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 bool at_ingress = skb_at_tc_ingress(skb);
83 struct cls_bpf_prog *prog;
86 if (unlikely(!skb_mac_header_was_set(skb)))
89 /* Needed here for accessing maps. */
91 list_for_each_entry_rcu(prog, &head->plist, link) {
94 qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
97 /* It is safe to push/pull even if skb_shared() */
98 __skb_push(skb, skb->mac_len);
99 filter_res = BPF_PROG_RUN(prog->filter, skb);
100 __skb_pull(skb, skb->mac_len);
102 filter_res = BPF_PROG_RUN(prog->filter, skb);
105 if (prog->exts_integrated) {
106 res->class = prog->res.class;
107 res->classid = qdisc_skb_cb(skb)->tc_classid;
109 ret = cls_bpf_exec_opcode(filter_res);
110 if (ret == TC_ACT_UNSPEC)
119 if (filter_res != -1)
120 res->classid = filter_res;
122 ret = tcf_exts_exec(skb, &prog->exts, res);
133 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
135 return !prog->bpf_ops;
138 static int cls_bpf_init(struct tcf_proto *tp)
140 struct cls_bpf_head *head;
142 head = kzalloc(sizeof(*head), GFP_KERNEL);
146 INIT_LIST_HEAD_RCU(&head->plist);
147 rcu_assign_pointer(tp->root, head);
152 static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
154 tcf_exts_destroy(&prog->exts);
156 if (cls_bpf_is_ebpf(prog))
157 bpf_prog_put(prog->filter);
159 bpf_prog_destroy(prog->filter);
161 kfree(prog->bpf_name);
162 kfree(prog->bpf_ops);
166 static void __cls_bpf_delete_prog(struct rcu_head *rcu)
168 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
170 cls_bpf_delete_prog(prog->tp, prog);
173 static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
175 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
177 list_del_rcu(&prog->link);
178 tcf_unbind_filter(tp, &prog->res);
179 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
184 static bool cls_bpf_destroy(struct tcf_proto *tp, bool force)
186 struct cls_bpf_head *head = rtnl_dereference(tp->root);
187 struct cls_bpf_prog *prog, *tmp;
189 if (!force && !list_empty(&head->plist))
192 list_for_each_entry_safe(prog, tmp, &head->plist, link) {
193 list_del_rcu(&prog->link);
194 tcf_unbind_filter(tp, &prog->res);
195 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
198 RCU_INIT_POINTER(tp->root, NULL);
199 kfree_rcu(head, rcu);
203 static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
205 struct cls_bpf_head *head = rtnl_dereference(tp->root);
206 struct cls_bpf_prog *prog;
207 unsigned long ret = 0UL;
212 list_for_each_entry(prog, &head->plist, link) {
213 if (prog->handle == handle) {
214 ret = (unsigned long) prog;
222 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
224 struct sock_filter *bpf_ops;
225 struct sock_fprog_kern fprog_tmp;
227 u16 bpf_size, bpf_num_ops;
230 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
231 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
234 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
235 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
238 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
242 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
244 fprog_tmp.len = bpf_num_ops;
245 fprog_tmp.filter = bpf_ops;
247 ret = bpf_prog_create(&fp, &fprog_tmp);
253 prog->bpf_ops = bpf_ops;
254 prog->bpf_num_ops = bpf_num_ops;
255 prog->bpf_name = NULL;
261 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
262 const struct tcf_proto *tp)
268 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
270 fp = bpf_prog_get(bpf_fd);
274 if (fp->type != BPF_PROG_TYPE_SCHED_CLS) {
279 if (tb[TCA_BPF_NAME]) {
280 name = kmemdup(nla_data(tb[TCA_BPF_NAME]),
281 nla_len(tb[TCA_BPF_NAME]),
289 prog->bpf_ops = NULL;
290 prog->bpf_fd = bpf_fd;
291 prog->bpf_name = name;
294 if (fp->dst_needed && !(tp->q->flags & TCQ_F_INGRESS))
295 netif_keep_dst(qdisc_dev(tp->q));
300 static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
301 struct cls_bpf_prog *prog,
302 unsigned long base, struct nlattr **tb,
303 struct nlattr *est, bool ovr)
305 bool is_bpf, is_ebpf, have_exts = false;
306 struct tcf_exts exts;
309 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
310 is_ebpf = tb[TCA_BPF_FD];
311 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
314 tcf_exts_init(&exts, TCA_BPF_ACT, TCA_BPF_POLICE);
