1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
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[1];
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(sizeof(*root_ht), 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 int u32_destroy_key(struct tc_u_knode *n, bool free_pf)
391 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
393 tcf_exts_destroy(&n->exts);
394 tcf_exts_put_net(&n->exts);
395 if (ht && --ht->refcnt == 0)
397 #ifdef CONFIG_CLS_U32_PERF
401 #ifdef CONFIG_CLS_U32_MARK
403 free_percpu(n->pcpu_success);
409 /* u32_delete_key_rcu should be called when free'ing a copied
410 * version of a tc_u_knode obtained from u32_init_knode(). When
411 * copies are obtained from u32_init_knode() the statistics are
412 * shared between the old and new copies to allow readers to
413 * continue to update the statistics during the copy. To support
414 * this the u32_delete_key_rcu variant does not free the percpu
417 static void u32_delete_key_work(struct work_struct *work)
419 struct tc_u_knode *key = container_of(to_rcu_work(work),
423 u32_destroy_key(key, false);
427 /* u32_delete_key_freepf_rcu is the rcu callback variant
428 * that free's the entire structure including the statistics
429 * percpu variables. Only use this if the key is not a copy
430 * returned by u32_init_knode(). See u32_delete_key_rcu()
431 * for the variant that should be used with keys return from
434 static void u32_delete_key_freepf_work(struct work_struct *work)
436 struct tc_u_knode *key = container_of(to_rcu_work(work),
440 u32_destroy_key(key, true);
444 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
446 struct tc_u_common *tp_c = tp->data;
447 struct tc_u_knode __rcu **kp;
448 struct tc_u_knode *pkp;
449 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
452 kp = &ht->ht[TC_U32_HASH(key->handle)];
453 for (pkp = rtnl_dereference(*kp); pkp;
454 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
456 RCU_INIT_POINTER(*kp, key->next);
459 tcf_unbind_filter(tp, &key->res);
460 idr_remove(&ht->handle_idr, key->handle);
461 tcf_exts_get_net(&key->exts);
462 tcf_queue_work(&key->rwork, u32_delete_key_freepf_work);
471 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
472 struct netlink_ext_ack *extack)
474 struct tcf_block *block = tp->chain->block;
475 struct tc_cls_u32_offload cls_u32 = {};
477 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
478 cls_u32.command = TC_CLSU32_DELETE_HNODE;
479 cls_u32.hnode.divisor = h->divisor;
480 cls_u32.hnode.handle = h->handle;
481 cls_u32.hnode.prio = h->prio;
483 tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
486 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
487 u32 flags, struct netlink_ext_ack *extack)
489 struct tcf_block *block = tp->chain->block;
490 struct tc_cls_u32_offload cls_u32 = {};
491 bool skip_sw = tc_skip_sw(flags);
492 bool offloaded = false;
495 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
496 cls_u32.command = TC_CLSU32_NEW_HNODE;
497 cls_u32.hnode.divisor = h->divisor;
498 cls_u32.hnode.handle = h->handle;
499 cls_u32.hnode.prio = h->prio;
501 err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
503 u32_clear_hw_hnode(tp, h, NULL);
505 } else if (err > 0) {
509 if (skip_sw && !offloaded)
515 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
516 struct netlink_ext_ack *extack)
518 struct tcf_block *block = tp->chain->block;
519 struct tc_cls_u32_offload cls_u32 = {};
521 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
522 cls_u32.command = TC_CLSU32_DELETE_KNODE;
523 cls_u32.knode.handle = n->handle;
525 tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, false);
526 tcf_block_offload_dec(block, &n->flags);
529 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
530 u32 flags, struct netlink_ext_ack *extack)
532 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
533 struct tcf_block *block = tp->chain->block;
534 struct tc_cls_u32_offload cls_u32 = {};
535 bool skip_sw = tc_skip_sw(flags);
538 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
539 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
540 cls_u32.knode.handle = n->handle;
541 cls_u32.knode.fshift = n->fshift;
542 #ifdef CONFIG_CLS_U32_MARK
543 cls_u32.knode.val = n->val;
544 cls_u32.knode.mask = n->mask;
546 cls_u32.knode.val = 0;
547 cls_u32.knode.mask = 0;
549 cls_u32.knode.sel = &n->sel;
550 cls_u32.knode.res = &n->res;
551 cls_u32.knode.exts = &n->exts;
553 cls_u32.knode.link_handle = ht->handle;
555 err = tc_setup_cb_call(block, TC_SETUP_CLSU32, &cls_u32, skip_sw);
557 u32_remove_hw_knode(tp, n, NULL);
559 } else if (err > 0) {
