Merge tag 'perf-tools-fixes-for-v6.4-1-2023-05-20' of git://git.kernel.org/pub/scm...
[linux-block.git] / net / sched / act_pedit.c
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/slab.h>
17 #include <net/netlink.h>
18 #include <net/pkt_sched.h>
19 #include <linux/tc_act/tc_pedit.h>
20 #include <net/tc_act/tc_pedit.h>
21 #include <uapi/linux/tc_act/tc_pedit.h>
22 #include <net/pkt_cls.h>
23 #include <net/tc_wrapper.h>
24
25 static struct tc_action_ops act_pedit_ops;
26
27 static const struct nla_policy pedit_policy[TCA_PEDIT_MAX + 1] = {
28         [TCA_PEDIT_PARMS]       = { .len = sizeof(struct tc_pedit) },
29         [TCA_PEDIT_KEYS_EX]   = { .type = NLA_NESTED },
30 };
31
32 static const struct nla_policy pedit_key_ex_policy[TCA_PEDIT_KEY_EX_MAX + 1] = {
33         [TCA_PEDIT_KEY_EX_HTYPE] =
34                 NLA_POLICY_MAX(NLA_U16, TCA_PEDIT_HDR_TYPE_MAX),
35         [TCA_PEDIT_KEY_EX_CMD] = NLA_POLICY_MAX(NLA_U16, TCA_PEDIT_CMD_MAX),
36 };
37
38 static struct tcf_pedit_key_ex *tcf_pedit_keys_ex_parse(struct nlattr *nla,
39                                                         u8 n, struct netlink_ext_ack *extack)
40 {
41         struct tcf_pedit_key_ex *keys_ex;
42         struct tcf_pedit_key_ex *k;
43         const struct nlattr *ka;
44         int err = -EINVAL;
45         int rem;
46
47         if (!nla)
48                 return NULL;
49
50         keys_ex = kcalloc(n, sizeof(*k), GFP_KERNEL);
51         if (!keys_ex)
52                 return ERR_PTR(-ENOMEM);
53
54         k = keys_ex;
55
56         nla_for_each_nested(ka, nla, rem) {
57                 struct nlattr *tb[TCA_PEDIT_KEY_EX_MAX + 1];
58
59                 if (!n) {
60                         NL_SET_ERR_MSG_MOD(extack, "Can't parse more extended keys than requested");
61                         err = -EINVAL;
62                         goto err_out;
63                 }
64                 n--;
65
66                 if (nla_type(ka) != TCA_PEDIT_KEY_EX) {
67                         NL_SET_ERR_MSG_ATTR(extack, ka, "Unknown attribute, expected extended key");
68                         err = -EINVAL;
69                         goto err_out;
70                 }
71
72                 err = nla_parse_nested_deprecated(tb, TCA_PEDIT_KEY_EX_MAX,
73                                                   ka, pedit_key_ex_policy,
74                                                   NULL);
75                 if (err)
76                         goto err_out;
77
78                 if (NL_REQ_ATTR_CHECK(extack, nla, tb, TCA_PEDIT_KEY_EX_HTYPE)) {
79                         NL_SET_ERR_MSG(extack, "Missing required attribute");
80                         err = -EINVAL;
81                         goto err_out;
82                 }
83
84                 if (NL_REQ_ATTR_CHECK(extack, nla, tb, TCA_PEDIT_KEY_EX_CMD)) {
85                         NL_SET_ERR_MSG(extack, "Missing required attribute");
86                         err = -EINVAL;
87                         goto err_out;
88                 }
89
90                 k->htype = nla_get_u16(tb[TCA_PEDIT_KEY_EX_HTYPE]);
91                 k->cmd = nla_get_u16(tb[TCA_PEDIT_KEY_EX_CMD]);
92
93                 k++;
94         }
95
96         if (n) {
97                 NL_SET_ERR_MSG_MOD(extack, "Not enough extended keys to parse");
98                 err = -EINVAL;
99                 goto err_out;
100         }
101
102         return keys_ex;
103
104 err_out:
105         kfree(keys_ex);
106         return ERR_PTR(err);
107 }
108
109 static int tcf_pedit_key_ex_dump(struct sk_buff *skb,
110                                  struct tcf_pedit_key_ex *keys_ex, int n)
111 {
112         struct nlattr *keys_start = nla_nest_start_noflag(skb,
113                                                           TCA_PEDIT_KEYS_EX);
114
115         if (!keys_start)
116                 goto nla_failure;
