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;
47 struct sock_filter *bpf_ops;
53 static const struct nla_policy bpf_policy[TCA_BPF_MAX + 1] = {
54 [TCA_BPF_CLASSID] = { .type = NLA_U32 },
55 [TCA_BPF_FD] = { .type = NLA_U32 },
56 [TCA_BPF_NAME] = { .type = NLA_NUL_STRING, .len = CLS_BPF_NAME_LEN },
57 [TCA_BPF_OPS_LEN] = { .type = NLA_U16 },
58 [TCA_BPF_OPS] = { .type = NLA_BINARY,
59 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
62 static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
63 struct tcf_result *res)
65 struct cls_bpf_head *head = rcu_dereference_bh(tp->root);
66 struct cls_bpf_prog *prog;
67 #ifdef CONFIG_NET_CLS_ACT
68 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
70 bool at_ingress = false;
74 if (unlikely(!skb_mac_header_was_set(skb)))
77 /* Needed here for accessing maps. */
79 list_for_each_entry_rcu(prog, &head->plist, link) {
83 /* It is safe to push/pull even if skb_shared() */
84 __skb_push(skb, skb->mac_len);
85 filter_res = BPF_PROG_RUN(prog->filter, skb);
86 __skb_pull(skb, skb->mac_len);
88 filter_res = BPF_PROG_RUN(prog->filter, skb);
96 res->classid = filter_res;
98 ret = tcf_exts_exec(skb, &prog->exts, res);
109 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
111 return !prog->bpf_ops;
114 static int cls_bpf_init(struct tcf_proto *tp)
116 struct cls_bpf_head *head;
118 head = kzalloc(sizeof(*head), GFP_KERNEL);
122 INIT_LIST_HEAD_RCU(&head->plist);
123 rcu_assign_pointer(tp->root, head);
128 static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
130 tcf_exts_destroy(&prog->exts);
132 if (cls_bpf_is_ebpf(prog))
133 bpf_prog_put(prog->filter);
135 bpf_prog_destroy(prog->filter);
137 kfree(prog->bpf_name);
138 kfree(prog->bpf_ops);
142 static void __cls_bpf_delete_prog(struct rcu_head *rcu)
144 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
146 cls_bpf_delete_prog(prog->tp, prog);
149 static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
151 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
153 list_del_rcu(&prog->link);
154 tcf_unbind_filter(tp, &prog->res);
155 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
160 static bool cls_bpf_destroy(struct tcf_proto *tp, bool force)
162 struct cls_bpf_head *head = rtnl_dereference(tp->root);
163 struct cls_bpf_prog *prog, *tmp;
165 if (!force && !list_empty(&head->plist))
168 list_for_each_entry_safe(prog, tmp, &head->plist, link) {
169 list_del_rcu(&prog->link);
170 tcf_unbind_filter(tp, &prog->res);
171 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
174 RCU_INIT_POINTER(tp->root, NULL);
175 kfree_rcu(head, rcu);
179 static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
181 struct cls_bpf_head *head = rtnl_dereference(tp->root);
182 struct cls_bpf_prog *prog;
183 unsigned long ret = 0UL;
188 list_for_each_entry(prog, &head->plist, link) {
189 if (prog->handle == handle) {
190 ret = (unsigned long) prog;
198 static int cls_bpf_prog_from_ops(struct nlattr **tb,
199 struct cls_bpf_prog *prog, u32 classid)
201 struct sock_filter *bpf_ops;
202 struct sock_fprog_kern fprog_tmp;
204 u16 bpf_size, bpf_num_ops;
207 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
208 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
211 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
212 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
215 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
219 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
221 fprog_tmp.len = bpf_num_ops;
222 fprog_tmp.filter = bpf_ops;
224 ret = bpf_prog_create(&fp, &fprog_tmp);
230 prog->bpf_ops = bpf_ops;
231 prog->bpf_num_ops = bpf_num_ops;
232 prog->bpf_name = NULL;
235 prog->res.classid = classid;
240 static int cls_bpf_prog_from_efd(struct nlattr **tb,
241 struct cls_bpf_prog *prog, u32 classid)
247 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
249 fp = bpf_prog_get(bpf_fd);
253 if (fp->type != BPF_PROG_TYPE_SCHED_CLS) {
258 if (tb[TCA_BPF_NAME]) {
259 name = kmemdup(nla_data(tb[TCA_BPF_NAME]),
260 nla_len(tb[TCA_BPF_NAME]),
268 prog->bpf_ops = NULL;
269 prog->bpf_fd = bpf_fd;
270 prog->bpf_name = name;
273 prog->res.classid = classid;
