Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / net / sched / cls_bpf.c
blobb07c1fa8bc0dd8ccf2a207e50aaeeaed4110b7a8
1 /*
2 * Berkeley Packet Filter based traffic classifier
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.
8 * (C) 2013 Daniel Borkmann <dborkman@redhat.com>
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.
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>
22 #include <net/rtnetlink.h>
23 #include <net/pkt_cls.h>
24 #include <net/sock.h>
26 MODULE_LICENSE("GPL");
27 MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
28 MODULE_DESCRIPTION("TC BPF based classifier");
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)
34 struct cls_bpf_head {
35 struct list_head plist;
36 struct idr handle_idr;
37 struct rcu_head rcu;
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 u32 gen_flags;
46 struct tcf_exts exts;
47 u32 handle;
48 u16 bpf_num_ops;
49 struct sock_filter *bpf_ops;
50 const char *bpf_name;
51 struct tcf_proto *tp;
52 union {
53 struct work_struct work;
54 struct rcu_head rcu;
58 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
59 [TCA_BPF_CLASSID] = { .type = NLA_U32 },
60 [TCA_BPF_FLAGS] = { .type = NLA_U32 },
61 [TCA_BPF_FLAGS_GEN] = { .type = NLA_U32 },
62 [TCA_BPF_FD] = { .type = NLA_U32 },
63 [TCA_BPF_NAME] = { .type = NLA_NUL_STRING,
64 .len = CLS_BPF_NAME_LEN },
65 [TCA_BPF_OPS_LEN] = { .type = NLA_U16 },
66 [TCA_BPF_OPS] = { .type = NLA_BINARY,
67 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
70 static int cls_bpf_exec_opcode(int code)
72 switch (code) {
73 case TC_ACT_OK:
74 case TC_ACT_SHOT:
75 case TC_ACT_STOLEN:
76 case TC_ACT_TRAP:
77 case TC_ACT_REDIRECT:
78 case TC_ACT_UNSPEC:
79 return code;
80 default:
81 return TC_ACT_UNSPEC;
85 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
86 struct tcf_result *res)
88 struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
89 bool at_ingress = skb_at_tc_ingress(skb);
90 struct cls_bpf_prog *prog;
91 int ret = -1;
93 /* Needed here for accessing maps. */
94 rcu_read_lock();
95 list_for_each_entry_rcu(prog, &head->plist, link) {
96 int filter_res;
98 qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
100 if (tc_skip_sw(prog->gen_flags)) {
101 filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
102 } else if (at_ingress) {
103 /* It is safe to push/pull even if skb_shared() */
104 __skb_push(skb, skb->mac_len);
105 bpf_compute_data_pointers(skb);
106 filter_res = BPF_PROG_RUN(prog->filter, skb);
107 __skb_pull(skb, skb->mac_len);
108 } else {
109 bpf_compute_data_pointers(skb);
110 filter_res = BPF_PROG_RUN(prog->filter, skb);
113 if (prog->exts_integrated) {
114 res->class = 0;
115 res->classid = TC_H_MAJ(prog->res.classid) |
116 qdisc_skb_cb(skb)->tc_classid;
118 ret = cls_bpf_exec_opcode(filter_res);
119 if (ret == TC_ACT_UNSPEC)
120 continue;
121 break;
124 if (filter_res == 0)
125 continue;
126 if (filter_res != -1) {
127 res->class = 0;
128 res->classid = filter_res;
129 } else {
130 *res = prog->res;
133 ret = tcf_exts_exec(skb, &prog->exts, res);
134 if (ret < 0)
135 continue;
137 break;
139 rcu_read_unlock();
141 return ret;
144 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
146 return !prog->bpf_ops;
149 static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
150 struct cls_bpf_prog *oldprog,
