]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/sched/cls_u32.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


8d25f38cc1ad4f39efba88c1f848ca0af523844f
[mirror_ubuntu-bionic-kernel.git] / net / sched / cls_u32.c
1 /*
2 * net/sched/cls_u32.c Ugly (or Universal) 32bit key Packet Classifier.
3 *
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.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 *
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.
14 *
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.
19 *
20 * It seems that it represents the best middle point between
21 * speed and manageability both by human and by machine.
22 *
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.
26 *
27 * JHS: We should remove the CONFIG_NET_CLS_IND from here
28 * eventually when the meta match extension is made available
29 *
30 * nfmark match added by Catalin(ux aka Dino) BOIE <catab at umbrella.ro>
31 */
32
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>
49
50 struct tc_u_knode {
51 struct tc_u_knode __rcu *next;
52 u32 handle;
53 struct tc_u_hnode __rcu *ht_up;
54 struct tcf_exts exts;
55 #ifdef CONFIG_NET_CLS_IND
56 int ifindex;
57 #endif
58 u8 fshift;
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;
63 #endif
64 u32 flags;
65 #ifdef CONFIG_CLS_U32_MARK
66 u32 val;
67 u32 mask;
68 u32 __percpu *pcpu_success;
69 #endif
70 struct tcf_proto *tp;
71 union {
72 struct work_struct work;
73 struct rcu_head rcu;
74 };
75 /* The 'sel' field MUST be the last field in structure to allow for
76 * tc_u32_keys allocated at end of structure.
77 */
78 struct tc_u32_sel sel;
79 };
80
81 struct tc_u_hnode {
82 struct tc_u_hnode __rcu *next;
83 u32 handle;
84 u32 prio;
85 struct tc_u_common *tp_c;
86 int refcnt;
87 unsigned int divisor;
88 struct idr handle_idr;
89 struct rcu_head rcu;
90 /* The 'ht' field MUST be the last field in structure to allow for
91 * more entries allocated at end of structure.
92 */
93 struct tc_u_knode __rcu *ht[1];
94 };
95
96 struct tc_u_common {
97 struct tc_u_hnode __rcu *hlist;
98 struct tcf_block *block;
99 int refcnt;
100 struct idr handle_idr;
101 struct hlist_node hnode;
102 struct rcu_head rcu;
103 };
104
105 static inline unsigned int u32_hash_fold(__be32 key,
106 const struct tc_u32_sel *sel,
107 u8 fshift)
108 {
109 unsigned int h = ntohl(key & sel->hmask) >> fshift;
110
111 return h;
112 }
113
114 static int u32_classify(struct sk_buff *skb, const struct tcf_proto *tp,
115 struct tcf_result *res)
116 {
117 struct {
118 struct tc_u_knode *knode;
119 unsigned int off;
120 } stack[TC_U32_MAXDEPTH];
121
122 struct tc_u_hnode *ht = rcu_dereference_bh(tp->root);
123 unsigned int off = skb_network_offset(skb);
124 struct tc_u_knode *n;
125 int sdepth = 0;
126 int off2 = 0;
127 int sel = 0;
128 #ifdef CONFIG_CLS_U32_PERF
129 int j;
130 #endif
131 int i, r;
132
133 next_ht:
134 n = rcu_dereference_bh(ht->ht[sel]);
135
136 next_knode:
137 if (n) {
138 struct tc_u32_key *key = n->sel.keys;
139
140 #ifdef CONFIG_CLS_U32_PERF
141 __this_cpu_inc(n->pf->rcnt);
142 j = 0;
143 #endif
144
145 if (tc_skip_sw(n->flags)) {
146 n = rcu_dereference_bh(n->next);
147 goto next_knode;
148 }
149
150 #ifdef CONFIG_CLS_U32_MARK
151 if ((skb->mark & n->mask) != n->val) {
152 n = rcu_dereference_bh(n->next);
153 goto next_knode;
154 } else {
155 __this_cpu_inc(*n->pcpu_success);
156 }
157 #endif
158
159 for (i = n->sel.nkeys; i > 0; i--, key++) {
160 int toff = off + key->off + (off2 & key->offmask);
161 __be32 *data, hdata;
162
163 if (skb_headroom(skb) + toff > INT_MAX)
164 goto out;
165
166 data = skb_header_pointer(skb, toff, 4, &hdata);
167 if (!data)
168 goto out;
169 if ((*data ^ key->val) & key->mask) {
170 n = rcu_dereference_bh(n->next);
171 goto next_knode;
172 }
173 #ifdef CONFIG_CLS_U32_PERF
174 __this_cpu_inc(n->pf->kcnts[j]);
175 j++;
176 #endif
177 }
178
179 ht = rcu_dereference_bh(n->ht_down);
180 if (!ht) {
181 check_terminal:
182 if (n->sel.flags & TC_U32_TERMINAL) {
183
184 *res = n->res;
185 #ifdef CONFIG_NET_CLS_IND
186 if (!tcf_match_indev(skb, n->ifindex)) {
187 n = rcu_dereference_bh(n->next);
188 goto next_knode;
189 }
190 #endif
191 #ifdef CONFIG_CLS_U32_PERF
192 __this_cpu_inc(n->pf->rhit);
193 #endif
194 r = tcf_exts_exec(skb, &n->exts, res);
195 if (r < 0) {
196 n = rcu_dereference_bh(n->next);
197 goto next_knode;
198 }
199
200 return r;
201 }
202 n = rcu_dereference_bh(n->next);
203 goto next_knode;
204 }
205
206 /* PUSH */
207 if (sdepth >= TC_U32_MAXDEPTH)
208 goto deadloop;
209 stack[sdepth].knode = n;
210 stack[sdepth].off = off;
211 sdepth++;
212
213 ht = rcu_dereference_bh(n->ht_down);
214 sel = 0;
215 if (ht->divisor) {
216 __be32 *data, hdata;
217
218 data = skb_header_pointer(skb, off + n->sel.hoff, 4,
219 &hdata);
220 if (!data)
221 goto out;
222 sel = ht->divisor & u32_hash_fold(*data, &n->sel,
223 n->fshift);
