *pkp;
453 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
456 kp = &ht->ht[TC_U32_HASH(key->handle)];
457 for (pkp = rtnl_dereference(*kp); pkp;
458 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
460 RCU_INIT_POINTER(*kp, key->next);
463 tcf_unbind_filter(tp, &key->res);
464 idr_remove(&ht->handle_idr, key->handle);
465 tcf_exts_get_net(&key->exts);
466 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
475 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
476 struct netlink_ext_ack *extack)
478 struct tcf_block *block = tp->chain->block;
479 struct tc_cls_u32_offload cls_u32 = {};
481 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
482 cls_u32.command = TC_CLSU32_DELETE_HNODE;
483 cls_u32.hnode.divisor = h->divisor;
484 cls_u32.hnode.handle = h->handle;
485 cls_u32.hnode.prio = h->prio;
487 tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false, true);
490 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
491 u32 flags, struct netlink_ext_ack *extack)
493 struct tcf_block *block = tp->chain->block;
494 struct tc_cls_u32_offload cls_u32 = {};
495 bool skip_sw = tc_skip_sw(flags);
496 bool offloaded = false;
499 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
500 cls_u32.command = TC_CLSU32_NEW_HNODE;
501 cls_u32.hnode.divisor = h->divisor;
502 cls_u32.hnode.handle = h->handle;
503 cls_u32.hnode.prio = h->prio;
505 err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw, true);
507 u32_clear_hw_hnode(tp, h, NULL);
509 } else if (err > 0) {
513 if (skip_sw && !offloaded)
519 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
520 struct netlink_ext_ack *extack)
522 struct tcf_block *block = tp->chain->block;
523 struct tc_cls_u32_offload cls_u32 = {};
525 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
526 cls_u32.command = TC_CLSU32_DELETE_KNODE;
527 cls_u32.knode.handle = n->handle;
529 tc_setup_cb_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
530 &n->flags, &n->in_hw_count, true);
533 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
534 u32 flags, struct netlink_ext_ack *extack)
536 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
537 struct tcf_block *block = tp->chain->block;
538 struct tc_cls_u32_offload cls_u32 = {};
539 bool skip_sw = tc_skip_sw(flags);
542 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
543 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
544 cls_u32.knode.handle = n->handle;
545 cls_u32.knode.fshift = n->fshift;
546 #ifdef CONFIG_CLS_U32_MARK
547 cls_u32.knode.val = n->val;
548 cls_u32.knode.mask = n->mask;
550 cls_u32.knode.val = 0;
551 cls_u32.knode.mask = 0;
553 cls_u32.knode.sel = &n->sel;
554 cls_u32.knode.res = &n->res;
555 cls_u32.knode.exts = &n->exts;
557 cls_u32.knode.link_handle = ht->handle;
559 err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
560 &n->flags, &n->in_hw_count, true);
562 u32_remove_hw_knode(tp, n, NULL);
566 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
572 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
573 struct netlink_ext_ack *extack)
575 struct tc_u_common *tp_c = tp->data;
576 struct tc_u_knode *n;
579 for (h = 0; h <= ht->divisor; h++) {
580 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
581 RCU_INIT_POINTER(ht->ht[h],
582 rtnl_dereference(n->next));
584 tcf_unbind_filter(tp, &n->res);
585 u32_remove_hw_knode(tp, n, extack);
586 idr_remove(&ht->handle_idr, n->handle);
587 if (tcf_exts_get_net(&n->exts))
588 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
590 u32_destroy_key(n, true);
595 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
596 struct netlink_ext_ack *extack)
598 struct tc_u_common *tp_c = tp->data;
599 struct tc_u_hnode __rcu **hn;
600 struct tc_u_hnode *phn;
602 WARN_ON(--ht->refcnt);
604 u32_clear_hnode(tp, ht, extack);
607 for (phn = rtnl_dereference(*hn);
609 hn = &phn->next, phn = rtnl_dereference(*hn)) {
611 u32_clear_hw_hnode(tp, ht, extack);
612 idr_destroy(&ht->handle_idr);
613 idr_remove(&tp_c->handle_idr, ht->handle);
614 RCU_INIT_POINTER(*hn, ht->next);
623 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
624 struct netlink_ext_ack *extack)
626 struct tc_u_common *tp_c = tp->data;
627 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
629 WARN_ON(root_ht == NULL);
