2 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
11 * The filters are packed to hash tables of key nodes
12 * with a set of 32bit key/mask pairs at every node.
13 * Nodes reference next level hash tables etc.
15 * This scheme is the best universal classifier I managed to
16 * invent; it is not super-fast, but it is not slow (provided you
17 * program it correctly), and general enough. And its relative
18 * speed grows as the number of rules becomes larger.
20 * It seems that it represents the best middle point between
21 * speed and manageability both by human and by machine.
23 * It is especially useful for link sharing combined with QoS;
24 * pure RSVP doesn't need such a general approach and can use
25 * much simpler (and faster) schemes, sort of cls_rsvp.c.
27 * JHS: We should remove the CONFIG_NET_CLS_IND from here
28 * eventually when the meta match extension is made available
30 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
33 #include <linux/module.h>
34 #include <linux/slab.h>
35 #include <linux/types.h>
36 #include <linux/kernel.h>
37 #include <linux/string.h>
38 #include <linux/errno.h>
39 #include <linux/percpu.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/skbuff.h>
42 #include <linux/bitmap.h>
43 #include <linux/netdevice.h>
44 #include <linux/hash.h>
45 #include <net/netlink.h>
46 #include <net/act_api.h>
47 #include <net/pkt_cls.h>
48 #include <linux/idr.h>
51 struct tc_u_knode __rcu *next;
53 struct tc_u_hnode __rcu *ht_up;
55 #ifdef CONFIG_NET_CLS_IND
59 struct tcf_result res;
60 struct tc_u_hnode __rcu *ht_down;
61 #ifdef CONFIG_CLS_U32_PERF
62 struct tc_u32_pcnt __percpu *pf;
65 #ifdef CONFIG_CLS_U32_MARK
68 u32 __percpu *pcpu_success;
72 struct work_struct work;
75 /* The 'sel' field MUST be the last field in structure to allow for
76 * tc_u32_keys allocated at end of structure.
78 struct tc_u32_sel sel;
82 struct tc_u_hnode __rcu *next;
85 struct tc_u_common *tp_c;
88 struct idr handle_idr;
91 /* The 'ht' field MUST be the last field in structure to allow for
92 * more entries allocated at end of structure.
94 struct tc_u_knode __rcu *ht[1];
98 struct tc_u_hnode __rcu *hlist;
99 struct tcf_block *block;
101 struct idr handle_idr;
102 struct hlist_node hnode;
106 static inline unsigned int u32_hash_fold(__be32 key,
107 const struct tc_u32_sel *sel,
110 unsigned int h = ntohl(key & sel->hmask) >> fshift;
115 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
116 struct tcf_result *res)
119 struct tc_u_knode *knode;
121 } stack[TC_U32_MAXDEPTH];
123 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
124 unsigned int off = skb_network_offset(skb);
125 struct tc_u_knode *n;
129 #ifdef CONFIG_CLS_U32_PERF
135 n = rcu_dereference_bh(ht->ht[sel]);
139 struct tc_u32_key *key = n->sel.keys;
141 #ifdef CONFIG_CLS_U32_PERF
142 __this_cpu_inc(n->pf->rcnt);
146 if (tc_skip_sw(n->flags)) {
147 n = rcu_dereference_bh(n->next);
151 #ifdef CONFIG_CLS_U32_MARK
152 if ((skb->mark & n->mask) != n->val) {
153 n = rcu_dereference_bh(n->next);
156 __this_cpu_inc(*n->pcpu_success);
160 for (i = n->sel.nkeys; i > 0; i--, key++) {
161 int toff = off + key->off + (off2 & key->offmask);
164 if (skb_headroom(skb) + toff > INT_MAX)
167 data = skb_header_pointer(skb, toff, 4, &hdata);
170 if ((*data ^ key->val) & key->mask) {
171 n = rcu_dereference_bh(n->next);
174 #ifdef CONFIG_CLS_U32_PERF
175 __this_cpu_inc(n->pf->kcnts[j]);
180 ht = rcu_dereference_bh(n->ht_down);
183 if (n->sel.flags & TC_U32_TERMINAL) {
186 #ifdef CONFIG_NET_CLS_IND
187 if (!tcf_match_indev(skb, n->ifindex)) {
188 n = rcu_dereference_bh(n->next);
192 #ifdef CONFIG_CLS_U32_PERF
193 __this_cpu_inc(n->pf->rhit);
195 r = tcf_exts_exec(skb, &n->exts, res);
197 n = rcu_dereference_bh(n->next);
203 n = rcu_dereference_bh(n->next);
208 if (sdepth >= TC_U32_MAXDEPTH)
210 stack[sdepth].knode = n;
211 stack[sdepth].off = off;
214 ht = rcu_dereference_bh(n->ht_down);
219 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
223 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
226 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