224 }
225 if (!(n->sel.flags & (TC_U32_VAROFFSET | TC_U32_OFFSET | TC_U32_EAT)))
226 goto next_ht;
227
228 if (n->sel.flags & (TC_U32_OFFSET | TC_U32_VAROFFSET)) {
229 off2 = n->sel.off + 3;
230 if (n->sel.flags & TC_U32_VAROFFSET) {
231 __be16 *data, hdata;
232
233 data = skb_header_pointer(skb,
234 off + n->sel.offoff,
235 2, &hdata);
236 if (!data)
237 goto out;
238 off2 += ntohs(n->sel.offmask & *data) >>
239 n->sel.offshift;
240 }
241 off2 &= ~3;
242 }
243 if (n->sel.flags & TC_U32_EAT) {
244 off += off2;
245 off2 = 0;
246 }
247
248 if (off < skb->len)
249 goto next_ht;
250 }
251
252 /* POP */
253 if (sdepth--) {
254 n = stack[sdepth].knode;
255 ht = rcu_dereference_bh(n->ht_up);
256 off = stack[sdepth].off;
257 goto check_terminal;
258 }
259 out:
260 return -1;
261
262 deadloop:
263 net_warn_ratelimited("cls_u32: dead loop\n");
264 return -1;
265 }
266
267 static struct tc_u_hnode *u32_lookup_ht(struct tc_u_common *tp_c, u32 handle)
268 {
269 struct tc_u_hnode *ht;
270
271 for (ht = rtnl_dereference(tp_c->hlist);
272 ht;
273 ht = rtnl_dereference(ht->next))
274 if (ht->handle == handle)
275 break;
276
277 return ht;
278 }
279
280 static struct tc_u_knode *u32_lookup_key(struct tc_u_hnode *ht, u32 handle)
281 {
282 unsigned int sel;
283 struct tc_u_knode *n = NULL;
284
285 sel = TC_U32_HASH(handle);
286 if (sel > ht->divisor)
287 goto out;
288
289 for (n = rtnl_dereference(ht->ht[sel]);
290 n;
291 n = rtnl_dereference(n->next))
292 if (n->handle == handle)
293 break;
294 out:
295 return n;
296 }
297
298
299 static void *u32_get(struct tcf_proto *tp, u32 handle)
300 {
301 struct tc_u_hnode *ht;
302 struct tc_u_common *tp_c = tp->data;
303
304 if (TC_U32_HTID(handle) == TC_U32_ROOT)
305 ht = rtnl_dereference(tp->root);
306 else
307 ht = u32_lookup_ht(tp_c, TC_U32_HTID(handle));
308
309 if (!ht)
310 return NULL;
311
312 if (TC_U32_KEY(handle) == 0)
313 return ht;
314
315 return u32_lookup_key(ht, handle);
316 }
317
318 static u32 gen_new_htid(struct tc_u_common *tp_c, struct tc_u_hnode *ptr)
319 {
320 unsigned long idr_index;
321 int err;
322
323 /* This is only used inside rtnl lock it is safe to increment
324 * without read _copy_ update semantics
325 */
326 err = idr_alloc_ext(&tp_c->handle_idr, ptr, &idr_index,
327 1, 0x7FF, GFP_KERNEL);
328 if (err)
329 return 0;
330 return (u32)(idr_index | 0x800) << 20;
331 }
332
333 static struct hlist_head *tc_u_common_hash;
334
335 #define U32_HASH_SHIFT 10
336 #define U32_HASH_SIZE (1 << U32_HASH_SHIFT)
337
338 static unsigned int tc_u_hash(const struct tcf_proto *tp)
339 {
340 return hash_ptr(tp->chain->block, U32_HASH_SHIFT);
341 }
342
343 static struct tc_u_common *tc_u_common_find(const struct tcf_proto *tp)
344 {
345 struct tc_u_common *tc;
346 unsigned int h;
347
348 h = tc_u_hash(tp);
349 hlist_for_each_entry(tc, &tc_u_common_hash[h], hnode) {
350 if (tc->block == tp->chain->block)
351 return tc;
352 }
353 return NULL;
354 }
355
356 static int u32_init(struct tcf_proto *tp)
357 {
358 struct tc_u_hnode *root_ht;
359 struct tc_u_common *tp_c;
360 unsigned int h;
361
362 tp_c = tc_u_common_find(tp);
363
364 root_ht = kzalloc(sizeof(*root_ht), GFP_KERNEL);
365 if (root_ht == NULL)
366 return -ENOBUFS;
367
368 root_ht->refcnt++;
369 root_ht->handle = tp_c ? gen_new_htid(tp_c, root_ht) : 0x80000000;
370 root_ht->prio = tp->prio;
371 idr_init(&root_ht->handle_idr);
372
373 if (tp_c == NULL) {
374 tp_c = kzalloc(sizeof(*tp_c), GFP_KERNEL);
375 if (tp_c == NULL) {
376 kfree(root_ht);
377 return -ENOBUFS;
378 }
379 tp_c->block = tp->chain->block;
380 INIT_HLIST_NODE(&tp_c->hnode);
381 idr_init(&tp_c->handle_idr);
382
383 h = tc_u_hash(tp);
384 hlist_add_head(&tp_c->hnode, &tc_u_common_hash[h]);
385 }
386
387 tp_c->refcnt++;
388 RCU_INIT_POINTER(root_ht->next, tp_c->hlist);
389 rcu_assign_pointer(tp_c->hlist, root_ht);
390 root_ht->tp_c = tp_c;
391
392 rcu_assign_pointer(tp->root, root_ht);
393 tp->data = tp_c;
394 return 0;
395 }
396
397 static int u32_destroy_key(struct tcf_proto *tp, struct tc_u_knode *n,
398 bool free_pf)
399 {
400 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
401
402 tcf_exts_destroy(&n->exts);
403 tcf_exts_put_net(&n->exts);
404 if (ht && --ht->refcnt == 0)
405 kfree(ht);
406 #ifdef CONFIG_CLS_U32_PERF
407 if (free_pf)
408 free_percpu(n->pf);
409 #endif
410 #ifdef CONFIG_CLS_U32_MARK
411 if (free_pf)
412 free_percpu(n->pcpu_success);
413 #endif
414 kfree(n);
415 return 0;
416 }
417
418 /* u32_delete_key_rcu should be called when free'ing a copied
419 * version of a tc_u_knode obtained from u32_init_knode(). When
420 * copies are obtained from u32_init_knode() the statistics are
421 * shared between the old and new copies to allow readers to
422 * continue to update the statistics during the copy. To support
423 * this the u32_delete_key_rcu variant does not free the percpu
424 * statistics.