117         for (; n > 0; n--) {
118                 struct nlattr *key_start;
119
120                 key_start = nla_nest_start_noflag(skb, TCA_PEDIT_KEY_EX);
121                 if (!key_start)
122                         goto nla_failure;
123
124                 if (nla_put_u16(skb, TCA_PEDIT_KEY_EX_HTYPE, keys_ex->htype) ||
125                     nla_put_u16(skb, TCA_PEDIT_KEY_EX_CMD, keys_ex->cmd))
126                         goto nla_failure;
127
128                 nla_nest_end(skb, key_start);
129
130                 keys_ex++;
131         }
132
133         nla_nest_end(skb, keys_start);
134
135         return 0;
136 nla_failure:
137         nla_nest_cancel(skb, keys_start);
138         return -EINVAL;
139 }
140
141 static void tcf_pedit_cleanup_rcu(struct rcu_head *head)
142 {
143         struct tcf_pedit_parms *parms =
144                 container_of(head, struct tcf_pedit_parms, rcu);
145
146         kfree(parms->tcfp_keys_ex);
147         kfree(parms->tcfp_keys);
148
149         kfree(parms);
150 }
151
152 static int tcf_pedit_init(struct net *net, struct nlattr *nla,
153                           struct nlattr *est, struct tc_action **a,
154                           struct tcf_proto *tp, u32 flags,
155                           struct netlink_ext_ack *extack)
156 {
157         struct tc_action_net *tn = net_generic(net, act_pedit_ops.net_id);
158         bool bind = flags & TCA_ACT_FLAGS_BIND;
159         struct tcf_chain *goto_ch = NULL;
160         struct tcf_pedit_parms *oparms, *nparms;
161         struct nlattr *tb[TCA_PEDIT_MAX + 1];
162         struct tc_pedit *parm;
163         struct nlattr *pattr;
164         struct tcf_pedit *p;
165         int ret = 0, err;
166         int i, ksize;
167         u32 index;
168
169         if (!nla) {
170                 NL_SET_ERR_MSG_MOD(extack, "Pedit requires attributes to be passed");
171                 return -EINVAL;
172         }
173
174         err = nla_parse_nested_deprecated(tb, TCA_PEDIT_MAX, nla,
175                                           pedit_policy, NULL);
176         if (err < 0)
177                 return err;
178
179         pattr = tb[TCA_PEDIT_PARMS];
180         if (!pattr)
181                 pattr = tb[TCA_PEDIT_PARMS_EX];
182         if (!pattr) {
183                 NL_SET_ERR_MSG_MOD(extack, "Missing required TCA_PEDIT_PARMS or TCA_PEDIT_PARMS_EX pedit attribute");
184                 return -EINVAL;
185         }
186
187         parm = nla_data(pattr);
188
189         index = parm->index;
190         err = tcf_idr_check_alloc(tn, &index, a, bind);
191         if (!err) {
192                 ret = tcf_idr_create_from_flags(tn, index, est, a,
193                                                 &act_pedit_ops, bind, flags);
194                 if (ret) {
195                         tcf_idr_cleanup(tn, index);
196                         return ret;
197                 }
198                 ret = ACT_P_CREATED;
199         } else if (err > 0) {
200                 if (bind)
201                         return 0;
202                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
203                         ret = -EEXIST;
204                         goto out_release;
205                 }
206         } else {
207                 return err;
208         }
209
210         if (!parm->nkeys) {
211                 NL_SET_ERR_MSG_MOD(extack, "Pedit requires keys to be passed");
212                 ret = -EINVAL;
213                 goto out_release;
214         }
215         ksize = parm->nkeys * sizeof(struct tc_pedit_key);
216         if (nla_len(pattr) < sizeof(*parm) + ksize) {
217                 NL_SET_ERR_MSG_ATTR(extack, pattr, "Length of TCA_PEDIT_PARMS or TCA_PEDIT_PARMS_EX pedit attribute is invalid");
218                 ret = -EINVAL;
219                 goto out_release;
