1 /*
2 * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
3 *
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.
8 */
9
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>
17
18 #include <net/netlink.h>
19 #include <net/pkt_sched.h>
20
21 #include <linux/tc_act/tc_bpf.h>
22 #include <net/tc_act/tc_bpf.h>
23
24 #define BPF_TAB_MASK 15
25 #define ACT_BPF_NAME_LEN 256
26
27 struct tcf_bpf_cfg {
28 struct bpf_prog *filter;
29 struct sock_filter *bpf_ops;
30 const char *bpf_name;
31 u32 bpf_fd;
32 u16 bpf_num_ops;
33 bool is_ebpf;
34 };
35
36 static int bpf_net_id;
37 static struct tc_action_ops act_bpf_ops;
38
tcf_bpf(struct sk_buff * skb,const struct tc_action * act,struct tcf_result * res)39 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
40 struct tcf_result *res)
41 {
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;
46
47 tcf_lastuse_update(&prog->tcf_tm);
48 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb);
49
50 rcu_read_lock();
51 filter = rcu_dereference(prog->filter);
52 if (at_ingress) {
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);
57 } else {
58 bpf_compute_data_end(skb);
59 filter_res = BPF_PROG_RUN(filter, skb);
60 }
61 rcu_read_unlock();
62
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.
66 *
67 * In case a different well-known TC_ACT opcode has been
68 * returned, it will overwrite the default one.
69 *
70 * For everything else that is unkown, TC_ACT_UNSPEC is
71 * returned.
72 */
73 switch (filter_res) {
74 case TC_ACT_PIPE:
75 case TC_ACT_RECLASSIFY:
76 case TC_ACT_OK:
77 case TC_ACT_REDIRECT:
78 action = filter_res;
79 break;
80 case TC_ACT_SHOT:
81 action = filter_res;
82 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats));
83 break;
84 case TC_ACT_UNSPEC:
85 action = prog->tcf_action;
86 break;
87 default:
88 action = TC_ACT_UNSPEC;
89 break;
90 }
91
92 return action;
93 }
94
tcf_bpf_is_ebpf(const struct tcf_bpf * prog)95 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
96 {
97 return !prog->bpf_ops;
98 }
99
tcf_bpf_dump_bpf_info(const struct tcf_bpf * prog,struct sk_buff * skb)100 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
101 struct sk_buff *skb)
102 {
103 struct nlattr *nla;
104
105 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
106 return -EMSGSIZE;
107
108 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
109 sizeof(struct sock_filter));
110 if (nla == NULL)
111 return -EMSGSIZE;
112
113 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
114
115 return 0;
116 }
117
tcf_bpf_dump_ebpf_info(const struct tcf_bpf * prog,struct sk_buff * skb)118 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
119 struct sk_buff *skb)
120 {
121 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
122 return -EMSGSIZE;
123
124 if (prog->bpf_name &&
125 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
126 return -EMSGSIZE;
127
128 return 0;
129 }
130
tcf_bpf_dump(struct sk_buff * skb,struct tc_action * act,int bind,int ref)131 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
132 int bind, int ref)
133 {
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,
141 };
142 struct tcf_t tm;
143 int ret;
144
145 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
146 goto nla_put_failure;
147
148 if (tcf_bpf_is_ebpf(prog))
149 ret = tcf_bpf_dump_ebpf_info(prog, skb);
150 else
151 ret = tcf_bpf_dump_bpf_info(prog, skb);
152 if (ret)
153 goto nla_put_failure;
154
155 tcf_tm_dump(&tm, &prog->tcf_tm);
156 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm,
157 TCA_ACT_BPF_PAD))
158 goto nla_put_failure;
159
160 return skb->len;
161
162 nla_put_failure:
163 nlmsg_trim(skb, tp);
164 return -1;
165 }
166
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 },
175 };
176
tcf_bpf_init_from_ops(struct nlattr ** tb,struct tcf_bpf_cfg * cfg)177 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
178 {
179 struct sock_filter *bpf_ops;
180 struct sock_fprog_kern fprog_tmp;
181 struct bpf_prog *fp;
182 u16 bpf_size, bpf_num_ops;
183 int ret;
184
185 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
186 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
187 return -EINVAL;
188
189 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
190 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
191 return -EINVAL;
192
193 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
194 if (bpf_ops == NULL)
195 return -ENOMEM;
196
197 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
198
199 fprog_tmp.len = bpf_num_ops;
200 fprog_tmp.filter = bpf_ops;
201
202 ret = bpf_prog_create(&fp, &fprog_tmp);
203 if (ret < 0) {
204 kfree(bpf_ops);
205 return ret;
206 }
207
208 cfg->bpf_ops = bpf_ops;
209 cfg->bpf_num_ops = bpf_num_ops;
210 cfg->filter = fp;
211 cfg->is_ebpf = false;
212
213 return 0;
214 }
215
tcf_bpf_init_from_efd(struct nlattr ** tb,struct tcf_bpf_cfg * cfg)216 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
217 {
218 struct bpf_prog *fp;
219 char *name = NULL;
220 u32 bpf_fd;
221
222 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
223
224 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT);
225 if (IS_ERR(fp))
226 return PTR_ERR(fp);
227
228 if (tb[TCA_ACT_BPF_NAME]) {
229 name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
