]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/blob - net/sched/cls_bpf.c
net: sched: cls_bpf: use bitwise & rather than logical && on gen_flags
[mirror_ubuntu-disco-kernel.git] / net / sched / cls_bpf.c
1 /*
2 * Berkeley Packet Filter based traffic classifier
3 *
4 * Might be used to classify traffic through flexible, user-defined and
5 * possibly JIT-ed BPF filters for traffic control as an alternative to
6 * ematches.
7 *
8 * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
13 */
14
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/skbuff.h>
18 #include <linux/filter.h>
19 #include <linux/bpf.h>
20 #include <linux/idr.h>
21
22 #include <net/rtnetlink.h>
23 #include <net/pkt_cls.h>
24 #include <net/sock.h>
25
26 MODULE_LICENSE("GPL");
27 MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
28 MODULE_DESCRIPTION("TC BPF based classifier");
29
30 #define CLS_BPF_NAME_LEN 256
31 #define CLS_BPF_SUPPORTED_GEN_FLAGS \
32 (TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)
33
34 struct cls_bpf_head {
35 struct list_head plist;
36 struct idr handle_idr;
37 struct rcu_head rcu;
38 };
39
40 struct cls_bpf_prog {
41 struct bpf_prog *filter;
42 struct list_head link;
43 struct tcf_result res;
44 bool exts_integrated;
45 bool offloaded;
46 u32 gen_flags;
47 struct tcf_exts exts;
48 u32 handle;
49 u16 bpf_num_ops;
50 struct sock_filter *bpf_ops;
51 const char *bpf_name;
52 struct tcf_proto *tp;
53 union {
54 struct work_struct work;
55 struct rcu_head rcu;
56 };
57 };
58
59 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
60 [TCA_BPF_CLASSID] = { .type = NLA_U32 },
61 [TCA_BPF_FLAGS] = { .type = NLA_U32 },
62 [TCA_BPF_FLAGS_GEN] = { .type = NLA_U32 },
63 [TCA_BPF_FD] = { .type = NLA_U32 },
64 [TCA_BPF_NAME] = { .type = NLA_NUL_STRING,
65 .len = CLS_BPF_NAME_LEN },
66 [TCA_BPF_OPS_LEN] = { .type = NLA_U16 },
67 [TCA_BPF_OPS] = { .type = NLA_BINARY,
68 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
69 };
70
71 static int cls_bpf_exec_opcode(int code)
72 {
73 switch (code) {
74 case TC_ACT_OK:
75 case TC_ACT_SHOT:
76 case TC_ACT_STOLEN:
77 case TC_ACT_TRAP:
78 case TC_ACT_REDIRECT:
79 case TC_ACT_UNSPEC:
80 return code;
81 default:
82 return TC_ACT_UNSPEC;
83 }
84 }
85
86 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
87 struct tcf_result *res)
88 {
89 struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
90 bool at_ingress = skb_at_tc_ingress(skb);
91 struct cls_bpf_prog *prog;
92 int ret = -1;
93
94 /* Needed here for accessing maps. */
95 rcu_read_lock();
96 list_for_each_entry_rcu(prog, &head->plist, link) {
97 int filter_res;
98
99 qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
100
101 if (tc_skip_sw(prog->gen_flags)) {
102 filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
103 } else if (at_ingress) {
104 /* It is safe to push/pull even if skb_shared() */
105 __skb_push(skb, skb->mac_len);
106 bpf_compute_data_pointers(skb);
107 filter_res = BPF_PROG_RUN(prog->filter, skb);
108 __skb_pull(skb, skb->mac_len);
109 } else {
110 bpf_compute_data_pointers(skb);
111 filter_res = BPF_PROG_RUN(prog->filter, skb);
112 }
113
114 if (prog->exts_integrated) {
115 res->class = 0;
116 res->classid = TC_H_MAJ(prog->res.classid) |
117 qdisc_skb_cb(skb)->tc_classid;
118
119 ret = cls_bpf_exec_opcode(filter_res);
120 if (ret == TC_ACT_UNSPEC)
121 continue;
122 break;
123 }
124
125 if (filter_res == 0)
126 continue;
127 if (filter_res != -1) {
128 res->class = 0;
