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;
38 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
39 struct tcf_result *res)
41 struct tcf_bpf *prog = act->priv;
42 struct bpf_prog *filter;
43 int action, filter_res;
44 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
46 if (unlikely(!skb_mac_header_was_set(skb)))
49 tcf_lastuse_update(&prog->tcf_tm);
50 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
53 filter = rcu_dereference(prog->filter);
55 __skb_push(skb, skb->mac_len);
56 bpf_compute_data_end(skb);
57 filter_res = BPF_PROG_RUN(filter, skb);
58 __skb_pull(skb, skb->mac_len);
60 bpf_compute_data_end(skb);
61 filter_res = BPF_PROG_RUN(filter, skb);
65 /* A BPF program may overwrite the default action opcode.
66 * Similarly as in cls_bpf, if filter_res == -1 we use the
67 * default action specified from tc.
69 * In case a different well-known TC_ACT opcode has been
70 * returned, it will overwrite the default one.
72 * For everything else that is unkown, TC_ACT_UNSPEC is
77 case TC_ACT_RECLASSIFY:
84 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
87 action = prog->tcf_action;
90 action = TC_ACT_UNSPEC;
97 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
99 return !prog->bpf_ops;
102 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
107 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
110 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
111 sizeof(struct sock_filter));
115 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
120 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
123 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
126 if (prog->bpf_name &&
127 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
133 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
136 unsigned char *tp = skb_tail_pointer(skb);
137 struct tcf_bpf *prog = act->priv;
138 struct tc_act_bpf opt = {
139 .index = prog->tcf_index,
140 .refcnt = prog->tcf_refcnt - ref,
141 .bindcnt = prog->tcf_bindcnt - bind,
142 .action = prog->tcf_action,
147 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
148 goto nla_put_failure;
150 if (tcf_bpf_is_ebpf(prog))
151 ret = tcf_bpf_dump_ebpf_info(prog, skb);
153 ret = tcf_bpf_dump_bpf_info(prog, skb);
155 goto nla_put_failure;
157 tcf_tm_dump(&tm, &prog->tcf_tm);
158 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
160 goto nla_put_failure;
169 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
170 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
171 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
172 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING,
173 .len = ACT_BPF_NAME_LEN },
174 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
175 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
176 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
179 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
181 struct sock_filter *bpf_ops;
182 struct sock_fprog_kern fprog_tmp;
184 u16 bpf_size, bpf_num_ops;
187 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
188 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
191 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
192 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
195 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
199 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
201 fprog_tmp.len = bpf_num_ops;
202 fprog_tmp.filter = bpf_ops;
204 ret = bpf_prog_create(&fp, &fprog_tmp);
210 cfg->bpf_ops = bpf_ops;
211 cfg->bpf_num_ops = bpf_num_ops;
213 cfg->is_ebpf = false;
218 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
224 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
226 fp = bpf_prog_get(bpf_fd);
230 if (fp->type != BPF_PROG_TYPE_SCHED_ACT) {
235 if (tb[TCA_ACT_BPF_NAME]) {
236 name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
237 nla_len(tb[TCA_ACT_BPF_NAME]),
245 cfg->bpf_fd = bpf_fd;
246 cfg->bpf_name = name;
253 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
256 bpf_prog_put(cfg->filter);
258 bpf_prog_destroy(cfg->filter);
261 kfree(cfg->bpf_name);
264 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
265 struct tcf_bpf_cfg *cfg)
267 cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
268 /* updates to prog->filter are prevented, since it's called either
269 * with rtnl lock or during final cleanup in rcu callback
271 cfg->filter = rcu_dereference_protected(prog->filter, 1);
273 cfg->bpf_ops = prog->bpf_ops;
274 cfg->bpf_name = prog->bpf_name;
277 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
278 struct nlattr *est, struct tc_action *act,
279 int replace, int bind)
281 struct tc_action_net *tn = net_generic(net, bpf_net_id);
282 struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
283 struct tcf_bpf_cfg cfg, old;
284 struct tc_act_bpf *parm;
285 struct tcf_bpf *prog;
286 bool is_bpf, is_ebpf;
292 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
296 if (!tb[TCA_ACT_BPF_PARMS])
299 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
301 if (!tcf_hash_check(tn, parm->index, act, bind)) {
302 ret = tcf_hash_create(tn, parm->index, est, act,
303 sizeof(*prog), bind, true);
309 /* Don't override defaults. */
313 tcf_hash_release(act, bind);
318 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
319 is_ebpf = tb[TCA_ACT_BPF_FD];
321 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
326 memset(&cfg, 0, sizeof(cfg));
328 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
329 tcf_bpf_init_from_efd(tb, &cfg);
336 if (res != ACT_P_CREATED)
337 tcf_bpf_prog_fill_cfg(prog, &old);
339 prog->bpf_ops = cfg.bpf_ops;
340 prog->bpf_name = cfg.bpf_name;
343 prog->bpf_num_ops = cfg.bpf_num_ops;
345 prog->bpf_fd = cfg.bpf_fd;
347 prog->tcf_action = parm->action;
348 rcu_assign_pointer(prog->filter, cfg.filter);
350 if (res == ACT_P_CREATED) {
351 tcf_hash_insert(tn, act);
353 /* make sure the program being replaced is no longer executing */
355 tcf_bpf_cfg_cleanup(&old);
360 if (res == ACT_P_CREATED)
361 tcf_hash_cleanup(act, est);
366 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
368 struct tcf_bpf_cfg tmp;
370 tcf_bpf_prog_fill_cfg(act->priv, &tmp);
371 tcf_bpf_cfg_cleanup(&tmp);
374 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
375 struct netlink_callback *cb, int type,
378 struct tc_action_net *tn = net_generic(net, bpf_net_id);
380 return tcf_generic_walker(tn, skb, cb, type, a);
383 static int tcf_bpf_search(struct net *net, struct tc_action *a, u32 index)
385 struct tc_action_net *tn = net_generic(net, bpf_net_id);
387 return tcf_hash_search(tn, a, index);
390 static struct tc_action_ops act_bpf_ops __read_mostly = {
393 .owner = THIS_MODULE,
395 .dump = tcf_bpf_dump,
396 .cleanup = tcf_bpf_cleanup,
397 .init = tcf_bpf_init,
398 .walk = tcf_bpf_walker,
399 .lookup = tcf_bpf_search,
402 static __net_init int bpf_init_net(struct net *net)
404 struct tc_action_net *tn = net_generic(net, bpf_net_id);
406 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
409 static void __net_exit bpf_exit_net(struct net *net)
411 struct tc_action_net *tn = net_generic(net, bpf_net_id);
413 tc_action_net_exit(tn);
416 static struct pernet_operations bpf_net_ops = {
417 .init = bpf_init_net,
418 .exit = bpf_exit_net,
420 .size = sizeof(struct tc_action_net),
423 static int __init bpf_init_module(void)
425 return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
428 static void __exit bpf_cleanup_module(void)
430 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
433 module_init(bpf_init_module);
434 module_exit(bpf_cleanup_module);
436 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
437 MODULE_DESCRIPTION("TC BPF based action");
438 MODULE_LICENSE("GPL v2");