631 if (root_ht && --root_ht->refcnt == 1)
632 u32_destroy_hnode(tp, root_ht, extack);
634 if (--tp_c->refcnt == 0) {
635 struct tc_u_hnode *ht;
637 hlist_del(&tp_c->hnode);
639 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
640 u32_clear_hnode(tp, ht, extack);
641 RCU_INIT_POINTER(tp_c->hlist, ht->next);
643 /* u32_destroy_key() will later free ht for us, if it's
644 * still referenced by some knode
646 if (--ht->refcnt == 0)
650 idr_destroy(&tp_c->handle_idr);
657 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
658 bool rtnl_held, struct netlink_ext_ack *extack)
660 struct tc_u_hnode *ht = arg;
661 struct tc_u_common *tp_c = tp->data;
664 if (TC_U32_KEY(ht->handle)) {
665 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
666 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
671 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
675 if (ht->refcnt == 1) {
676 u32_destroy_hnode(tp, ht, extack);
678 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
683 *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
687 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
689 u32 index = htid | 0x800;
690 u32 max = htid | 0xFFF;
692 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
694 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
702 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
703 [TCA_U32_CLASSID] = { .type = NLA_U32 },
704 [TCA_U32_HASH] = { .type = NLA_U32 },
705 [TCA_U32_LINK] = { .type = NLA_U32 },
706 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
707 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
708 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
709 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
710 [TCA_U32_FLAGS] = { .type = NLA_U32 },
713 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
715 struct tc_u_knode *n, struct nlattr **tb,
716 struct nlattr *est, bool ovr,
717 struct netlink_ext_ack *extack)
719 int err, ifindex = -1;
721 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
725 if (tb[TCA_U32_INDEV]) {
726 ifindex = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
731 if (tb[TCA_U32_LINK]) {
732 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
733 struct tc_u_hnode *ht_down = NULL, *ht_old;
735 if (TC_U32_KEY(handle)) {
736 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
741 ht_down = u32_lookup_ht(tp->data, handle);
744 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
747 if (ht_down->is_root) {
748 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
754 ht_old = rtnl_dereference(n->ht_down);
755 rcu_assign_pointer(n->ht_down, ht_down);
760 if (tb[TCA_U32_CLASSID]) {
761 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
762 tcf_bind_filter(tp, &n->res, base);
766 n->ifindex = ifindex;
771 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
772 struct tc_u_knode *n)
774 struct tc_u_knode __rcu **ins;
775 struct tc_u_knode *pins;
776 struct tc_u_hnode *ht;
778 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
779 ht = rtnl_dereference(tp->root);
781 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
783 ins = &ht->ht[TC_U32_HASH(n->handle)];
785 /* The node must always exist for it to be replaced if this is not the
786 * case then something went very wrong elsewhere.
788 for (pins = rtnl_dereference(*ins); ;
789 ins = &pins->next, pins = rtnl_dereference(*ins))
790 if (pins->handle == n->handle)
793 idr_replace(&ht->handle_idr, n, n->handle);
794 RCU_INIT_POINTER(n->next, pins->next);
795 rcu_assign_pointer(*ins, n);
798 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
799 struct tc_u_knode *n)
801 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
802 struct tc_u32_sel *s = &n->sel;
803 struct tc_u_knode *new;
805 new = kzalloc(struct_size(new, sel.keys, s->nkeys), GFP_KERNEL);
809 RCU_INIT_POINTER(new->next, n->next);
810 new->handle = n->handle;
811 RCU_INIT_POINTER(new->ht_up, n->ht_up);
813 new->ifindex = n->ifindex;
814 new->fshift = n->fshift;
815 new->flags = n->flags;
816 RCU_INIT_POINTER(new->ht_down, ht);
818 #ifdef CONFIG_CLS_U32_PERF
819 /* Statistics may be incremented by readers during update
820 * so we must keep them in tact. When the node is later destroyed
821 * a special destroy call must be made to not free the pf memory.