220         }
221
222         nparms = kzalloc(sizeof(*nparms), GFP_KERNEL);
223         if (!nparms) {
224                 ret = -ENOMEM;
225                 goto out_release;
226         }
227
228         nparms->tcfp_keys_ex =
229                 tcf_pedit_keys_ex_parse(tb[TCA_PEDIT_KEYS_EX], parm->nkeys, extack);
230         if (IS_ERR(nparms->tcfp_keys_ex)) {
231                 ret = PTR_ERR(nparms->tcfp_keys_ex);
232                 goto out_free;
233         }
234
235         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
236         if (err < 0) {
237                 ret = err;
238                 goto out_free_ex;
239         }
240
241         nparms->tcfp_off_max_hint = 0;
242         nparms->tcfp_flags = parm->flags;
243         nparms->tcfp_nkeys = parm->nkeys;
244
245         nparms->tcfp_keys = kmalloc(ksize, GFP_KERNEL);
246         if (!nparms->tcfp_keys) {
247                 ret = -ENOMEM;
248                 goto put_chain;
249         }
250
251         memcpy(nparms->tcfp_keys, parm->keys, ksize);
252
253         for (i = 0; i < nparms->tcfp_nkeys; ++i) {
254                 u32 offmask = nparms->tcfp_keys[i].offmask;
255                 u32 cur = nparms->tcfp_keys[i].off;
256
257                 /* The AT option can be added to static offsets in the datapath */
258                 if (!offmask && cur % 4) {
259                         NL_SET_ERR_MSG_MOD(extack, "Offsets must be on 32bit boundaries");
260                         ret = -EINVAL;
261                         goto out_free_keys;
262                 }
263
264                 /* sanitize the shift value for any later use */
265                 nparms->tcfp_keys[i].shift = min_t(size_t,
266                                                    BITS_PER_TYPE(int) - 1,
267                                                    nparms->tcfp_keys[i].shift);
268
269                 /* The AT option can read a single byte, we can bound the actual
270                  * value with uchar max.
271                  */
272                 cur += (0xff & offmask) >> nparms->tcfp_keys[i].shift;
273
274                 /* Each key touches 4 bytes starting from the computed offset */
275                 nparms->tcfp_off_max_hint =
276                         max(nparms->tcfp_off_max_hint, cur + 4);
277         }
278
279         p = to_pedit(*a);
280
281         spin_lock_bh(&p->tcf_lock);
282         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
283         oparms = rcu_replace_pointer(p->parms, nparms, 1);
284         spin_unlock_bh(&p->tcf_lock);
285
286         if (oparms)
287                 call_rcu(&oparms->rcu, tcf_pedit_cleanup_rcu);
288
289         if (goto_ch)
290                 tcf_chain_put_by_act(goto_ch);
291
292         return ret;
293
294 out_free_keys:
295         kfree(nparms->tcfp_keys);
296 put_chain:
297         if (goto_ch)
298                 tcf_chain_put_by_act(goto_ch);
299 out_free_ex:
300         kfree(nparms->tcfp_keys_ex);
301 out_free:
302         kfree(nparms);
303 out_release:
304         tcf_idr_release(*a, bind);
305         return ret;
306 }
307
308 static void tcf_pedit_cleanup(struct tc_action *a)
309 {
310         struct tcf_pedit *p = to_pedit(a);
311         struct tcf_pedit_parms *parms;
312
313         parms = rcu_dereference_protected(p->parms, 1);
314
315         if (parms)
316                 call_rcu(&parms->rcu, tcf_pedit_cleanup_rcu);
317 }
318
319 static bool offset_valid(struct sk_buff *skb, int offset)
320 {
321         if (offset > 0 && offset > skb->len)
322                 return false;
323
324         if  (offset < 0 && -offset > skb_headroom(skb))
325                 return false;
326
327         return true;
328 }
329
330 static void pedit_skb_hdr_offset(struct sk_buff *skb,
331                                  enum pedit_header_type htype, int *hoffset)