151 struct netlink_ext_ack *extack)
153 struct tcf_block *block = tp->chain->block;
154 struct tc_cls_bpf_offload cls_bpf = {};
155 struct cls_bpf_prog *obj;
156 bool skip_sw;
157 int err;
159 skip_sw = prog && tc_skip_sw(prog->gen_flags);
160 obj = prog ?: oldprog;
162 tc_cls_common_offload_init(&cls_bpf.common, tp, obj->gen_flags,
163 extack);
164 cls_bpf.command = TC_CLSBPF_OFFLOAD;
165 cls_bpf.exts = &obj->exts;
166 cls_bpf.prog = prog ? prog->filter : NULL;
167 cls_bpf.oldprog = oldprog ? oldprog->filter : NULL;
168 cls_bpf.name = obj->bpf_name;
169 cls_bpf.exts_integrated = obj->exts_integrated;
171 if (oldprog)
172 tcf_block_offload_dec(block, &oldprog->gen_flags);
174 err = tc_setup_cb_call(block, NULL, TC_SETUP_CLSBPF, &cls_bpf, skip_sw);
175 if (prog) {
176 if (err < 0) {
177 cls_bpf_offload_cmd(tp, oldprog, prog, extack);
178 return err;
179 } else if (err > 0) {
180 tcf_block_offload_inc(block, &prog->gen_flags);
184 if (prog && skip_sw && !(prog->gen_flags & TCA_CLS_FLAGS_IN_HW))
185 return -EINVAL;
187 return 0;
190 static u32 cls_bpf_flags(u32 flags)
192 return flags & CLS_BPF_SUPPORTED_GEN_FLAGS;
195 static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
196 struct cls_bpf_prog *oldprog,
197 struct netlink_ext_ack *extack)
199 if (prog && oldprog &&
200 cls_bpf_flags(prog->gen_flags) !=
201 cls_bpf_flags(oldprog->gen_flags))
202 return -EINVAL;
204 if (prog && tc_skip_hw(prog->gen_flags))
205 prog = NULL;
206 if (oldprog && tc_skip_hw(oldprog->gen_flags))
207 oldprog = NULL;
208 if (!prog && !oldprog)
209 return 0;
211 return cls_bpf_offload_cmd(tp, prog, oldprog, extack);
214 static void cls_bpf_stop_offload(struct tcf_proto *tp,
215 struct cls_bpf_prog *prog,
216 struct netlink_ext_ack *extack)
218 int err;
220 err = cls_bpf_offload_cmd(tp, NULL, prog, extack);
221 if (err)
222 pr_err("Stopping hardware offload failed: %d\n", err);
225 static void cls_bpf_offload_update_stats(struct tcf_proto *tp,
226 struct cls_bpf_prog *prog)
228 struct tcf_block *block = tp->chain->block;
229 struct tc_cls_bpf_offload cls_bpf = {};
231 tc_cls_common_offload_init(&cls_bpf.common, tp, prog->gen_flags, NULL);
232 cls_bpf.command = TC_CLSBPF_STATS;
233 cls_bpf.exts = &prog->exts;
234 cls_bpf.prog = prog->filter;
235 cls_bpf.name = prog->bpf_name;
236 cls_bpf.exts_integrated = prog->exts_integrated;
238 tc_setup_cb_call(block, NULL, TC_SETUP_CLSBPF, &cls_bpf, false);
241 static int cls_bpf_init(struct tcf_proto *tp)
243 struct cls_bpf_head *head;
245 head = kzalloc(sizeof(*head), GFP_KERNEL);
246 if (head == NULL)
247 return -ENOBUFS;
249 INIT_LIST_HEAD_RCU(&head->plist);
250 idr_init(&head->handle_idr);
251 rcu_assign_pointer(tp->root, head);
253 return 0;
256 static void cls_bpf_free_parms(struct cls_bpf_prog *prog)
258 if (cls_bpf_is_ebpf(prog))
259 bpf_prog_put(prog->filter);
260 else
261 bpf_prog_destroy(prog->filter);
263 kfree(prog->bpf_name);
264 kfree(prog->bpf_ops);
267 static void __cls_bpf_delete_prog(struct cls_bpf_prog *prog)
269 tcf_exts_destroy(&prog->exts);
270 tcf_exts_put_net(&prog->exts);
272 cls_bpf_free_parms(prog);
273 kfree(prog);
276 static void cls_bpf_delete_prog_work(struct work_struct *work)
278 struct cls_bpf_prog *prog = container_of(work, struct cls_bpf_prog, work);
280 rtnl_lock();