425 */
426 static void u32_delete_key_work(struct work_struct *work)
427 {
428 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
429
430 rtnl_lock();
431 u32_destroy_key(key->tp, key, false);
432 rtnl_unlock();
433 }
434
435 static void u32_delete_key_rcu(struct rcu_head *rcu)
436 {
437 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
438
439 INIT_WORK(&key->work, u32_delete_key_work);
440 tcf_queue_work(&key->work);
441 }
442
443 /* u32_delete_key_freepf_rcu is the rcu callback variant
444 * that free's the entire structure including the statistics
445 * percpu variables. Only use this if the key is not a copy
446 * returned by u32_init_knode(). See u32_delete_key_rcu()
447 * for the variant that should be used with keys return from
448 * u32_init_knode()
449 */
450 static void u32_delete_key_freepf_work(struct work_struct *work)
451 {
452 struct tc_u_knode *key = container_of(work, struct tc_u_knode, work);
453
454 rtnl_lock();
455 u32_destroy_key(key->tp, key, true);
456 rtnl_unlock();
457 }
458
459 static void u32_delete_key_freepf_rcu(struct rcu_head *rcu)
460 {
461 struct tc_u_knode *key = container_of(rcu, struct tc_u_knode, rcu);
462
463 INIT_WORK(&key->work, u32_delete_key_freepf_work);
464 tcf_queue_work(&key->work);
465 }
466
467 static int u32_delete_key(struct tcf_proto *tp, struct tc_u_knode *key)
468 {
469 struct tc_u_knode __rcu **kp;
470 struct tc_u_knode *pkp;
471 struct tc_u_hnode *ht = rtnl_dereference(key->ht_up);
472
473 if (ht) {
474 kp = &ht->ht[TC_U32_HASH(key->handle)];
475 for (pkp = rtnl_dereference(*kp); pkp;
476 kp = &pkp->next, pkp = rtnl_dereference(*kp)) {
477 if (pkp == key) {
478 RCU_INIT_POINTER(*kp, key->next);
479
480 tcf_unbind_filter(tp, &key->res);
481 idr_remove(&ht->handle_idr, key->handle);
482 tcf_exts_get_net(&key->exts);
483 call_rcu(&key->rcu, u32_delete_key_freepf_rcu);
484 return 0;
485 }
486 }
487 }
488 WARN_ON(1);
489 return 0;
490 }
491
492 static void u32_clear_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h)
493 {
494 struct tcf_block *block = tp->chain->block;
495 struct tc_cls_u32_offload cls_u32 = {};
496
497 tc_cls_common_offload_init(&cls_u32.common, tp);
498 cls_u32.command = TC_CLSU32_DELETE_HNODE;
499 cls_u32.hnode.divisor = h->divisor;
500 cls_u32.hnode.handle = h->handle;
501 cls_u32.hnode.prio = h->prio;
502
503 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
504 }
505
506 static int u32_replace_hw_hnode(struct tcf_proto *tp, struct tc_u_hnode *h,
507 u32 flags)
508 {
509 struct tcf_block *block = tp->chain->block;
510 struct tc_cls_u32_offload cls_u32 = {};
511 bool skip_sw = tc_skip_sw(flags);
512 bool offloaded = false;
513 int err;
514
515 tc_cls_common_offload_init(&cls_u32.common, tp);
516 cls_u32.command = TC_CLSU32_NEW_HNODE;
517 cls_u32.hnode.divisor = h->divisor;
518 cls_u32.hnode.handle = h->handle;
519 cls_u32.hnode.prio = h->prio;
520
521 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
522 if (err < 0) {
523 u32_clear_hw_hnode(tp, h);
524 return err;
525 } else if (err > 0) {
526 offloaded = true;
527 }
528
529 if (skip_sw && !offloaded)
530 return -EINVAL;
531
532 return 0;
533 }
534
535 static void u32_remove_hw_knode(struct tcf_proto *tp, u32 handle)
536 {
537 struct tcf_block *block = tp->chain->block;
538 struct tc_cls_u32_offload cls_u32 = {};
539
540 tc_cls_common_offload_init(&cls_u32.common, tp);
541 cls_u32.command = TC_CLSU32_DELETE_KNODE;
542 cls_u32.knode.handle = handle;
543
544 tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, false);
545 }
546
547 static int u32_replace_hw_knode(struct tcf_proto *tp, struct tc_u_knode *n,
548 u32 flags)
549 {
550 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
551 struct tcf_block *block = tp->chain->block;
552 struct tc_cls_u32_offload cls_u32 = {};
553 bool skip_sw = tc_skip_sw(flags);
554 int err;
555
556 tc_cls_common_offload_init(&cls_u32.common, tp);
557 cls_u32.command = TC_CLSU32_REPLACE_KNODE;