230 nla_len(tb[TCA_ACT_BPF_NAME]),
231 GFP_KERNEL);
232 if (!name) {
233 bpf_prog_put(fp);
234 return -ENOMEM;
235 }
236 }
237
238 cfg->bpf_fd = bpf_fd;
239 cfg->bpf_name = name;
240 cfg->filter = fp;
241 cfg->is_ebpf = true;
242
243 return 0;
244 }
245
tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg * cfg)246 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg)
247 {
248 struct bpf_prog *filter = cfg->filter;
249
250 if (filter) {
251 if (cfg->is_ebpf)
252 bpf_prog_put(filter);
253 else
254 bpf_prog_destroy(filter);
255 }
256
257 kfree(cfg->bpf_ops);
258 kfree(cfg->bpf_name);
259 }
260
tcf_bpf_prog_fill_cfg(const struct tcf_bpf * prog,struct tcf_bpf_cfg * cfg)261 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog,
262 struct tcf_bpf_cfg *cfg)
263 {
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
267 */
268 cfg->filter = rcu_dereference_protected(prog->filter, 1);
269
270 cfg->bpf_ops = prog->bpf_ops;
271 cfg->bpf_name = prog->bpf_name;
272 }
273
tcf_bpf_init(struct net * net,struct nlattr * nla,struct nlattr * est,struct tc_action ** act,int replace,int bind)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)
277 {
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;
284 int ret, res = 0;
285
286 if (!nla)
287 return -EINVAL;
288
289 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
290 if (ret < 0)
291 return ret;
292
293 if (!tb[TCA_ACT_BPF_PARMS])
294 return -EINVAL;
295
296 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
297
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);
301 if (ret < 0)
302 return ret;
303
304 res = ACT_P_CREATED;
305 } else {
306 /* Don't override defaults. */
307 if (bind)
308 return 0;
309
310 tcf_hash_release(*act, bind);
311 if (!replace)
312 return -EEXIST;
313 }
314
315 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
316 is_ebpf = tb[TCA_ACT_BPF_FD];
317
318 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) {
319 ret = -EINVAL;
320 goto out;
321 }
322
323 memset(&cfg, 0, sizeof(cfg));
324
325 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
326 tcf_bpf_init_from_efd(tb, &cfg);
327 if (ret < 0)
328 goto out;
329
330 prog = to_bpf(*act);
331 ASSERT_RTNL();
332
333 if (res != ACT_P_CREATED)
334 tcf_bpf_prog_fill_cfg(prog, &old);
335
336 prog->bpf_ops = cfg.bpf_ops;
337 prog->bpf_name = cfg.bpf_name;
338
339 if (cfg.bpf_num_ops)
340 prog->bpf_num_ops = cfg.bpf_num_ops;
341 if (cfg.bpf_fd)
342 prog->bpf_fd = cfg.bpf_fd;
343
344 prog->tcf_action = parm->action;
345 rcu_assign_pointer(prog->filter, cfg.filter);
346
347 if (res == ACT_P_CREATED) {
348 tcf_hash_insert(tn, *act);
349 } else {
350 /* make sure the program being replaced is no longer executing */
351 synchronize_rcu();
352 tcf_bpf_cfg_cleanup(&old);
353 }
354
355 return res;
356 out:
357 if (res == ACT_P_CREATED)
358 tcf_hash_cleanup(*act, est);
359
360 return ret;
361 }
362
tcf_bpf_cleanup(struct tc_action * act,int bind)363 static void tcf_bpf_cleanup(struct tc_action *act, int bind)
364 {
365 struct tcf_bpf_cfg tmp;
366
367 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp);
368 tcf_bpf_cfg_cleanup(&tmp);
369 }
370
tcf_bpf_walker(struct net * net,struct sk_buff * skb,struct netlink_callback * cb,int type,const struct tc_action_ops * ops)371 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb,
372 struct netlink_callback *cb, int type,
373 const struct tc_action_ops *ops)
374 {
375 struct tc_action_net *tn = net_generic(net, bpf_net_id);
376
377 return tcf_generic_walker(tn, skb, cb, type, ops);
378 }
379
tcf_bpf_search(struct net * net,struct tc_action ** a,u32 index)380 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index)
381 {
382 struct tc_action_net *tn = net_generic(net, bpf_net_id);
383
384 return tcf_hash_search(tn, a, index);
385 }
386
387 static struct tc_action_ops act_bpf_ops __read_mostly = {
388 .kind = "bpf",
389 .type = TCA_ACT_BPF,
390 .owner = THIS_MODULE,
391 .act = tcf_bpf,
392 .dump = tcf_bpf_dump,
393 .cleanup = tcf_bpf_cleanup,
394 .init = tcf_bpf_init,
395 .walk = tcf_bpf_walker,
396 .lookup = tcf_bpf_search,
397 .size = sizeof(struct tcf_bpf),
398 };
399
bpf_init_net(struct net * net)400 static __net_init int bpf_init_net(struct net *net)
401 {
402 struct tc_action_net *tn = net_generic(net, bpf_net_id);
403
404 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK);
405 }
406
bpf_exit_net(struct net * net)407 static void __net_exit bpf_exit_net(struct net *net)
408 {
409 struct tc_action_net *tn = net_generic(net, bpf_net_id);
410
411 tc_action_net_exit(tn);
412 }
413
414 static struct pernet_operations bpf_net_ops = {
415 .init = bpf_init_net,
416 .exit = bpf_exit_net,
417 .id = &bpf_net_id,
418 .size = sizeof(struct tc_action_net),
419 };
420
bpf_init_module(void)421 static int __init bpf_init_module(void)
422 {
423 return tcf_register_action(&act_bpf_ops, &bpf_net_ops);
424 }
425
bpf_cleanup_module(void)426 static void __exit bpf_cleanup_module(void)
427 {
428 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops);
429 }
430
431 module_init(bpf_init_module);
432 module_exit(bpf_cleanup_module);
433
434 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
435 MODULE_DESCRIPTION("TC BPF based action");
436 MODULE_LICENSE("GPL v2");
437