1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * net/sched/act_pedit.c Generic packet editor
4 *
5 * Authors: Jamal Hadi Salim (2002-4)
6 */
7
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/string.h>
11 #include <linux/errno.h>
12 #include <linux/skbuff.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/ip.h>
17 #include <linux/ipv6.h>
18 #include <linux/slab.h>
19 #include <net/ipv6.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
22 #include <linux/tc_act/tc_pedit.h>
23 #include <net/tc_act/tc_pedit.h>
24 #include <uapi/linux/tc_act/tc_pedit.h>
25 #include <net/pkt_cls.h>
26 #include <net/tc_wrapper.h>
27
28 static struct tc_action_ops act_pedit_ops;
29
30 static const struct nla_policy pedit_policy[TCA_PEDIT_MAX + 1] = {
31 [TCA_PEDIT_PARMS] = { .len = sizeof(struct tc_pedit) },
32 [TCA_PEDIT_PARMS_EX] = { .len = sizeof(struct tc_pedit) },
33 [TCA_PEDIT_KEYS_EX] = { .type = NLA_NESTED },
34 };
35
36 static const struct nla_policy pedit_key_ex_policy[TCA_PEDIT_KEY_EX_MAX + 1] = {
37 [TCA_PEDIT_KEY_EX_HTYPE] =
38 NLA_POLICY_MAX(NLA_U16, TCA_PEDIT_HDR_TYPE_MAX),
39 [TCA_PEDIT_KEY_EX_CMD] = NLA_POLICY_MAX(NLA_U16, TCA_PEDIT_CMD_MAX),
40 };
41
42 static struct tcf_pedit_key_ex *tcf_pedit_keys_ex_parse(struct nlattr *nla,
43 u8 n, struct netlink_ext_ack *extack)
44 {
45 struct tcf_pedit_key_ex *keys_ex;
46 struct tcf_pedit_key_ex *k;
47 const struct nlattr *ka;
48 int err = -EINVAL;
49 int rem;
50
51 if (!nla)
52 return NULL;
53
54 keys_ex = kcalloc(n, sizeof(*k), GFP_KERNEL);
55 if (!keys_ex)
56 return ERR_PTR(-ENOMEM);
57
58 k = keys_ex;
59
60 nla_for_each_nested(ka, nla, rem) {
61 struct nlattr *tb[TCA_PEDIT_KEY_EX_MAX + 1];
62
63 if (!n) {
64 NL_SET_ERR_MSG_MOD(extack, "Can't parse more extended keys than requested");
65 err = -EINVAL;
66 goto err_out;
67 }
68 n--;
69
70 if (nla_type(ka) != TCA_PEDIT_KEY_EX) {
71 NL_SET_ERR_MSG_ATTR(extack, ka, "Unknown attribute, expected extended key");
72 err = -EINVAL;
73 goto err_out;
74 }
75
76 err = nla_parse_nested_deprecated(tb, TCA_PEDIT_KEY_EX_MAX,
77 ka, pedit_key_ex_policy,
78 NULL);
79 if (err)
80 goto err_out;
81
82 if (NL_REQ_ATTR_CHECK(extack, nla, tb, TCA_PEDIT_KEY_EX_HTYPE)) {
83 NL_SET_ERR_MSG(extack, "Missing required attribute");
84 err = -EINVAL;
85 goto err_out;
86 }
87
88 if (NL_REQ_ATTR_CHECK(extack, nla, tb, TCA_PEDIT_KEY_EX_CMD)) {
89 NL_SET_ERR_MSG(extack, "Missing required attribute");
90 err = -EINVAL;
91 goto err_out;
92 }
93
94 k->htype = nla_get_u16(tb[TCA_PEDIT_KEY_EX_HTYPE]);
95 k->cmd = nla_get_u16(tb[TCA_PEDIT_KEY_EX_CMD]);
96
97 k++;
98 }
99
100 if (n) {
101 NL_SET_ERR_MSG_MOD(extack, "Not enough extended keys to parse");
102 err = -EINVAL;
103 goto err_out;
104 }
105
106 return keys_ex;
107
108 err_out:
109 kfree(keys_ex);
110 return ERR_PTR(err);
111 }
112
113 static int tcf_pedit_key_ex_dump(struct sk_buff *skb,
114 struct tcf_pedit_key_ex *keys_ex, int n)
115 {
