Linux 5.7.7
[linux/fpc-iii.git] / net / sched / cls_u32.c
blobe15ff335953deef36e55901d8f4ce34e15ed676f
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
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>
43 struct tc_u_knode {
44 struct tc_u_knode __rcu *next;
45 u32 handle;
46 struct tc_u_hnode __rcu *ht_up;
47 struct tcf_exts exts;
48 int ifindex;
49 u8 fshift;
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;
54 #endif
55 u32 flags;
56 unsigned int in_hw_count;
57 #ifdef CONFIG_CLS_U32_MARK
58 u32 val;
59 u32 mask;
60 u32 __percpu *pcpu_success;
61 #endif
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;
69 struct tc_u_hnode {
70 struct tc_u_hnode __rcu *next;
71 u32 handle;
72 u32 prio;
73 int refcnt;
74 unsigned int divisor;
75 struct idr handle_idr;
76 bool is_root;
77 struct rcu_head rcu;
78 u32 flags;
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];
85 struct tc_u_common {
86 struct tc_u_hnode __rcu *hlist;
87 void *ptr;
88 int refcnt;
89 struct idr handle_idr;
90 struct hlist_node hnode;
91 long knodes;
94 static inline unsigned int u32_hash_fold(__be32 key,
95 const struct tc_u32_sel *sel,
96 u8 fshift)
98 unsigned int h = ntohl(key & sel->hmask) >> fshift;
100 return h;
103 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
104 struct tcf_result *res)
106 struct {
107 struct tc_u_knode *knode;
108 unsigned int off;
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;
114 int sdepth = 0;
115 int off2 = 0;
116 int sel = 0;
117 #ifdef CONFIG_CLS_U32_PERF
118 int j;
119 #endif
120 int i, r;
122 next_ht:
123 n = rcu_dereference_bh(ht->ht[sel]);
125 next_knode:
126 if (n) {
127 struct tc_u32_key *key = n->sel.keys;
129 #ifdef CONFIG_CLS_U32_PERF
130 __this_cpu_inc(n->pf->rcnt);
131 j = 0;
132 #endif
134 if (tc_skip_sw(n->flags)) {
135 n = rcu_dereference_bh(n->next);
136 goto next_knode;
139 #ifdef CONFIG_CLS_U32_MARK
140 if ((skb->mark & n->mask) != n->val) {
141 n = rcu_dereference_bh(n->next);
142 goto next_knode;
143 } else {
144 __this_cpu_inc(*n->pcpu_success);
146 #endif
148 for (i = n->sel.nkeys; i > 0; i--, key++) {
149 int toff = off + key->off + (off2 & key->offmask);
150 __be32 *data, hdata;
152 if (skb_headroom(skb) + toff > INT_MAX)
153 goto out;
155 data = skb_header_pointer(skb, toff, 4, &hdata);
156 if (!data)
157 goto out;
158 if ((*data ^ key->val) & key->mask) {
159 n = rcu_dereference_bh(n->next);
160 goto next_knode;
162 #ifdef CONFIG_CLS_U32_PERF
163 __this_cpu_inc(n->pf->kcnts[j]);
164 j++;
165 #endif
168 ht = rcu_dereference_bh(n->ht_down);
169 if (!ht) {
170 check_terminal:
171 if (n->sel.flags & TC_U32_TERMINAL) {
173 *res = n->res;
174 if (!tcf_match_indev(skb, n->ifindex)) {
175 n = rcu_dereference_bh(n->next);
176 goto next_knode;
178 #ifdef CONFIG_CLS_U32_PERF
179 __this_cpu_inc(n->pf->rhit);
180 #endif
181 r = tcf_exts_exec(skb, &n->exts, res);
182 if (r < 0) {
183 n = rcu_dereference_bh(n->next);
184 goto next_knode;
187 return r;
189 n = rcu_dereference_bh(n->next);
190 goto next_knode;
