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 bool at_ingress = skb_at_tc_ingress(skb);
43 struct tcf_bpf *prog = to_bpf(act);
44 struct bpf_prog *filter;
45 int action, filter_res;
47 tcf_lastuse_update(&prog->tcf_tm);
48 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
51 filter = rcu_dereference(prog->filter);
53 __skb_push(skb, skb->mac_len);
54 bpf_compute_data_end(skb);
55 filter_res = BPF_PROG_RUN(filter, skb);
56 __skb_pull(skb, skb->mac_len);
58 bpf_compute_data_end(skb);
59 filter_res = BPF_PROG_RUN(filter, skb);
63 /* A BPF program may overwrite the default action opcode.
64 * Similarly as in cls_bpf, if filter_res == -1 we use the
65 * default action specified from tc.
67 * In case a different well-known TC_ACT opcode has been
68 * returned, it will overwrite the default one.
70 * For everything else that is unkown, TC_ACT_UNSPEC is
75 case TC_ACT_RECLASSIFY:
82 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
85 action = prog->tcf_action;
88 action = TC_ACT_UNSPEC;
95 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
97 return !prog->bpf_ops;
100 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
105 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
108 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
109 sizeof(struct sock_filter));
113 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
118 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
121 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
124 if (prog->bpf_name &&
125 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
131 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
134 unsigned char *tp = skb_tail_pointer(skb);
135 struct tcf_bpf *prog = to_bpf(act);
136 struct tc_act_bpf opt = {
137 .index = prog->tcf_index,
138 .refcnt = prog->tcf_refcnt - ref,
139 .bindcnt = prog->tcf_bindcnt - bind,
140 .action = prog->tcf_action,
145 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
146 goto nla_put_failure;
148 if (tcf_bpf_is_ebpf(prog))
149 ret = tcf_bpf_dump_ebpf_info(prog, skb);
151 ret = tcf_bpf_dump_bpf_info(prog, skb);
153 goto nla_put_failure;
155 tcf_tm_dump(&tm, &prog->tcf_tm);
156 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
158 goto nla_put_failure;
167 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
168 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
169 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
170 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING,
171 .len = ACT_BPF_NAME_LEN },
172 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
173 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
174 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
177 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
179 struct sock_filter *bpf_ops;
180 struct sock_fprog_kern fprog_tmp;
182 u16 bpf_size, bpf_num_ops;
185 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
186 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
189 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
190 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
193 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
197 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
199 fprog_tmp.len = bpf_num_ops;
200 fprog_tmp.filter = bpf_ops;
202 ret = bpf_prog_create(&fp, &fprog_tmp);
208 cfg->bpf_ops = bpf_ops;
209 cfg->bpf_num_ops = bpf_num_ops;
211 cfg->is_ebpf = false;
216 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
222 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
224 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
228 if (tb[TCA_ACT_BPF_NAME]) {
229 name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL);
236 cfg->bpf_fd = bpf_fd;
237 cfg->bpf_name = name;
244 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
247 bpf_prog_put(cfg->filter);
249 bpf_prog_destroy(cfg->filter);
252 kfree(cfg->bpf_name);
255 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
256 struct tcf_bpf_cfg *cfg)
258 cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
259 /* updates to prog->filter are prevented, since it's called either
260 * with rtnl lock or during final cleanup in rcu callback
262 cfg->filter = rcu_dereference_protected(prog->filter, 1);
264 cfg->bpf_ops = prog->bpf_ops;
265 cfg->bpf_name = prog->bpf_name;
268 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
269 struct nlattr *est, struct tc_action **act,
270 int replace, int bind)
272 struct tc_action_net *tn = net_generic(net, bpf_net_id);
273 struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
274 struct tcf_bpf_cfg cfg, old;
275 struct tc_act_bpf *parm;
276 struct tcf_bpf *prog;
277 bool is_bpf, is_ebpf;
283 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
287 if (!tb[TCA_ACT_BPF_PARMS])
290 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
292 if (!tcf_hash_check(tn, parm->index, act, bind)) {
293 ret = tcf_hash_create(tn, parm->index, est, act,
294 &act_bpf_ops, bind, true);
300 /* Don't override defaults. */
304 tcf_hash_release(*act, bind);
309 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
310 is_ebpf = tb[TCA_ACT_BPF_FD];
312 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
317 memset(&cfg, 0, sizeof(cfg));
319 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
320 tcf_bpf_init_from_efd(tb, &cfg);
327 if (res != ACT_P_CREATED)
328 tcf_bpf_prog_fill_cfg(prog, &old);
330 prog->bpf_ops = cfg.bpf_ops;
331 prog->bpf_name = cfg.bpf_name;
334 prog->bpf_num_ops = cfg.bpf_num_ops;
336 prog->bpf_fd = cfg.bpf_fd;
338 prog->tcf_action = parm->action;
339 rcu_assign_pointer(prog->filter, cfg.filter);
341 if (res == ACT_P_CREATED) {
342 tcf_hash_insert(tn, *act);
344 /* make sure the program being replaced is no longer executing */
346 tcf_bpf_cfg_cleanup(&old);
351 if (res == ACT_P_CREATED)
352 tcf_hash_cleanup(*act, est);
357 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
359 struct tcf_bpf_cfg tmp;
361 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
362 tcf_bpf_cfg_cleanup(&tmp);
365 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
366 struct netlink_callback *cb, int type,
367 const struct tc_action_ops *ops)
369 struct tc_action_net *tn = net_generic(net, bpf_net_id);
371 return tcf_generic_walker(tn, skb, cb, type, ops);
374 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
376 struct tc_action_net *tn = net_generic(net, bpf_net_id);
378 return tcf_hash_search(tn, a, index);
381 static struct tc_action_ops act_bpf_ops __read_mostly = {
384 .owner = THIS_MODULE,
386 .dump = tcf_bpf_dump,
387 .cleanup = tcf_bpf_cleanup,
388 .init = tcf_bpf_init,
389 .walk = tcf_bpf_walker,
390 .lookup = tcf_bpf_search,
391 .size = sizeof(struct tcf_bpf),
394 static __net_init int bpf_init_net(struct net *net)
396 struct tc_action_net *tn = net_generic(net, bpf_net_id);
398 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
401 static void __net_exit bpf_exit_net(struct net *net)
403 struct tc_action_net *tn = net_generic(net, bpf_net_id);
405 tc_action_net_exit(tn);
408 static struct pernet_operations bpf_net_ops = {
409 .init = bpf_init_net,
410 .exit = bpf_exit_net,
412 .size = sizeof(struct tc_action_net),
415 static int __init bpf_init_module(void)
417 return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
420 static void __exit bpf_cleanup_module(void)
422 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
425 module_init(bpf_init_module);
426 module_exit(bpf_cleanup_module);
428 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
429 MODULE_DESCRIPTION("TC BPF based action");
430 MODULE_LICENSE("GPL v2");