281 __cls_bpf_delete_prog(prog);
282 rtnl_unlock();
285 static void cls_bpf_delete_prog_rcu(struct rcu_head *rcu)
287 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
289 INIT_WORK(&prog->work, cls_bpf_delete_prog_work);
290 tcf_queue_work(&prog->work);
293 static void __cls_bpf_delete(struct tcf_proto *tp, struct cls_bpf_prog *prog,
294 struct netlink_ext_ack *extack)
296 struct cls_bpf_head *head = rtnl_dereference(tp->root);
298 idr_remove(&head->handle_idr, prog->handle);
299 cls_bpf_stop_offload(tp, prog, extack);
300 list_del_rcu(&prog->link);
301 tcf_unbind_filter(tp, &prog->res);
302 if (tcf_exts_get_net(&prog->exts))
303 call_rcu(&prog->rcu, cls_bpf_delete_prog_rcu);
304 else
305 __cls_bpf_delete_prog(prog);
308 static int cls_bpf_delete(struct tcf_proto *tp, void *arg, bool *last,
309 struct netlink_ext_ack *extack)
311 struct cls_bpf_head *head = rtnl_dereference(tp->root);
313 __cls_bpf_delete(tp, arg, extack);
314 *last = list_empty(&head->plist);
315 return 0;
318 static void cls_bpf_destroy(struct tcf_proto *tp,
319 struct netlink_ext_ack *extack)
321 struct cls_bpf_head *head = rtnl_dereference(tp->root);
322 struct cls_bpf_prog *prog, *tmp;
324 list_for_each_entry_safe(prog, tmp, &head->plist, link)
325 __cls_bpf_delete(tp, prog, extack);
327 idr_destroy(&head->handle_idr);
328 kfree_rcu(head, rcu);
331 static void *cls_bpf_get(struct tcf_proto *tp, u32 handle)
333 struct cls_bpf_head *head = rtnl_dereference(tp->root);
334 struct cls_bpf_prog *prog;
336 list_for_each_entry(prog, &head->plist, link) {
337 if (prog->handle == handle)
338 return prog;
341 return NULL;
344 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
346 struct sock_filter *bpf_ops;
347 struct sock_fprog_kern fprog_tmp;
348 struct bpf_prog *fp;
349 u16 bpf_size, bpf_num_ops;
350 int ret;
352 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
353 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
354 return -EINVAL;
356 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
357 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
358 return -EINVAL;
360 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
361 if (bpf_ops == NULL)
362 return -ENOMEM;
364 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
366 fprog_tmp.len = bpf_num_ops;
367 fprog_tmp.filter = bpf_ops;
369 ret = bpf_prog_create(&fp, &fprog_tmp);
370 if (ret < 0) {
371 kfree(bpf_ops);
372 return ret;
375 prog->bpf_ops = bpf_ops;
376 prog->bpf_num_ops = bpf_num_ops;
377 prog->bpf_name = NULL;
378 prog->filter = fp;
380 return 0;
383 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
384 u32 gen_flags, const struct tcf_proto *tp)
386 struct bpf_prog *fp;
387 char *name = NULL;
388 bool skip_sw;
389 u32 bpf_fd;
391 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
392 skip_sw = gen_flags & TCA_CLS_FLAGS_SKIP_SW;
394 fp = bpf_prog_get_type_dev(bpf_fd, BPF_PROG_TYPE_SCHED_CLS, skip_sw);
395 if (IS_ERR(fp))
396 return PTR_ERR(fp);
398 if (tb[TCA_BPF_NAME]) {
399 name = nla_memdup(tb[TCA_BPF_NAME], GFP_KERNEL);
400 if (!name) {
401 bpf_prog_put(fp);
402 return -ENOMEM;
406 prog->bpf_ops = NULL;
407 prog->bpf_name = name;
408 prog->filter = fp;
410 if (fp->dst_needed)
411 tcf_block_netif_keep_dst(tp->chain->block);
413 return 0;