278 static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
279 struct cls_bpf_prog *prog,
280 unsigned long base, struct nlattr **tb,
281 struct nlattr *est, bool ovr)
283 struct tcf_exts exts;
284 bool is_bpf, is_ebpf;
288 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
289 is_ebpf = tb[TCA_BPF_FD];
291 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf) ||
292 !tb[TCA_BPF_CLASSID])
295 tcf_exts_init(&exts, TCA_BPF_ACT, TCA_BPF_POLICE);
296 ret = tcf_exts_validate(net, tp, tb, est, &exts, ovr);
300 classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
302 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog, classid) :
303 cls_bpf_prog_from_efd(tb, prog, classid);
305 tcf_exts_destroy(&exts);
309 tcf_bind_filter(tp, &prog->res, base);
310 tcf_exts_change(tp, &prog->exts, &exts);
315 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
316 struct cls_bpf_head *head)
318 unsigned int i = 0x80000000;
322 if (++head->hgen == 0x7FFFFFFF)
324 } while (--i > 0 && cls_bpf_get(tp, head->hgen));
326 if (unlikely(i == 0)) {
327 pr_err("Insufficient number of handles\n");
336 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
337 struct tcf_proto *tp, unsigned long base,
338 u32 handle, struct nlattr **tca,
339 unsigned long *arg, bool ovr)
341 struct cls_bpf_head *head = rtnl_dereference(tp->root);
342 struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
343 struct nlattr *tb[TCA_BPF_MAX + 1];
344 struct cls_bpf_prog *prog;
347 if (tca[TCA_OPTIONS] == NULL)
350 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
354 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
358 tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
361 if (handle && oldprog->handle != handle) {
368 prog->handle = cls_bpf_grab_new_handle(tp, head);
370 prog->handle = handle;
371 if (prog->handle == 0) {
376 ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
381 list_replace_rcu(&prog->link, &oldprog->link);
382 tcf_unbind_filter(tp, &oldprog->res);
383 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
385 list_add_rcu(&prog->link, &head->plist);
388 *arg = (unsigned long) prog;
396 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
401 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
404 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
405 sizeof(struct sock_filter));
409 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
414 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
417 if (nla_put_u32(skb, TCA_BPF_FD, prog->bpf_fd))
420 if (prog->bpf_name &&
421 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
427 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
428 struct sk_buff *skb, struct tcmsg *tm)
430 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
437 tm->tcm_handle = prog->handle;
439 nest = nla_nest_start(skb, TCA_OPTIONS);
441 goto nla_put_failure;
443 if (nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
444 goto nla_put_failure;
446 if (cls_bpf_is_ebpf(prog))
447 ret = cls_bpf_dump_ebpf_info(prog, skb);
449 ret = cls_bpf_dump_bpf_info(prog, skb);
451 goto nla_put_failure;
453 if (tcf_exts_dump(skb, &prog->exts) < 0)
454 goto nla_put_failure;
456 nla_nest_end(skb, nest);
458 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
459 goto nla_put_failure;
464 nla_nest_cancel(skb, nest);
468 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
470 struct cls_bpf_head *head = rtnl_dereference(tp->root);
471 struct cls_bpf_prog *prog;
473 list_for_each_entry(prog, &head->plist, link) {
474 if (arg->count < arg->skip)
476 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
485 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
487 .owner = THIS_MODULE,
488 .classify = cls_bpf_classify,
489 .init = cls_bpf_init,
490 .destroy = cls_bpf_destroy,
492 .change = cls_bpf_change,
493 .delete = cls_bpf_delete,
494 .walk = cls_bpf_walk,
495 .dump = cls_bpf_dump,
498 static int __init cls_bpf_init_mod(void)
500 return register_tcf_proto_ops(&cls_bpf_ops);
503 static void __exit cls_bpf_exit_mod(void)
505 unregister_tcf_proto_ops(&cls_bpf_ops);
508 module_init(cls_bpf_init_mod);
509 module_exit(cls_bpf_exit_mod);