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 bpf_compute_data_end(skb);
100 filter_res = BPF_PROG_RUN(prog->filter, skb);
101 __skb_pull(skb, skb->mac_len);
103 bpf_compute_data_end(skb);
104 filter_res = BPF_PROG_RUN(prog->filter, skb);
107 if (prog->exts_integrated) {
109 res->classid = TC_H_MAJ(prog->res.classid) |
110 qdisc_skb_cb(skb)->tc_classid;
112 ret = cls_bpf_exec_opcode(filter_res);
113 if (ret == TC_ACT_UNSPEC)
120 if (filter_res != -1) {
122 res->classid = filter_res;
127 ret = tcf_exts_exec(skb, &prog->exts, res);
138 static bool cls_bpf_is_ebpf(const struct cls_bpf_prog *prog)
140 return !prog->bpf_ops;
143 static int cls_bpf_init(struct tcf_proto *tp)
145 struct cls_bpf_head *head;
147 head = kzalloc(sizeof(*head), GFP_KERNEL);
151 INIT_LIST_HEAD_RCU(&head->plist);
152 rcu_assign_pointer(tp->root, head);
157 static void cls_bpf_delete_prog(struct tcf_proto *tp, struct cls_bpf_prog *prog)
159 tcf_exts_destroy(&prog->exts);
161 if (cls_bpf_is_ebpf(prog))
162 bpf_prog_put(prog->filter);
164 bpf_prog_destroy(prog->filter);
166 kfree(prog->bpf_name);
167 kfree(prog->bpf_ops);
171 static void __cls_bpf_delete_prog(struct rcu_head *rcu)
173 struct cls_bpf_prog *prog = container_of(rcu, struct cls_bpf_prog, rcu);
175 cls_bpf_delete_prog(prog->tp, prog);
178 static int cls_bpf_delete(struct tcf_proto *tp, unsigned long arg)
180 struct cls_bpf_prog *prog = (struct cls_bpf_prog *) arg;
182 list_del_rcu(&prog->link);
183 tcf_unbind_filter(tp, &prog->res);
184 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
189 static bool cls_bpf_destroy(struct tcf_proto *tp, bool force)
191 struct cls_bpf_head *head = rtnl_dereference(tp->root);
192 struct cls_bpf_prog *prog, *tmp;
194 if (!force && !list_empty(&head->plist))
197 list_for_each_entry_safe(prog, tmp, &head->plist, link) {
198 list_del_rcu(&prog->link);
199 tcf_unbind_filter(tp, &prog->res);
200 call_rcu(&prog->rcu, __cls_bpf_delete_prog);
203 RCU_INIT_POINTER(tp->root, NULL);
204 kfree_rcu(head, rcu);
208 static unsigned long cls_bpf_get(struct tcf_proto *tp, u32 handle)
210 struct cls_bpf_head *head = rtnl_dereference(tp->root);
211 struct cls_bpf_prog *prog;
212 unsigned long ret = 0UL;
217 list_for_each_entry(prog, &head->plist, link) {
218 if (prog->handle == handle) {
219 ret = (unsigned long) prog;
227 static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)
229 struct sock_filter *bpf_ops;
230 struct sock_fprog_kern fprog_tmp;
232 u16 bpf_size, bpf_num_ops;
235 bpf_num_ops = nla_get_u16(tb[TCA_BPF_OPS_LEN]);
236 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
239 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
240 if (bpf_size != nla_len(tb[TCA_BPF_OPS]))
243 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
247 memcpy(bpf_ops, nla_data(tb[TCA_BPF_OPS]), bpf_size);
249 fprog_tmp.len = bpf_num_ops;
250 fprog_tmp.filter = bpf_ops;
252 ret = bpf_prog_create(&fp, &fprog_tmp);
258 prog->bpf_ops = bpf_ops;
259 prog->bpf_num_ops = bpf_num_ops;
260 prog->bpf_name = NULL;
266 static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
267 const struct tcf_proto *tp)
273 bpf_fd = nla_get_u32(tb[TCA_BPF_FD]);
275 fp = bpf_prog_get(bpf_fd);
279 if (fp->type != BPF_PROG_TYPE_SCHED_CLS) {
284 if (tb[TCA_BPF_NAME]) {
285 name = kmemdup(nla_data(tb[TCA_BPF_NAME]),
286 nla_len(tb[TCA_BPF_NAME]),
294 prog->bpf_ops = NULL;
295 prog->bpf_fd = bpf_fd;
296 prog->bpf_name = name;
299 if (fp->dst_needed && !(tp->q->flags & TCQ_F_INGRESS))
300 netif_keep_dst(qdisc_dev(tp->q));
305 static int cls_bpf_modify_existing(struct net *net, struct tcf_proto *tp,
306 struct cls_bpf_prog *prog,
307 unsigned long base, struct nlattr **tb,
308 struct nlattr *est, bool ovr)
310 bool is_bpf, is_ebpf, have_exts = false;
311 struct tcf_exts exts;
314 is_bpf = tb[TCA_BPF_OPS_LEN] && tb[TCA_BPF_OPS];
315 is_ebpf = tb[TCA_BPF_FD];
316 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf))
319 tcf_exts_init(&exts, TCA_BPF_ACT, TCA_BPF_POLICE);
320 ret = tcf_exts_validate(net, tp, tb, est, &exts, ovr);
