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 /* Protected by rtnl lock */
320 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
322 int id = idr_alloc_cyclic(&tp_c->handle_idr, ptr, 1, 0x7FF, GFP_KERNEL);
325 return (id | 0x800U) << 20;
328 static struct hlist_head *tc_u_common_hash;
330 #define U32_HASH_SHIFT 10
331 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
333 static unsigned int tc_u_hash(const struct tcf_proto *tp)
335 return hash_ptr(tp->chain->block, U32_HASH_SHIFT);
338 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
340 struct tc_u_common *tc;
344 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
345 if (tc->block == tp->chain->block)
351 static int u32_init(struct tcf_proto *tp)
353 struct tc_u_hnode *root_ht;
354 struct tc_u_common *tp_c;
357 tp_c = tc_u_common_find(tp);
359 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
364 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
365 root_ht->prio = tp->prio;
366 idr_init(&root_ht->handle_idr);
369 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
374 tp_c->block = tp->chain->block;
375 INIT_HLIST_NODE(&tp_c->hnode);
376 idr_init(&tp_c->handle_idr);
379 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
383 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
384 rcu_assign_pointer(tp_c->hlist, root_ht);
385 root_ht->tp_c = tp_c;
387 rcu_assign_pointer(tp->root, root_ht);
392 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
395 tcf_exts_destroy(&n->exts);
396 tcf_exts_put_net(&n->exts);
398 n->ht_down->refcnt--;
399 #ifdef CONFIG_CLS_U32_PERF
403 #ifdef CONFIG_CLS_U32_MARK
405 free_percpu(n->pcpu_success);
411 /* u32_delete_key_rcu should be called when free'ing a copied
412 * version of a tc_u_knode obtained from u32_init_knode(). When
413 * copies are obtained from u32_init_knode() the statistics are
414 * shared between the old and new copies to allow readers to
415 * continue to update the statistics during the copy. To support
416 * this the u32_delete_key_rcu variant does not free the percpu
419 static void u32_delete_key_work(struct work_struct *work)
421 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
424 u32_destroy_key(key->tp, key, false);
428 static void u32_delete_key_rcu(struct rcu_head *rcu)
430 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
432 INIT_WORK(&key->work, u32_delete_key_work);
433 tcf_queue_work(&key->work);
436 /* u32_delete_key_freepf_rcu is the rcu callback variant
437 * that free's the entire structure including the statistics
438 * percpu variables. Only use this if the key is not a copy
439 * returned by u32_init_knode(). See u32_delete_key_rcu()
440 * for the variant that should be used with keys return from
443 static void u32_delete_key_freepf_work(struct work_struct *work)
445 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
448 u32_destroy_key(key->tp, key, true);
452 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
454 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
456 INIT_WORK(&key->work, u32_delete_key_freepf_work);
457 tcf_queue_work(&key->work);
460 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
462 struct tc_u_knode __rcu **kp;
463 struct tc_u_knode *pkp;
464 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
467 kp = &ht->ht[TC_U32_HASH(key->handle)];
468 for (pkp = rtnl_dereference(*kp); pkp;
469 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
471 RCU_INIT_POINTER(*kp, key->next);
473 tcf_unbind_filter(tp, &key->res);
474 tcf_exts_get_net(&key->exts);
475 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
484 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
485 struct netlink_ext_ack *extack)
487 struct tcf_block *block = tp->chain->block;
488 struct tc_cls_u32_offload cls_u32 = {};
490 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
491 cls_u32.command = TC_CLSU32_DELETE_HNODE;
492 cls_u32.hnode.divisor = h->divisor;
493 cls_u32.hnode.handle = h->handle;