129 res->classid = filter_res;
130 } else {
131 *res = prog->res;
132 }
133
134 ret = tcf_exts_exec(skb, &prog->exts, res);
135 if (ret < 0)
136 continue;
137
138 break;
139 }
140 rcu_read_unlock();
141
142 return ret;
143 }
144
145 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
146 {
147 return !prog->bpf_ops;
148 }
149
150 static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
151 enum tc_clsbpf_command cmd)
152 {
153 bool addorrep = cmd == TC_CLSBPF_ADD || cmd == TC_CLSBPF_REPLACE;
154 struct tcf_block *block = tp->chain->block;
155 bool skip_sw = tc_skip_sw(prog->gen_flags);
156 struct tc_cls_bpf_offload cls_bpf = {};
157 int err;
158
159 tc_cls_common_offload_init(&cls_bpf.common, tp);
160 cls_bpf.command = cmd;
161 cls_bpf.exts = &prog->exts;
162 cls_bpf.prog = prog->filter;
163 cls_bpf.name = prog->bpf_name;
164 cls_bpf.exts_integrated = prog->exts_integrated;
165 cls_bpf.gen_flags = prog->gen_flags;
166
167 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSBPF, &cls_bpf, skip_sw);
168 if (addorrep) {
169 if (err < 0) {
170 cls_bpf_offload_cmd(tp, prog, TC_CLSBPF_DESTROY);
171 return err;
172 } else if (err > 0) {
173 prog->gen_flags |= TCA_CLS_FLAGS_IN_HW;
174 }
175 }
176
177 if (addorrep && skip_sw && !(prog->gen_flags & TCA_CLS_FLAGS_IN_HW))
178 return -EINVAL;
179
180 return 0;
181 }
182
183 static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
184 struct cls_bpf_prog *oldprog)
185 {
186 struct cls_bpf_prog *obj = prog;
187 enum tc_clsbpf_command cmd;
188 bool skip_sw;
189 int ret;
190
191 skip_sw = tc_skip_sw(prog->gen_flags) ||
192 (oldprog && tc_skip_sw(oldprog->gen_flags));
193
194 if (oldprog && oldprog->offloaded) {
195 if (!tc_skip_hw(prog->gen_flags)) {
196 cmd = TC_CLSBPF_REPLACE;
197 } else if (!tc_skip_sw(prog->gen_flags)) {
198 obj = oldprog;
199 cmd = TC_CLSBPF_DESTROY;
200 } else {
201 return -EINVAL;
202 }
203 } else {
204 if (tc_skip_hw(prog->gen_flags))
205 return skip_sw ? -EINVAL : 0;
206 cmd = TC_CLSBPF_ADD;
207 }
208
209 ret = cls_bpf_offload_cmd(tp, obj, cmd);
210 if (ret)
211 return ret;
212
213 obj->offloaded = true;
214 if (oldprog)
215 oldprog->offloaded = false;
216
217 return 0;
218 }
219
220 static void cls_bpf_stop_offload(struct tcf_proto *tp,
221 struct cls_bpf_prog *prog)
222 {
223 int err;
224
225 if (!prog->offloaded)
226 return;
227
228 err = cls_bpf_offload_cmd(tp, prog, TC_CLSBPF_DESTROY);
229 if (err) {
230 pr_err("Stopping hardware offload failed: %d\n", err);
231 return;
232 }
233
234 prog->offloaded = false;
235 }
236
237 static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
238 struct cls_bpf_prog *prog)
239 {
240 if (!prog->offloaded)
241 return;
242
243 cls_bpf_offload_cmd(tp, prog, TC_CLSBPF_STATS);
244 }
245
246 static int cls_bpf_init(struct tcf_proto *tp)
247 {
248 struct cls_bpf_head *head;
249
250 head = kzalloc(sizeof(*head), GFP_KERNEL);
251 if (head == NULL)
252 return -ENOBUFS;
253
254 INIT_LIST_HEAD_RCU(&head->plist);
255 idr_init(&head->handle_idr);
256 rcu_assign_pointer(tp->root, head);
257
258 return 0;
259 }
260
261 static void __cls_bpf_delete_prog(struct cls_bpf_prog *prog)
262 {
263 tcf_exts_destroy(&prog->exts);
264
265 if (cls_bpf_is_ebpf(prog))
266 bpf_prog_put(prog->filter);
267 else
268 bpf_prog_destroy(prog->filter);
269
270 kfree(prog->bpf_name);
271 kfree(prog->bpf_ops);
272 kfree(prog);
273 }
274
275 static void cls_bpf_delete_prog_work(struct work_struct *work)
276 {