116 struct nlattr *keys_start = nla_nest_start_noflag(skb,
117 TCA_PEDIT_KEYS_EX);
118
119 if (!keys_start)
120 goto nla_failure;
121 for (; n > 0; n--) {
122 struct nlattr *key_start;
123
124 key_start = nla_nest_start_noflag(skb, TCA_PEDIT_KEY_EX);
125 if (!key_start)
126 goto nla_failure;
127
128 if (nla_put_u16(skb, TCA_PEDIT_KEY_EX_HTYPE, keys_ex->htype) ||
129 nla_put_u16(skb, TCA_PEDIT_KEY_EX_CMD, keys_ex->cmd))
130 goto nla_failure;
131
132 nla_nest_end(skb, key_start);
133
134 keys_ex++;
135 }
136
137 nla_nest_end(skb, keys_start);
138
139 return 0;
140 nla_failure:
141 nla_nest_cancel(skb, keys_start);
142 return -EINVAL;
143 }
144
145 static void tcf_pedit_cleanup_rcu(struct rcu_head *head)
146 {
147 struct tcf_pedit_parms *parms =
148 container_of(head, struct tcf_pedit_parms, rcu);
149
150 kfree(parms->tcfp_keys_ex);
151 kfree(parms->tcfp_keys);
152
153 kfree(parms);
154 }
155
156 static int tcf_pedit_init(struct net *net, struct nlattr *nla,
157 struct nlattr *est, struct tc_action **a,
158 struct tcf_proto *tp, u32 flags,
159 struct netlink_ext_ack *extack)
160 {
161 struct tc_action_net *tn = net_generic(net, act_pedit_ops.net_id);
162 bool bind = flags & TCA_ACT_FLAGS_BIND;
163 struct tcf_chain *goto_ch = NULL;
164 struct tcf_pedit_parms *oparms, *nparms;
165 struct nlattr *tb[TCA_PEDIT_MAX + 1];
166 struct tc_pedit *parm;
167 struct nlattr *pattr;
168 struct tcf_pedit *p;
169 int ret = 0, err;
170 int i, ksize;
171 u32 index;
172
173 if (!nla) {
174 NL_SET_ERR_MSG_MOD(extack, "Pedit requires attributes to be passed");
175 return -EINVAL;
176 }
177
178 err = nla_parse_nested_deprecated(tb, TCA_PEDIT_MAX, nla,
179 pedit_policy, NULL);
180 if (err < 0)
181 return err;
182
183 pattr = tb[TCA_PEDIT_PARMS];
184 if (!pattr)
185 pattr = tb[TCA_PEDIT_PARMS_EX];
186 if (!pattr) {
187 NL_SET_ERR_MSG_MOD(extack, "Missing required TCA_PEDIT_PARMS or TCA_PEDIT_PARMS_EX pedit attribute");
188 return -EINVAL;
189 }
190
191 parm = nla_data(pattr);
192
193 index = parm->index;
194 err = tcf_idr_check_alloc(tn, &index, a, bind);
195 if (!err) {
196 ret = tcf_idr_create_from_flags(tn, index, est, a,
197 &act_pedit_ops, bind, flags);
198 if (ret) {
199 tcf_idr_cleanup(tn, index);
200 return ret;
201 }
202 ret = ACT_P_CREATED;
203 } else if (err > 0) {
204 if (bind)
205 return ACT_P_BOUND;
206 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
207 ret = -EEXIST;
208 goto out_release;
209 }
210 } else {
211 return err;
212 }
213
214 if (!parm->nkeys) {
215 NL_SET_ERR_MSG_MOD(extack, "Pedit requires keys to be passed");
216 ret = -EINVAL;
217 goto out_release;
218 }
219 ksize = parm->nkeys * sizeof(struct tc_pedit_key);
220 if (nla_len(pattr) < sizeof(*parm) + ksize) {
221 NL_SET_ERR_MSG_ATTR(extack, pattr, "Length of TCA_PEDIT_PARMS or TCA_PEDIT_PARMS_EX pedit attribute is invalid");
222 ret = -EINVAL;
223 goto out_release;
224 }
225
226 nparms = kzalloc(sizeof(*nparms), GFP_KERNEL);
