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;
35 static unsigned int bpf_net_id;
36 static struct tc_action_ops act_bpf_ops;
38 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
39 struct tcf_result *res)
41 bool at_ingress = skb_at_tc_ingress(skb);
42 struct tcf_bpf *prog = to_bpf(act);
43 struct bpf_prog *filter;
44 int action, filter_res;
46 tcf_lastuse_update(&prog->tcf_tm);
47 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
50 filter = rcu_dereference(prog->filter);
52 __skb_push(skb, skb->mac_len);
53 bpf_compute_data_end(skb);
54 filter_res = BPF_PROG_RUN(filter, skb);
55 __skb_pull(skb, skb->mac_len);
57 bpf_compute_data_end(skb);
58 filter_res = BPF_PROG_RUN(filter, skb);
62 /* A BPF program may overwrite the default action opcode.
63 * Similarly as in cls_bpf, if filter_res == -1 we use the
64 * default action specified from tc.
66 * In case a different well-known TC_ACT opcode has been
67 * returned, it will overwrite the default one.
69 * For everything else that is unkown, TC_ACT_UNSPEC is
74 case TC_ACT_RECLASSIFY:
81 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
84 action = prog->tcf_action;
87 action = TC_ACT_UNSPEC;
94 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
96 return !prog->bpf_ops;
99 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
104 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
107 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
108 sizeof(struct sock_filter));
112 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
117 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
122 if (prog->bpf_name &&
123 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
126 if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id))
129 nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag));
133 memcpy(nla_data(nla), prog->filter->tag, nla_len(nla));
138 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
141 unsigned char *tp = skb_tail_pointer(skb);
142 struct tcf_bpf *prog = to_bpf(act);
143 struct tc_act_bpf opt = {
144 .index = prog->tcf_index,
145 .refcnt = prog->tcf_refcnt - ref,
146 .bindcnt = prog->tcf_bindcnt - bind,
147 .action = prog->tcf_action,
152 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
153 goto nla_put_failure;
155 if (tcf_bpf_is_ebpf(prog))
156 ret = tcf_bpf_dump_ebpf_info(prog, skb);
158 ret = tcf_bpf_dump_bpf_info(prog, skb);
160 goto nla_put_failure;
162 tcf_tm_dump(&tm, &prog->tcf_tm);
163 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
165 goto nla_put_failure;
174 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
175 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
176 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
177 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING,
178 .len = ACT_BPF_NAME_LEN },
179 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
180 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
181 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
184 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
186 struct sock_filter *bpf_ops;
187 struct sock_fprog_kern fprog_tmp;
189 u16 bpf_size, bpf_num_ops;
192 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
193 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
196 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
197 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
200 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
204 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
206 fprog_tmp.len = bpf_num_ops;
207 fprog_tmp.filter = bpf_ops;
209 ret = bpf_prog_create(&fp, &fprog_tmp);
215 cfg->bpf_ops = bpf_ops;
216 cfg->bpf_num_ops = bpf_num_ops;
218 cfg->is_ebpf = false;
223 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
229 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
231 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
235 if (tb[TCA_ACT_BPF_NAME]) {
236 name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL);
243 cfg->bpf_name = name;
250 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
253 bpf_prog_put(cfg->filter);
255 bpf_prog_destroy(cfg->filter);
258 kfree(cfg->bpf_name);
261 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
262 struct tcf_bpf_cfg *cfg)
264 cfg->is_ebpf = tcf_bpf_is_ebpf(prog);
265 /* updates to prog->filter are prevented, since it's called either
266 * with rtnl lock or during final cleanup in rcu callback
268 cfg->filter = rcu_dereference_protected(prog->filter, 1);
270 cfg->bpf_ops = prog->bpf_ops;
271 cfg->bpf_name = prog->bpf_name;
274 static int tcf_bpf_init(struct net *net, struct nlattr *nla,
275 struct nlattr *est, struct tc_action **act,
276 int replace, int bind)
278 struct tc_action_net *tn = net_generic(net, bpf_net_id);
279 struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
280 struct tcf_bpf_cfg cfg, old;
281 struct tc_act_bpf *parm;
282 struct tcf_bpf *prog;
283 bool is_bpf, is_ebpf;
289 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy, NULL);
293 if (!tb[TCA_ACT_BPF_PARMS])
296 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
298 if (!tcf_hash_check(tn, parm->index, act, bind)) {
299 ret = tcf_hash_create(tn, parm->index, est, act,
300 &act_bpf_ops, bind, true);
306 /* Don't override defaults. */
310 tcf_hash_release(*act, bind);
315 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
316 is_ebpf = tb[TCA_ACT_BPF_FD];
318 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
323 memset(&cfg, 0, sizeof(cfg));
325 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
326 tcf_bpf_init_from_efd(tb, &cfg);
333 if (res != ACT_P_CREATED)
334 tcf_bpf_prog_fill_cfg(prog, &old);
336 prog->bpf_ops = cfg.bpf_ops;
337 prog->bpf_name = cfg.bpf_name;
340 prog->bpf_num_ops = cfg.bpf_num_ops;
342 prog->tcf_action = parm->action;
343 rcu_assign_pointer(prog->filter, cfg.filter);
345 if (res == ACT_P_CREATED) {
346 tcf_hash_insert(tn, *act);
348 /* make sure the program being replaced is no longer executing */
350 tcf_bpf_cfg_cleanup(&old);
355 if (res == ACT_P_CREATED)
356 tcf_hash_cleanup(*act, est);
361 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
363 struct tcf_bpf_cfg tmp;
365 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
366 tcf_bpf_cfg_cleanup(&tmp);
369 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
370 struct netlink_callback *cb, int type,
371 const struct tc_action_ops *ops)
373 struct tc_action_net *tn = net_generic(net, bpf_net_id);
375 return tcf_generic_walker(tn, skb, cb, type, ops);
378 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
380 struct tc_action_net *tn = net_generic(net, bpf_net_id);
382 return tcf_hash_search(tn, a, index);
385 static struct tc_action_ops act_bpf_ops __read_mostly = {
388 .owner = THIS_MODULE,
390 .dump = tcf_bpf_dump,
391 .cleanup = tcf_bpf_cleanup,
392 .init = tcf_bpf_init,
393 .walk = tcf_bpf_walker,
394 .lookup = tcf_bpf_search,
395 .size = sizeof(struct tcf_bpf),
398 static __net_init int bpf_init_net(struct net *net)
400 struct tc_action_net *tn = net_generic(net, bpf_net_id);
402 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
405 static void __net_exit bpf_exit_net(struct net *net)
407 struct tc_action_net *tn = net_generic(net, bpf_net_id);
409 tc_action_net_exit(tn);
412 static struct pernet_operations bpf_net_ops = {
413 .init = bpf_init_net,
414 .exit = bpf_exit_net,
416 .size = sizeof(struct tc_action_net),
419 static int __init bpf_init_module(void)
421 return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
424 static void __exit bpf_cleanup_module(void)
426 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
429 module_init(bpf_init_module);
430 module_exit(bpf_cleanup_module);
432 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
433 MODULE_DESCRIPTION("TC BPF based action");
434 MODULE_LICENSE("GPL v2");