229 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
230 off2 = n->sel.off + 3;
231 if (n->sel.flags & TC_U32_VAROFFSET) {
234 data = skb_header_pointer(skb,
239 off2 += ntohs(n->sel.offmask & *data) >>
244 if (n->sel.flags & TC_U32_EAT) {
255 n = stack[sdepth].knode;
256 ht = rcu_dereference_bh(n->ht_up);
257 off = stack[sdepth].off;
264 net_warn_ratelimited("cls_u32: dead loop\n");
268 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
270 struct tc_u_hnode *ht;
272 for (ht = rtnl_dereference(tp_c->hlist);
274 ht = rtnl_dereference(ht->next))
275 if (ht->handle == handle)
281 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
284 struct tc_u_knode *n = NULL;
286 sel = TC_U32_HASH(handle);
287 if (sel > ht->divisor)
290 for (n = rtnl_dereference(ht->ht[sel]);
292 n = rtnl_dereference(n->next))
293 if (n->handle == handle)
300 static void *u32_get(struct tcf_proto *tp, u32 handle)
302 struct tc_u_hnode *ht;
303 struct tc_u_common *tp_c = tp->data;
305 if (TC_U32_HTID(handle) == TC_U32_ROOT)
306 ht = rtnl_dereference(tp->root);
308 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
313 if (TC_U32_KEY(handle) == 0)
316 return u32_lookup_key(ht, handle);
319 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
321 unsigned long idr_index;
324 /* This is only used inside rtnl lock it is safe to increment
325 * without read _copy_ update semantics
327 err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index,
328 1, 0x7FF, GFP_KERNEL);
331 return (u32)(idr_index | 0x800) << 20;
334 static struct hlist_head *tc_u_common_hash;
336 #define U32_HASH_SHIFT 10
337 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
339 static unsigned int tc_u_hash(const struct tcf_proto *tp)
341 return hash_ptr(tp->chain->block, U32_HASH_SHIFT);
344 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
346 struct tc_u_common *tc;
350 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
351 if (tc->block == tp->chain->block)
357 static int u32_init(struct tcf_proto *tp)
359 struct tc_u_hnode *root_ht;
360 struct tc_u_common *tp_c;
363 tp_c = tc_u_common_find(tp);
365 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
370 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
371 root_ht->prio = tp->prio;
372 idr_init(&root_ht->handle_idr);
375 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
380 tp_c->block = tp->chain->block;
381 INIT_HLIST_NODE(&tp_c->hnode);
382 idr_init(&tp_c->handle_idr);
385 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
389 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
390 rcu_assign_pointer(tp_c->hlist, root_ht);
391 root_ht->tp_c = tp_c;
393 rcu_assign_pointer(tp->root, root_ht);
398 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
401 tcf_exts_destroy(&n->exts);
402 tcf_exts_put_net(&n->exts);
404 n->ht_down->refcnt--;
405 #ifdef CONFIG_CLS_U32_PERF
409 #ifdef CONFIG_CLS_U32_MARK
411 free_percpu(n->pcpu_success);
417 /* u32_delete_key_rcu should be called when free'ing a copied
418 * version of a tc_u_knode obtained from u32_init_knode(). When
419 * copies are obtained from u32_init_knode() the statistics are
420 * shared between the old and new copies to allow readers to
421 * continue to update the statistics during the copy. To support
422 * this the u32_delete_key_rcu variant does not free the percpu
425 static void u32_delete_key_work(struct work_struct *work)
427 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
430 u32_destroy_key(key->tp, key, false);
434 static void u32_delete_key_rcu(struct rcu_head *rcu)
436 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
438 INIT_WORK(&key->work, u32_delete_key_work);
439 tcf_queue_work(&key->work);
442 /* u32_delete_key_freepf_rcu is the rcu callback variant
443 * that free's the entire structure including the statistics
444 * percpu variables. Only use this if the key is not a copy
445 * returned by u32_init_knode(). See u32_delete_key_rcu()
446 * for the variant that should be used with keys return from
449 static void u32_delete_key_freepf_work(struct work_struct *work)
451 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