324 if (tb[TCA_BPF_FLAGS]) {
325 u32 bpf_flags = nla_get_u32(tb[TCA_BPF_FLAGS]);
327 if (bpf_flags & ~TCA_BPF_FLAG_ACT_DIRECT) {
328 tcf_exts_destroy(&exts);
332 have_exts = bpf_flags & TCA_BPF_FLAG_ACT_DIRECT;
335 prog->exts_integrated = have_exts;
337 ret = is_bpf ? cls_bpf_prog_from_ops(tb, prog) :
338 cls_bpf_prog_from_efd(tb, prog, tp);
340 tcf_exts_destroy(&exts);
344 if (tb[TCA_BPF_CLASSID]) {
345 prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
346 tcf_bind_filter(tp, &prog->res, base);
349 tcf_exts_change(tp, &prog->exts, &exts);
353 static u32 cls_bpf_grab_new_handle(struct tcf_proto *tp,
354 struct cls_bpf_head *head)
356 unsigned int i = 0x80000000;
360 if (++head->hgen == 0x7FFFFFFF)
362 } while (--i > 0 && cls_bpf_get(tp, head->hgen));
364 if (unlikely(i == 0)) {
365 pr_err("Insufficient number of handles\n");
374 static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
375 struct tcf_proto *tp, unsigned long base,
376 u32 handle, struct nlattr **tca,
377 unsigned long *arg, bool ovr)
379 struct cls_bpf_head *head = rtnl_dereference(tp->root);
380 struct cls_bpf_prog *oldprog = (struct cls_bpf_prog *) *arg;
381 struct nlattr *tb[TCA_BPF_MAX + 1];
382 struct cls_bpf_prog *prog;
385 if (tca[TCA_OPTIONS] == NULL)
388 ret = nla_parse_nested(tb, TCA_BPF_MAX, tca[TCA_OPTIONS], bpf_policy);
392 prog = kzalloc(sizeof(*prog), GFP_KERNEL);
396 tcf_exts_init(&prog->exts, TCA_BPF_ACT, TCA_BPF_POLICE);
399 if (handle && oldprog->handle != handle) {
406 prog->handle = cls_bpf_grab_new_handle(tp, head);
408 prog->handle = handle;
409 if (prog->handle == 0) {
414 ret = cls_bpf_modify_existing(net, tp, prog, base, tb, tca[TCA_RATE], ovr);
419 list_replace_rcu(&oldprog->link, &prog->link);
420 tcf_unbind_filter(tp, &oldprog->res);
421 call_rcu(&oldprog->rcu, __cls_bpf_delete_prog);
423 list_add_rcu(&prog->link, &head->plist);
426 *arg = (unsigned long) prog;
434 static int cls_bpf_dump_bpf_info(const struct cls_bpf_prog *prog,
439 if (nla_put_u16(skb, TCA_BPF_OPS_LEN, prog->bpf_num_ops))
442 nla = nla_reserve(skb, TCA_BPF_OPS, prog->bpf_num_ops *
443 sizeof(struct sock_filter));
447 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
452 static int cls_bpf_dump_ebpf_info(const struct cls_bpf_prog *prog,
455 if (nla_put_u32(skb, TCA_BPF_FD, prog->bpf_fd))
458 if (prog->bpf_name &&
459 nla_put_string(skb, TCA_BPF_NAME, prog->bpf_name))
465 static int cls_bpf_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
466 struct sk_buff *skb, struct tcmsg *tm)
468 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 (prog->res.classid &&
483 nla_put_u32(skb, TCA_BPF_CLASSID, prog->res.classid))
484 goto nla_put_failure;
486 if (cls_bpf_is_ebpf(prog))
487 ret = cls_bpf_dump_ebpf_info(prog, skb);
489 ret = cls_bpf_dump_bpf_info(prog, skb);
491 goto nla_put_failure;
493 if (tcf_exts_dump(skb, &prog->exts) < 0)
494 goto nla_put_failure;
496 if (prog->exts_integrated)
497 bpf_flags |= TCA_BPF_FLAG_ACT_DIRECT;
498 if (bpf_flags && nla_put_u32(skb, TCA_BPF_FLAGS, bpf_flags))
499 goto nla_put_failure;
501 nla_nest_end(skb, nest);
503 if (tcf_exts_dump_stats(skb, &prog->exts) < 0)
504 goto nla_put_failure;
509 nla_nest_cancel(skb, nest);
513 static void cls_bpf_walk(struct tcf_proto *tp, struct tcf_walker *arg)
515 struct cls_bpf_head *head = rtnl_dereference(tp->root);
516 struct cls_bpf_prog *prog;
518 list_for_each_entry(prog, &head->plist, link) {
519 if (arg->count < arg->skip)
521 if (arg->fn(tp, (unsigned long) prog, arg) < 0) {
530 static struct tcf_proto_ops cls_bpf_ops __read_mostly = {
532 .owner = THIS_MODULE,
533 .classify = cls_bpf_classify,
534 .init = cls_bpf_init,
535 .destroy = cls_bpf_destroy,
537 .change = cls_bpf_change,
538 .delete = cls_bpf_delete,
539 .walk = cls_bpf_walk,
540 .dump = cls_bpf_dump,
543 static int __init cls_bpf_init_mod(void)
545 return register_tcf_proto_ops(&cls_bpf_ops);
548 static void __exit cls_bpf_exit_mod(void)
550 unregister_tcf_proto_ops(&cls_bpf_ops);
553 module_init(cls_bpf_init_mod);
554 module_exit(cls_bpf_exit_mod);