494 cls_u32.hnode.prio = h->prio;
496 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
499 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
500 u32 flags, struct netlink_ext_ack *extack)
502 struct tcf_block *block = tp->chain->block;
503 struct tc_cls_u32_offload cls_u32 = {};
504 bool skip_sw = tc_skip_sw(flags);
505 bool offloaded = false;
508 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
509 cls_u32.command = TC_CLSU32_NEW_HNODE;
510 cls_u32.hnode.divisor = h->divisor;
511 cls_u32.hnode.handle = h->handle;
512 cls_u32.hnode.prio = h->prio;
514 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
516 u32_clear_hw_hnode(tp, h, NULL);
518 } else if (err > 0) {
522 if (skip_sw && !offloaded)
528 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
529 struct netlink_ext_ack *extack)
531 struct tcf_block *block = tp->chain->block;
532 struct tc_cls_u32_offload cls_u32 = {};
534 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
535 cls_u32.command = TC_CLSU32_DELETE_KNODE;
536 cls_u32.knode.handle = n->handle;
538 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
539 tcf_block_offload_dec(block, &n->flags);
542 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
543 u32 flags, struct netlink_ext_ack *extack)
545 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
546 struct tcf_block *block = tp->chain->block;
547 struct tc_cls_u32_offload cls_u32 = {};
548 bool skip_sw = tc_skip_sw(flags);
551 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
552 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
553 cls_u32.knode.handle = n->handle;
554 cls_u32.knode.fshift = n->fshift;
555 #ifdef CONFIG_CLS_U32_MARK
556 cls_u32.knode.val = n->val;
557 cls_u32.knode.mask = n->mask;
559 cls_u32.knode.val = 0;
560 cls_u32.knode.mask = 0;
562 cls_u32.knode.sel = &n->sel;
563 cls_u32.knode.exts = &n->exts;
565 cls_u32.knode.link_handle = ht->handle;
567 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
569 u32_remove_hw_knode(tp, n, NULL);
571 } else if (err > 0) {
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_knode *n;
587 for (h = 0; h <= ht->divisor; h++) {
588 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
589 RCU_INIT_POINTER(ht->ht[h],
590 rtnl_dereference(n->next));
591 tcf_unbind_filter(tp, &n->res);
592 u32_remove_hw_knode(tp, n, extack);
593 idr_remove(&ht->handle_idr, n->handle);
594 if (tcf_exts_get_net(&n->exts))
595 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
597 u32_destroy_key(n->tp, n, true);
602 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
603 struct netlink_ext_ack *extack)
605 struct tc_u_common *tp_c = tp->data;
606 struct tc_u_hnode __rcu **hn;
607 struct tc_u_hnode *phn;
611 u32_clear_hnode(tp, ht, extack);
614 for (phn = rtnl_dereference(*hn);
616 hn = &phn->next, phn = rtnl_dereference(*hn)) {
618 u32_clear_hw_hnode(tp, ht, extack);
619 idr_destroy(&ht->handle_idr);
620 idr_remove(&tp_c->handle_idr, ht->handle);
621 RCU_INIT_POINTER(*hn, ht->next);
630 static bool ht_empty(struct tc_u_hnode *ht)
634 for (h = 0; h <= ht->divisor; h++)
635 if (rcu_access_pointer(ht->ht[h]))
641 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
643 struct tc_u_common *tp_c = tp->data;
644 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
646 WARN_ON(root_ht == NULL);
648 if (root_ht && --root_ht->refcnt == 0)
649 u32_destroy_hnode(tp, root_ht, extack);
651 if (--tp_c->refcnt == 0) {
652 struct tc_u_hnode *ht;
654 hlist_del(&tp_c->hnode);
656 for (ht = rtnl_dereference(tp_c->hlist);
658 ht = rtnl_dereference(ht->next)) {
660 u32_clear_hnode(tp, ht, extack);
663 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
664 RCU_INIT_POINTER(tp_c->hlist, ht->next);
668 idr_destroy(&tp_c->handle_idr);
675 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
676 struct netlink_ext_ack *extack)
678 struct tc_u_hnode *ht = arg;
679 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
680 struct tc_u_common *tp_c = tp->data;
