2 * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/skbuff.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/filter.h>
16 #include <linux/bpf.h>
18 #include <net/netlink.h>
19 #include <net/pkt_sched.h>
21 #include <linux/tc_act/tc_bpf.h>
22 #include <net/tc_act/tc_bpf.h>
24 #define BPF_TAB_MASK 15
25 #define ACT_BPF_NAME_LEN 256
28 struct bpf_prog *filter;
29 struct sock_filter *bpf_ops;
36 static int bpf_net_id;
37 static struct tc_action_ops act_bpf_ops;
39 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
40 struct tcf_result *res)
42 struct tcf_bpf *prog = to_bpf(act);
43 struct bpf_prog *filter;
44 int action, filter_res;
45 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
47 if (unlikely(!skb_mac_header_was_set(skb)))
50 tcf_lastuse_update(&prog->tcf_tm);
51 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
54 filter = rcu_dereference(prog->filter);
56 __skb_push(skb, skb->mac_len);
57 bpf_compute_data_end(skb);
58 filter_res = BPF_PROG_RUN(filter, skb);
59 __skb_pull(skb, skb->mac_len);
61 bpf_compute_data_end(skb);
62 filter_res = BPF_PROG_RUN(filter, skb);
66 /* A BPF program may overwrite the default action opcode.
67 * Similarly as in cls_bpf, if filter_res == -1 we use the
68 * default action specified from tc.
70 * In case a different well-known TC_ACT opcode has been
71 * returned, it will overwrite the default one.
73 * For everything else that is unkown, TC_ACT_UNSPEC is
78 case TC_ACT_RECLASSIFY:
85 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
88 action = prog->tcf_action;
91 action = TC_ACT_UNSPEC;
98 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
100 return !prog->bpf_ops;
103 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
108 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
111 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
112 sizeof(struct sock_filter));
116 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
121 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
124 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
127 if (prog->bpf_name &&
128 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
134 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
137 unsigned char *tp = skb_tail_pointer(skb);
138 struct tcf_bpf *prog = to_bpf(act);
139 struct tc_act_bpf opt = {
140 .index = prog->tcf_index,
141 .refcnt = prog->tcf_refcnt - ref,
142 .bindcnt = prog->tcf_bindcnt - bind,
143 .action = prog->tcf_action,
148 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
149 goto nla_put_failure;
151 if (tcf_bpf_is_ebpf(prog))
152 ret = tcf_bpf_dump_ebpf_info(prog, skb);
154 ret = tcf_bpf_dump_bpf_info(prog, skb);
156 goto nla_put_failure;
158 tcf_tm_dump(&tm, &prog->tcf_tm);
159 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
161 goto nla_put_failure;
170 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
171 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
172 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
173 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING,
174 .len = ACT_BPF_NAME_LEN },
175 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
176 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
177 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
180 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
182 struct sock_filter *bpf_ops;
183 struct sock_fprog_kern fprog_tmp;
185 u16 bpf_size, bpf_num_ops;
188 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
189 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
192 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
193 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
196 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
200 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
202 fprog_tmp.len = bpf_num_ops;
203 fprog_tmp.filter = bpf_ops;
205 ret = bpf_prog_create(&fp, &fprog_tmp);
211 cfg->bpf_ops = bpf_ops;
212 cfg->bpf_num_ops = bpf_num_ops;
214 cfg->is_ebpf = false;
219 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
225 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
227 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
231 if (tb[TCA_ACT_BPF_NAME]) {
232 name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
233 nla_len(tb[TCA_ACT_BPF_NAME]),
241 cfg->bpf_fd = bpf_fd;
242 cfg->bpf_name = name;
249 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
252 bpf_prog_put(cfg->filter);
254 bpf_prog_destroy(cfg->filter);
257 kfree(cfg->bpf_name);
260 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
261 struct tcf_bpf_cfg *cfg)
263 cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
264 /* updates to prog->filter are prevented, since it's called either
265 * with rtnl lock or during final cleanup in rcu callback
267 cfg->filter = rcu_dereference_protected(prog->filter, 1);
269 cfg->bpf_ops = prog->bpf_ops;
270 cfg->bpf_name = prog->bpf_name;
273 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
274 struct nlattr *est, struct tc_action **act,
275 int replace, int bind)
277 struct tc_action_net *tn = net_generic(net, bpf_net_id);
278 struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
279 struct tcf_bpf_cfg cfg, old;
280 struct tc_act_bpf *parm;
281 struct tcf_bpf *prog;
282 bool is_bpf, is_ebpf;
288 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
292 if (!tb[TCA_ACT_BPF_PARMS])
295 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
297 if (!tcf_hash_check(tn, parm->index, act, bind)) {
298 ret = tcf_hash_create(tn, parm->index, est, act,
299 &act_bpf_ops, bind, true);
305 /* Don't override defaults. */
309 tcf_hash_release(*act, bind);
314 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
315 is_ebpf = tb[TCA_ACT_BPF_FD];
317 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
322 memset(&cfg, 0, sizeof(cfg));
324 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
325 tcf_bpf_init_from_efd(tb, &cfg);
332 if (res != ACT_P_CREATED)
333 tcf_bpf_prog_fill_cfg(prog, &old);
335 prog->bpf_ops = cfg.bpf_ops;
336 prog->bpf_name = cfg.bpf_name;
339 prog->bpf_num_ops = cfg.bpf_num_ops;
341 prog->bpf_fd = cfg.bpf_fd;
343 prog->tcf_action = parm->action;
344 rcu_assign_pointer(prog->filter, cfg.filter);
346 if (res == ACT_P_CREATED) {
347 tcf_hash_insert(tn, *act);
349 /* make sure the program being replaced is no longer executing */
351 tcf_bpf_cfg_cleanup(&old);
356 if (res == ACT_P_CREATED)
357 tcf_hash_cleanup(*act, est);
362 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
364 struct tcf_bpf_cfg tmp;
366 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
367 tcf_bpf_cfg_cleanup(&tmp);
370 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
371 struct netlink_callback *cb, int type,
372 const struct tc_action_ops *ops)
374 struct tc_action_net *tn = net_generic(net, bpf_net_id);
376 return tcf_generic_walker(tn, skb, cb, type, ops);
379 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
381 struct tc_action_net *tn = net_generic(net, bpf_net_id);
383 return tcf_hash_search(tn, a, index);
386 static struct tc_action_ops act_bpf_ops __read_mostly = {
389 .owner = THIS_MODULE,
391 .dump = tcf_bpf_dump,
392 .cleanup = tcf_bpf_cleanup,
393 .init = tcf_bpf_init,
394 .walk = tcf_bpf_walker,
395 .lookup = tcf_bpf_search,
396 .size = sizeof(struct tcf_bpf),
399 static __net_init int bpf_init_net(struct net *net)
401 struct tc_action_net *tn = net_generic(net, bpf_net_id);
403 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
406 static void __net_exit bpf_exit_net(struct net *net)
408 struct tc_action_net *tn = net_generic(net, bpf_net_id);
410 tc_action_net_exit(tn);
413 static struct pernet_operations bpf_net_ops = {
414 .init = bpf_init_net,
415 .exit = bpf_exit_net,
417 .size = sizeof(struct tc_action_net),
420 static int __init bpf_init_module(void)
422 return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
425 static void __exit bpf_cleanup_module(void)
427 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
430 module_init(bpf_init_module);
431 module_exit(bpf_cleanup_module);
433 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
434 MODULE_DESCRIPTION("TC BPF based action");
435 MODULE_LICENSE("GPL v2");