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)
68 case TC_ACT_RECLASSIFY:
81 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
82 struct tcf_result *res)
84 struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
85 struct cls_bpf_prog *prog;
86 #ifdef CONFIG_NET_CLS_ACT
87 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
89 bool at_ingress = false;
93 if (unlikely(!skb_mac_header_was_set(skb)))
96 /* Needed here for accessing maps. */
98 list_for_each_entry_rcu(prog, &head->plist, link) {
101 qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
104 /* It is safe to push/pull even if skb_shared() */
105 __skb_push(skb, skb->mac_len);
106 filter_res = BPF_PROG_RUN(prog->filter, skb);
107 __skb_pull(skb, skb->mac_len);
109 filter_res = BPF_PROG_RUN(prog->filter, skb);
112 if (prog->exts_integrated) {
113 res->class = prog->res.class;
114 res->classid = qdisc_skb_cb(skb)->tc_classid;
116 ret = cls_bpf_exec_opcode(filter_res);
117 if (ret == TC_ACT_UNSPEC)
126 if (filter_res != -1)
127 res->classid = filter_res;
129 ret = tcf_exts_exec(skb, &prog->exts, res);
140 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
142 return !prog->bpf_ops;
145 static int cls_bpf_init(struct tcf_proto *tp)
147 struct cls_bpf_head *head;
149 head = kzalloc(sizeof(*head), GFP_KERNEL);
153 INIT_LIST_HEAD_RCU(&head->plist);
154 rcu_assign_pointer(tp->root, head);
159 static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
161 tcf_exts_destroy(&prog->exts);
163 if (cls_bpf_is_ebpf(prog))
164 bpf_prog_put(prog->filter);
166 bpf_prog_destroy(prog->filter);
168 kfree(prog->bpf_name);
169 kfree(prog->bpf_ops);
173 static void __cls_bpf_delete_prog(struct rcu_head *rcu)
175 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
177 cls_bpf_delete_prog(prog->tp, prog);
180 static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
182 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
184 list_del_rcu(&prog->link);
185 tcf_unbind_filter(tp, &prog->res);
186 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
191 static bool cls_bpf_destroy(struct tcf_proto *tp, bool force)
193 struct cls_bpf_head *head = rtnl_dereference(tp->root);
194 struct cls_bpf_prog *prog, *tmp;
196 if (!force && !list_empty(&head->plist))
199 list_for_each_entry_safe(prog, tmp, &head->plist, link) {
200 list_del_rcu(&prog->link);
201 tcf_unbind_filter(tp, &prog->res);
202 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
205 RCU_INIT_POINTER(tp->root, NULL);
206 kfree_rcu(head, rcu);
210 static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
212 struct cls_bpf_head *head = rtnl_dereference(tp->root);
213 struct cls_bpf_prog *prog;
214 unsigned long ret = 0UL;
219 list_for_each_entry(prog, &head->plist, link) {
220 if (prog->handle == handle) {
221 ret = (unsigned long) prog;
229 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
231 struct sock_filter *bpf_ops;
232 struct sock_fprog_kern fprog_tmp;
234 u16 bpf_size, bpf_num_ops;
237 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
238 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
241 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
242 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
245 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
249 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
251 fprog_tmp.len = bpf_num_ops;
252 fprog_tmp.filter = bpf_ops;
254 ret = bpf_prog_create(&fp, &fprog_tmp);
260 prog->bpf_ops = bpf_ops;
261 prog->bpf_num_ops = bpf_num_ops;
262 prog->bpf_name = NULL;
268 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
269 const struct tcf_proto *tp)
275 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
277 fp = bpf_prog_get(bpf_fd);
281 if (fp->type != BPF_PROG_TYPE_SCHED_CLS) {
286 if (tb[TCA_BPF_NAME]) {
287 name = kmemdup(nla_data(tb[TCA_BPF_NAME]),
288 nla_len(tb[TCA_BPF_NAME]),
296 prog->bpf_ops = NULL;
297 prog->bpf_fd = bpf_fd;
298 prog->bpf_name = name;
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;
314 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
315 is_ebpf = tb[TCA_BPF_FD];
317 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf) ||
318 !tb[TCA_BPF_CLASSID])
321 tcf_exts_init(&exts, TCA_BPF_ACT, TCA_BPF_POLICE);
322 ret = tcf_exts_validate(net, tp, tb, est, &exts, ovr);
326 classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
327 if (tb[TCA_BPF_FLAGS]) {
328 u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
330 if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT) {
331 tcf_exts_destroy(&exts);
335 have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
338 prog->res.classid = classid;
339 prog->exts_integrated = have_exts;
341 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
342 cls_bpf_prog_from_efd(tb, prog, tp);
344 tcf_exts_destroy(&exts);
348 tcf_bind_filter(tp, &prog->res, base);
349 tcf_exts_change(tp, &prog->exts, &exts);
354 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
355 struct cls_bpf_head *head)
357 unsigned int i = 0x80000000;
361 if (++head->hgen == 0x7FFFFFFF)
363 } while (--i > 0 && cls_bpf_get(tp, head->hgen));
365 if (unlikely(i == 0)) {
366 pr_err("Insufficient number of handles\n");
375 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
376 struct tcf_proto *tp, unsigned long base,
377 u32 handle, struct nlattr **tca,
378 unsigned long *arg, bool ovr)
380 struct cls_bpf_head *head = rtnl_dereference(tp->root);
381 struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
382 struct nlattr *tb[TCA_BPF_MAX + 1];
383 struct cls_bpf_prog *prog;
386 if (tca[TCA_OPTIONS] == NULL)
389 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
393 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
397 tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
400 if (handle && oldprog->handle != handle) {
407 prog->handle = cls_bpf_grab_new_handle(tp, head);
409 prog->handle = handle;
410 if (prog->handle == 0) {
415 ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
420 list_replace_rcu(&oldprog->link, &prog->link);
421 tcf_unbind_filter(tp, &oldprog->res);
422 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
424 list_add_rcu(&prog->link, &head->plist);
427 *arg = (unsigned long) prog;
435 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
440 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
443 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
444 sizeof(struct sock_filter));
448 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
453 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
456 if (nla_put_u32(skb, TCA_BPF_FD, prog->bpf_fd))
459 if (prog->bpf_name &&
460 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
466 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
467 struct sk_buff *skb, struct tcmsg *tm)
469 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
476 tm->tcm_handle = prog->handle;
478 nest = nla_nest_start(skb, TCA_OPTIONS);
480 goto nla_put_failure;
482 if (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 nla_nest_end(skb, nest);
497 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
498 goto nla_put_failure;
503 nla_nest_cancel(skb, nest);
507 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
509 struct cls_bpf_head *head = rtnl_dereference(tp->root);
510 struct cls_bpf_prog *prog;
512 list_for_each_entry(prog, &head->plist, link) {
513 if (arg->count < arg->skip)
515 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
524 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
526 .owner = THIS_MODULE,
527 .classify = cls_bpf_classify,
528 .init = cls_bpf_init,
529 .destroy = cls_bpf_destroy,
531 .change = cls_bpf_change,
532 .delete = cls_bpf_delete,
533 .walk = cls_bpf_walk,
534 .dump = cls_bpf_dump,
537 static int __init cls_bpf_init_mod(void)
539 return register_tcf_proto_ops(&cls_bpf_ops);
542 static void __exit cls_bpf_exit_mod(void)
544 unregister_tcf_proto_ops(&cls_bpf_ops);
547 module_init(cls_bpf_init_mod);
548 module_exit(cls_bpf_exit_mod);