560 n->in_hw_count = err;
561 tcf_block_offload_inc(block, &n->flags);
564 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
570 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
571 struct netlink_ext_ack *extack)
573 struct tc_u_common *tp_c = tp->data;
574 struct tc_u_knode *n;
577 for (h = 0; h <= ht->divisor; h++) {
578 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
579 RCU_INIT_POINTER(ht->ht[h],
580 rtnl_dereference(n->next));
582 tcf_unbind_filter(tp, &n->res);
583 u32_remove_hw_knode(tp, n, extack);
584 idr_remove(&ht->handle_idr, n->handle);
585 if (tcf_exts_get_net(&n->exts))
586 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
588 u32_destroy_key(n, true);
593 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
594 struct netlink_ext_ack *extack)
596 struct tc_u_common *tp_c = tp->data;
597 struct tc_u_hnode __rcu **hn;
598 struct tc_u_hnode *phn;
600 WARN_ON(--ht->refcnt);
602 u32_clear_hnode(tp, ht, extack);
605 for (phn = rtnl_dereference(*hn);
607 hn = &phn->next, phn = rtnl_dereference(*hn)) {
609 u32_clear_hw_hnode(tp, ht, extack);
610 idr_destroy(&ht->handle_idr);
611 idr_remove(&tp_c->handle_idr, ht->handle);
612 RCU_INIT_POINTER(*hn, ht->next);
621 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
622 struct netlink_ext_ack *extack)
624 struct tc_u_common *tp_c = tp->data;
625 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
627 WARN_ON(root_ht == NULL);
629 if (root_ht && --root_ht->refcnt == 1)
630 u32_destroy_hnode(tp, root_ht, extack);
632 if (--tp_c->refcnt == 0) {
633 struct tc_u_hnode *ht;
635 hlist_del(&tp_c->hnode);
637 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
638 u32_clear_hnode(tp, ht, extack);
639 RCU_INIT_POINTER(tp_c->hlist, ht->next);
641 /* u32_destroy_key() will later free ht for us, if it's
642 * still referenced by some knode
644 if (--ht->refcnt == 0)
648 idr_destroy(&tp_c->handle_idr);
655 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
656 bool rtnl_held, struct netlink_ext_ack *extack)
658 struct tc_u_hnode *ht = arg;
659 struct tc_u_common *tp_c = tp->data;
662 if (TC_U32_KEY(ht->handle)) {
663 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
664 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
669 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
673 if (ht->refcnt == 1) {
674 u32_destroy_hnode(tp, ht, extack);
676 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
681 *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
685 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
687 u32 index = htid | 0x800;
688 u32 max = htid | 0xFFF;
690 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
692 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
700 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
701 [TCA_U32_CLASSID] = { .type = NLA_U32 },
702 [TCA_U32_HASH] = { .type = NLA_U32 },
703 [TCA_U32_LINK] = { .type = NLA_U32 },
704 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
705 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
706 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
707 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
708 [TCA_U32_FLAGS] = { .type = NLA_U32 },
711 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
713 struct tc_u_knode *n, struct nlattr **tb,
714 struct nlattr *est, bool ovr,
715 struct netlink_ext_ack *extack)
719 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
723 if (tb[TCA_U32_LINK]) {
724 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
725 struct tc_u_hnode *ht_down = NULL, *ht_old;
727 if (TC_U32_KEY(handle)) {
728 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
733 ht_down = u32_lookup_ht(tp->data, handle);
736 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
739 if (ht_down->is_root) {
740 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
746 ht_old = rtnl_dereference(n->ht_down);
747 rcu_assign_pointer(n->ht_down, ht_down);
752 if (tb[TCA_U32_CLASSID]) {
753 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
754 tcf_bind_filter(tp, &n->res, base);
757 if (tb[TCA_U32_INDEV]) {
759 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
767 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
768 struct tc_u_knode *n)
770 struct tc_u_knode __rcu **ins;
771 struct tc_u_knode *pins;
772 struct tc_u_hnode *ht;
774 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
775 ht = rtnl_dereference(tp->root);
777 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
779 ins = &ht->ht[TC_U32_HASH(n->handle)];
781 /* The node must always exist for it to be replaced if this is not the
782 * case then something went very wrong elsewhere.