227 if (!nparms) {
228 ret = -ENOMEM;
229 goto out_release;
230 }
231
232 nparms->tcfp_keys_ex =
233 tcf_pedit_keys_ex_parse(tb[TCA_PEDIT_KEYS_EX], parm->nkeys, extack);
234 if (IS_ERR(nparms->tcfp_keys_ex)) {
235 ret = PTR_ERR(nparms->tcfp_keys_ex);
236 goto out_free;
237 }
238
239 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
240 if (err < 0) {
241 ret = err;
242 goto out_free_ex;
243 }
244
245 nparms->tcfp_off_max_hint = 0;
246 nparms->tcfp_flags = parm->flags;
247 nparms->tcfp_nkeys = parm->nkeys;
248
249 nparms->tcfp_keys = kmemdup(parm->keys, ksize, GFP_KERNEL);
250 if (!nparms->tcfp_keys) {
251 ret = -ENOMEM;
252 goto put_chain;
253 }
254
255 for (i = 0; i < nparms->tcfp_nkeys; ++i) {
256 u32 offmask = nparms->tcfp_keys[i].offmask;
257 u32 cur = nparms->tcfp_keys[i].off;
258
259 /* The AT option can be added to static offsets in the datapath */
260 if (!offmask && cur % 4) {
261 NL_SET_ERR_MSG_MOD(extack, "Offsets must be on 32bit boundaries");
262 ret = -EINVAL;
263 goto out_free_keys;
264 }
265
266 /* sanitize the shift value for any later use */
267 nparms->tcfp_keys[i].shift = min_t(size_t,
268 BITS_PER_TYPE(int) - 1,
269 nparms->tcfp_keys[i].shift);
270
271 /* The AT option can read a single byte, we can bound the actual
272 * value with uchar max.
273 */
274 cur += (0xff & offmask) >> nparms->tcfp_keys[i].shift;
275
276 /* Each key touches 4 bytes starting from the computed offset */
277 nparms->tcfp_off_max_hint =
278 max(nparms->tcfp_off_max_hint, cur + 4);
279 }
280
281 p = to_pedit(*a);
282
283 spin_lock_bh(&p->tcf_lock);
284 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
285 oparms = rcu_replace_pointer(p->parms, nparms, 1);
286 spin_unlock_bh(&p->tcf_lock);
287
288 if (oparms)
289 call_rcu(&oparms->rcu, tcf_pedit_cleanup_rcu);
290
291 if (goto_ch)
292 tcf_chain_put_by_act(goto_ch);
293
294 return ret;
295
296 out_free_keys:
297 kfree(nparms->tcfp_keys);
298 put_chain:
299 if (goto_ch)
300 tcf_chain_put_by_act(goto_ch);
301 out_free_ex:
302 kfree(nparms->tcfp_keys_ex);
303 out_free:
304 kfree(nparms);
305 out_release:
306 tcf_idr_release(*a, bind);
307 return ret;
308 }
309
310 static void tcf_pedit_cleanup(struct tc_action *a)
311 {
312 struct tcf_pedit *p = to_pedit(a);
313 struct tcf_pedit_parms *parms;
314
315 parms = rcu_dereference_protected(p->parms, 1);
316
317 if (parms)
318 call_rcu(&parms->rcu, tcf_pedit_cleanup_rcu);
319 }
320
321 static bool offset_valid(struct sk_buff *skb, int offset)
322 {
323 if (offset > 0 && offset > skb->len)
324 return false;
325
326 if (offset < 0 && -offset > skb_headroom(skb))
327 return false;
328
329 return true;
330 }
331
332 static int pedit_l4_skb_offset(struct sk_buff *skb, int *hoffset, const int header_type)
333 {
334 const int noff = skb_network_offset(skb);
335 int ret = -EINVAL;
336 struct iphdr _iph;
337
338 switch (skb->protocol) {
339 case htons(ETH_P_IP): {