416 static int cls_bpf_set_parms(struct net *net, struct tcf_proto *tp,
417 struct cls_bpf_prog *prog, unsigned long base,
418 struct nlattr **tb, struct nlattr *est, bool ovr,
419 struct netlink_ext_ack *extack)
421 bool is_bpf, is_ebpf, have_exts = false;
422 u32 gen_flags = 0;
423 int ret;
425 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
426 is_ebpf = tb[TCA_BPF_FD];
427 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
428 return -EINVAL;
430 ret = tcf_exts_validate(net, tp, tb, est, &prog->exts, ovr, extack);
431 if (ret < 0)
432 return ret;
434 if (tb[TCA_BPF_FLAGS]) {
435 u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
437 if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT)
438 return -EINVAL;
440 have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
442 if (tb[TCA_BPF_FLAGS_GEN]) {
443 gen_flags = nla_get_u32(tb[TCA_BPF_FLAGS_GEN]);
444 if (gen_flags & ~CLS_BPF_SUPPORTED_GEN_FLAGS ||
445 !tc_flags_valid(gen_flags))
446 return -EINVAL;
449 prog->exts_integrated = have_exts;
450 prog->gen_flags = gen_flags;
452 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
453 cls_bpf_prog_from_efd(tb, prog, gen_flags, tp);
454 if (ret < 0)
455 return ret;
457 if (tb[TCA_BPF_CLASSID]) {
458 prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
459 tcf_bind_filter(tp, &prog->res, base);
462 return 0;
465 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
466 struct tcf_proto *tp, unsigned long base,
467 u32 handle, struct nlattr **tca,
468 void **arg, bool ovr, struct netlink_ext_ack *extack)
470 struct cls_bpf_head *head = rtnl_dereference(tp->root);
471 struct cls_bpf_prog *oldprog = *arg;
472 struct nlattr *tb[TCA_BPF_MAX + 1];
473 struct cls_bpf_prog *prog;
474 int ret;
476 if (tca[TCA_OPTIONS] == NULL)
477 return -EINVAL;
479 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy,
480 NULL);
481 if (ret < 0)
482 return ret;
484 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
485 if (!prog)
486 return -ENOBUFS;
488 ret = tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
489 if (ret < 0)
490 goto errout;
492 if (oldprog) {
493 if (handle && oldprog->handle != handle) {
494 ret = -EINVAL;
495 goto errout;
499 if (handle == 0) {
500 handle = 1;
501 ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
502 INT_MAX, GFP_KERNEL);
503 } else if (!oldprog) {
504 ret = idr_alloc_u32(&head->handle_idr, prog, &handle,
505 handle, GFP_KERNEL);
508 if (ret)
509 goto errout;
510 prog->handle = handle;
512 ret = cls_bpf_set_parms(net, tp, prog, base, tb, tca[TCA_RATE], ovr,
513 extack);
514 if (ret < 0)
515 goto errout_idr;
517 ret = cls_bpf_offload(tp, prog, oldprog, extack);
518 if (ret)
519 goto errout_parms;
521 if (!tc_in_hw(prog->gen_flags))
522 prog->gen_flags |= TCA_CLS_FLAGS_NOT_IN_HW;
524 if (oldprog) {
525 idr_replace(&head->handle_idr, prog, handle);
526 list_replace_rcu(&oldprog->link, &prog->link);
527 tcf_unbind_filter(tp, &oldprog->res);
528 tcf_exts_get_net(&oldprog->exts);
529 call_rcu(&oldprog->rcu, cls_bpf_delete_prog_rcu);
530 } else {
531 list_add_rcu(&prog->link, &head->plist);
534 *arg = prog;
535 return 0;
537 errout_parms:
538 cls_bpf_free_parms(prog);
539 errout_idr:
540 if (!oldprog)
541 idr_remove(&head->handle_idr, prog->handle);
542 errout:
543 tcf_exts_destroy(&prog->exts);
544 kfree(prog);
545 return ret;