826 #ifdef CONFIG_CLS_U32_MARK
829 /* Similarly success statistics must be moved as pointers */
830 new->pcpu_success = n->pcpu_success;
832 memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
834 if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
839 /* bump reference count as long as we hold pointer to structure */
846 static int u32_change(struct net *net, struct sk_buff *in_skb,
847 struct tcf_proto *tp, unsigned long base, u32 handle,
848 struct nlattr **tca, void **arg, bool ovr, bool rtnl_held,
849 struct netlink_ext_ack *extack)
851 struct tc_u_common *tp_c = tp->data;
852 struct tc_u_hnode *ht;
853 struct tc_u_knode *n;
854 struct tc_u32_sel *s;
855 struct nlattr *opt = tca[TCA_OPTIONS];
856 struct nlattr *tb[TCA_U32_MAX + 1];
863 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
870 err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
875 if (tb[TCA_U32_FLAGS]) {
876 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
877 if (!tc_flags_valid(flags)) {
878 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
885 struct tc_u_knode *new;
887 if (TC_U32_KEY(n->handle) == 0) {
888 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
892 if ((n->flags ^ flags) &
893 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
894 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
898 new = u32_init_knode(net, tp, n);
902 err = u32_set_parms(net, tp, base, new, tb,
903 tca[TCA_RATE], ovr, extack);
906 __u32_destroy_key(new);
910 err = u32_replace_hw_knode(tp, new, flags, extack);
912 __u32_destroy_key(new);
916 if (!tc_in_hw(new->flags))
917 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
919 u32_replace_knode(tp, tp_c, new);
920 tcf_unbind_filter(tp, &n->res);
921 tcf_exts_get_net(&n->exts);
922 tcf_queue_work(&n->rwork, u32_delete_key_work);
926 if (tb[TCA_U32_DIVISOR]) {
927 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
929 if (!is_power_of_2(divisor)) {
930 NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
933 if (divisor-- > 0x100) {
934 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
937 if (TC_U32_KEY(handle)) {
938 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
941 ht = kzalloc(struct_size(ht, ht, divisor + 1), GFP_KERNEL);
945 handle = gen_new_htid(tp->data, ht);
951 err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
959 ht->divisor = divisor;
962 idr_init(&ht->handle_idr);
965 err = u32_replace_hw_hnode(tp, ht, flags, extack);
967 idr_remove(&tp_c->handle_idr, handle);
972 RCU_INIT_POINTER(ht->next, tp_c->hlist);
973 rcu_assign_pointer(tp_c->hlist, ht);
979 if (tb[TCA_U32_HASH]) {
980 htid = nla_get_u32(tb[TCA_U32_HASH]);
981 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
982 ht = rtnl_dereference(tp->root);
985 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
987 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
992 ht = rtnl_dereference(tp->root);
996 if (ht->divisor < TC_U32_HASH(htid)) {
997 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1001 /* At this point, we need to derive the new handle that will be used to
1002 * uniquely map the identity of this table match entry. The
1003 * identity of the entry that we need to construct is 32 bits made of:
1004 * htid(12b):bucketid(8b):node/entryid(12b)
1006 * At this point _we have the table(ht)_ in which we will insert this
1007 * entry. We carry the table's id in variable "htid".
1008 * Note that earlier code picked the ht selection either by a) the user
1009 * providing the htid specified via TCA_U32_HASH attribute or b) when
1010 * no such attribute is passed then the root ht, is default to at ID
1011 * 0x[800][00][000]. Rule: the root table has a single bucket with ID 0.
1012 * If OTOH the user passed us the htid, they may also pass a bucketid of
1013 * choice. 0 is fine. For example a user htid is 0x[600][01][000] it is
1014 * indicating hash bucketid of 1. Rule: the entry/node ID _cannot_ be
1015 * passed via the htid, so even if it was non-zero it will be ignored.
1017 * We may also have a handle, if the user passed one. The handle also
1018 * carries the same addressing of htid(12b):bucketid(8b):node/entryid(12b).
1019 * Rule: the bucketid on the handle is ignored even if one was passed;
1020 * rather the value on "htid" is always assumed to be the bucketid.
1023 /* Rule: The htid from handle and tableid from htid must match */
1024 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1025 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1028 /* Ok, so far we have a valid htid(12b):bucketid(8b) but we
1029 * need to finalize the table entry identification with the last
1030 * part - the node/entryid(12b)). Rule: Nodeid _cannot be 0_ for
1031 * entries. Rule: nodeid of 0 is reserved only for tables(see
1032 * earlier code which processes TC_U32_DIVISOR attribute).
1033 * Rule: The nodeid can only be derived from the handle (and not
1035 * Rule: if the handle specified zero for the node id example
1036 * 0x60000000, then pick a new nodeid from the pool of IDs
1037 * this hash table has been allocating from.
1038 * If OTOH it is specified (i.e for example the user passed a
1039 * handle such as 0x60000123), then we use it generate our final
1040 * handle which is used to uniquely identify the match entry.
1042 if (!TC_U32_NODE(handle)) {
1043 handle = gen_new_kid(ht, htid);
1045 handle = htid | TC_U32_NODE(handle);
1046 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle,
1047 handle, GFP_KERNEL);
1052 /* The user did not give us a handle; lets just generate one
1053 * from the table's pool of nodeids.