784 for (pins = rtnl_dereference(*ins); ;
785 ins = &pins->next, pins = rtnl_dereference(*ins))
786 if (pins->handle == n->handle)
789 idr_replace(&ht->handle_idr, n, n->handle);
790 RCU_INIT_POINTER(n->next, pins->next);
791 rcu_assign_pointer(*ins, n);
794 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
795 struct tc_u_knode *n)
797 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
798 struct tc_u32_sel *s = &n->sel;
799 struct tc_u_knode *new;
801 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
807 RCU_INIT_POINTER(new->next, n->next);
808 new->handle = n->handle;
809 RCU_INIT_POINTER(new->ht_up, n->ht_up);
811 new->ifindex = n->ifindex;
812 new->fshift = n->fshift;
814 new->flags = n->flags;
815 RCU_INIT_POINTER(new->ht_down, ht);
817 /* bump reference count as long as we hold pointer to structure */
821 #ifdef CONFIG_CLS_U32_PERF
822 /* Statistics may be incremented by readers during update
823 * so we must keep them in tact. When the node is later destroyed
824 * a special destroy call must be made to not free the pf memory.
829 #ifdef CONFIG_CLS_U32_MARK
832 /* Similarly success statistics must be moved as pointers */
833 new->pcpu_success = n->pcpu_success;
835 memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
837 if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
845 static int u32_change(struct net *net, struct sk_buff *in_skb,
846 struct tcf_proto *tp, unsigned long base, u32 handle,
847 struct nlattr **tca, void **arg, bool ovr, bool rtnl_held,
848 struct netlink_ext_ack *extack)
850 struct tc_u_common *tp_c = tp->data;
851 struct tc_u_hnode *ht;
852 struct tc_u_knode *n;
853 struct tc_u32_sel *s;
854 struct nlattr *opt = tca[TCA_OPTIONS];
855 struct nlattr *tb[TCA_U32_MAX + 1];
859 #ifdef CONFIG_CLS_U32_PERF
865 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
872 err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
877 if (tb[TCA_U32_FLAGS]) {
878 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
879 if (!tc_flags_valid(flags)) {
880 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
887 struct tc_u_knode *new;
889 if (TC_U32_KEY(n->handle) == 0) {
890 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
894 if ((n->flags ^ flags) &
895 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW)) {
896 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
900 new = u32_init_knode(net, tp, n);
904 err = u32_set_parms(net, tp, base, new, tb,
905 tca[TCA_RATE], ovr, extack);
908 u32_destroy_key(new, false);
912 err = u32_replace_hw_knode(tp, new, flags, extack);
914 u32_destroy_key(new, false);
918 if (!tc_in_hw(new->flags))
919 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
921 u32_replace_knode(tp, tp_c, new);
922 tcf_unbind_filter(tp, &n->res);
923 tcf_exts_get_net(&n->exts);
924 tcf_queue_work(&n->rwork, u32_delete_key_work);
928 if (tb[TCA_U32_DIVISOR]) {
929 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
931 if (!is_power_of_2(divisor)) {
932 NL_SET_ERR_MSG_MOD(extack, "Divisor is not a power of 2");
935 if (divisor-- > 0x100) {
936 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
939 if (TC_U32_KEY(handle)) {
940 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
943 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
947 handle = gen_new_htid(tp->data, ht);
953 err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
961 ht->divisor = divisor;
964 idr_init(&ht->handle_idr);
967 err = u32_replace_hw_hnode(tp, ht, flags, extack);
969 idr_remove(&tp_c->handle_idr, handle);
974 RCU_INIT_POINTER(ht->next, tp_c->hlist);
975 rcu_assign_pointer(tp_c->hlist, ht);
981 if (tb[TCA_U32_HASH]) {
982 htid = nla_get_u32(tb[TCA_U32_HASH]);
983 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
984 ht = rtnl_dereference(tp->root);
987 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
989 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
994 ht = rtnl_dereference(tp->root);
998 if (ht->divisor < TC_U32_HASH(htid)) {