332 {
333         /* 'htype' is validated in the netlink parsing */
334         switch (htype) {
335         case TCA_PEDIT_KEY_EX_HDR_TYPE_ETH:
336                 if (skb_mac_header_was_set(skb))
337                         *hoffset = skb_mac_offset(skb);
338                 break;
339         case TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK:
340         case TCA_PEDIT_KEY_EX_HDR_TYPE_IP4:
341         case TCA_PEDIT_KEY_EX_HDR_TYPE_IP6:
342                 *hoffset = skb_network_offset(skb);
343                 break;
344         case TCA_PEDIT_KEY_EX_HDR_TYPE_TCP:
345         case TCA_PEDIT_KEY_EX_HDR_TYPE_UDP:
346                 if (skb_transport_header_was_set(skb))
347                         *hoffset = skb_transport_offset(skb);
348                 break;
349         default:
350                 break;
351         }
352 }
353
354 TC_INDIRECT_SCOPE int tcf_pedit_act(struct sk_buff *skb,
355                                     const struct tc_action *a,
356                                     struct tcf_result *res)
357 {
358         enum pedit_header_type htype = TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK;
359         enum pedit_cmd cmd = TCA_PEDIT_KEY_EX_CMD_SET;
360         struct tcf_pedit *p = to_pedit(a);
361         struct tcf_pedit_key_ex *tkey_ex;
362         struct tcf_pedit_parms *parms;
363         struct tc_pedit_key *tkey;
364         u32 max_offset;
365         int i;
366
367         parms = rcu_dereference_bh(p->parms);
368
369         max_offset = (skb_transport_header_was_set(skb) ?
370                       skb_transport_offset(skb) :
371                       skb_network_offset(skb)) +
372                      parms->tcfp_off_max_hint;
373         if (skb_ensure_writable(skb, min(skb->len, max_offset)))
374                 goto done;
375
376         tcf_lastuse_update(&p->tcf_tm);
377         tcf_action_update_bstats(&p->common, skb);
378
379         tkey = parms->tcfp_keys;
380         tkey_ex = parms->tcfp_keys_ex;
381
382         for (i = parms->tcfp_nkeys; i > 0; i--, tkey++) {
383                 int offset = tkey->off;
384                 int hoffset = 0;
385                 u32 *ptr, hdata;
386                 u32 val;
387
388                 if (tkey_ex) {
389                         htype = tkey_ex->htype;
390                         cmd = tkey_ex->cmd;
391
392                         tkey_ex++;
393                 }
394
395                 pedit_skb_hdr_offset(skb, htype, &hoffset);
396
397                 if (tkey->offmask) {
398                         u8 *d, _d;
399
400                         if (!offset_valid(skb, hoffset + tkey->at)) {
401                                 pr_info_ratelimited("tc action pedit 'at' offset %d out of bounds\n",
402                                                     hoffset + tkey->at);
403                                 goto bad;
404                         }
405                         d = skb_header_pointer(skb, hoffset + tkey->at,
406                                                sizeof(_d), &_d);
407                         if (!d)
408                                 goto bad;
409
410                         offset += (*d & tkey->offmask) >> tkey->shift;
411                         if (offset % 4) {
412                                 pr_info_ratelimited("tc action pedit offset must be on 32 bit boundaries\n");
413                                 goto bad;
414                         }
415                 }
416
417                 if (!offset_valid(skb, hoffset + offset)) {
418                         pr_info_ratelimited("tc action pedit offset %d out of bounds\n", hoffset + offset);
419                         goto bad;
420                 }
421
422                 ptr = skb_header_pointer(skb, hoffset + offset,
423                                          sizeof(hdata), &hdata);