193 /* PUSH */
194 if (sdepth >= TC_U32_MAXDEPTH)
195 goto deadloop;
196 stack[sdepth].knode = n;
197 stack[sdepth].off = off;
198 sdepth++;
200 ht = rcu_dereference_bh(n->ht_down);
201 sel = 0;
202 if (ht->divisor) {
203 __be32 *data, hdata;
205 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
206 &hdata);
207 if (!data)
208 goto out;
209 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
210 n->fshift);
212 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
213 goto next_ht;
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) {
218 __be16 *data, hdata;
220 data = skb_header_pointer(skb,
221 off + n->sel.offoff,
222 2, &hdata);
223 if (!data)
224 goto out;
225 off2 += ntohs(n->sel.offmask & *data) >>
226 n->sel.offshift;
228 off2 &= ~3;
230 if (n->sel.flags & TC_U32_EAT) {
231 off += off2;
232 off2 = 0;
235 if (off < skb->len)
236 goto next_ht;
239 /* POP */
240 if (sdepth--) {
241 n = stack[sdepth].knode;
242 ht = rcu_dereference_bh(n->ht_up);
243 off = stack[sdepth].off;
244 goto check_terminal;
246 out:
247 return -1;
249 deadloop:
250 net_warn_ratelimited("cls_u32: dead loop\n");
251 return -1;
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)
262 break;
264 return ht;
267 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
269 unsigned int sel;
270 struct tc_u_knode *n = NULL;
272 sel = TC_U32_HASH(handle);
273 if (sel > ht->divisor)
274 goto out;
276 for (n = rtnl_dereference(ht->ht[sel]);
278 n = rtnl_dereference(n->next))
279 if (n->handle == handle)
280 break;
281 out:
282 return n;
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);
293 else
294 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
296 if (!ht)
297 return NULL;
299 if (TC_U32_KEY(handle) == 0)
300 return ht;
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);
309 if (id < 0)
310 return 0;
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))
330 return block;
331 else
332 return block->q;
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) {
344 if (tc->ptr == key)
345 return tc;
347 return NULL;
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);
357 if (root_ht == NULL)
358 return -ENOBUFS;
360 root_ht->refcnt++;
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);
366 if (tp_c == NULL) {
367 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
368 if (tp_c == NULL) {
369 kfree(root_ht);
370 return -ENOBUFS;
372 tp_c->ptr = key;
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));
379 tp_c->refcnt++;
380 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
381 rcu_assign_pointer(tp_c->hlist, root_ht);
383 root_ht->refcnt++;
384 rcu_assign_pointer(tp->root, root_ht);
385 tp->data = tp_c;
386 return 0;
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)
396 kfree(ht);
397 #ifdef CONFIG_CLS_U32_PERF
398 if (free_pf)
399 free_percpu(n->pf);
400 #endif
401 #ifdef CONFIG_CLS_U32_MARK
402 if (free_pf)
403 free_percpu(n->pcpu_success);
404 #endif
405 kfree(n);
406 return 0;
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
415 * statistics.