454 u32_destroy_key(key->tp, key, true);
458 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
460 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
462 INIT_WORK(&key->work, u32_delete_key_freepf_work);
463 tcf_queue_work(&key->work);
466 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
468 struct tc_u_knode __rcu **kp;
469 struct tc_u_knode *pkp;
470 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
473 kp = &ht->ht[TC_U32_HASH(key->handle)];
474 for (pkp = rtnl_dereference(*kp); pkp;
475 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
477 RCU_INIT_POINTER(*kp, key->next);
479 tcf_unbind_filter(tp, &key->res);
480 tcf_exts_get_net(&key->exts);
481 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
490 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
491 struct netlink_ext_ack *extack)
493 struct tcf_block *block = tp->chain->block;
494 struct tc_cls_u32_offload cls_u32 = {};
496 tc_cls_common_offload_init(&cls_u32.common, tp, h->flags, extack);
497 cls_u32.command = TC_CLSU32_DELETE_HNODE;
498 cls_u32.hnode.divisor = h->divisor;
499 cls_u32.hnode.handle = h->handle;
500 cls_u32.hnode.prio = h->prio;
502 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
505 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
506 u32 flags, struct netlink_ext_ack *extack)
508 struct tcf_block *block = tp->chain->block;
509 struct tc_cls_u32_offload cls_u32 = {};
510 bool skip_sw = tc_skip_sw(flags);
511 bool offloaded = false;
514 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
515 cls_u32.command = TC_CLSU32_NEW_HNODE;
516 cls_u32.hnode.divisor = h->divisor;
517 cls_u32.hnode.handle = h->handle;
518 cls_u32.hnode.prio = h->prio;
520 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
522 u32_clear_hw_hnode(tp, h, NULL);
524 } else if (err > 0) {
528 if (skip_sw && !offloaded)
534 static void u32_remove_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
535 struct netlink_ext_ack *extack)
537 struct tcf_block *block = tp->chain->block;
538 struct tc_cls_u32_offload cls_u32 = {};
540 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
541 cls_u32.command = TC_CLSU32_DELETE_KNODE;
542 cls_u32.knode.handle = n->handle;
544 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
545 tcf_block_offload_dec(block, &n->flags);
548 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
549 u32 flags, struct netlink_ext_ack *extack)
551 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
552 struct tcf_block *block = tp->chain->block;
553 struct tc_cls_u32_offload cls_u32 = {};
554 bool skip_sw = tc_skip_sw(flags);
557 tc_cls_common_offload_init(&cls_u32.common, tp, flags, extack);
558 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
559 cls_u32.knode.handle = n->handle;
560 cls_u32.knode.fshift = n->fshift;
561 #ifdef CONFIG_CLS_U32_MARK
562 cls_u32.knode.val = n->val;
563 cls_u32.knode.mask = n->mask;
565 cls_u32.knode.val = 0;
566 cls_u32.knode.mask = 0;
568 cls_u32.knode.sel = &n->sel;
569 cls_u32.knode.exts = &n->exts;
571 cls_u32.knode.link_handle = ht->handle;
573 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
575 u32_remove_hw_knode(tp, n, NULL);
577 } else if (err > 0) {
578 tcf_block_offload_inc(block, &n->flags);
581 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
587 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
588 struct netlink_ext_ack *extack)
590 struct tc_u_knode *n;
593 for (h = 0; h <= ht->divisor; h++) {
594 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
595 RCU_INIT_POINTER(ht->ht[h],
596 rtnl_dereference(n->next));
597 tcf_unbind_filter(tp, &n->res);
598 u32_remove_hw_knode(tp, n, extack);
599 idr_remove_ext(&ht->handle_idr, n->handle);
600 if (tcf_exts_get_net(&n->exts))
601 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
603 u32_destroy_key(n->tp, n, true);
608 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
609 struct netlink_ext_ack *extack)
611 struct tc_u_common *tp_c = tp->data;
612 struct tc_u_hnode __rcu **hn;
613 struct tc_u_hnode *phn;
617 u32_clear_hnode(tp, ht, extack);
620 for (phn = rtnl_dereference(*hn);