558 cls_u32.knode.handle = n->handle;
559 cls_u32.knode.fshift = n->fshift;
560 #ifdef CONFIG_CLS_U32_MARK
561 cls_u32.knode.val = n->val;
562 cls_u32.knode.mask = n->mask;
563 #else
564 cls_u32.knode.val = 0;
565 cls_u32.knode.mask = 0;
566 #endif
567 cls_u32.knode.sel = &n->sel;
568 cls_u32.knode.exts = &n->exts;
569 if (n->ht_down)
570 cls_u32.knode.link_handle = ht->handle;
571
572 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSU32, &cls_u32, skip_sw);
573 if (err < 0) {
574 u32_remove_hw_knode(tp, n->handle);
575 return err;
576 } else if (err > 0) {
577 n->flags |= TCA_CLS_FLAGS_IN_HW;
578 }
579
580 if (skip_sw && !(n->flags & TCA_CLS_FLAGS_IN_HW))
581 return -EINVAL;
582
583 return 0;
584 }
585
586 static void u32_clear_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
587 {
588 struct tc_u_knode *n;
589 unsigned int h;
590
591 for (h = 0; h <= ht->divisor; h++) {
592 while ((n = rtnl_dereference(ht->ht[h])) != NULL) {
593 RCU_INIT_POINTER(ht->ht[h],
594 rtnl_dereference(n->next));
595 tcf_unbind_filter(tp, &n->res);
596 u32_remove_hw_knode(tp, n->handle);
597 idr_remove_ext(&ht->handle_idr, n->handle);
598 if (tcf_exts_get_net(&n->exts))
599 call_rcu(&n->rcu, u32_delete_key_freepf_rcu);
600 else
601 u32_destroy_key(n->tp, n, true);
602 }
603 }
604 }
605
606 static int u32_destroy_hnode(struct tcf_proto *tp, struct tc_u_hnode *ht)
607 {
608 struct tc_u_common *tp_c = tp->data;
609 struct tc_u_hnode __rcu **hn;
610 struct tc_u_hnode *phn;
611
612 WARN_ON(ht->refcnt);
613
614 u32_clear_hnode(tp, ht);
615
616 hn = &tp_c->hlist;
617 for (phn = rtnl_dereference(*hn);
618 phn;
619 hn = &phn->next, phn = rtnl_dereference(*hn)) {
620 if (phn == ht) {
621 u32_clear_hw_hnode(tp, ht);
622 idr_destroy(&ht->handle_idr);
623 idr_remove_ext(&tp_c->handle_idr, ht->handle);
624 RCU_INIT_POINTER(*hn, ht->next);
625 kfree_rcu(ht, rcu);
626 return 0;
627 }
628 }
629
630 return -ENOENT;
631 }
632
633 static bool ht_empty(struct tc_u_hnode *ht)
634 {
635 unsigned int h;
636
637 for (h = 0; h <= ht->divisor; h++)
638 if (rcu_access_pointer(ht->ht[h]))
639 return false;
640
641 return true;
642 }
643
644 static void u32_destroy(struct tcf_proto *tp)
645 {
646 struct tc_u_common *tp_c = tp->data;
647 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
648
649 WARN_ON(root_ht == NULL);
650
651 if (root_ht && --root_ht->refcnt == 0)
652 u32_destroy_hnode(tp, root_ht);
653
654 if (--tp_c->refcnt == 0) {
655 struct tc_u_hnode *ht;
656
657 hlist_del(&tp_c->hnode);
658
659 while ((ht = rtnl_dereference(tp_c->hlist)) != NULL) {
660 u32_clear_hnode(tp, ht);
661 RCU_INIT_POINTER(tp_c->hlist, ht->next);
662
663 /* u32_destroy_key() will later free ht for us, if it's
664 * still referenced by some knode
665 */
666 if (--ht->refcnt == 0)
667 kfree_rcu(ht, rcu);
668 }
669
670 idr_destroy(&tp_c->handle_idr);
671 kfree(tp_c);
672 }
673
674 tp->data = NULL;
675 }
676
677 static int u32_delete(struct tcf_proto *tp, void *arg, bool *last)
678 {
679 struct tc_u_hnode *ht = arg;
680 struct tc_u_hnode *root_ht = rtnl_dereference(tp->root);
681 struct tc_u_common *tp_c = tp->data;
682 int ret = 0;
683
684 if (ht == NULL)
685 goto out;
686
687 if (TC_U32_KEY(ht->handle)) {
688 u32_remove_hw_knode(tp, ht->handle);
689 ret = u32_delete_key(tp, (struct tc_u_knode *)ht);
690 goto out;
691 }
692
693 if (root_ht == ht)
694 return -EINVAL;
695
696 if (ht->refcnt == 1) {
697 ht->refcnt--;
698 u32_destroy_hnode(tp, ht);
699 } else {
700 return -EBUSY;
701 }
702
703 out:
704 *last = true;
705 if (root_ht) {
706 if (root_ht->refcnt > 1) {
707 *last = false;
708 goto ret;
709 }
710 if (root_ht->refcnt == 1) {
711 if (!ht_empty(root_ht)) {
712 *last = false;
713 goto ret;
714 }
715 }
716 }
717
718 if (tp_c->refcnt > 1) {
719 *last = false;
720 goto ret;
721 }
722
723 if (tp_c->refcnt == 1) {
724 struct tc_u_hnode *ht;
725
726 for (ht = rtnl_dereference(tp_c->hlist);
727 ht;
728 ht = rtnl_dereference(ht->next))
729 if (!ht_empty(ht)) {