417 static void u32_delete_key_work(struct work_struct *work)
419 struct tc_u_knode *key = container_of(to_rcu_work(work),
420 struct tc_u_knode,
421 rwork);
422 rtnl_lock();
423 u32_destroy_key(key, false);
424 rtnl_unlock();
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
432 * u32_init_knode()
434 static void u32_delete_key_freepf_work(struct work_struct *work)
436 struct tc_u_knode *key = container_of(to_rcu_work(work),
437 struct tc_u_knode,
438 rwork);
439 rtnl_lock();
440 u32_destroy_key(key, true);
441 rtnl_unlock();
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);
451 if (ht) {
452 kp = &ht->ht[TC_U32_HASH(key->handle)];
453 for (pkp = rtnl_dereference(*kp); pkp;
454 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
455 if (pkp == key) {
456 RCU_INIT_POINTER(*kp, key->next);
457 tp_c->knodes--;
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);
463 return 0;
467 WARN_ON(1);
468 return 0;
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, true);
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;
493 int err;
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, true);
502 if (err < 0) {
503 u32_clear_hw_hnode(tp, h, NULL);
504 return err;
505 } else if (err > 0) {
506 offloaded = true;
509 if (skip_sw && !offloaded)
510 return -EINVAL;
512 return 0;
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_destroy(block, tp, TC_SETUP_CLSU32, &cls_u32, false,
526 &n->flags, &n->in_hw_count, true);
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);
536 int err;
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;
545 #else
546 cls_u32.knode.val = 0;
547 cls_u32.knode.mask = 0;
548 #endif
549 cls_u32.knode.sel = &n->sel;
550 cls_u32.knode.res = &n->res;
551 cls_u32.knode.exts = &n->exts;
552 if (n->ht_down)
553 cls_u32.knode.link_handle = ht->handle;
555 err = tc_setup_cb_add(block, tp, TC_SETUP_CLSU32, &cls_u32, skip_sw,
556 &n->flags, &n->in_hw_count, true);
557 if (err) {
558 u32_remove_hw_knode(tp, n, NULL);
559 return err;
562 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
563 return -EINVAL;
565 return 0;
568 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
569 struct netlink_ext_ack *extack)
571 struct tc_u_common *tp_c = tp->data;
572 struct tc_u_knode *n;
573 unsigned int h;
575 for (h = 0; h <= ht->divisor; h++) {
576 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
577 RCU_INIT_POINTER(ht->ht[h],
578 rtnl_dereference(n->next));
579 tp_c->knodes--;
580 tcf_unbind_filter(tp, &n->res);
581 u32_remove_hw_knode(tp, n, extack);
582 idr_remove(&ht->handle_idr, n->handle);
583 if (tcf_exts_get_net(&n->exts))
584 tcf_queue_work(&n->rwork, u32_delete_key_freepf_work);
585 else
586 u32_destroy_key(n, true);
591 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
592 struct netlink_ext_ack *extack)
594 struct tc_u_common *tp_c = tp->data;
595 struct tc_u_hnode __rcu **hn;
596 struct tc_u_hnode *phn;
598 WARN_ON(--ht->refcnt);
600 u32_clear_hnode(tp, ht, extack);
602 hn = &tp_c->hlist;
603 for (phn = rtnl_dereference(*hn);
604 phn;
605 hn = &phn->next, phn = rtnl_dereference(*hn)) {
606 if (phn == ht) {
607 u32_clear_hw_hnode(tp, ht, extack);
608 idr_destroy(&ht->handle_idr);
609 idr_remove(&tp_c->handle_idr, ht->handle);
610 RCU_INIT_POINTER(*hn, ht->next);
611 kfree_rcu(ht, rcu);
612 return 0;
616 return -ENOENT;
619 static void u32_destroy(struct tcf_proto *tp, bool rtnl_held,
620 struct netlink_ext_ack *extack)
622 struct tc_u_common *tp_c = tp->data;
623 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
625 WARN_ON(root_ht == NULL);
627 if (root_ht && --root_ht->refcnt == 1)
628 u32_destroy_hnode(tp, root_ht, extack);
630 if (--tp_c->refcnt == 0) {
631 struct tc_u_hnode *ht;
633 hlist_del(&tp_c->hnode);
635 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
636 u32_clear_hnode(tp, ht, extack);
637 RCU_INIT_POINTER(tp_c->hlist, ht->next);
639 /* u32_destroy_key() will later free ht for us, if it's
640 * still referenced by some knode
642 if (--ht->refcnt == 0)
643 kfree_rcu(ht, rcu);
646 idr_destroy(&tp_c->handle_idr);
647 kfree(tp_c);
650 tp->data = NULL;
653 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last,
654 bool rtnl_held, struct netlink_ext_ack *extack)
656 struct tc_u_hnode *ht = arg;
657 struct tc_u_common *tp_c = tp->data;
658 int ret = 0;