548 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
549 struct sk_buff *skb)
551 struct nlattr *nla;
553 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
554 return -EMSGSIZE;
556 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
557 sizeof(struct sock_filter));
558 if (nla == NULL)
559 return -EMSGSIZE;
561 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
563 return 0;
566 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
567 struct sk_buff *skb)
569 struct nlattr *nla;
571 if (prog->bpf_name &&
572 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
573 return -EMSGSIZE;
575 if (nla_put_u32(skb, TCA_BPF_ID, prog->filter->aux->id))
576 return -EMSGSIZE;
578 nla = nla_reserve(skb, TCA_BPF_TAG, sizeof(prog->filter->tag));
579 if (nla == NULL)
580 return -EMSGSIZE;
582 memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
584 return 0;
587 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, void *fh,
588 struct sk_buff *skb, struct tcmsg *tm)
590 struct cls_bpf_prog *prog = fh;
591 struct nlattr *nest;
592 u32 bpf_flags = 0;
593 int ret;
595 if (prog == NULL)
596 return skb->len;
598 tm->tcm_handle = prog->handle;
600 cls_bpf_offload_update_stats(tp, prog);
602 nest = nla_nest_start(skb, TCA_OPTIONS);
603 if (nest == NULL)
604 goto nla_put_failure;
606 if (prog->res.classid &&
607 nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
608 goto nla_put_failure;
610 if (cls_bpf_is_ebpf(prog))
611 ret = cls_bpf_dump_ebpf_info(prog, skb);
612 else
613 ret = cls_bpf_dump_bpf_info(prog, skb);
614 if (ret)
615 goto nla_put_failure;
617 if (tcf_exts_dump(skb, &prog->exts) < 0)
618 goto nla_put_failure;
620 if (prog->exts_integrated)
621 bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
622 if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
623 goto nla_put_failure;
624 if (prog->gen_flags &&
625 nla_put_u32(skb, TCA_BPF_FLAGS_GEN, prog->gen_flags))
626 goto nla_put_failure;
628 nla_nest_end(skb, nest);
630 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
631 goto nla_put_failure;
633 return skb->len;
635 nla_put_failure:
636 nla_nest_cancel(skb, nest);
637 return -1;
640 static void cls_bpf_bind_class(void *fh, u32 classid, unsigned long cl)
642 struct cls_bpf_prog *prog = fh;
644 if (prog && prog->res.classid == classid)
645 prog->res.class = cl;
648 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
650 struct cls_bpf_head *head = rtnl_dereference(tp->root);
651 struct cls_bpf_prog *prog;
653 list_for_each_entry(prog, &head->plist, link) {
654 if (arg->count < arg->skip)
655 goto skip;
656 if (arg->fn(tp, prog, arg) < 0) {
657 arg->stop = 1;
658 break;
660 skip:
661 arg->count++;
665 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
666 .kind = "bpf",
667 .owner = THIS_MODULE,
668 .classify = cls_bpf_classify,
669 .init = cls_bpf_init,
670 .destroy = cls_bpf_destroy,
671 .get = cls_bpf_get,
672 .change = cls_bpf_change,
673 .delete = cls_bpf_delete,
674 .walk = cls_bpf_walk,
675 .dump = cls_bpf_dump,
676 .bind_class = cls_bpf_bind_class,
679 static int __init cls_bpf_init_mod(void)
681 return register_tcf_proto_ops(&cls_bpf_ops);
684 static void __exit cls_bpf_exit_mod(void)
686 unregister_tcf_proto_ops(&cls_bpf_ops);
689 module_init(cls_bpf_init_mod);
690 module_exit(cls_bpf_exit_mod);