340 const struct iphdr *iph = skb_header_pointer(skb, noff, sizeof(_iph), &_iph);
341
342 if (!iph)
343 goto out;
344 *hoffset = noff + iph->ihl * 4;
345 ret = 0;
346 break;
347 }
348 case htons(ETH_P_IPV6):
349 ret = ipv6_find_hdr(skb, hoffset, header_type, NULL, NULL) == header_type ? 0 : -EINVAL;
350 break;
351 }
352 out:
353 return ret;
354 }
355
356 static int pedit_skb_hdr_offset(struct sk_buff *skb,
357 enum pedit_header_type htype, int *hoffset)
358 {
359 int ret = -EINVAL;
360 /* 'htype' is validated in the netlink parsing */
361 switch (htype) {
362 case TCA_PEDIT_KEY_EX_HDR_TYPE_ETH:
363 if (skb_mac_header_was_set(skb)) {
364 *hoffset = skb_mac_offset(skb);
365 ret = 0;
366 }
367 break;
368 case TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK:
369 case TCA_PEDIT_KEY_EX_HDR_TYPE_IP4:
370 case TCA_PEDIT_KEY_EX_HDR_TYPE_IP6:
371 *hoffset = skb_network_offset(skb);
372 ret = 0;
373 break;
374 case TCA_PEDIT_KEY_EX_HDR_TYPE_TCP:
375 ret = pedit_l4_skb_offset(skb, hoffset, IPPROTO_TCP);
376 break;
377 case TCA_PEDIT_KEY_EX_HDR_TYPE_UDP:
378 ret = pedit_l4_skb_offset(skb, hoffset, IPPROTO_UDP);
379 break;
380 default:
381 break;
382 }
383 return ret;
384 }
385
386 TC_INDIRECT_SCOPE int tcf_pedit_act(struct sk_buff *skb,
387 const struct tc_action *a,
388 struct tcf_result *res)
389 {
390 enum pedit_header_type htype = TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK;
391 enum pedit_cmd cmd = TCA_PEDIT_KEY_EX_CMD_SET;
392 struct tcf_pedit *p = to_pedit(a);
393 struct tcf_pedit_key_ex *tkey_ex;
394 struct tcf_pedit_parms *parms;
395 struct tc_pedit_key *tkey;
396 u32 max_offset;
397 int i;
398
399 parms = rcu_dereference_bh(p->parms);
400
401 max_offset = (skb_transport_header_was_set(skb) ?
402 skb_transport_offset(skb) :
403 skb_network_offset(skb)) +
404 parms->tcfp_off_max_hint;
405 if (skb_ensure_writable(skb, min(skb->len, max_offset)))
406 goto done;
407
408 tcf_lastuse_update(&p->tcf_tm);
409 tcf_action_update_bstats(&p->common, skb);
410
411 tkey = parms->tcfp_keys;
412 tkey_ex = parms->tcfp_keys_ex;
413
414 for (i = parms->tcfp_nkeys; i > 0; i--, tkey++) {
415 int offset = tkey->off;
416 int hoffset = 0;
417 u32 *ptr, hdata;
418 u32 val;
419 int rc;
420
421 if (tkey_ex) {
422 htype = tkey_ex->htype;
423 cmd = tkey_ex->cmd;
424
425 tkey_ex++;
426 }
427
428 rc = pedit_skb_hdr_offset(skb, htype, &hoffset);
429 if (rc) {
430 pr_info_ratelimited("tc action pedit unable to extract header offset for header type (0x%x)\n", htype);
431 goto bad;
432 }
433
434 if (tkey->offmask) {
435 u8 *d, _d;
436
437 if (!offset_valid(skb, hoffset + tkey->at)) {
438 pr_info_ratelimited("tc action pedit 'at' offset %d out of bounds\n",
439 hoffset + tkey->at);
440 goto bad;
441 }
442 d = skb_header_pointer(skb, hoffset + tkey->at,
443 sizeof(_d), &_d);
444 if (!d)
445 goto bad;
446
447 offset += (*d & tkey->offmask) >> tkey->shift;
448 if (offset % 4) {