730 *last = false;
731 break;
732 }
733 }
734
735 ret:
736 return ret;
737 }
738
739 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 htid)
740 {
741 unsigned long idr_index;
742 u32 start = htid | 0x800;
743 u32 max = htid | 0xFFF;
744 u32 min = htid;
745
746 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
747 start, max + 1, GFP_KERNEL)) {
748 if (idr_alloc_ext(&ht->handle_idr, NULL, &idr_index,
749 min + 1, max + 1, GFP_KERNEL))
750 return max;
751 }
752
753 return (u32)idr_index;
754 }
755
756 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {
757 [TCA_U32_CLASSID] = { .type = NLA_U32 },
758 [TCA_U32_HASH] = { .type = NLA_U32 },
759 [TCA_U32_LINK] = { .type = NLA_U32 },
760 [TCA_U32_DIVISOR] = { .type = NLA_U32 },
761 [TCA_U32_SEL] = { .len = sizeof(struct tc_u32_sel) },
762 [TCA_U32_INDEV] = { .type = NLA_STRING, .len = IFNAMSIZ },
763 [TCA_U32_MARK] = { .len = sizeof(struct tc_u32_mark) },
764 [TCA_U32_FLAGS] = { .type = NLA_U32 },
765 };
766
767 static int u32_set_parms(struct net *net, struct tcf_proto *tp,
768 unsigned long base, struct tc_u_hnode *ht,
769 struct tc_u_knode *n, struct nlattr **tb,
770 struct nlattr *est, bool ovr)
771 {
772 int err;
773
774 err = tcf_exts_validate(net, tp, tb, est, &n->exts, ovr);
775 if (err < 0)
776 return err;
777
778 if (tb[TCA_U32_LINK]) {
779 u32 handle = nla_get_u32(tb[TCA_U32_LINK]);
780 struct tc_u_hnode *ht_down = NULL, *ht_old;
781
782 if (TC_U32_KEY(handle))
783 return -EINVAL;
784
785 if (handle) {
786 ht_down = u32_lookup_ht(ht->tp_c, handle);
787
788 if (ht_down == NULL)
789 return -EINVAL;
790 ht_down->refcnt++;
791 }
792
793 ht_old = rtnl_dereference(n->ht_down);
794 rcu_assign_pointer(n->ht_down, ht_down);
795
796 if (ht_old)
797 ht_old->refcnt--;
798 }
799 if (tb[TCA_U32_CLASSID]) {
800 n->res.classid = nla_get_u32(tb[TCA_U32_CLASSID]);
801 tcf_bind_filter(tp, &n->res, base);
802 }
803
804 #ifdef CONFIG_NET_CLS_IND
805 if (tb[TCA_U32_INDEV]) {
806 int ret;
807 ret = tcf_change_indev(net, tb[TCA_U32_INDEV]);
808 if (ret < 0)
809 return -EINVAL;
810 n->ifindex = ret;
811 }
812 #endif
813 return 0;
814 }
815
816 static void u32_replace_knode(struct tcf_proto *tp, struct tc_u_common *tp_c,
817 struct tc_u_knode *n)
818 {
819 struct tc_u_knode __rcu **ins;
820 struct tc_u_knode *pins;
821 struct tc_u_hnode *ht;
822
823 if (TC_U32_HTID(n->handle) == TC_U32_ROOT)
824 ht = rtnl_dereference(tp->root);
825 else
826 ht = u32_lookup_ht(tp_c, TC_U32_HTID(n->handle));
827
828 ins = &ht->ht[TC_U32_HASH(n->handle)];
829
830 /* The node must always exist for it to be replaced if this is not the
831 * case then something went very wrong elsewhere.
832 */
833 for (pins = rtnl_dereference(*ins); ;
834 ins = &pins->next, pins = rtnl_dereference(*ins))
835 if (pins->handle == n->handle)
836 break;
837
838 idr_replace_ext(&ht->handle_idr, n, n->handle);
839 RCU_INIT_POINTER(n->next, pins->next);
840 rcu_assign_pointer(*ins, n);
841 }
842
843 static struct tc_u_knode *u32_init_knode(struct tcf_proto *tp,
844 struct tc_u_knode *n)
845 {
846 struct tc_u_hnode *ht = rtnl_dereference(n->ht_down);
847 struct tc_u32_sel *s = &n->sel;
848 struct tc_u_knode *new;
849
850 new = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key),
851 GFP_KERNEL);
852
853 if (!new)
854 return NULL;
855
856 RCU_INIT_POINTER(new->next, n->next);
857 new->handle = n->handle;
858 RCU_INIT_POINTER(new->ht_up, n->ht_up);
859
860 #ifdef CONFIG_NET_CLS_IND
861 new->ifindex = n->ifindex;
862 #endif
863 new->fshift = n->fshift;
864 new->res = n->res;
865 new->flags = n->flags;
866 RCU_INIT_POINTER(new->ht_down, ht);
867
868 /* bump reference count as long as we hold pointer to structure */
869 if (ht)
870 ht->refcnt++;
871
872 #ifdef CONFIG_CLS_U32_PERF
873 /* Statistics may be incremented by readers during update
874 * so we must keep them in tact. When the node is later destroyed
875 * a special destroy call must be made to not free the pf memory.