277 struct cls_bpf_prog *prog = container_of(work, struct cls_bpf_prog, work);
278
279 rtnl_lock();
280 __cls_bpf_delete_prog(prog);
281 rtnl_unlock();
282 }
283
284 static void cls_bpf_delete_prog_rcu(struct rcu_head *rcu)
285 {
286 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
287
288 INIT_WORK(&prog->work, cls_bpf_delete_prog_work);
289 tcf_queue_work(&prog->work);
290 }
291
292 static void __cls_bpf_delete(struct tcf_proto *tp, struct cls_bpf_prog *prog)
293 {
294 struct cls_bpf_head *head = rtnl_dereference(tp->root);
295
296 idr_remove_ext(&head->handle_idr, prog->handle);
297 cls_bpf_stop_offload(tp, prog);
298 list_del_rcu(&prog->link);
299 tcf_unbind_filter(tp, &prog->res);
300 call_rcu(&prog->rcu, cls_bpf_delete_prog_rcu);
301 }
302
303 static int cls_bpf_delete(struct tcf_proto *tp, void *arg, bool *last)
304 {
305 struct cls_bpf_head *head = rtnl_dereference(tp->root);
306
307 __cls_bpf_delete(tp, arg);
308 *last = list_empty(&head->plist);
309 return 0;
310 }
311
312 static void cls_bpf_destroy(struct tcf_proto *tp)
313 {
314 struct cls_bpf_head *head = rtnl_dereference(tp->root);
315 struct cls_bpf_prog *prog, *tmp;
316
317 list_for_each_entry_safe(prog, tmp, &head->plist, link)
318 __cls_bpf_delete(tp, prog);
319
320 idr_destroy(&head->handle_idr);
321 kfree_rcu(head, rcu);
322 }
323
324 static void *cls_bpf_get(struct tcf_proto *tp, u32 handle)
325 {
326 struct cls_bpf_head *head = rtnl_dereference(tp->root);
327 struct cls_bpf_prog *prog;
328
329 list_for_each_entry(prog, &head->plist, link) {
330 if (prog->handle == handle)
331 return prog;
332 }
333
334 return NULL;
335 }
336
337 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
338 {
339 struct sock_filter *bpf_ops;
340 struct sock_fprog_kern fprog_tmp;
341 struct bpf_prog *fp;
342 u16 bpf_size, bpf_num_ops;
343 int ret;
344
345 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
346 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
347 return -EINVAL;
348
349 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
350 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
351 return -EINVAL;
352
353 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
354 if (bpf_ops == NULL)
355 return -ENOMEM;
356
357 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
358
359 fprog_tmp.len = bpf_num_ops;
360 fprog_tmp.filter = bpf_ops;
361
362 ret = bpf_prog_create(&fp, &fprog_tmp);
363 if (ret < 0) {
364 kfree(bpf_ops);
365 return ret;
366 }
367
368 prog->bpf_ops = bpf_ops;
369 prog->bpf_num_ops = bpf_num_ops;
370 prog->bpf_name = NULL;
371 prog->filter = fp;
372
373 return 0;
374 }
375
376 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
377 const struct tcf_proto *tp)
378 {
379 struct bpf_prog *fp;
380 char *name = NULL;
381 u32 bpf_fd;
382
383 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
384
385 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_CLS);
386 if (IS_ERR(fp))
387 return PTR_ERR(fp);
388
389 if (tb[TCA_BPF_NAME]) {
390 name = nla_memdup(tb[TCA_BPF_NAME], GFP_KERNEL);
391 if (!name) {
392 bpf_prog_put(fp);
393 return -ENOMEM;
394 }
395 }
396
397 prog->bpf_ops = NULL;
398 prog->bpf_name = name;
399 prog->filter = fp;
400
401 if (fp->dst_needed && !(tp->q->flags & TCQ_F_INGRESS))
402 netif_keep_dst(qdisc_dev(tp->q));
403
404 return 0;
405 }
406
407 static int cls_bpf_set_parms(struct net *net, struct tcf_proto *tp,
408 struct cls_bpf_prog *prog, unsigned long base,