999 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1004 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1005 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1008 handle = htid | TC_U32_NODE(handle);
1009 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1014 handle = gen_new_kid(ht, htid);
1016 if (tb[TCA_U32_SEL] == NULL) {
1017 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1022 s = nla_data(tb[TCA_U32_SEL]);
1023 sel_size = struct_size(s, keys, s->nkeys);
1024 if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1029 n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1035 #ifdef CONFIG_CLS_U32_PERF
1036 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1037 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1044 memcpy(&n->sel, s, sel_size);
1045 RCU_INIT_POINTER(n->ht_up, ht);
1047 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1050 err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1054 #ifdef CONFIG_CLS_U32_MARK
1055 n->pcpu_success = alloc_percpu(u32);
1056 if (!n->pcpu_success) {
1061 if (tb[TCA_U32_MARK]) {
1062 struct tc_u32_mark *mark;
1064 mark = nla_data(tb[TCA_U32_MARK]);
1066 n->mask = mark->mask;
1070 err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1073 struct tc_u_knode __rcu **ins;
1074 struct tc_u_knode *pins;
1076 err = u32_replace_hw_knode(tp, n, flags, extack);
1080 if (!tc_in_hw(n->flags))
1081 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1083 ins = &ht->ht[TC_U32_HASH(handle)];
1084 for (pins = rtnl_dereference(*ins); pins;
1085 ins = &pins->next, pins = rtnl_dereference(*ins))
1086 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1089 RCU_INIT_POINTER(n->next, pins);
1090 rcu_assign_pointer(*ins, n);
1097 #ifdef CONFIG_CLS_U32_MARK
1098 free_percpu(n->pcpu_success);
1102 tcf_exts_destroy(&n->exts);
1103 #ifdef CONFIG_CLS_U32_PERF
1109 idr_remove(&ht->handle_idr, handle);
1113 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 int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1155 bool add, tc_setup_cb_t *cb, void *cb_priv,
1156 struct netlink_ext_ack *extack)
1158 struct tc_cls_u32_offload cls_u32 = {};
1161 tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1162 cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1163 cls_u32.hnode.divisor = ht->divisor;
1164 cls_u32.hnode.handle = ht->handle;
1165 cls_u32.hnode.prio = ht->prio;
1167 err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1168 if (err && add && tc_skip_sw(ht->flags))
1174 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1175 bool add, tc_setup_cb_t *cb, void *cb_priv,
1176 struct netlink_ext_ack *extack)
1178 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1179 struct tcf_block *block = tp->chain->block;
1180 struct tc_cls_u32_offload cls_u32 = {};
1183 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1184 cls_u32.command = add ?
1185 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1186 cls_u32.knode.handle = n->handle;
1189 cls_u32.knode.fshift = n->fshift;
1190 #ifdef CONFIG_CLS_U32_MARK
1191 cls_u32.knode.val = n->val;
1192 cls_u32.knode.mask = n->mask;
1194 cls_u32.knode.val = 0;
1195 cls_u32.knode.mask = 0;
1197 cls_u32.knode.sel = &n->sel;
1198 cls_u32.knode.res = &n->res;
1199 cls_u32.knode.exts = &n->exts;
1201 cls_u32.knode.link_handle = ht->handle;
1204 err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1206 if (add && tc_skip_sw(n->flags))
1211 tc_cls_offload_cnt_update(block, &n->in_hw_count, &n->flags, add);
1216 static int u32_reoffload(struct tcf_proto *tp, bool add, tc_setup_cb_t *cb,
1217 void *cb_priv, struct netlink_ext_ack *extack)
1219 struct tc_u_common *tp_c = tp->data;
1220 struct tc_u_hnode *ht;
1221 struct tc_u_knode *n;
1225 for (ht = rtnl_dereference(tp_c->hlist);
1227 ht = rtnl_dereference(ht->next)) {
1228 if (ht->prio != tp->prio)
1231 /* When adding filters to a new dev, try to offload the
1232 * hashtable first. When removing, do the filters before the
1235 if (add && !tc_skip_hw(ht->flags)) {