660 if (TC_U32_KEY(ht->handle)) {
661 u32_remove_hw_knode(tp, (struct tc_u_knode *)ht, extack);
662 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
663 goto out;
666 if (ht->is_root) {
667 NL_SET_ERR_MSG_MOD(extack, "Not allowed to delete root node");
668 return -EINVAL;
671 if (ht->refcnt == 1) {
672 u32_destroy_hnode(tp, ht, extack);
673 } else {
674 NL_SET_ERR_MSG_MOD(extack, "Can not delete in-use filter");
675 return -EBUSY;
678 out:
679 *last = tp_c->refcnt == 1 && tp_c->knodes == 0;
680 return ret;
683 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
685 u32 index = htid | 0x800;
686 u32 max = htid | 0xFFF;
688 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max, GFP_KERNEL)) {
689 index = htid + 1;
690 if (idr_alloc_u32(&ht->handle_idr, NULL, &index, max,
691 GFP_KERNEL))
692 index = max;
695 return index;
698 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
699 [TCA_U32_CLASSID] = { .type = NLA_U32 },
700 [TCA_U32_HASH] = { .type = NLA_U32 },
701 [TCA_U32_LINK] = { .type = NLA_U32 },
702 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
703 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
704 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
705 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
706 [TCA_U32_FLAGS] = { .type = NLA_U32 },
709 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
710 unsigned long base,
711 struct tc_u_knode *n, struct nlattr **tb,
712 struct nlattr *est, bool ovr,
713 struct netlink_ext_ack *extack)
715 int err;
717 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr, true, extack);
718 if (err < 0)
719 return err;
721 if (tb[TCA_U32_LINK]) {
722 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
723 struct tc_u_hnode *ht_down = NULL, *ht_old;
725 if (TC_U32_KEY(handle)) {
726 NL_SET_ERR_MSG_MOD(extack, "u32 Link handle must be a hash table");
727 return -EINVAL;
730 if (handle) {
731 ht_down = u32_lookup_ht(tp->data, handle);
733 if (!ht_down) {
734 NL_SET_ERR_MSG_MOD(extack, "Link hash table not found");
735 return -EINVAL;
737 if (ht_down->is_root) {
738 NL_SET_ERR_MSG_MOD(extack, "Not linking to root node");
739 return -EINVAL;
741 ht_down->refcnt++;
744 ht_old = rtnl_dereference(n->ht_down);
745 rcu_assign_pointer(n->ht_down, ht_down);
747 if (ht_old)
748 ht_old->refcnt--;
750 if (tb[TCA_U32_CLASSID]) {
751 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
752 tcf_bind_filter(tp, &n->res, base);
755 if (tb[TCA_U32_INDEV]) {
756 int ret;
757 ret = tcf_change_indev(net, tb[TCA_U32_INDEV], extack);
758 if (ret < 0)
759 return -EINVAL;
760 n->ifindex = ret;
762 return 0;
765 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
766 struct tc_u_knode *n)
768 struct tc_u_knode __rcu **ins;
769 struct tc_u_knode *pins;
770 struct tc_u_hnode *ht;
772 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
773 ht = rtnl_dereference(tp->root);
774 else
775 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
777 ins = &ht->ht[TC_U32_HASH(n->handle)];
779 /* The node must always exist for it to be replaced if this is not the
780 * case then something went very wrong elsewhere.
782 for (pins = rtnl_dereference(*ins); ;
783 ins = &pins->next, pins = rtnl_dereference(*ins))
784 if (pins->handle == n->handle)
785 break;
787 idr_replace(&ht->handle_idr, n, n->handle);
788 RCU_INIT_POINTER(n->next, pins->next);
789 rcu_assign_pointer(*ins, n);
792 static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,
793 struct tc_u_knode *n)
795 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
796 struct tc_u32_sel *s = &n->sel;
797 struct tc_u_knode *new;
799 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
800 GFP_KERNEL);
802 if (!new)
803 return NULL;
805 RCU_INIT_POINTER(new->next, n->next);
806 new->handle = n->handle;
807 RCU_INIT_POINTER(new->ht_up, n->ht_up);
809 new->ifindex = n->ifindex;
810 new->fshift = n->fshift;
811 new->res = n->res;
812 new->flags = n->flags;
813 RCU_INIT_POINTER(new->ht_down, ht);
815 /* bump reference count as long as we hold pointer to structure */
816 if (ht)
817 ht->refcnt++;
819 #ifdef CONFIG_CLS_U32_PERF
820 /* Statistics may be incremented by readers during update
821 * so we must keep them in tact. When the node is later destroyed
822 * a special destroy call must be made to not free the pf memory.