686 if (TC_U32_KEY(ht->handle)) {
687 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
688 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
693 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
697 if (ht->refcnt == 1) {
699 u32_destroy_hnode(tp, ht, extack);
701 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
708 if (root_ht->refcnt > 1) {
712 if (root_ht->refcnt == 1) {
713 if (!ht_empty(root_ht)) {
720 if (tp_c->refcnt > 1) {
725 if (tp_c->refcnt == 1) {
726 struct tc_u_hnode *ht;
728 for (ht = rtnl_dereference(tp_c->hlist);
730 ht = rtnl_dereference(ht->next))
741 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
743 unsigned long idr_index;
744 u32 start = htid | 0x800;
745 u32 max = htid | 0xFFF;
748 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
749 start, max + 1, GFP_KERNEL)) {
750 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
751 min + 1, max + 1, GFP_KERNEL))
755 return (u32)idr_index;
758 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
759 [TCA_U32_CLASSID] = { .type = NLA_U32 },
760 [TCA_U32_HASH] = { .type = NLA_U32 },
761 [TCA_U32_LINK] = { .type = NLA_U32 },
762 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
763 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
764 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
765 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
766 [TCA_U32_FLAGS] = { .type = NLA_U32 },
769 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
770 unsigned long base, struct tc_u_hnode *ht,
771 struct tc_u_knode *n, struct nlattr **tb,
772 struct nlattr *est, bool ovr,
773 struct netlink_ext_ack *extack)
777 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack);
781 if (tb[TCA_U32_LINK]) {
782 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
783 struct tc_u_hnode *ht_down = NULL, *ht_old;
785 if (TC_U32_KEY(handle)) {
786 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
791 ht_down = u32_lookup_ht(ht->tp_c, handle);
794 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
800 ht_old = rtnl_dereference(n->ht_down);
801 rcu_assign_pointer(n->ht_down, ht_down);
806 if (tb[TCA_U32_CLASSID]) {
807 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
808 tcf_bind_filter(tp, &n->res, base);
811 #ifdef CONFIG_NET_CLS_IND
812 if (tb[TCA_U32_INDEV]) {
814 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
823 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
824 struct tc_u_knode *n)
826 struct tc_u_knode __rcu **ins;
827 struct tc_u_knode *pins;
828 struct tc_u_hnode *ht;
830 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
831 ht = rtnl_dereference(tp->root);
833 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
835 ins = &ht->ht[TC_U32_HASH(n->handle)];
837 /* The node must always exist for it to be replaced if this is not the
838 * case then something went very wrong elsewhere.
840 for (pins = rtnl_dereference(*ins); ;
841 ins = &pins->next, pins = rtnl_dereference(*ins))
842 if (pins->handle == n->handle)
845 idr_replace(&ht->handle_idr, n, n->handle);
846 RCU_INIT_POINTER(n->next, pins->next);
847 rcu_assign_pointer(*ins, n);
850 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
851 struct tc_u_knode *n)
853 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
854 struct tc_u32_sel *s = &n->sel;
855 struct tc_u_knode *new;
857 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
863 RCU_INIT_POINTER(new->next, n->next);
864 new->handle = n->handle;
865 RCU_INIT_POINTER(new->ht_up, n->ht_up);
867 #ifdef CONFIG_NET_CLS_IND
868 new->ifindex = n->ifindex;
870 new->fshift = n->fshift;
872 new->flags = n->flags;
873 RCU_INIT_POINTER(new->ht_down, ht);
875 /* bump reference count as long as we hold pointer to structure */
879 #ifdef CONFIG_CLS_U32_PERF
880 /* Statistics may be incremented by readers during update
881 * so we must keep them in tact. When the node is later destroyed
882 * a special destroy call must be made to not free the pf memory.