1236 err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1242 for (h = 0; h <= ht->divisor; h++) {
1243 for (n = rtnl_dereference(ht->ht[h]);
1245 n = rtnl_dereference(n->next)) {
1246 if (tc_skip_hw(n->flags))
1249 err = u32_reoffload_knode(tp, n, add, cb,
1256 if (!add && !tc_skip_hw(ht->flags))
1257 u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1263 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1265 struct tc_u_knode *n = fh;
1267 if (n && n->res.classid == classid)
1271 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1272 struct sk_buff *skb, struct tcmsg *t, bool rtnl_held)
1274 struct tc_u_knode *n = fh;
1275 struct tc_u_hnode *ht_up, *ht_down;
1276 struct nlattr *nest;
1281 t->tcm_handle = n->handle;
1283 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1285 goto nla_put_failure;
1287 if (TC_U32_KEY(n->handle) == 0) {
1288 struct tc_u_hnode *ht = fh;
1289 u32 divisor = ht->divisor + 1;
1291 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1292 goto nla_put_failure;
1294 #ifdef CONFIG_CLS_U32_PERF
1295 struct tc_u32_pcnt *gpf;
1299 if (nla_put(skb, TCA_U32_SEL,
1300 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1302 goto nla_put_failure;
1304 ht_up = rtnl_dereference(n->ht_up);
1306 u32 htid = n->handle & 0xFFFFF000;
1307 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1308 goto nla_put_failure;
1310 if (n->res.classid &&
1311 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1312 goto nla_put_failure;
1314 ht_down = rtnl_dereference(n->ht_down);
1316 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1317 goto nla_put_failure;
1319 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1320 goto nla_put_failure;
1322 #ifdef CONFIG_CLS_U32_MARK
1323 if ((n->val || n->mask)) {
1324 struct tc_u32_mark mark = {.val = n->val,
1329 for_each_possible_cpu(cpum) {
1330 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1332 mark.success += cnt;
1335 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1336 goto nla_put_failure;
1340 if (tcf_exts_dump(skb, &n->exts) < 0)
1341 goto nla_put_failure;
1344 struct net_device *dev;
1345 dev = __dev_get_by_index(net, n->ifindex);
1346 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1347 goto nla_put_failure;
1349 #ifdef CONFIG_CLS_U32_PERF
1350 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1351 n->sel.nkeys * sizeof(u64),
1354 goto nla_put_failure;
1356 for_each_possible_cpu(cpu) {
1358 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1360 gpf->rcnt += pf->rcnt;
1361 gpf->rhit += pf->rhit;
1362 for (i = 0; i < n->sel.nkeys; i++)
1363 gpf->kcnts[i] += pf->kcnts[i];
1366 if (nla_put_64bit(skb, TCA_U32_PCNT,
1367 sizeof(struct tc_u32_pcnt) +
1368 n->sel.nkeys * sizeof(u64),
1369 gpf, TCA_U32_PAD)) {
1371 goto nla_put_failure;
1377 nla_nest_end(skb, nest);
1379 if (TC_U32_KEY(n->handle))
1380 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1381 goto nla_put_failure;
1385 nla_nest_cancel(skb, nest);
1389 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1391 .classify = u32_classify,
1393 .destroy = u32_destroy,
1395 .change = u32_change,
1396 .delete = u32_delete,
1398 .reoffload = u32_reoffload,
1400 .bind_class = u32_bind_class,
1401 .owner = THIS_MODULE,
1404 static int __init init_u32(void)
1408 pr_info("u32 classifier\n");
1409 #ifdef CONFIG_CLS_U32_PERF
1410 pr_info(" Performance counters on\n");
1412 pr_info(" input device check on\n");
1413 #ifdef CONFIG_NET_CLS_ACT
1414 pr_info(" Actions configured\n");
1416 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1417 sizeof(struct hlist_head),
1419 if (!tc_u_common_hash)
1422 for (i = 0; i < U32_HASH_SIZE; i++)
1423 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1425 ret = register_tcf_proto_ops(&cls_u32_ops);
1427 kvfree(tc_u_common_hash);
1431 static void __exit exit_u32(void)
1433 unregister_tcf_proto_ops(&cls_u32_ops);
1434 kvfree(tc_u_common_hash);
1437 module_init(init_u32)
1438 module_exit(exit_u32)
1439 MODULE_LICENSE("GPL");