824 new->pf = n->pf;
825 #endif
827 #ifdef CONFIG_CLS_U32_MARK
828 new->val = n->val;
829 new->mask = n->mask;
830 /* Similarly success statistics must be moved as pointers */
831 new->pcpu_success = n->pcpu_success;
832 #endif
833 memcpy(&new->sel, s, struct_size(s, keys, s->nkeys));
835 if (tcf_exts_init(&new->exts, net, TCA_U32_ACT, TCA_U32_POLICE)) {
836 kfree(new);
837 return NULL;
840 return new;
843 static int u32_change(struct net *net, struct sk_buff *in_skb,
844 struct tcf_proto *tp, unsigned long base, u32 handle,
845 struct nlattr **tca, void **arg, bool ovr, bool rtnl_held,
846 struct netlink_ext_ack *extack)
848 struct tc_u_common *tp_c = tp->data;
849 struct tc_u_hnode *ht;
850 struct tc_u_knode *n;
851 struct tc_u32_sel *s;
852 struct nlattr *opt = tca[TCA_OPTIONS];
853 struct nlattr *tb[TCA_U32_MAX + 1];
854 u32 htid, flags = 0;
855 size_t sel_size;
856 int err;
857 #ifdef CONFIG_CLS_U32_PERF
858 size_t size;
859 #endif
861 if (!opt) {
862 if (handle) {
863 NL_SET_ERR_MSG_MOD(extack, "Filter handle requires options");
864 return -EINVAL;
865 } else {
866 return 0;
870 err = nla_parse_nested_deprecated(tb, TCA_U32_MAX, opt, u32_policy,
871 extack);
872 if (err < 0)
873 return err;
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");
879 return -EINVAL;
883 n = *arg;
884 if (n) {
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");
889 return -EINVAL;
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");
895 return -EINVAL;
898 new = u32_init_knode(net, tp, n);
899 if (!new)
900 return -ENOMEM;
902 err = u32_set_parms(net, tp, base, new, tb,
903 tca[TCA_RATE], ovr, extack);
905 if (err) {
906 u32_destroy_key(new, false);
907 return err;
910 err = u32_replace_hw_knode(tp, new, flags, extack);
911 if (err) {
912 u32_destroy_key(new, false);
913 return err;
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);
923 return 0;
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");
931 return -EINVAL;
933 if (divisor-- > 0x100) {
934 NL_SET_ERR_MSG_MOD(extack, "Exceeded maximum 256 hash buckets");
935 return -EINVAL;
937 if (TC_U32_KEY(handle)) {
938 NL_SET_ERR_MSG_MOD(extack, "Divisor can only be used on a hash table");
939 return -EINVAL;
941 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
942 if (ht == NULL)
943 return -ENOBUFS;
944 if (handle == 0) {
945 handle = gen_new_htid(tp->data, ht);
946 if (handle == 0) {
947 kfree(ht);
948 return -ENOMEM;
950 } else {
951 err = idr_alloc_u32(&tp_c->handle_idr, ht, &handle,
952 handle, GFP_KERNEL);
953 if (err) {
954 kfree(ht);
955 return err;
958 ht->refcnt = 1;
959 ht->divisor = divisor;
960 ht->handle = handle;
961 ht->prio = tp->prio;
962 idr_init(&ht->handle_idr);
963 ht->flags = flags;
965 err = u32_replace_hw_hnode(tp, ht, flags, extack);
966 if (err) {
967 idr_remove(&tp_c->handle_idr, handle);
968 kfree(ht);
969 return err;
972 RCU_INIT_POINTER(ht->next, tp_c->hlist);
973 rcu_assign_pointer(tp_c->hlist, ht);
974 *arg = ht;
976 return 0;
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);