876 */
877 new->pf = n->pf;
878 #endif
879
880 #ifdef CONFIG_CLS_U32_MARK
881 new->val = n->val;
882 new->mask = n->mask;
883 /* Similarly success statistics must be moved as pointers */
884 new->pcpu_success = n->pcpu_success;
885 #endif
886 new->tp = tp;
887 memcpy(&new->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
888
889 if (tcf_exts_init(&new->exts, TCA_U32_ACT, TCA_U32_POLICE)) {
890 kfree(new);
891 return NULL;
892 }
893
894 return new;
895 }
896
897 static int u32_change(struct net *net, struct sk_buff *in_skb,
898 struct tcf_proto *tp, unsigned long base, u32 handle,
899 struct nlattr **tca, void **arg, bool ovr)
900 {
901 struct tc_u_common *tp_c = tp->data;
902 struct tc_u_hnode *ht;
903 struct tc_u_knode *n;
904 struct tc_u32_sel *s;
905 struct nlattr *opt = tca[TCA_OPTIONS];
906 struct nlattr *tb[TCA_U32_MAX + 1];
907 u32 htid, flags = 0;
908 int err;
909 #ifdef CONFIG_CLS_U32_PERF
910 size_t size;
911 #endif
912
913 if (opt == NULL)
914 return handle ? -EINVAL : 0;
915
916 err = nla_parse_nested(tb, TCA_U32_MAX, opt, u32_policy, NULL);
917 if (err < 0)
918 return err;
919
920 if (tb[TCA_U32_FLAGS]) {
921 flags = nla_get_u32(tb[TCA_U32_FLAGS]);
922 if (!tc_flags_valid(flags))
923 return -EINVAL;
924 }
925
926 n = *arg;
927 if (n) {
928 struct tc_u_knode *new;
929
930 if (TC_U32_KEY(n->handle) == 0)
931 return -EINVAL;
932
933 if ((n->flags ^ flags) &
934 ~(TCA_CLS_FLAGS_IN_HW | TCA_CLS_FLAGS_NOT_IN_HW))
935 return -EINVAL;
936
937 new = u32_init_knode(tp, n);
938 if (!new)
939 return -ENOMEM;
940
941 err = u32_set_parms(net, tp, base,
942 rtnl_dereference(n->ht_up), new, tb,
943 tca[TCA_RATE], ovr);
944
945 if (err) {
946 u32_destroy_key(tp, new, false);
947 return err;
948 }
949
950 err = u32_replace_hw_knode(tp, new, flags);
951 if (err) {
952 u32_destroy_key(tp, new, false);
953 return err;
954 }
955
956 if (!tc_in_hw(new->flags))
957 new->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
958
959 u32_replace_knode(tp, tp_c, new);
960 tcf_unbind_filter(tp, &n->res);
961 tcf_exts_get_net(&n->exts);
962 call_rcu(&n->rcu, u32_delete_key_rcu);
963 return 0;
964 }
965
966 if (tb[TCA_U32_DIVISOR]) {
967 unsigned int divisor = nla_get_u32(tb[TCA_U32_DIVISOR]);
968
969 if (--divisor > 0x100)
970 return -EINVAL;
971 if (TC_U32_KEY(handle))
972 return -EINVAL;
973 ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);
974 if (ht == NULL)
975 return -ENOBUFS;
976 if (handle == 0) {
977 handle = gen_new_htid(tp->data, ht);
978 if (handle == 0) {
979 kfree(ht);
980 return -ENOMEM;
981 }
982 } else {
983 err = idr_alloc_ext(&tp_c->handle_idr, ht, NULL,
984 handle, handle + 1, GFP_KERNEL);
985 if (err) {
986 kfree(ht);
987 return err;
988 }
989 }
990 ht->tp_c = tp_c;
991 ht->refcnt = 1;
992 ht->divisor = divisor;
993 ht->handle = handle;
994 ht->prio = tp->prio;
995 idr_init(&ht->handle_idr);
996
997 err = u32_replace_hw_hnode(tp, ht, flags);
998 if (err) {
999 idr_remove_ext(&tp_c->handle_idr, handle);
1000 kfree(ht);
1001 return err;
1002 }
1003
1004 RCU_INIT_POINTER(ht->next, tp_c->hlist);
1005 rcu_assign_pointer(tp_c->hlist, ht);
1006 *arg = ht;
1007
1008 return 0;
1009 }
1010
1011 if (tb[TCA_U32_HASH]) {
1012 htid = nla_get_u32(tb[TCA_U32_HASH]);
1013 if (TC_U32_HTID(htid) == TC_U32_ROOT) {
1014 ht = rtnl_dereference(tp->root);
1015 htid = ht->handle;
1016 } else {
1017 ht = u32_lookup_ht(tp->data, TC_U32_HTID(htid));
1018 if (ht == NULL)
1019 return -EINVAL;
1020 }
1021 } else {
1022 ht = rtnl_dereference(tp->root);
1023 htid = ht->handle;
1024 }
1025
1026 if (ht->divisor < TC_U32_HASH(htid))
1027 return -EINVAL;
1028
1029 if (handle) {
1030 if (TC_U32_HTID(handle) && TC_U32_HTID(handle^htid))
1031 return -EINVAL;
1032 handle = htid | TC_U32_NODE(handle);
1033 err = idr_alloc_ext(&ht->handle_idr, NULL, NULL,
1034 handle, handle + 1,
1035 GFP_KERNEL);