622 hn = &phn->next, phn = rtnl_dereference(*hn)) {
624 u32_clear_hw_hnode(tp, ht, extack);
625 idr_destroy(&ht->handle_idr);
626 idr_remove_ext(&tp_c->handle_idr, ht->handle);
627 RCU_INIT_POINTER(*hn, ht->next);
636 static bool ht_empty(struct tc_u_hnode *ht)
640 for (h = 0; h <= ht->divisor; h++)
641 if (rcu_access_pointer(ht->ht[h]))
647 static void u32_destroy(struct tcf_proto *tp, struct netlink_ext_ack *extack)
649 struct tc_u_common *tp_c = tp->data;
650 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
652 WARN_ON(root_ht == NULL);
654 if (root_ht && --root_ht->refcnt == 0)
655 u32_destroy_hnode(tp, root_ht, extack);
657 if (--tp_c->refcnt == 0) {
658 struct tc_u_hnode *ht;
660 hlist_del(&tp_c->hnode);
662 for (ht = rtnl_dereference(tp_c->hlist);
664 ht = rtnl_dereference(ht->next)) {
666 u32_clear_hnode(tp, ht, extack);
669 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
670 RCU_INIT_POINTER(tp_c->hlist, ht->next);
674 idr_destroy(&tp_c->handle_idr);
681 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
682 struct netlink_ext_ack *extack)
684 struct tc_u_hnode *ht = arg;
685 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
686 struct tc_u_common *tp_c = tp->data;
692 if (TC_U32_KEY(ht->handle)) {
693 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
694 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
699 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
703 if (ht->refcnt == 1) {
705 u32_destroy_hnode(tp, ht, extack);
707 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
714 if (root_ht->refcnt > 1) {
718 if (root_ht->refcnt == 1) {
719 if (!ht_empty(root_ht)) {
726 if (tp_c->refcnt > 1) {
731 if (tp_c->refcnt == 1) {
732 struct tc_u_hnode *ht;
734 for (ht = rtnl_dereference(tp_c->hlist);
736 ht = rtnl_dereference(ht->next))
747 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
749 unsigned long idr_index;
750 u32 start = htid | 0x800;
751 u32 max = htid | 0xFFF;
754 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
755 start, max + 1, GFP_KERNEL)) {
756 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
757 min + 1, max + 1, GFP_KERNEL))
761 return (u32)idr_index;
764 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
765 [TCA_U32_CLASSID] = { .type = NLA_U32 },
766 [TCA_U32_HASH] = { .type = NLA_U32 },
767 [TCA_U32_LINK] = { .type = NLA_U32 },
768 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
769 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
770 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
771 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
772 [TCA_U32_FLAGS] = { .type = NLA_U32 },
775 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
776 unsigned long base, struct tc_u_hnode *ht,
777 struct tc_u_knode *n, struct nlattr **tb,
778 struct nlattr *est, bool ovr,
779 struct netlink_ext_ack *extack)
783 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, extack);
787 if (tb[TCA_U32_LINK]) {
788 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
789 struct tc_u_hnode *ht_down = NULL, *ht_old;
791 if (TC_U32_KEY(handle)) {
792 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
797 ht_down = u32_lookup_ht(ht->tp_c, handle);
800 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
806 ht_old = rtnl_dereference(n->ht_down);
807 rcu_assign_pointer(n->ht_down, ht_down);
812 if (tb[TCA_U32_CLASSID]) {
813 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
814 tcf_bind_filter(tp, &n->res, base);
817 #ifdef CONFIG_NET_CLS_IND
818 if (tb[TCA_U32_INDEV]) {
820 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
829 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
830 struct tc_u_knode *n)
832 struct tc_u_knode __rcu **ins;
833 struct tc_u_knode *pins;
834 struct tc_u_hnode *ht;
836 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
837 ht = rtnl_dereference(tp->root);
839 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
841 ins = &ht->ht[TC_U32_HASH(n->handle)];
843 /* The node must always exist for it to be replaced if this is not the
844 * case then something went very wrong elsewhere.