887 #ifdef CONFIG_CLS_U32_MARK
890 /* Similarly success statistics must be moved as pointers */
891 new->pcpu_success = n->pcpu_success;
894 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
896 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
904 static int u32_change(struct net *net, struct sk_buff *in_skb,
905 struct tcf_proto *tp, unsigned long base, u32 handle,
906 struct nlattr **tca, void **arg, bool ovr,
907 struct netlink_ext_ack *extack)
909 struct tc_u_common *tp_c = tp->data;
910 struct tc_u_hnode *ht;
911 struct tc_u_knode *n;
912 struct tc_u32_sel *s;
913 struct nlattr *opt = tca[TCA_OPTIONS];
914 struct nlattr *tb[TCA_U32_MAX + 1];
917 #ifdef CONFIG_CLS_U32_PERF
923 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
930 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack);
934 if (tb[TCA_U32_FLAGS]) {
935 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
936 if (!tc_flags_valid(flags)) {
937 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
944 struct tc_u_knode *new;
946 if (TC_U32_KEY(n->handle) == 0) {
947 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
951 if (n->flags != flags) {
952 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
956 new = u32_init_knode(tp, n);
960 err = u32_set_parms(net, tp, base,
961 rtnl_dereference(n->ht_up), new, tb,
962 tca[TCA_RATE], ovr, extack);
965 u32_destroy_key(tp, new, false);
969 err = u32_replace_hw_knode(tp, new, flags, extack);
971 u32_destroy_key(tp, new, false);
975 if (!tc_in_hw(new->flags))
976 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
978 u32_replace_knode(tp, tp_c, new);
979 tcf_unbind_filter(tp, &n->res);
980 tcf_exts_get_net(&n->exts);
981 call_rcu(&n->rcu, u32_delete_key_rcu);
985 if (tb[TCA_U32_DIVISOR]) {
986 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
988 if (--divisor > 0x100) {
989 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
992 if (TC_U32_KEY(handle)) {
993 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
996 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
1000 handle = gen_new_htid(tp->data, ht);
1006 err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL,
1007 handle, handle + 1, GFP_KERNEL);
1015 ht->divisor = divisor;
1016 ht->handle = handle;
1017 ht->prio = tp->prio;
1018 idr_init(&ht->handle_idr);
1021 err = u32_replace_hw_hnode(tp, ht, flags, extack);
1023 idr_remove(&tp_c->handle_idr, handle);
1028 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1029 rcu_assign_pointer(tp_c->hlist, ht);
1035 if (tb[TCA_U32_HASH]) {
1036 htid = nla_get_u32(tb[TCA_U32_HASH]);
1037 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1038 ht = rtnl_dereference(tp->root);
1041 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1043 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1048 ht = rtnl_dereference(tp->root);
1052 if (ht->divisor < TC_U32_HASH(htid)) {
1053 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1058 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1059 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1062 handle = htid | TC_U32_NODE(handle);
1063 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL,
1069 handle = gen_new_kid(ht, htid);
1071 if (tb[TCA_U32_SEL] == NULL) {
1072 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1077 s = nla_data(tb[TCA_U32_SEL]);
1079 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1085 #ifdef CONFIG_CLS_U32_PERF
1086 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1087 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1094 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1095 RCU_INIT_POINTER(n->ht_up, ht);
1097 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1101 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1105 #ifdef CONFIG_CLS_U32_MARK
1106 n->pcpu_success = alloc_percpu(u32);
1107 if (!n->pcpu_success) {
1112 if (tb[TCA_U32_MARK]) {
1113 struct tc_u32_mark *mark;
1115 mark = nla_data(tb[TCA_U32_MARK]);
1117 n->mask = mark->mask;
1121 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr,
1124 struct tc_u_knode __rcu **ins;
1125 struct tc_u_knode *pins;
1127 err = u32_replace_hw_knode(tp, n, flags, extack);
1131 if (!tc_in_hw(n->flags))
1132 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1134 ins = &ht->ht[TC_U32_HASH(handle)];
1135 for (pins = rtnl_dereference(*ins); pins;
1136 ins = &pins->next, pins = rtnl_dereference(*ins))
