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 ACT_BPF_NAME_LEN 256
27 struct bpf_prog *filter;
28 struct sock_filter *bpf_ops;
34 static unsigned int bpf_net_id;
35 static struct tc_action_ops act_bpf_ops;
37 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
38 struct tcf_result *res)
40 bool at_ingress = skb_at_tc_ingress(skb);
41 struct tcf_bpf *prog = to_bpf(act);
42 struct bpf_prog *filter;
43 int action, filter_res;
45 tcf_lastuse_update(&prog->tcf_tm);
46 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
49 filter = rcu_dereference(prog->filter);
51 __skb_push(skb, skb->mac_len);
52 bpf_compute_data_end(skb);
53 filter_res = BPF_PROG_RUN(filter, skb);
54 __skb_pull(skb, skb->mac_len);
56 bpf_compute_data_end(skb);
57 filter_res = BPF_PROG_RUN(filter, skb);
61 /* A BPF program may overwrite the default action opcode.
62 * Similarly as in cls_bpf, if filter_res == -1 we use the
63 * default action specified from tc.
65 * In case a different well-known TC_ACT opcode has been
66 * returned, it will overwrite the default one.
68 * For everything else that is unkown, TC_ACT_UNSPEC is
73 case TC_ACT_RECLASSIFY:
80 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
83 action = prog->tcf_action;
86 action = TC_ACT_UNSPEC;
93 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
95 return !prog->bpf_ops;
98 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
103 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
106 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
107 sizeof(struct sock_filter));
111 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
116 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
121 if (prog->bpf_name &&
122 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
125 if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id))
128 nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag));
132 memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
137 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
140 unsigned char *tp = skb_tail_pointer(skb);
141 struct tcf_bpf *prog = to_bpf(act);
142 struct tc_act_bpf opt = {
143 .index = prog->tcf_index,
144 .refcnt = prog->tcf_refcnt - ref,
145 .bindcnt = prog->tcf_bindcnt - bind,
146 .action = prog->tcf_action,
151 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
152 goto nla_put_failure;
154 if (tcf_bpf_is_ebpf(prog))
155 ret = tcf_bpf_dump_ebpf_info(prog, skb);
157 ret = tcf_bpf_dump_bpf_info(prog, skb);
159 goto nla_put_failure;
161 tcf_tm_dump(&tm, &prog->tcf_tm);
162 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
164 goto nla_put_failure;
173 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
174 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
175 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
176 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING,
177 .len = ACT_BPF_NAME_LEN },
178 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
179 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
180 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
183 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
185 struct sock_filter *bpf_ops;
186 struct sock_fprog_kern fprog_tmp;
188 u16 bpf_size, bpf_num_ops;
191 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
192 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
195 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
196 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
199 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
203 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
205 fprog_tmp.len = bpf_num_ops;
206 fprog_tmp.filter = bpf_ops;
208 ret = bpf_prog_create(&fp, &fprog_tmp);
214 cfg->bpf_ops = bpf_ops;
215 cfg->bpf_num_ops = bpf_num_ops;
217 cfg->is_ebpf = false;
222 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
228 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
230 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
234 if (tb[TCA_ACT_BPF_NAME]) {
235 name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL);
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, NULL);
292 if (!tb[TCA_ACT_BPF_PARMS])
295 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
297 if (!tcf_idr_check(tn, parm->index, act, bind)) {
298 ret = tcf_idr_create(tn, parm->index, est, act,
299 &act_bpf_ops, bind, true);
305 /* Don't override defaults. */
309 tcf_idr_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->tcf_action = parm->action;
342 rcu_assign_pointer(prog->filter, cfg.filter);
344 if (res == ACT_P_CREATED) {
345 tcf_idr_insert(tn, *act);
347 /* make sure the program being replaced is no longer executing */
349 tcf_bpf_cfg_cleanup(&old);
354 if (res == ACT_P_CREATED)
355 tcf_idr_cleanup(*act, est);
360 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
362 struct tcf_bpf_cfg tmp;
364 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
365 tcf_bpf_cfg_cleanup(&tmp);
368 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
369 struct netlink_callback *cb, int type,
370 const struct tc_action_ops *ops)
372 struct tc_action_net *tn = net_generic(net, bpf_net_id);
374 return tcf_generic_walker(tn, skb, cb, type, ops);
377 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
379 struct tc_action_net *tn = net_generic(net, bpf_net_id);
381 return tcf_idr_search(tn, a, index);
384 static struct tc_action_ops act_bpf_ops __read_mostly = {
387 .owner = THIS_MODULE,
389 .dump = tcf_bpf_dump,
390 .cleanup = tcf_bpf_cleanup,
391 .init = tcf_bpf_init,
392 .walk = tcf_bpf_walker,
393 .lookup = tcf_bpf_search,
394 .size = sizeof(struct tcf_bpf),
397 static __net_init int bpf_init_net(struct net *net)
399 struct tc_action_net *tn = net_generic(net, bpf_net_id);
401 return tc_action_net_init(tn, &act_bpf_ops, net);
404 static void __net_exit bpf_exit_net(struct net *net)
406 struct tc_action_net *tn = net_generic(net, bpf_net_id);
408 tc_action_net_exit(tn);
411 static struct pernet_operations bpf_net_ops = {
412 .init = bpf_init_net,
413 .exit = bpf_exit_net,
415 .size = sizeof(struct tc_action_net),
418 static int __init bpf_init_module(void)
420 return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
423 static void __exit bpf_cleanup_module(void)
425 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
428 module_init(bpf_init_module);
429 module_exit(bpf_cleanup_module);
431 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
432 MODULE_DESCRIPTION("TC BPF based action");
433 MODULE_LICENSE("GPL v2");