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
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>
21 #include <net/rtnetlink.h>
22 #include <net/pkt_cls.h>
25 MODULE_LICENSE("GPL");
26 MODULE_AUTHOR("Daniel Borkmann <dborkman@redhat.com>");
27 MODULE_DESCRIPTION("TC BPF based classifier");
29 #define CLS_BPF_NAME_LEN 256
32 struct list_head plist;
38 struct bpf_prog *filter;
39 struct list_head link;
40 struct tcf_result res;
48 struct sock_filter *bpf_ops;
54 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
55 [TCA_BPF_CLASSID] = { .type = NLA_U32 },
56 [TCA_BPF_FLAGS] = { .type = NLA_U32 },
57 [TCA_BPF_FD] = { .type = NLA_U32 },
58 [TCA_BPF_NAME] = { .type = NLA_NUL_STRING, .len = CLS_BPF_NAME_LEN },
59 [TCA_BPF_OPS_LEN] = { .type = NLA_U16 },
60 [TCA_BPF_OPS] = { .type = NLA_BINARY,
61 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
64 static int cls_bpf_exec_opcode(int code)
78 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
79 struct tcf_result *res)
81 struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
82 struct cls_bpf_prog *prog;
83 #ifdef CONFIG_NET_CLS_ACT
84 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
86 bool at_ingress = false;
90 if (unlikely(!skb_mac_header_was_set(skb)))
93 /* Needed here for accessing maps. */
95 list_for_each_entry_rcu(prog, &head->plist, link) {
98 qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
101 /* It is safe to push/pull even if skb_shared() */
102 __skb_push(skb, skb->mac_len);
103 filter_res = BPF_PROG_RUN(prog->filter, skb);
104 __skb_pull(skb, skb->mac_len);
106 filter_res = BPF_PROG_RUN(prog->filter, skb);
109 if (prog->exts_integrated) {
110 res->class = prog->res.class;
111 res->classid = qdisc_skb_cb(skb)->tc_classid;
113 ret = cls_bpf_exec_opcode(filter_res);
114 if (ret == TC_ACT_UNSPEC)
123 if (filter_res != -1)
124 res->classid = filter_res;
126 ret = tcf_exts_exec(skb, &prog->exts, res);
137 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
139 return !prog->bpf_ops;
142 static int cls_bpf_init(struct tcf_proto *tp)
144 struct cls_bpf_head *head;
146 head = kzalloc(sizeof(*head), GFP_KERNEL);
150 INIT_LIST_HEAD_RCU(&head->plist);
151 rcu_assign_pointer(tp->root, head);
156 static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
158 tcf_exts_destroy(&prog->exts);
160 if (cls_bpf_is_ebpf(prog))
161 bpf_prog_put(prog->filter);
163 bpf_prog_destroy(prog->filter);
165 kfree(prog->bpf_name);
166 kfree(prog->bpf_ops);
170 static void __cls_bpf_delete_prog(struct rcu_head *rcu)
172 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
174 cls_bpf_delete_prog(prog->tp, prog);
177 static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
179 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
181 list_del_rcu(&prog->link);
182 tcf_unbind_filter(tp, &prog->res);
183 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
188 static bool cls_bpf_destroy(struct tcf_proto *tp, bool force)
190 struct cls_bpf_head *head = rtnl_dereference(tp->root);
191 struct cls_bpf_prog *prog, *tmp;
193 if (!force && !list_empty(&head->plist))
196 list_for_each_entry_safe(prog, tmp, &head->plist, link) {
197 list_del_rcu(&prog->link);
198 tcf_unbind_filter(tp, &prog->res);
199 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
202 RCU_INIT_POINTER(tp->root, NULL);
203 kfree_rcu(head, rcu);
207 static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
209 struct cls_bpf_head *head = rtnl_dereference(tp->root);
210 struct cls_bpf_prog *prog;
211 unsigned long ret = 0UL;
216 list_for_each_entry(prog, &head->plist, link) {
217 if (prog->handle == handle) {
218 ret = (unsigned long) prog;
226 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
228 struct sock_filter *bpf_ops;
229 struct sock_fprog_kern fprog_tmp;
231 u16 bpf_size, bpf_num_ops;
234 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
235 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
238 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
239 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
242 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
246 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
248 fprog_tmp.len = bpf_num_ops;
249 fprog_tmp.filter = bpf_ops;
251 ret = bpf_prog_create(&fp, &fprog_tmp);
257 prog->bpf_ops = bpf_ops;
258 prog->bpf_num_ops = bpf_num_ops;
259 prog->bpf_name = NULL;
265 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
266 const struct tcf_proto *tp)
272 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
274 fp = bpf_prog_get(bpf_fd);
278 if (fp->type != BPF_PROG_TYPE_SCHED_CLS) {
283 if (tb[TCA_BPF_NAME]) {
284 name = kmemdup(nla_data(tb[TCA_BPF_NAME]),
285 nla_len(tb[TCA_BPF_NAME]),
293 prog->bpf_ops = NULL;