449 pr_info_ratelimited("tc action pedit offset must be on 32 bit boundaries\n");
450 goto bad;
451 }
452 }
453
454 if (!offset_valid(skb, hoffset + offset)) {
455 pr_info_ratelimited("tc action pedit offset %d out of bounds\n", hoffset + offset);
456 goto bad;
457 }
458
459 ptr = skb_header_pointer(skb, hoffset + offset,
460 sizeof(hdata), &hdata);
461 if (!ptr)
462 goto bad;
463 /* just do it, baby */
464 switch (cmd) {
465 case TCA_PEDIT_KEY_EX_CMD_SET:
466 val = tkey->val;
467 break;
468 case TCA_PEDIT_KEY_EX_CMD_ADD:
469 val = (*ptr + tkey->val) & ~tkey->mask;
470 break;
471 default:
472 pr_info_ratelimited("tc action pedit bad command (%d)\n", cmd);
473 goto bad;
474 }
475
476 *ptr = ((*ptr & tkey->mask) ^ val);
477 if (ptr == &hdata)
478 skb_store_bits(skb, hoffset + offset, ptr, 4);
479 }
480
481 goto done;
482
483 bad:
484 tcf_action_inc_overlimit_qstats(&p->common);
485 done:
486 return p->tcf_action;
487 }
488
489 static void tcf_pedit_stats_update(struct tc_action *a, u64 bytes, u64 packets,
490 u64 drops, u64 lastuse, bool hw)
491 {
492 struct tcf_pedit *d = to_pedit(a);
493 struct tcf_t *tm = &d->tcf_tm;
494
495 tcf_action_update_stats(a, bytes, packets, drops, hw);
496 tm->lastuse = max_t(u64, tm->lastuse, lastuse);
497 }
498
499 static int tcf_pedit_dump(struct sk_buff *skb, struct tc_action *a,
500 int bind, int ref)
501 {
502 unsigned char *b = skb_tail_pointer(skb);
503 struct tcf_pedit *p = to_pedit(a);
504 struct tcf_pedit_parms *parms;
505 struct tc_pedit *opt;
506 struct tcf_t t;
507 int s;
508
509 spin_lock_bh(&p->tcf_lock);
510 parms = rcu_dereference_protected(p->parms, 1);
511 s = struct_size(opt, keys, parms->tcfp_nkeys);
512
513 opt = kzalloc(s, GFP_ATOMIC);
514 if (unlikely(!opt)) {
515 spin_unlock_bh(&p->tcf_lock);
516 return -ENOBUFS;
517 }
518 opt->nkeys = parms->tcfp_nkeys;
519
520 memcpy(opt->keys, parms->tcfp_keys,
521 flex_array_size(opt, keys, parms->tcfp_nkeys));
522 opt->index = p->tcf_index;
523 opt->flags = parms->tcfp_flags;
524 opt->action = p->tcf_action;
525 opt->refcnt = refcount_read(&p->tcf_refcnt) - ref;
526 opt->bindcnt = atomic_read(&p->tcf_bindcnt) - bind;
527
528 if (parms->tcfp_keys_ex) {
529 if (tcf_pedit_key_ex_dump(skb, parms->tcfp_keys_ex,
530 parms->tcfp_nkeys))
531 goto nla_put_failure;
532
533 if (nla_put(skb, TCA_PEDIT_PARMS_EX, s, opt))
534 goto nla_put_failure;
535 } else {
536 if (nla_put(skb, TCA_PEDIT_PARMS, s, opt))
537 goto nla_put_failure;
538 }
539
540 tcf_tm_dump(&t, &p->tcf_tm);
541 if (nla_put_64bit(skb, TCA_PEDIT_TM, sizeof(t), &t, TCA_PEDIT_PAD))
542 goto nla_put_failure;
543 spin_unlock_bh(&p->tcf_lock);
544
545 kfree(opt);
546 return skb->len;
547
548 nla_put_failure:
549 spin_unlock_bh(&p->tcf_lock);
550 nlmsg_trim(skb, b);
551 kfree(opt);
552 return -1;
553 }
554
555 static int tcf_pedit_offload_act_setup(struct tc_action *act, void *entry_data,
556 u32 *index_inc, bool bind,
557 struct netlink_ext_ack *extack)