846 for (pins = rtnl_dereference(*ins); ;
847 ins = &pins->next, pins = rtnl_dereference(*ins))
848 if (pins->handle == n->handle)
851 idr_replace_ext(&ht->handle_idr, n, n->handle);
852 RCU_INIT_POINTER(n->next, pins->next);
853 rcu_assign_pointer(*ins, n);
856 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
857 struct tc_u_knode *n)
859 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
860 struct tc_u32_sel *s = &n->sel;
861 struct tc_u_knode *new;
863 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
869 RCU_INIT_POINTER(new->next, n->next);
870 new->handle = n->handle;
871 RCU_INIT_POINTER(new->ht_up, n->ht_up);
873 #ifdef CONFIG_NET_CLS_IND
874 new->ifindex = n->ifindex;
876 new->fshift = n->fshift;
878 new->flags = n->flags;
879 RCU_INIT_POINTER(new->ht_down, ht);
881 /* bump reference count as long as we hold pointer to structure */
885 #ifdef CONFIG_CLS_U32_PERF
886 /* Statistics may be incremented by readers during update
887 * so we must keep them in tact. When the node is later destroyed
888 * a special destroy call must be made to not free the pf memory.
893 #ifdef CONFIG_CLS_U32_MARK
896 /* Similarly success statistics must be moved as pointers */
897 new->pcpu_success = n->pcpu_success;
900 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
902 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
910 static int u32_change(struct net *net, struct sk_buff *in_skb,
911 struct tcf_proto *tp, unsigned long base, u32 handle,
912 struct nlattr **tca, void **arg, bool ovr,
913 struct netlink_ext_ack *extack)
915 struct tc_u_common *tp_c = tp->data;
916 struct tc_u_hnode *ht;
917 struct tc_u_knode *n;
918 struct tc_u32_sel *s;
919 struct nlattr *opt = tca[TCA_OPTIONS];
920 struct nlattr *tb[TCA_U32_MAX + 1];
923 #ifdef CONFIG_CLS_U32_PERF
929 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
936 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, extack);
940 if (tb[TCA_U32_FLAGS]) {
941 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
942 if (!tc_flags_valid(flags)) {
943 NL_SET_ERR_MSG_MOD(extack, "Invalid filter flags");
950 struct tc_u_knode *new;
952 if (TC_U32_KEY(n->handle) == 0) {
953 NL_SET_ERR_MSG_MOD(extack, "Key node id cannot be zero");
957 if (n->flags != flags) {
958 NL_SET_ERR_MSG_MOD(extack, "Key node flags do not match passed flags");
962 new = u32_init_knode(tp, n);
966 err = u32_set_parms(net, tp, base,
967 rtnl_dereference(n->ht_up), new, tb,
968 tca[TCA_RATE], ovr, extack);
971 u32_destroy_key(tp, new, false);
975 err = u32_replace_hw_knode(tp, new, flags, extack);
977 u32_destroy_key(tp, new, false);
981 if (!tc_in_hw(new->flags))
982 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
984 u32_replace_knode(tp, tp_c, new);
985 tcf_unbind_filter(tp, &n->res);
986 tcf_exts_get_net(&n->exts);
987 call_rcu(&n->rcu, u32_delete_key_rcu);
991 if (tb[TCA_U32_DIVISOR]) {
992 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
994 if (--divisor > 0x100) {
995 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
998 if (TC_U32_KEY(handle)) {
999 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
1002 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
1006 handle = gen_new_htid(tp->data, ht);
1012 err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL,
1013 handle, handle + 1, GFP_KERNEL);