294 prog->bpf_fd = bpf_fd;
295 prog->bpf_name = name;
299 netif_keep_dst(qdisc_dev(tp->q));
304 static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
305 struct cls_bpf_prog *prog,
306 unsigned long base, struct nlattr **tb,
307 struct nlattr *est, bool ovr)
309 bool is_bpf, is_ebpf, have_exts = false;
310 struct tcf_exts exts;
313 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
314 is_ebpf = tb[TCA_BPF_FD];
315 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
318 tcf_exts_init(&exts, TCA_BPF_ACT, TCA_BPF_POLICE);
319 ret = tcf_exts_validate(net, tp, tb, est, &exts, ovr);
323 if (tb[TCA_BPF_FLAGS]) {
324 u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
326 if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT) {
327 tcf_exts_destroy(&exts);
331 have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
334 prog->exts_integrated = have_exts;
336 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
337 cls_bpf_prog_from_efd(tb, prog, tp);
339 tcf_exts_destroy(&exts);
343 if (tb[TCA_BPF_CLASSID]) {
344 prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
345 tcf_bind_filter(tp, &prog->res, base);
348 tcf_exts_change(tp, &prog->exts, &exts);
352 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
353 struct cls_bpf_head *head)
355 unsigned int i = 0x80000000;
359 if (++head->hgen == 0x7FFFFFFF)
361 } while (--i > 0 && cls_bpf_get(tp, head->hgen));
363 if (unlikely(i == 0)) {
364 pr_err("Insufficient number of handles\n");
373 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
374 struct tcf_proto *tp, unsigned long base,
375 u32 handle, struct nlattr **tca,
376 unsigned long *arg, bool ovr)
378 struct cls_bpf_head *head = rtnl_dereference(tp->root);
379 struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
380 struct nlattr *tb[TCA_BPF_MAX + 1];
381 struct cls_bpf_prog *prog;
384 if (tca[TCA_OPTIONS] == NULL)
387 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
391 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
395 tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
398 if (handle && oldprog->handle != handle) {
405 prog->handle = cls_bpf_grab_new_handle(tp, head);
407 prog->handle = handle;
408 if (prog->handle == 0) {
413 ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
418 list_replace_rcu(&oldprog->link, &prog->link);
419 tcf_unbind_filter(tp, &oldprog->res);
420 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
422 list_add_rcu(&prog->link, &head->plist);
425 *arg = (unsigned long) prog;
433 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
438 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
441 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
442 sizeof(struct sock_filter));
446 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
451 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
454 if (nla_put_u32(skb, TCA_BPF_FD, prog->bpf_fd))
457 if (prog->bpf_name &&
458 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
464 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
465 struct sk_buff *skb, struct tcmsg *tm)
467 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
475 tm->tcm_handle = prog->handle;
477 nest = nla_nest_start(skb, TCA_OPTIONS);
479 goto nla_put_failure;
481 if (prog->res.classid &&
482 nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
483 goto nla_put_failure;
485 if (cls_bpf_is_ebpf(prog))
486 ret = cls_bpf_dump_ebpf_info(prog, skb);
488 ret = cls_bpf_dump_bpf_info(prog, skb);
490 goto nla_put_failure;
492 if (tcf_exts_dump(skb, &prog->exts) < 0)
493 goto nla_put_failure;
495 if (prog->exts_integrated)
496 bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
497 if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
498 goto nla_put_failure;
500 nla_nest_end(skb, nest);
502 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
503 goto nla_put_failure;
508 nla_nest_cancel(skb, nest);
512 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
514 struct cls_bpf_head *head = rtnl_dereference(tp->root);
515 struct cls_bpf_prog *prog;
517 list_for_each_entry(prog, &head->plist, link) {
518 if (arg->count < arg->skip)
520 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
529 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
531 .owner = THIS_MODULE,
532 .classify = cls_bpf_classify,
533 .init = cls_bpf_init,
534 .destroy = cls_bpf_destroy,
536 .change = cls_bpf_change,
537 .delete = cls_bpf_delete,
538 .walk = cls_bpf_walk,
539 .dump = cls_bpf_dump,
542 static int __init cls_bpf_init_mod(void)
544 return register_tcf_proto_ops(&cls_bpf_ops);
547 static void __exit cls_bpf_exit_mod(void)
549 unregister_tcf_proto_ops(&cls_bpf_ops);
552 module_init(cls_bpf_init_mod);
553 module_exit(cls_bpf_exit_mod);