1055 handle = gen_new_kid(ht, htid);
1058 if (tb[TCA_U32_SEL] == NULL) {
1059 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1064 s = nla_data(tb[TCA_U32_SEL]);
1065 sel_size = struct_size(s, keys, s->nkeys);
1066 if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1071 n = kzalloc(struct_size(n, sel.keys, s->nkeys), GFP_KERNEL);
1077 #ifdef CONFIG_CLS_U32_PERF
1078 n->pf = __alloc_percpu(struct_size(n->pf, kcnts, s->nkeys),
1079 __alignof__(struct tc_u32_pcnt));
1086 memcpy(&n->sel, s, sel_size);
1087 RCU_INIT_POINTER(n->ht_up, ht);
1089 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1092 err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1096 #ifdef CONFIG_CLS_U32_MARK
1097 n->pcpu_success = alloc_percpu(u32);
1098 if (!n->pcpu_success) {
1103 if (tb[TCA_U32_MARK]) {
1104 struct tc_u32_mark *mark;
1106 mark = nla_data(tb[TCA_U32_MARK]);
1108 n->mask = mark->mask;
1112 err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1115 struct tc_u_knode __rcu **ins;
1116 struct tc_u_knode *pins;
1118 err = u32_replace_hw_knode(tp, n, flags, extack);
1122 if (!tc_in_hw(n->flags))
1123 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1125 ins = &ht->ht[TC_U32_HASH(handle)];
1126 for (pins = rtnl_dereference(*ins); pins;
1127 ins = &pins->next, pins = rtnl_dereference(*ins))
1128 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1131 RCU_INIT_POINTER(n->next, pins);
1132 rcu_assign_pointer(*ins, n);
1139 #ifdef CONFIG_CLS_U32_MARK
1140 free_percpu(n->pcpu_success);
1144 tcf_exts_destroy(&n->exts);
1145 #ifdef CONFIG_CLS_U32_PERF
1151 idr_remove(&ht->handle_idr, handle);
1155 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1158 struct tc_u_common *tp_c = tp->data;
1159 struct tc_u_hnode *ht;
1160 struct tc_u_knode *n;
1166 for (ht = rtnl_dereference(tp_c->hlist);
1168 ht = rtnl_dereference(ht->next)) {
1169 if (ht->prio != tp->prio)
1171 if (arg->count >= arg->skip) {
1172 if (arg->fn(tp, ht, arg) < 0) {
1178 for (h = 0; h <= ht->divisor; h++) {
1179 for (n = rtnl_dereference(ht->ht[h]);
1181 n = rtnl_dereference(n->next)) {
1182 if (arg->count < arg->skip) {
1186 if (arg->fn(tp, n, arg) < 0) {
1196 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1197 bool add, flow_setup_cb_t *cb, void *cb_priv,
1198 struct netlink_ext_ack *extack)
1200 struct tc_cls_u32_offload cls_u32 = {};
1203 tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1204 cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1205 cls_u32.hnode.divisor = ht->divisor;
1206 cls_u32.hnode.handle = ht->handle;
1207 cls_u32.hnode.prio = ht->prio;
1209 err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1210 if (err && add && tc_skip_sw(ht->flags))
1216 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1217 bool add, flow_setup_cb_t *cb, void *cb_priv,
1218 struct netlink_ext_ack *extack)
1220 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1221 struct tcf_block *block = tp->chain->block;
1222 struct tc_cls_u32_offload cls_u32 = {};
1225 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1226 cls_u32.command = add ?