1036 if (err)
1037 return err;
1038 } else
1039 handle = gen_new_kid(ht, htid);
1040
1041 if (tb[TCA_U32_SEL] == NULL) {
1042 err = -EINVAL;
1043 goto erridr;
1044 }
1045
1046 s = nla_data(tb[TCA_U32_SEL]);
1047
1048 n = kzalloc(sizeof(*n) + s->nkeys*sizeof(struct tc_u32_key), GFP_KERNEL);
1049 if (n == NULL) {
1050 err = -ENOBUFS;
1051 goto erridr;
1052 }
1053
1054 #ifdef CONFIG_CLS_U32_PERF
1055 size = sizeof(struct tc_u32_pcnt) + s->nkeys * sizeof(u64);
1056 n->pf = __alloc_percpu(size, __alignof__(struct tc_u32_pcnt));
1057 if (!n->pf) {
1058 err = -ENOBUFS;
1059 goto errfree;
1060 }
1061 #endif
1062
1063 memcpy(&n->sel, s, sizeof(*s) + s->nkeys*sizeof(struct tc_u32_key));
1064 RCU_INIT_POINTER(n->ht_up, ht);
1065 n->handle = handle;
1066 n->fshift = s->hmask ? ffs(ntohl(s->hmask)) - 1 : 0;
1067 n->flags = flags;
1068 n->tp = tp;
1069
1070 err = tcf_exts_init(&n->exts, TCA_U32_ACT, TCA_U32_POLICE);
1071 if (err < 0)
1072 goto errout;
1073
1074 #ifdef CONFIG_CLS_U32_MARK
1075 n->pcpu_success = alloc_percpu(u32);
1076 if (!n->pcpu_success) {
1077 err = -ENOMEM;
1078 goto errout;
1079 }
1080
1081 if (tb[TCA_U32_MARK]) {
1082 struct tc_u32_mark *mark;
1083
1084 mark = nla_data(tb[TCA_U32_MARK]);
1085 n->val = mark->val;
1086 n->mask = mark->mask;
1087 }
1088 #endif
1089
1090 err = u32_set_parms(net, tp, base, ht, n, tb, tca[TCA_RATE], ovr);
1091 if (err == 0) {
1092 struct tc_u_knode __rcu **ins;
1093 struct tc_u_knode *pins;
1094
1095 err = u32_replace_hw_knode(tp, n, flags);
1096 if (err)
1097 goto errhw;
1098
1099 if (!tc_in_hw(n->flags))
1100 n->flags |= TCA_CLS_FLAGS_NOT_IN_HW;
1101
1102 ins = &ht->ht[TC_U32_HASH(handle)];
1103 for (pins = rtnl_dereference(*ins); pins;
1104 ins = &pins->next, pins = rtnl_dereference(*ins))
1105 if (TC_U32_NODE(handle) < TC_U32_NODE(pins->handle))
1106 break;
1107
1108 RCU_INIT_POINTER(n->next, pins);
1109 rcu_assign_pointer(*ins, n);
1110 *arg = n;
1111 return 0;
1112 }
1113
1114 errhw:
1115 #ifdef CONFIG_CLS_U32_MARK
1116 free_percpu(n->pcpu_success);
1117 #endif
1118
1119 errout:
1120 tcf_exts_destroy(&n->exts);
1121 #ifdef CONFIG_CLS_U32_PERF
1122 errfree:
1123 free_percpu(n->pf);
1124 #endif
1125 kfree(n);
1126 erridr:
1127 idr_remove_ext(&ht->handle_idr, handle);
1128 return err;
1129 }
1130
1131 static void u32_walk(struct tcf_proto *tp, struct tcf_walker *arg)
1132 {
1133 struct tc_u_common *tp_c = tp->data;
1134 struct tc_u_hnode *ht;
1135 struct tc_u_knode *n;
1136 unsigned int h;
1137
1138 if (arg->stop)
1139 return;
1140
1141 for (ht = rtnl_dereference(tp_c->hlist);
1142 ht;
1143 ht = rtnl_dereference(ht->next)) {
1144 if (ht->prio != tp->prio)
1145 continue;
1146 if (arg->count >= arg->skip) {
1147 if (arg->fn(tp, ht, arg) < 0) {
1148 arg->stop = 1;
1149 return;
1150 }
1151 }
1152 arg->count++;
1153 for (h = 0; h <= ht->divisor; h++) {
1154 for (n = rtnl_dereference(ht->ht[h]);
1155 n;
1156 n = rtnl_dereference(n->next)) {
1157 if (arg->count < arg->skip) {
1158 arg->count++;
1159 continue;
1160 }
1161 if (arg->fn(tp, n, arg) < 0) {
1162 arg->stop = 1;
1163 return;
1164 }
1165 arg->count++;
1166 }
1167 }
1168 }
1169 }
1170
1171 static void u32_bind_class(void *fh, u32 classid, unsigned long cl)
1172 {
1173 struct tc_u_knode *n = fh;
1174
1175 if (n && n->res.classid == classid)
1176 n->res.class = cl;
1177 }
1178
1179 static int u32_dump(struct net *net, struct tcf_proto *tp, void *fh,
1180 struct sk_buff *skb, struct tcmsg *t)
1181 {
1182 struct tc_u_knode *n = fh;
1183 struct tc_u_hnode *ht_up, *ht_down;
1184 struct nlattr *nest;
1185
1186 if (n == NULL)
1187 return skb->len;
1188
1189 t->tcm_handle = n->handle;
1190
1191 nest = nla_nest_start(skb, TCA_OPTIONS);
1192 if (nest == NULL)
1193 goto nla_put_failure;
1194
1195 if (TC_U32_KEY(n->handle) == 0) {
1196 struct tc_u_hnode *ht = fh;
1197 u32 divisor = ht->divisor + 1;
1198