983 htid = ht->handle;
984 } else {
985 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
986 if (!ht) {
987 NL_SET_ERR_MSG_MOD(extack, "Specified hash table not found");
988 return -EINVAL;
991 } else {
992 ht = rtnl_dereference(tp->root);
993 htid = ht->handle;
996 if (ht->divisor < TC_U32_HASH(htid)) {
997 NL_SET_ERR_MSG_MOD(extack, "Specified hash table buckets exceed configured value");
998 return -EINVAL;
1001 if (handle) {
1002 if (TC_U32_HTID(handle) && TC_U32_HTID(handle ^ htid)) {
1003 NL_SET_ERR_MSG_MOD(extack, "Handle specified hash table address mismatch");
1004 return -EINVAL;
1006 handle = htid | TC_U32_NODE(handle);
1007 err = idr_alloc_u32(&ht->handle_idr, NULL, &handle, handle,
1008 GFP_KERNEL);
1009 if (err)
1010 return err;
1011 } else
1012 handle = gen_new_kid(ht, htid);
1014 if (tb[TCA_U32_SEL] == NULL) {
1015 NL_SET_ERR_MSG_MOD(extack, "Selector not specified");
1016 err = -EINVAL;
1017 goto erridr;
1020 s = nla_data(tb[TCA_U32_SEL]);
1021 sel_size = struct_size(s, keys, s->nkeys);
1022 if (nla_len(tb[TCA_U32_SEL]) < sel_size) {
1023 err = -EINVAL;
1024 goto erridr;
1027 n = kzalloc(offsetof(typeof(*n), sel) + sel_size, GFP_KERNEL);
1028 if (n == NULL) {
1029 err = -ENOBUFS;
1030 goto erridr;
1033 #ifdef CONFIG_CLS_U32_PERF
1034 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1035 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1036 if (!n->pf) {
1037 err = -ENOBUFS;
1038 goto errfree;
1040 #endif
1042 memcpy(&n->sel, s, sel_size);
1043 RCU_INIT_POINTER(n->ht_up, ht);
1044 n->handle = handle;
1045 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1046 n->flags = flags;
1048 err = tcf_exts_init(&n->exts, net, TCA_U32_ACT, TCA_U32_POLICE);
1049 if (err < 0)
1050 goto errout;
1052 #ifdef CONFIG_CLS_U32_MARK
1053 n->pcpu_success = alloc_percpu(u32);
1054 if (!n->pcpu_success) {
1055 err = -ENOMEM;
1056 goto errout;
1059 if (tb[TCA_U32_MARK]) {
1060 struct tc_u32_mark *mark;
1062 mark = nla_data(tb[TCA_U32_MARK]);
1063 n->val = mark->val;
1064 n->mask = mark->mask;
1066 #endif
1068 err = u32_set_parms(net, tp, base, n, tb, tca[TCA_RATE], ovr,
1069 extack);
1070 if (err == 0) {
1071 struct tc_u_knode __rcu **ins;
1072 struct tc_u_knode *pins;
1074 err = u32_replace_hw_knode(tp, n, flags, extack);
1075 if (err)
1076 goto errhw;
1078 if (!tc_in_hw(n->flags))
1079 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1081 ins = &ht->ht[TC_U32_HASH(handle)];
1082 for (pins = rtnl_dereference(*ins); pins;
1083 ins = &pins->next, pins = rtnl_dereference(*ins))
1084 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1085 break;
1087 RCU_INIT_POINTER(n->next, pins);
1088 rcu_assign_pointer(*ins, n);
1089 tp_c->knodes++;
1090 *arg = n;
1091 return 0;
1094 errhw:
1095 #ifdef CONFIG_CLS_U32_MARK
1096 free_percpu(n->pcpu_success);
1097 #endif
1099 errout:
1100 tcf_exts_destroy(&n->exts);
1101 #ifdef CONFIG_CLS_U32_PERF
1102 errfree:
1103 free_percpu(n->pf);
1104 #endif
1105 kfree(n);
1106 erridr:
1107 idr_remove(&ht->handle_idr, handle);
1108 return err;