1137 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1140 RCU_INIT_POINTER(n->next, pins);
1141 rcu_assign_pointer(*ins, n);
1147 #ifdef CONFIG_CLS_U32_MARK
1148 free_percpu(n->pcpu_success);
1152 tcf_exts_destroy(&n->exts);
1153 #ifdef CONFIG_CLS_U32_PERF
1159 idr_remove(&ht->handle_idr, handle);
1163 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1165 struct tc_u_common *tp_c = tp->data;
1166 struct tc_u_hnode *ht;
1167 struct tc_u_knode *n;
1173 for (ht = rtnl_dereference(tp_c->hlist);
1175 ht = rtnl_dereference(ht->next)) {
1176 if (ht->prio != tp->prio)
1178 if (arg->count >= arg->skip) {
1179 if (arg->fn(tp, ht, arg) < 0) {
1185 for (h = 0; h <= ht->divisor; h++) {
1186 for (n = rtnl_dereference(ht->ht[h]);
1188 n = rtnl_dereference(n->next)) {
1189 if (arg->count < arg->skip) {
1193 if (arg->fn(tp, n, arg) < 0) {
1203 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1205 struct tc_u_knode *n = fh;
1207 if (n && n->res.classid == classid)
1211 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1212 struct sk_buff *skb, struct tcmsg *t)
1214 struct tc_u_knode *n = fh;
1215 struct tc_u_hnode *ht_up, *ht_down;
1216 struct nlattr *nest;
1221 t->tcm_handle = n->handle;
1223 nest = nla_nest_start(skb, TCA_OPTIONS);
1225 goto nla_put_failure;
1227 if (TC_U32_KEY(n->handle) == 0) {
1228 struct tc_u_hnode *ht = fh;
1229 u32 divisor = ht->divisor + 1;
1231 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1232 goto nla_put_failure;
1234 #ifdef CONFIG_CLS_U32_PERF
1235 struct tc_u32_pcnt *gpf;
1239 if (nla_put(skb, TCA_U32_SEL,
1240 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1242 goto nla_put_failure;
1244 ht_up = rtnl_dereference(n->ht_up);
1246 u32 htid = n->handle & 0xFFFFF000;
1247 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1248 goto nla_put_failure;
1250 if (n->res.classid &&
1251 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1252 goto nla_put_failure;
1254 ht_down = rtnl_dereference(n->ht_down);
1256 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1257 goto nla_put_failure;
1259 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1260 goto nla_put_failure;
1262 #ifdef CONFIG_CLS_U32_MARK
1263 if ((n->val || n->mask)) {
1264 struct tc_u32_mark mark = {.val = n->val,
1269 for_each_possible_cpu(cpum) {
1270 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1272 mark.success += cnt;
1275 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1276 goto nla_put_failure;
1280 if (tcf_exts_dump(skb, &n->exts) < 0)
1281 goto nla_put_failure;
1283 #ifdef CONFIG_NET_CLS_IND
1285 struct net_device *dev;
1286 dev = __dev_get_by_index(net, n->ifindex);
1287 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1288 goto nla_put_failure;
1291 #ifdef CONFIG_CLS_U32_PERF
1292 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1293 n->sel.nkeys * sizeof(u64),
1296 goto nla_put_failure;
1298 for_each_possible_cpu(cpu) {
1300 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1302 gpf->rcnt += pf->rcnt;
1303 gpf->rhit += pf->rhit;
1304 for (i = 0; i < n->sel.nkeys; i++)
1305 gpf->kcnts[i] += pf->kcnts[i];
1308 if (nla_put_64bit(skb, TCA_U32_PCNT,
1309 sizeof(struct tc_u32_pcnt) +
1310 n->sel.nkeys * sizeof(u64),
1311 gpf, TCA_U32_PAD)) {
1313 goto nla_put_failure;
1319 nla_nest_end(skb, nest);
1321 if (TC_U32_KEY(n->handle))
1322 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1323 goto nla_put_failure;
1327 nla_nest_cancel(skb, nest);
1331 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1333 .classify = u32_classify,
1335 .destroy = u32_destroy,
1337 .change = u32_change,
1338 .delete = u32_delete,
1341 .bind_class = u32_bind_class,
1342 .owner = THIS_MODULE,
1345 static int __init init_u32(void)
1349 pr_info("u32 classifier\n");
1350 #ifdef CONFIG_CLS_U32_PERF
1351 pr_info(" Performance counters on\n");
1353 #ifdef CONFIG_NET_CLS_IND
1354 pr_info(" input device check on\n");
1356 #ifdef CONFIG_NET_CLS_ACT
1357 pr_info(" Actions configured\n");
1359 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1360 sizeof(struct hlist_head),
1362 if (!tc_u_common_hash)
1365 for (i = 0; i < U32_HASH_SIZE; i++)
1366 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1368 ret = register_tcf_proto_ops(&cls_u32_ops);
1370 kvfree(tc_u_common_hash);
1374 static void __exit exit_u32(void)
1376 unregister_tcf_proto_ops(&cls_u32_ops);
1377 kvfree(tc_u_common_hash);
1380 module_init(init_u32)
1381 module_exit(exit_u32)
1382 MODULE_LICENSE("GPL");