424                 if (!ptr)
425                         goto bad;
426                 /* just do it, baby */
427                 switch (cmd) {
428                 case TCA_PEDIT_KEY_EX_CMD_SET:
429                         val = tkey->val;
430                         break;
431                 case TCA_PEDIT_KEY_EX_CMD_ADD:
432                         val = (*ptr + tkey->val) & ~tkey->mask;
433                         break;
434                 default:
435                         pr_info_ratelimited("tc action pedit bad command (%d)\n", cmd);
436                         goto bad;
437                 }
438
439                 *ptr = ((*ptr & tkey->mask) ^ val);
440                 if (ptr == &hdata)
441                         skb_store_bits(skb, hoffset + offset, ptr, 4);
442         }
443
444         goto done;
445
446 bad:
447         tcf_action_inc_overlimit_qstats(&p->common);
448 done:
449         return p->tcf_action;
450 }
451
452 static void tcf_pedit_stats_update(struct tc_action *a, u64 bytes, u64 packets,
453                                    u64 drops, u64 lastuse, bool hw)
454 {
455         struct tcf_pedit *d = to_pedit(a);
456         struct tcf_t *tm = &d->tcf_tm;
457
458         tcf_action_update_stats(a, bytes, packets, drops, hw);
459         tm->lastuse = max_t(u64, tm->lastuse, lastuse);
460 }
461
462 static int tcf_pedit_dump(struct sk_buff *skb, struct tc_action *a,
463                           int bind, int ref)
464 {
465         unsigned char *b = skb_tail_pointer(skb);
466         struct tcf_pedit *p = to_pedit(a);
467         struct tcf_pedit_parms *parms;
468         struct tc_pedit *opt;
469         struct tcf_t t;
470         int s;
471
472         spin_lock_bh(&p->tcf_lock);
473         parms = rcu_dereference_protected(p->parms, 1);
474         s = struct_size(opt, keys, parms->tcfp_nkeys);
475
476         opt = kzalloc(s, GFP_ATOMIC);
477         if (unlikely(!opt)) {
478                 spin_unlock_bh(&p->tcf_lock);
479                 return -ENOBUFS;
480         }
481
482         memcpy(opt->keys, parms->tcfp_keys,
483                flex_array_size(opt, keys, parms->tcfp_nkeys));
484         opt->index = p->tcf_index;
485         opt->nkeys = parms->tcfp_nkeys;
486         opt->flags = parms->tcfp_flags;
487         opt->action = p->tcf_action;
488         opt->refcnt = refcount_read(&p->tcf_refcnt) - ref;
489         opt->bindcnt = atomic_read(&p->tcf_bindcnt) - bind;
490
491         if (parms->tcfp_keys_ex) {
492                 if (tcf_pedit_key_ex_dump(skb, parms->tcfp_keys_ex,
493                                           parms->tcfp_nkeys))
494                         goto nla_put_failure;
495
496                 if (nla_put(skb, TCA_PEDIT_PARMS_EX, s, opt))
497                         goto nla_put_failure;
498         } else {
499                 if (nla_put(skb, TCA_PEDIT_PARMS, s, opt))
500                         goto nla_put_failure;
501         }
502
503         tcf_tm_dump(&t, &p->tcf_tm);
504         if (nla_put_64bit(skb, TCA_PEDIT_TM, sizeof(t), &t, TCA_PEDIT_PAD))
505                 goto nla_put_failure;
506         spin_unlock_bh(&p->tcf_lock);
507
508         kfree(opt);
509         return skb->len;
510
511 nla_put_failure:
512         spin_unlock_bh(&p->tcf_lock);
513         nlmsg_trim(skb, b);
514         kfree(opt);
515         return -1;
516 }
517
518 static int tcf_pedit_offload_act_setup(struct tc_action *act, void *entry_data,
519                                        u32 *index_inc, bool bind,
520                                        struct netlink_ext_ack *extack)
521 {
522         if (bind) {
523                 struct flow_action_entry *entry = entry_data;
524                 int k;
525
526                 for (k = 0; k < tcf_pedit_nkeys(act); k++) {
527                         switch (tcf_pedit_cmd(act, k)) {