1111 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg,
1112 bool rtnl_held)
1114 struct tc_u_common *tp_c = tp->data;
1115 struct tc_u_hnode *ht;
1116 struct tc_u_knode *n;
1117 unsigned int h;
1119 if (arg->stop)
1120 return;
1122 for (ht = rtnl_dereference(tp_c->hlist);
1124 ht = rtnl_dereference(ht->next)) {
1125 if (ht->prio != tp->prio)
1126 continue;
1127 if (arg->count >= arg->skip) {
1128 if (arg->fn(tp, ht, arg) < 0) {
1129 arg->stop = 1;
1130 return;
1133 arg->count++;
1134 for (h = 0; h <= ht->divisor; h++) {
1135 for (n = rtnl_dereference(ht->ht[h]);
1137 n = rtnl_dereference(n->next)) {
1138 if (arg->count < arg->skip) {
1139 arg->count++;
1140 continue;
1142 if (arg->fn(tp, n, arg) < 0) {
1143 arg->stop = 1;
1144 return;
1146 arg->count++;
1152 static int u32_reoffload_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht,
1153 bool add, flow_setup_cb_t *cb, void *cb_priv,
1154 struct netlink_ext_ack *extack)
1156 struct tc_cls_u32_offload cls_u32 = {};
1157 int err;
1159 tc_cls_common_offload_init(&cls_u32.common, tp, ht->flags, extack);
1160 cls_u32.command = add ? TC_CLSU32_NEW_HNODE : TC_CLSU32_DELETE_HNODE;
1161 cls_u32.hnode.divisor = ht->divisor;
1162 cls_u32.hnode.handle = ht->handle;
1163 cls_u32.hnode.prio = ht->prio;
1165 err = cb(TC_SETUP_CLSU32, &cls_u32, cb_priv);
1166 if (err && add && tc_skip_sw(ht->flags))
1167 return err;
1169 return 0;
1172 static int u32_reoffload_knode(struct tcf_proto *tp, struct tc_u_knode *n,
1173 bool add, flow_setup_cb_t *cb, void *cb_priv,
1174 struct netlink_ext_ack *extack)
1176 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
1177 struct tcf_block *block = tp->chain->block;
1178 struct tc_cls_u32_offload cls_u32 = {};
1179 int err;
1181 tc_cls_common_offload_init(&cls_u32.common, tp, n->flags, extack);
1182 cls_u32.command = add ?
1183 TC_CLSU32_REPLACE_KNODE : TC_CLSU32_DELETE_KNODE;
1184 cls_u32.knode.handle = n->handle;
1186 if (add) {
1187 cls_u32.knode.fshift = n->fshift;
1188 #ifdef CONFIG_CLS_U32_MARK
1189 cls_u32.knode.val = n->val;
1190 cls_u32.knode.mask = n->mask;
1191 #else
1192 cls_u32.knode.val = 0;
1193 cls_u32.knode.mask = 0;
1194 #endif
1195 cls_u32.knode.sel = &n->sel;
1196 cls_u32.knode.res = &n->res;
1197 cls_u32.knode.exts = &n->exts;
1198 if (n->ht_down)
1199 cls_u32.knode.link_handle = ht->handle;
1202 err = tc_setup_cb_reoffload(block, tp, add, cb, TC_SETUP_CLSU32,
1203 &cls_u32, cb_priv, &n->flags,
1204 &n->in_hw_count);
1205 if (err)
1206 return err;
1208 return 0;
1211 static int u32_reoffload(struct tcf_proto *tp, bool add, flow_setup_cb_t *cb,
1212 void *cb_priv, struct netlink_ext_ack *extack)
1214 struct tc_u_common *tp_c = tp->data;
1215 struct tc_u_hnode *ht;
1216 struct tc_u_knode *n;
1217 unsigned int h;
1218 int err;
1220 for (ht = rtnl_dereference(tp_c->hlist);
1222 ht = rtnl_dereference(ht->next)) {
1223 if (ht->prio != tp->prio)
1224 continue;
1226 /* When adding filters to a new dev, try to offload the
1227 * hashtable first. When removing, do the filters before the
1228 * hashtable.