1199 if (nla_put_u32(skb, TCA_U32_DIVISOR, divisor))
1200 goto nla_put_failure;
1201 } else {
1202 #ifdef CONFIG_CLS_U32_PERF
1203 struct tc_u32_pcnt *gpf;
1204 int cpu;
1205 #endif
1206
1207 if (nla_put(skb, TCA_U32_SEL,
1208 sizeof(n->sel) + n->sel.nkeys*sizeof(struct tc_u32_key),
1209 &n->sel))
1210 goto nla_put_failure;
1211
1212 ht_up = rtnl_dereference(n->ht_up);
1213 if (ht_up) {
1214 u32 htid = n->handle & 0xFFFFF000;
1215 if (nla_put_u32(skb, TCA_U32_HASH, htid))
1216 goto nla_put_failure;
1217 }
1218 if (n->res.classid &&
1219 nla_put_u32(skb, TCA_U32_CLASSID, n->res.classid))
1220 goto nla_put_failure;
1221
1222 ht_down = rtnl_dereference(n->ht_down);
1223 if (ht_down &&
1224 nla_put_u32(skb, TCA_U32_LINK, ht_down->handle))
1225 goto nla_put_failure;
1226
1227 if (n->flags && nla_put_u32(skb, TCA_U32_FLAGS, n->flags))
1228 goto nla_put_failure;
1229
1230 #ifdef CONFIG_CLS_U32_MARK
1231 if ((n->val || n->mask)) {
1232 struct tc_u32_mark mark = {.val = n->val,
1233 .mask = n->mask,
1234 .success = 0};
1235 int cpum;
1236
1237 for_each_possible_cpu(cpum) {
1238 __u32 cnt = *per_cpu_ptr(n->pcpu_success, cpum);
1239
1240 mark.success += cnt;
1241 }
1242
1243 if (nla_put(skb, TCA_U32_MARK, sizeof(mark), &mark))
1244 goto nla_put_failure;
1245 }
1246 #endif
1247
1248 if (tcf_exts_dump(skb, &n->exts) < 0)
1249 goto nla_put_failure;
1250
1251 #ifdef CONFIG_NET_CLS_IND
1252 if (n->ifindex) {
1253 struct net_device *dev;
1254 dev = __dev_get_by_index(net, n->ifindex);
1255 if (dev && nla_put_string(skb, TCA_U32_INDEV, dev->name))
1256 goto nla_put_failure;
1257 }
1258 #endif
1259 #ifdef CONFIG_CLS_U32_PERF
1260 gpf = kzalloc(sizeof(struct tc_u32_pcnt) +
1261 n->sel.nkeys * sizeof(u64),
1262 GFP_KERNEL);
1263 if (!gpf)
1264 goto nla_put_failure;
1265
1266 for_each_possible_cpu(cpu) {
1267 int i;
1268 struct tc_u32_pcnt *pf = per_cpu_ptr(n->pf, cpu);
1269
1270 gpf->rcnt += pf->rcnt;
1271 gpf->rhit += pf->rhit;
1272 for (i = 0; i < n->sel.nkeys; i++)
1273 gpf->kcnts[i] += pf->kcnts[i];
1274 }
1275
1276 if (nla_put_64bit(skb, TCA_U32_PCNT,
1277 sizeof(struct tc_u32_pcnt) +
1278 n->sel.nkeys * sizeof(u64),
1279 gpf, TCA_U32_PAD)) {
1280 kfree(gpf);
1281 goto nla_put_failure;
1282 }
1283 kfree(gpf);
1284 #endif
1285 }
1286
1287 nla_nest_end(skb, nest);
1288
1289 if (TC_U32_KEY(n->handle))
1290 if (tcf_exts_dump_stats(skb, &n->exts) < 0)
1291 goto nla_put_failure;
1292 return skb->len;
1293
1294 nla_put_failure:
1295 nla_nest_cancel(skb, nest);
1296 return -1;
1297 }
1298
1299 static struct tcf_proto_ops cls_u32_ops __read_mostly = {
1300 .kind = "u32",
1301 .classify = u32_classify,
1302 .init = u32_init,
1303 .destroy = u32_destroy,
1304 .get = u32_get,
1305 .change = u32_change,
1306 .delete = u32_delete,
1307 .walk = u32_walk,
1308 .dump = u32_dump,
1309 .bind_class = u32_bind_class,
1310 .owner = THIS_MODULE,
1311 };
1312
1313 static int __init init_u32(void)
1314 {
1315 int i, ret;
1316
1317 pr_info("u32 classifier\n");
1318 #ifdef CONFIG_CLS_U32_PERF
1319 pr_info(" Performance counters on\n");
1320 #endif
1321 #ifdef CONFIG_NET_CLS_IND
1322 pr_info(" input device check on\n");
1323 #endif
1324 #ifdef CONFIG_NET_CLS_ACT
1325 pr_info(" Actions configured\n");
1326 #endif
1327 tc_u_common_hash = kvmalloc_array(U32_HASH_SIZE,
1328 sizeof(struct hlist_head),
1329 GFP_KERNEL);
1330 if (!tc_u_common_hash)
1331 return -ENOMEM;
1332
1333 for (i = 0; i < U32_HASH_SIZE; i++)
1334 INIT_HLIST_HEAD(&tc_u_common_hash[i]);
1335
1336 ret = register_tcf_proto_ops(&cls_u32_ops);
1337 if (ret)
1338 kvfree(tc_u_common_hash);
1339 return ret;
1340 }
1341
1342 static void __exit exit_u32(void)
1343 {
1344 unregister_tcf_proto_ops(&cls_u32_ops);
1345 kvfree(tc_u_common_hash);
1346 }
1347
1348 module_init(init_u32)
1349 module_exit(exit_u32)
1350 MODULE_LICENSE("GPL");
ubuntu-bionic-kernel mirror