1227 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1228 cls_u32.knode.handle = n->handle;
1231 cls_u32.knode.fshift = n->fshift;
1232 #ifdef CONFIG_CLS_U32_MARK
1233 cls_u32.knode.val = n->val;
1234 cls_u32.knode.mask = n->mask;
1236 cls_u32.knode.val = 0;
1237 cls_u32.knode.mask = 0;
1239 cls_u32.knode.sel = &n->sel;
1240 cls_u32.knode.res = &n->res;
1241 cls_u32.knode.exts = &n->exts;
1243 cls_u32.knode.link_handle = ht->handle;
1246 err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1247 &cls_u32, cb_priv, &n->flags,
1255 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1256 void *cb_priv, struct netlink_ext_ack *extack)
1258 struct tc_u_common *tp_c = tp->data;
1259 struct tc_u_hnode *ht;
1260 struct tc_u_knode *n;
1264 for (ht = rtnl_dereference(tp_c->hlist);
1266 ht = rtnl_dereference(ht->next)) {
1267 if (ht->prio != tp->prio)
1270 /* When adding filters to a new dev, try to offload the
1271 * hashtable first. When removing, do the filters before the
1274 if (add && !tc_skip_hw(ht->flags)) {
1275 err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1281 for (h = 0; h <= ht->divisor; h++) {
1282 for (n = rtnl_dereference(ht->ht[h]);
1284 n = rtnl_dereference(n->next)) {
1285 if (tc_skip_hw(n->flags))
1288 err = u32_reoffload_knode(tp, n, add, cb,
1295 if (!add && !tc_skip_hw(ht->flags))
1296 u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1302 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1305 struct tc_u_knode *n = fh;
1307 if (n && n->res.classid == classid) {
1309 __tcf_bind_filter(q, &n->res, base);
1311 __tcf_unbind_filter(q, &n->res);
1315 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1316 struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1318 struct tc_u_knode *n = fh;
1319 struct tc_u_hnode *ht_up, *ht_down;
1320 struct nlattr *nest;
1325 t->tcm_handle = n->handle;
1327 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1329 goto nla_put_failure;
1331 if (TC_U32_KEY(n->handle) == 0) {
1332 struct tc_u_hnode *ht = fh;
1333 u32 divisor = ht->divisor + 1;
1335 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1336 goto nla_put_failure;
1338 #ifdef CONFIG_CLS_U32_PERF
1339 struct tc_u32_pcnt *gpf;
1343 if (nla_put(skb, TCA_U32_SEL, struct_size(&n->sel, keys, n->sel.nkeys),
1345 goto nla_put_failure;
1347 ht_up = rtnl_dereference(n->ht_up);
1349 u32 htid = n->handle & 0xFFFFF000;
1350 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1351 goto nla_put_failure;
1353 if (n->res.classid &&
1354 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1355 goto nla_put_failure;
1357 ht_down = rtnl_dereference(n->ht_down);
1359 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1360 goto nla_put_failure;
1362 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1363 goto nla_put_failure;
1365 #ifdef CONFIG_CLS_U32_MARK
1366 if ((n->val || n->mask)) {
1367 struct tc_u32_mark mark = {.val = n->val,
1372 for_each_possible_cpu(cpum) {
1373 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1375 mark.success += cnt;
1378 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1379 goto nla_put_failure;
1383 if (tcf_exts_dump(skb, &n->exts) < 0)
1384 goto nla_put_failure;
1387 struct net_device *dev;
1388 dev = __dev_get_by_index(net, n->ifindex);
1389 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1390 goto nla_put_failure;
1392 #ifdef CONFIG_CLS_U32_PERF
1393 gpf = kzalloc(struct_size(gpf, kcnts, n->sel.nkeys), GFP_KERNEL);
1395 goto nla_put_failure;
1397 for_each_possible_cpu(cpu) {
1399 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1401 gpf->rcnt += pf->rcnt;
1402 gpf->rhit += pf->rhit;
1403 for (i = 0; i < n->sel.nkeys; i++)
1404 gpf->kcnts[i] += pf->kcnts[i];
1407 if (nla_put_64bit(skb, TCA_U32_PCNT, struct_size(gpf, kcnts, n->sel.nkeys),
1408 gpf, TCA_U32_PAD)) {
1410 goto nla_put_failure;
1416 nla_nest_end(skb, nest);
1418 if (TC_U32_KEY(n->handle))
1419 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1420 goto nla_put_failure;
1424 nla_nest_cancel(skb, nest);
1428 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1430 .classify = u32_classify,
1432 .destroy = u32_destroy,
1434 .change = u32_change,
1435 .delete = u32_delete,
1437 .reoffload = u32_reoffload,
1439 .bind_class = u32_bind_class,
1440 .owner = THIS_MODULE,
1443 static int __init init_u32(void)
1447 pr_info("u32 classifier\n");
1448 #ifdef CONFIG_CLS_U32_PERF
1449 pr_info(" Performance counters on\n");
1451 pr_info(" input device check on\n");
1452 #ifdef CONFIG_NET_CLS_ACT
1453 pr_info(" Actions configured\n");
1455 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1456 sizeof(struct hlist_head),
1458 if (!tc_u_common_hash)
1461 for (i = 0; i < U32_HASH_SIZE; i++)
1462 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1464 ret = register_tcf_proto_ops(&cls_u32_ops);
1466 kvfree(tc_u_common_hash);
1470 static void __exit exit_u32(void)
1472 unregister_tcf_proto_ops(&cls_u32_ops);
1473 kvfree(tc_u_common_hash);
1476 module_init(init_u32)
1477 module_exit(exit_u32)
1478 MODULE_LICENSE("GPL");