1021 ht->divisor = divisor;
1022 ht->handle = handle;
1023 ht->prio = tp->prio;
1024 idr_init(&ht->handle_idr);
1027 err = u32_replace_hw_hnode(tp, ht, flags, extack);
1029 idr_remove_ext(&tp_c->handle_idr, handle);
1034 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1035 rcu_assign_pointer(tp_c->hlist, ht);
1041 if (tb[TCA_U32_HASH]) {
1042 htid = nla_get_u32(tb[TCA_U32_HASH]);
1043 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1044 ht = rtnl_dereference(tp->root);
1047 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1049 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
1054 ht = rtnl_dereference(tp->root);
1058 if (ht->divisor < TC_U32_HASH(htid)) {
1059 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
1064 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1065 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1068 handle = htid | TC_U32_NODE(handle);
1069 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL,
1075 handle = gen_new_kid(ht, htid);
1077 if (tb[TCA_U32_SEL] == NULL) {
1078 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1083 s = nla_data(tb[TCA_U32_SEL]);
1085 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1091 #ifdef CONFIG_CLS_U32_PERF
1092 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1093 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1100 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1101 RCU_INIT_POINTER(n->ht_up, ht);
1103 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1107 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1111 #ifdef CONFIG_CLS_U32_MARK
1112 n->pcpu_success = alloc_percpu(u32);
1113 if (!n->pcpu_success) {
1118 if (tb[TCA_U32_MARK]) {
1119 struct tc_u32_mark *mark;
1121 mark = nla_data(tb[TCA_U32_MARK]);
1123 n->mask = mark->mask;
1127 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr,
1130 struct tc_u_knode __rcu **ins;
1131 struct tc_u_knode *pins;
1133 err = u32_replace_hw_knode(tp, n, flags, extack);
1137 if (!tc_in_hw(n->flags))
1138 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1140 ins = &ht->ht[TC_U32_HASH(handle)];
1141 for (pins = rtnl_dereference(*ins); pins;
1142 ins = &pins->next, pins = rtnl_dereference(*ins))
1143 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1146 RCU_INIT_POINTER(n->next, pins);
1147 rcu_assign_pointer(*ins, n);
1153 #ifdef CONFIG_CLS_U32_MARK
1154 free_percpu(n->pcpu_success);
1158 tcf_exts_destroy(&n->exts);
1159 #ifdef CONFIG_CLS_U32_PERF
1165 idr_remove_ext(&ht->handle_idr, handle);
1169 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1171 struct tc_u_common *tp_c = tp->data;
1172 struct tc_u_hnode *ht;
1173 struct tc_u_knode *n;
1179 for (ht = rtnl_dereference(tp_c->hlist);
1181 ht = rtnl_dereference(ht->next)) {
1182 if (ht->prio != tp->prio)
1184 if (arg->count >= arg->skip) {
1185 if (arg->fn(tp, ht, arg) < 0) {
1191 for (h = 0; h <= ht->divisor; h++) {
1192 for (n = rtnl_dereference(ht->ht[h]);
1194 n = rtnl_dereference(n->next)) {
1195 if (arg->count < arg->skip) {
1199 if (arg->fn(tp, n, arg) < 0) {
1209 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1211 struct tc_u_knode *n = fh;
1213 if (n && n->res.classid == classid)
1217 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1218 struct sk_buff *skb, struct tcmsg *t)
1220 struct tc_u_knode *n = fh;
1221 struct tc_u_hnode *ht_up, *ht_down;