409 struct nlattr **tb, struct nlattr *est, bool ovr)
410 {
411 bool is_bpf, is_ebpf, have_exts = false;
412 u32 gen_flags = 0;
413 int ret;
414
415 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
416 is_ebpf = tb[TCA_BPF_FD];
417 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
418 return -EINVAL;
419
420 ret = tcf_exts_validate(net, tp, tb, est, &prog->exts, ovr);
421 if (ret < 0)
422 return ret;
423
424 if (tb[TCA_BPF_FLAGS]) {
425 u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
426
427 if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT)
428 return -EINVAL;
429
430 have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
431 }
432 if (tb[TCA_BPF_FLAGS_GEN]) {
433 gen_flags = nla_get_u32(tb[TCA_BPF_FLAGS_GEN]);
434 if (gen_flags & ~CLS_BPF_SUPPORTED_GEN_FLAGS ||
435 !tc_flags_valid(gen_flags))
436 return -EINVAL;
437 }
438
439 prog->exts_integrated = have_exts;
440 prog->gen_flags = gen_flags;
441
442 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
443 cls_bpf_prog_from_efd(tb, prog, tp);
444 if (ret < 0)
445 return ret;
446
447 if (tb[TCA_BPF_CLASSID]) {
448 prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
449 tcf_bind_filter(tp, &prog->res, base);
450 }
451
452 return 0;
453 }
454
455 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
456 struct tcf_proto *tp, unsigned long base,
457 u32 handle, struct nlattr **tca,
458 void **arg, bool ovr)
459 {
460 struct cls_bpf_head *head = rtnl_dereference(tp->root);
461 struct cls_bpf_prog *oldprog = *arg;
462 struct nlattr *tb[TCA_BPF_MAX + 1];
463 struct cls_bpf_prog *prog;
464 unsigned long idr_index;
465 int ret;
466
467 if (tca[TCA_OPTIONS] == NULL)
468 return -EINVAL;
469
470 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy,
471 NULL);
472 if (ret < 0)
473 return ret;
474
475 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
476 if (!prog)
477 return -ENOBUFS;
478
479 ret = tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
480 if (ret < 0)
481 goto errout;
482
483 if (oldprog) {
484 if (handle && oldprog->handle != handle) {
485 ret = -EINVAL;
486 goto errout;
487 }
488 }
489
490 if (handle == 0) {
491 ret = idr_alloc_ext(&head->handle_idr, prog, &idr_index,
492 1, 0x7FFFFFFF, GFP_KERNEL);
493 if (ret)
494 goto errout;
495 prog->handle = idr_index;
496 } else {
497 if (!oldprog) {
498 ret = idr_alloc_ext(&head->handle_idr, prog, &idr_index,
499 handle, handle + 1, GFP_KERNEL);
500 if (ret)
501 goto errout;
502 }
503 prog->handle = handle;
504 }
505
506 ret = cls_bpf_set_parms(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
507 if (ret < 0)
508 goto errout_idr;
509
510 ret = cls_bpf_offload(tp, prog, oldprog);
511 if (ret) {
512 if (!oldprog)
513 idr_remove_ext(&head->handle_idr, prog->handle);
514 __cls_bpf_delete_prog(prog);
515 return ret;
516 }
517
518 if (!tc_in_hw(prog->gen_flags))
519 prog->gen_flags |= TCA_CLS_FLAGS_NOT_IN_HW;
520
521 if (oldprog) {
522 idr_replace_ext(&head->handle_idr, prog, handle);
523 list_replace_rcu(&oldprog->link, &prog->link);
524 tcf_unbind_filter(tp, &oldprog->res);
525 call_rcu(&oldprog->rcu, cls_bpf_delete_prog_rcu);
526 } else {
527 list_add_rcu(&prog->link, &head->plist);
528 }
529
530 *arg = prog;
531 return 0;
532
533 errout_idr:
534 if (!oldprog)
535 idr_remove_ext(&head->handle_idr, prog->handle);
536 errout:
537 tcf_exts_destroy(&prog->exts);
538 kfree(prog);
539 return ret;
540 }
541
542 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