528                         case TCA_PEDIT_KEY_EX_CMD_SET:
529                                 entry->id = FLOW_ACTION_MANGLE;
530                                 break;
531                         case TCA_PEDIT_KEY_EX_CMD_ADD:
532                                 entry->id = FLOW_ACTION_ADD;
533                                 break;
534                         default:
535                                 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit command offload");
536                                 return -EOPNOTSUPP;
537                         }
538                         entry->mangle.htype = tcf_pedit_htype(act, k);
539                         entry->mangle.mask = tcf_pedit_mask(act, k);
540                         entry->mangle.val = tcf_pedit_val(act, k);
541                         entry->mangle.offset = tcf_pedit_offset(act, k);
542                         entry->hw_stats = tc_act_hw_stats(act->hw_stats);
543                         entry++;
544                 }
545                 *index_inc = k;
546         } else {
547                 struct flow_offload_action *fl_action = entry_data;
548                 u32 cmd = tcf_pedit_cmd(act, 0);
549                 int k;
550
551                 switch (cmd) {
552                 case TCA_PEDIT_KEY_EX_CMD_SET:
553                         fl_action->id = FLOW_ACTION_MANGLE;
554                         break;
555                 case TCA_PEDIT_KEY_EX_CMD_ADD:
556                         fl_action->id = FLOW_ACTION_ADD;
557                         break;
558                 default:
559                         NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit command offload");
560                         return -EOPNOTSUPP;
561                 }
562
563                 for (k = 1; k < tcf_pedit_nkeys(act); k++) {
564                         if (cmd != tcf_pedit_cmd(act, k)) {
565                                 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit command offload");
566                                 return -EOPNOTSUPP;
567                         }
568                 }
569         }
570
571         return 0;
572 }
573
574 static struct tc_action_ops act_pedit_ops = {
575         .kind           =       "pedit",
576         .id             =       TCA_ID_PEDIT,
577         .owner          =       THIS_MODULE,
578         .act            =       tcf_pedit_act,
579         .stats_update   =       tcf_pedit_stats_update,
580         .dump           =       tcf_pedit_dump,
581         .cleanup        =       tcf_pedit_cleanup,
582         .init           =       tcf_pedit_init,
583         .offload_act_setup =    tcf_pedit_offload_act_setup,
584         .size           =       sizeof(struct tcf_pedit),
585 };
586
587 static __net_init int pedit_init_net(struct net *net)
588 {
589         struct tc_action_net *tn = net_generic(net, act_pedit_ops.net_id);
590
591         return tc_action_net_init(net, tn, &act_pedit_ops);
592 }
593
594 static void __net_exit pedit_exit_net(struct list_head *net_list)
595 {
596         tc_action_net_exit(net_list, act_pedit_ops.net_id);
597 }
598
599 static struct pernet_operations pedit_net_ops = {
600         .init = pedit_init_net,
601         .exit_batch = pedit_exit_net,
602         .id   = &act_pedit_ops.net_id,
603         .size = sizeof(struct tc_action_net),
604 };
605
606 MODULE_AUTHOR("Jamal Hadi Salim(2002-4)");
607 MODULE_DESCRIPTION("Generic Packet Editor actions");
608 MODULE_LICENSE("GPL");
609
610 static int __init pedit_init_module(void)
611 {
612         return tcf_register_action(&act_pedit_ops, &pedit_net_ops);
613 }
614
615 static void __exit pedit_cleanup_module(void)
616 {
617         tcf_unregister_action(&act_pedit_ops, &pedit_net_ops);
618 }
619
620 module_init(pedit_init_module);
621 module_exit(pedit_cleanup_module);