1222 struct nlattr *nest;
1227 t->tcm_handle = n->handle;
1229 nest = nla_nest_start(skb, TCA_OPTIONS);
1231 goto nla_put_failure;
1233 if (TC_U32_KEY(n->handle) == 0) {
1234 struct tc_u_hnode *ht = fh;
1235 u32 divisor = ht->divisor + 1;
1237 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1238 goto nla_put_failure;
1240 #ifdef CONFIG_CLS_U32_PERF
1241 struct tc_u32_pcnt *gpf;
1245 if (nla_put(skb, TCA_U32_SEL,
1246 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1248 goto nla_put_failure;
1250 ht_up = rtnl_dereference(n->ht_up);
1252 u32 htid = n->handle & 0xFFFFF000;
1253 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1254 goto nla_put_failure;
1256 if (n->res.classid &&
1257 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1258 goto nla_put_failure;
1260 ht_down = rtnl_dereference(n->ht_down);
1262 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1263 goto nla_put_failure;
1265 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1266 goto nla_put_failure;
1268 #ifdef CONFIG_CLS_U32_MARK
1269 if ((n->val || n->mask)) {
1270 struct tc_u32_mark mark = {.val = n->val,
1275 for_each_possible_cpu(cpum) {
1276 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1278 mark.success += cnt;
1281 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1282 goto nla_put_failure;
1286 if (tcf_exts_dump(skb, &n->exts) < 0)
1287 goto nla_put_failure;
1289 #ifdef CONFIG_NET_CLS_IND
1291 struct net_device *dev;
1292 dev = __dev_get_by_index(net, n->ifindex);
1293 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1294 goto nla_put_failure;
1297 #ifdef CONFIG_CLS_U32_PERF
1298 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1299 n->sel.nkeys * sizeof(u64),
1302 goto nla_put_failure;
1304 for_each_possible_cpu(cpu) {
1306 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1308 gpf->rcnt += pf->rcnt;
1309 gpf->rhit += pf->rhit;
1310 for (i = 0; i < n->sel.nkeys; i++)
1311 gpf->kcnts[i] += pf->kcnts[i];
1314 if (nla_put_64bit(skb, TCA_U32_PCNT,
1315 sizeof(struct tc_u32_pcnt) +
1316 n->sel.nkeys * sizeof(u64),
1317 gpf, TCA_U32_PAD)) {
1319 goto nla_put_failure;
1325 nla_nest_end(skb, nest);
1327 if (TC_U32_KEY(n->handle))
1328 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1329 goto nla_put_failure;
1333 nla_nest_cancel(skb, nest);
1337 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1339 .classify = u32_classify,
1341 .destroy = u32_destroy,
1343 .change = u32_change,
1344 .delete = u32_delete,
1347 .bind_class = u32_bind_class,
1348 .owner = THIS_MODULE,
1351 static int __init init_u32(void)
1355 pr_info("u32 classifier\n");
1356 #ifdef CONFIG_CLS_U32_PERF
1357 pr_info(" Performance counters on\n");
1359 #ifdef CONFIG_NET_CLS_IND
1360 pr_info(" input device check on\n");
1362 #ifdef CONFIG_NET_CLS_ACT
1363 pr_info(" Actions configured\n");
1365 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1366 sizeof(struct hlist_head),
1368 if (!tc_u_common_hash)
1371 for (i = 0; i < U32_HASH_SIZE; i++)
1372 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1374 ret = register_tcf_proto_ops(&cls_u32_ops);
1376 kvfree(tc_u_common_hash);
1380 static void __exit exit_u32(void)
1382 unregister_tcf_proto_ops(&cls_u32_ops);
1383 kvfree(tc_u_common_hash);
1386 module_init(init_u32)
1387 module_exit(exit_u32)
1388 MODULE_LICENSE("GPL");