1230 if (add && !tc_skip_hw(ht->flags)) {
1231 err = u32_reoffload_hnode(tp, ht, add, cb, cb_priv,
1232 extack);
1233 if (err)
1234 return err;
1237 for (h = 0; h <= ht->divisor; h++) {
1238 for (n = rtnl_dereference(ht->ht[h]);
1240 n = rtnl_dereference(n->next)) {
1241 if (tc_skip_hw(n->flags))
1242 continue;
1244 err = u32_reoffload_knode(tp, n, add, cb,
1245 cb_priv, extack);
1246 if (err)
1247 return err;
1251 if (!add && !tc_skip_hw(ht->flags))
1252 u32_reoffload_hnode(tp, ht, add, cb, cb_priv, extack);
1255 return 0;
1258 static void u32_bind_class(void *fh, u32 classid, unsigned long cl, void *q,
1259 unsigned long base)
1261 struct tc_u_knode *n = fh;
1263 if (n && n->res.classid == classid) {
1264 if (cl)
1265 __tcf_bind_filter(q, &n->res, base);
1266 else
1267 __tcf_unbind_filter(q, &n->res);
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;
1278 if (n == NULL)
1279 return skb->len;
1281 t->tcm_handle = n->handle;
1283 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1284 if (nest == NULL)
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;
1293 } else {
1294 #ifdef CONFIG_CLS_U32_PERF
1295 struct tc_u32_pcnt *gpf;
1296 int cpu;
1297 #endif
1299 if (nla_put(skb, TCA_U32_SEL,
1300 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1301 &n->sel))
1302 goto nla_put_failure;
1304 ht_up = rtnl_dereference(n->ht_up);
1305 if (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);
1315 if (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,
1325 .mask = n->mask,
1326 .success = 0};
1327 int cpum;
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;
1338 #endif
1340 if (tcf_exts_dump(skb, &n->exts) < 0)
1341 goto nla_put_failure;
1343 if (n->ifindex) {
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),
1352 GFP_KERNEL);
1353 if (!gpf)
1354 goto nla_put_failure;
1356 for_each_possible_cpu(cpu) {
1357 int i;
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)) {
1370 kfree(gpf);
1371 goto nla_put_failure;
1373 kfree(gpf);
1374 #endif
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;
1382 return skb->len;
1384 nla_put_failure:
1385 nla_nest_cancel(skb, nest);
1386 return -1;
1389 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1390 .kind = "u32",
1391 .classify = u32_classify,
1392 .init = u32_init,
1393 .destroy = u32_destroy,
1394 .get = u32_get,
1395 .change = u32_change,
1396 .delete = u32_delete,
1397 .walk = u32_walk,
1398 .reoffload = u32_reoffload,
1399 .dump = u32_dump,
1400 .bind_class = u32_bind_class,
1401 .owner = THIS_MODULE,
1404 static int __init init_u32(void)
1406 int i, ret;
1408 pr_info("u32 classifier\n");
1409 #ifdef CONFIG_CLS_U32_PERF
1410 pr_info(" Performance counters on\n");
1411 #endif
1412 pr_info(" input device check on\n");
1413 #ifdef CONFIG_NET_CLS_ACT
1414 pr_info(" Actions configured\n");
1415 #endif
1416 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1417 sizeof(struct hlist_head),
1418 GFP_KERNEL);
1419 if (!tc_u_common_hash)
1420 return -ENOMEM;
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);
1426 if (ret)
1427 kvfree(tc_u_common_hash);
1428 return ret;
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");