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 tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
37 struct tcf_result *res)
39 struct tcf_bpf *prog = act->priv;
40 struct bpf_prog *filter;
41 int action, filter_res;
42 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
44 if (unlikely(!skb_mac_header_was_set(skb)))
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 filter_res = BPF_PROG_RUN(filter, skb);
55 __skb_pull(skb, skb->mac_len);
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:
79 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
82 action = prog->tcf_action;
85 action = TC_ACT_UNSPEC;
92 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
94 return !prog->bpf_ops;
97 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
102 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
105 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
106 sizeof(struct sock_filter));
110 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
115 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
118 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
121 if (prog->bpf_name &&
122 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
128 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
131 unsigned char *tp = skb_tail_pointer(skb);
132 struct tcf_bpf *prog = act->priv;
133 struct tc_act_bpf opt = {
134 .index = prog->tcf_index,
135 .refcnt = prog->tcf_refcnt - ref,
136 .bindcnt = prog->tcf_bindcnt - bind,
137 .action = prog->tcf_action,
142 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
143 goto nla_put_failure;
145 if (tcf_bpf_is_ebpf(prog))
146 ret = tcf_bpf_dump_ebpf_info(prog, skb);
148 ret = tcf_bpf_dump_bpf_info(prog, skb);
150 goto nla_put_failure;
152 tm.install = jiffies_to_clock_t(jiffies - prog->tcf_tm.install);
153 tm.lastuse = jiffies_to_clock_t(jiffies - prog->tcf_tm.lastuse);
154 tm.expires = jiffies_to_clock_t(prog->tcf_tm.expires);
156 if (nla_put(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm))
157 goto nla_put_failure;
166 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
167 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
168 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
169 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, .len = ACT_BPF_NAME_LEN },
170 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
171 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
172 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
175 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
177 struct sock_filter *bpf_ops;
178 struct sock_fprog_kern fprog_tmp;
180 u16 bpf_size, bpf_num_ops;
183 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
184 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
187 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
188 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
191 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
195 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
197 fprog_tmp.len = bpf_num_ops;
198 fprog_tmp.filter = bpf_ops;
200 ret = bpf_prog_create(&fp, &fprog_tmp);
206 cfg->bpf_ops = bpf_ops;
207 cfg->bpf_num_ops = bpf_num_ops;
209 cfg->is_ebpf = false;
214 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
220 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
222 fp = bpf_prog_get(bpf_fd);
226 if (fp->type != 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 nlattr *tb[TCA_ACT_BPF_MAX + 1];
278 struct tcf_bpf_cfg cfg, old;
279 struct tc_act_bpf *parm;
280 struct tcf_bpf *prog;
281 bool is_bpf, is_ebpf;
287 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
291 if (!tb[TCA_ACT_BPF_PARMS])
294 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
296 if (!tcf_hash_check(parm->index, act, bind)) {
297 ret = tcf_hash_create(parm->index, est, act,
298 sizeof(*prog), bind, true);
304 /* Don't override defaults. */
308 tcf_hash_release(act, bind);
313 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
314 is_ebpf = tb[TCA_ACT_BPF_FD];
316 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
321 memset(&cfg, 0, sizeof(cfg));
323 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
324 tcf_bpf_init_from_efd(tb, &cfg);
331 if (res != ACT_P_CREATED)
332 tcf_bpf_prog_fill_cfg(prog, &old);
334 prog->bpf_ops = cfg.bpf_ops;
335 prog->bpf_name = cfg.bpf_name;
338 prog->bpf_num_ops = cfg.bpf_num_ops;
340 prog->bpf_fd = cfg.bpf_fd;
342 prog->tcf_action = parm->action;
343 rcu_assign_pointer(prog->filter, cfg.filter);
345 if (res == ACT_P_CREATED) {
346 tcf_hash_insert(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(act->priv, &tmp);
366 tcf_bpf_cfg_cleanup(&tmp);
369 static struct tc_action_ops act_bpf_ops __read_mostly = {
372 .owner = THIS_MODULE,
374 .dump = tcf_bpf_dump,
375 .cleanup = tcf_bpf_cleanup,
376 .init = tcf_bpf_init,
379 static int __init bpf_init_module(void)
381 return tcf_register_action(&act_bpf_ops, BPF_TAB_MASK);
384 static void __exit bpf_cleanup_module(void)
386 tcf_unregister_action(&act_bpf_ops);
389 module_init(bpf_init_module);
390 module_exit(bpf_cleanup_module);
392 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
393 MODULE_DESCRIPTION("TC BPF based action");
394 MODULE_LICENSE("GPL v2");