558 {
559 if (bind) {
560 struct flow_action_entry *entry = entry_data;
561 int k;
562
563 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
564 switch (tcf_pedit_cmd(act, k)) {
565 case TCA_PEDIT_KEY_EX_CMD_SET:
566 entry->id = FLOW_ACTION_MANGLE;
567 break;
568 case TCA_PEDIT_KEY_EX_CMD_ADD:
569 entry->id = FLOW_ACTION_ADD;
570 break;
571 default:
572 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit command offload");
573 return -EOPNOTSUPP;
574 }
575 entry->mangle.htype = tcf_pedit_htype(act, k);
576 entry->mangle.mask = tcf_pedit_mask(act, k);
577 entry->mangle.val = tcf_pedit_val(act, k);
578 entry->mangle.offset = tcf_pedit_offset(act, k);
579 entry->hw_stats = tc_act_hw_stats(act->hw_stats);
580 entry++;
581 }
582 *index_inc = k;
583 } else {
584 struct flow_offload_action *fl_action = entry_data;
585 u32 cmd = tcf_pedit_cmd(act, 0);
586 int k;
587
588 switch (cmd) {
589 case TCA_PEDIT_KEY_EX_CMD_SET:
590 fl_action->id = FLOW_ACTION_MANGLE;
591 break;
592 case TCA_PEDIT_KEY_EX_CMD_ADD:
593 fl_action->id = FLOW_ACTION_ADD;
594 break;
595 default:
596 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit command offload");
597 return -EOPNOTSUPP;
598 }
599
600 for (k = 1; k < tcf_pedit_nkeys(act); k++) {
601 if (cmd != tcf_pedit_cmd(act, k)) {
602 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit command offload");
603 return -EOPNOTSUPP;
604 }
605 }
606 }
607
608 return 0;
609 }
610
611 static struct tc_action_ops act_pedit_ops = {
612 .kind = "pedit",
613 .id = TCA_ID_PEDIT,
614 .owner = THIS_MODULE,
615 .act = tcf_pedit_act,
616 .stats_update = tcf_pedit_stats_update,
617 .dump = tcf_pedit_dump,
618 .cleanup = tcf_pedit_cleanup,
619 .init = tcf_pedit_init,
620 .offload_act_setup = tcf_pedit_offload_act_setup,
621 .size = sizeof(struct tcf_pedit),
622 };
623 MODULE_ALIAS_NET_ACT("pedit");
624
625 static __net_init int pedit_init_net(struct net *net)
626 {
627 struct tc_action_net *tn = net_generic(net, act_pedit_ops.net_id);
628
629 return tc_action_net_init(net, tn, &act_pedit_ops);
630 }
631
632 static void __net_exit pedit_exit_net(struct list_head *net_list)
633 {
634 tc_action_net_exit(net_list, act_pedit_ops.net_id);
635 }
636
637 static struct pernet_operations pedit_net_ops = {
638 .init = pedit_init_net,
639 .exit_batch = pedit_exit_net,
640 .id = &act_pedit_ops.net_id,
641 .size = sizeof(struct tc_action_net),
642 };
643
644 MODULE_AUTHOR("Jamal Hadi Salim(2002-4)");
645 MODULE_DESCRIPTION("Generic Packet Editor actions");
646 MODULE_LICENSE("GPL");
647
648 static int __init pedit_init_module(void)
649 {
650 return tcf_register_action(&act_pedit_ops, &pedit_net_ops);
651 }
652
653 static void __exit pedit_cleanup_module(void)
654 {
655 tcf_unregister_action(&act_pedit_ops, &pedit_net_ops);
656 }
657
658 module_init(pedit_init_module);
659 module_exit(pedit_cleanup_module);
660
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.