543 struct sk_buff *skb)
544 {
545 struct nlattr *nla;
546
547 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
548 return -EMSGSIZE;
549
550 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
551 sizeof(struct sock_filter));
552 if (nla == NULL)
553 return -EMSGSIZE;
554
555 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
556
557 return 0;
558 }
559
560 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
561 struct sk_buff *skb)
562 {
563 struct nlattr *nla;
564
565 if (prog->bpf_name &&
566 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
567 return -EMSGSIZE;
568
569 if (nla_put_u32(skb, TCA_BPF_ID, prog->filter->aux->id))
570 return -EMSGSIZE;
571
572 nla = nla_reserve(skb, TCA_BPF_TAG, sizeof(prog->filter->tag));
573 if (nla == NULL)
574 return -EMSGSIZE;
575
576 memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
577
578 return 0;
579 }
580
581 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, void *fh,
582 struct sk_buff *skb, struct tcmsg *tm)
583 {
584 struct cls_bpf_prog *prog = fh;
585 struct nlattr *nest;
586 u32 bpf_flags = 0;
587 int ret;
588
589 if (prog == NULL)
590 return skb->len;
591
592 tm->tcm_handle = prog->handle;
593
594 cls_bpf_offload_update_stats(tp, prog);
595
596 nest = nla_nest_start(skb, TCA_OPTIONS);
597 if (nest == NULL)
598 goto nla_put_failure;
599
600 if (prog->res.classid &&
601 nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
602 goto nla_put_failure;
603
604 if (cls_bpf_is_ebpf(prog))
605 ret = cls_bpf_dump_ebpf_info(prog, skb);
606 else
607 ret = cls_bpf_dump_bpf_info(prog, skb);
608 if (ret)
609 goto nla_put_failure;
610
611 if (tcf_exts_dump(skb, &prog->exts) < 0)
612 goto nla_put_failure;
613
614 if (prog->exts_integrated)
615 bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
616 if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
617 goto nla_put_failure;
618 if (prog->gen_flags &&
619 nla_put_u32(skb, TCA_BPF_FLAGS_GEN, prog->gen_flags))
620 goto nla_put_failure;
621
622 nla_nest_end(skb, nest);
623
624 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
625 goto nla_put_failure;
626
627 return skb->len;
628
629 nla_put_failure:
630 nla_nest_cancel(skb, nest);
631 return -1;
632 }
633
634 static void cls_bpf_bind_class(void *fh, u32 classid, unsigned long cl)
635 {
636 struct cls_bpf_prog *prog = fh;
637
638 if (prog && prog->res.classid == classid)
639 prog->res.class = cl;
640 }
641
642 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
643 {
644 struct cls_bpf_head *head = rtnl_dereference(tp->root);
645 struct cls_bpf_prog *prog;
646
647 list_for_each_entry(prog, &head->plist, link) {
648 if (arg->count < arg->skip)
649 goto skip;
650 if (arg->fn(tp, prog, arg) < 0) {
651 arg->stop = 1;
652 break;
653 }
654 skip:
655 arg->count++;
656 }
657 }
658
659 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
660 .kind = "bpf",
661 .owner = THIS_MODULE,
662 .classify = cls_bpf_classify,
663 .init = cls_bpf_init,
664 .destroy = cls_bpf_destroy,
665 .get = cls_bpf_get,
666 .change = cls_bpf_change,
667 .delete = cls_bpf_delete,
668 .walk = cls_bpf_walk,
669 .dump = cls_bpf_dump,
670 .bind_class = cls_bpf_bind_class,
671 };
672
673 static int __init cls_bpf_init_mod(void)
674 {
675 return register_tcf_proto_ops(&cls_bpf_ops);
676 }
677
678 static void __exit cls_bpf_exit_mod(void)
679 {
680 unregister_tcf_proto_ops(&cls_bpf_ops);
681 }
682
683 module_init(cls_bpf_init_mod);
684 module_exit(cls_bpf_exit_mod);