315 ret = tcf_exts_validate(net, tp, tb, est, &exts, ovr);
319 if (tb[TCA_BPF_FLAGS]) {
320 u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
322 if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT) {
323 tcf_exts_destroy(&exts);
327 have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
330 prog->exts_integrated = have_exts;
332 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
333 cls_bpf_prog_from_efd(tb, prog, tp);
335 tcf_exts_destroy(&exts);
339 if (tb[TCA_BPF_CLASSID]) {
340 prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
341 tcf_bind_filter(tp, &prog->res, base);
344 tcf_exts_change(tp, &prog->exts, &exts);
348 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
349 struct cls_bpf_head *head)
351 unsigned int i = 0x80000000;
355 if (++head->hgen == 0x7FFFFFFF)
357 } while (--i > 0 && cls_bpf_get(tp, head->hgen));
359 if (unlikely(i == 0)) {
360 pr_err("Insufficient number of handles\n");
369 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
370 struct tcf_proto *tp, unsigned long base,
371 u32 handle, struct nlattr **tca,
372 unsigned long *arg, bool ovr)
374 struct cls_bpf_head *head = rtnl_dereference(tp->root);
375 struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
376 struct nlattr *tb[TCA_BPF_MAX + 1];
377 struct cls_bpf_prog *prog;
380 if (tca[TCA_OPTIONS] == NULL)
383 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
387 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
391 tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
394 if (handle && oldprog->handle != handle) {
401 prog->handle = cls_bpf_grab_new_handle(tp, head);
403 prog->handle = handle;
404 if (prog->handle == 0) {
409 ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
414 list_replace_rcu(&oldprog->link, &prog->link);
415 tcf_unbind_filter(tp, &oldprog->res);
416 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
418 list_add_rcu(&prog->link, &head->plist);
421 *arg = (unsigned long) prog;
429 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
434 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
437 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
438 sizeof(struct sock_filter));
442 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
447 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
450 if (nla_put_u32(skb, TCA_BPF_FD, prog->bpf_fd))
453 if (prog->bpf_name &&
454 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
460 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
461 struct sk_buff *skb, struct tcmsg *tm)
463 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) fh;
471 tm->tcm_handle = prog->handle;
473 nest = nla_nest_start(skb, TCA_OPTIONS);
475 goto nla_put_failure;
477 if (prog->res.classid &&
478 nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
479 goto nla_put_failure;
481 if (cls_bpf_is_ebpf(prog))
482 ret = cls_bpf_dump_ebpf_info(prog, skb);
484 ret = cls_bpf_dump_bpf_info(prog, skb);
486 goto nla_put_failure;
488 if (tcf_exts_dump(skb, &prog->exts) < 0)
489 goto nla_put_failure;
491 if (prog->exts_integrated)
492 bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
493 if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
494 goto nla_put_failure;
496 nla_nest_end(skb, nest);
498 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
499 goto nla_put_failure;
504 nla_nest_cancel(skb, nest);
508 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
510 struct cls_bpf_head *head = rtnl_dereference(tp->root);
511 struct cls_bpf_prog *prog;
513 list_for_each_entry(prog, &head->plist, link) {
514 if (arg->count < arg->skip)
516 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
525 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
527 .owner = THIS_MODULE,
528 .classify = cls_bpf_classify,
529 .init = cls_bpf_init,
530 .destroy = cls_bpf_destroy,
532 .change = cls_bpf_change,
533 .delete = cls_bpf_delete,
534 .walk = cls_bpf_walk,
535 .dump = cls_bpf_dump,
538 static int __init cls_bpf_init_mod(void)
540 return register_tcf_proto_ops(&cls_bpf_ops);
543 static void __exit cls_bpf_exit_mod(void)
545 unregister_tcf_proto_ops(&cls_bpf_ops);
548 module_init(cls_bpf_init_mod);
549 module_exit(cls_bpf_exit_mod);