]>
Commit | Line | Data |
---|---|---|
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 ACT_BPF_NAME_LEN 256 | |
25 | ||
26 | struct tcf_bpf_cfg { | |
27 | struct bpf_prog *filter; | |
28 | struct sock_filter *bpf_ops; | |
29 | const char *bpf_name; | |
30 | u16 bpf_num_ops; | |
31 | bool is_ebpf; | |
32 | }; | |
33 | ||
34 | static unsigned int bpf_net_id; | |
35 | static struct tc_action_ops act_bpf_ops; | |
36 | ||
37 | static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act, | |
38 | struct tcf_result *res) | |
39 | { | |
40 | bool at_ingress = skb_at_tc_ingress(skb); | |
41 | struct tcf_bpf *prog = to_bpf(act); | |
42 | struct bpf_prog *filter; | |
43 | int action, filter_res; | |
44 | ||
45 | tcf_lastuse_update(&prog->tcf_tm); | |
46 | bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb); | |
47 | ||
48 | rcu_read_lock(); | |
49 | filter = rcu_dereference(prog->filter); | |
50 | if (at_ingress) { | |
51 | __skb_push(skb, skb->mac_len); | |
52 | bpf_compute_data_end(skb); | |
53 | filter_res = BPF_PROG_RUN(filter, skb); | |
54 | __skb_pull(skb, skb->mac_len); | |
55 | } else { | |
56 | bpf_compute_data_end(skb); | |
57 | filter_res = BPF_PROG_RUN(filter, skb); | |
58 | } | |
59 | rcu_read_unlock(); | |
60 | ||
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. | |
64 | * | |
65 | * In case a different well-known TC_ACT opcode has been | |
66 | * returned, it will overwrite the default one. | |
67 | * | |
68 | * For everything else that is unkown, TC_ACT_UNSPEC is | |
69 | * returned. | |
70 | */ | |
71 | switch (filter_res) { | |
72 | case TC_ACT_PIPE: | |
73 | case TC_ACT_RECLASSIFY: | |
74 | case TC_ACT_OK: | |
75 | case TC_ACT_REDIRECT: | |
76 | action = filter_res; | |
77 | break; | |
78 | case TC_ACT_SHOT: | |
79 | action = filter_res; | |
80 | qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats)); | |
81 | break; | |
82 | case TC_ACT_UNSPEC: | |
83 | action = prog->tcf_action; | |
84 | break; | |
85 | default: | |
86 | action = TC_ACT_UNSPEC; | |
87 | break; | |
88 | } | |
89 | ||
90 | return action; | |
91 | } | |
92 | ||
93 | static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog) | |
94 | { | |
95 | return !prog->bpf_ops; | |
96 | } | |
97 | ||
98 | static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog, | |
99 | struct sk_buff *skb) | |
100 | { | |
101 | struct nlattr *nla; | |
102 | ||
103 | if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops)) | |
104 | return -EMSGSIZE; | |
105 | ||
106 | nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops * | |
107 | sizeof(struct sock_filter)); | |
108 | if (nla == NULL) | |
109 | return -EMSGSIZE; | |
110 | ||
111 | memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla)); | |
112 | ||
113 | return 0; | |
114 | } | |
115 | ||
116 | static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog, | |
117 | struct sk_buff *skb) | |
118 | { | |
119 | struct nlattr *nla; | |
120 | ||
121 | if (prog->bpf_name && | |
122 | nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name)) | |
123 | return -EMSGSIZE; | |
124 | ||
125 | if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id)) | |
126 | return -EMSGSIZE; | |
127 | ||
128 | nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag)); | |
129 | if (nla == NULL) | |
130 | return -EMSGSIZE; | |
131 | ||
132 | memcpy(nla_data(nla), prog->filter->tag, nla_len(nla)); | |
133 | ||
134 | return 0; | |
135 | } | |
136 | ||
137 | static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act, | |
138 | int bind, int ref) | |
139 | { | |
140 | unsigned char *tp = skb_tail_pointer(skb); | |
141 | struct tcf_bpf *prog = to_bpf(act); | |
142 | struct tc_act_bpf opt = { | |
143 | .index = prog->tcf_index, | |
144 | .refcnt = prog->tcf_refcnt - ref, | |
145 | .bindcnt = prog->tcf_bindcnt - bind, | |
146 | .action = prog->tcf_action, | |
147 | }; | |
148 | struct tcf_t tm; | |
149 | int ret; | |
150 | ||
151 | if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt)) | |
152 | goto nla_put_failure; | |
153 | ||
154 | if (tcf_bpf_is_ebpf(prog)) | |
155 | ret = tcf_bpf_dump_ebpf_info(prog, skb); | |
156 | else | |
157 | ret = tcf_bpf_dump_bpf_info(prog, skb); | |
158 | if (ret) | |
159 | goto nla_put_failure; | |
160 | ||
161 | tcf_tm_dump(&tm, &prog->tcf_tm); | |
162 | if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm, | |
163 | TCA_ACT_BPF_PAD)) | |
164 | goto nla_put_failure; | |
165 | ||
166 | return skb->len; | |
167 | ||
168 | nla_put_failure: | |
169 | nlmsg_trim(skb, tp); | |
170 | return -1; | |
171 | } | |
172 | ||
173 | static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = { | |
174 | [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) }, | |
175 | [TCA_ACT_BPF_FD] = { .type = NLA_U32 }, | |
176 | [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, | |
177 | .len = ACT_BPF_NAME_LEN }, | |
178 | [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 }, | |
179 | [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY, | |
180 | .len = sizeof(struct sock_filter) * BPF_MAXINSNS }, | |
181 | }; | |
182 | ||
183 | static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg) | |
184 | { | |
185 | struct sock_filter *bpf_ops; | |
186 | struct sock_fprog_kern fprog_tmp; | |
187 | struct bpf_prog *fp; | |
188 | u16 bpf_size, bpf_num_ops; | |
189 | int ret; | |
190 | ||
191 | bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]); | |
192 | if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0) | |
193 | return -EINVAL; | |
194 | ||
195 | bpf_size = bpf_num_ops * sizeof(*bpf_ops); | |
196 | if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS])) | |
197 | return -EINVAL; | |
198 | ||
199 | bpf_ops = kzalloc(bpf_size, GFP_KERNEL); | |
200 | if (bpf_ops == NULL) | |
201 | return -ENOMEM; | |
202 | ||
203 | memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size); | |
204 | ||
205 | fprog_tmp.len = bpf_num_ops; | |
206 | fprog_tmp.filter = bpf_ops; | |
207 | ||
208 | ret = bpf_prog_create(&fp, &fprog_tmp); | |
209 | if (ret < 0) { | |
210 | kfree(bpf_ops); | |
211 | return ret; | |
212 | } | |
213 | ||
214 | cfg->bpf_ops = bpf_ops; | |
215 | cfg->bpf_num_ops = bpf_num_ops; | |
216 | cfg->filter = fp; | |
217 | cfg->is_ebpf = false; | |
218 | ||
219 | return 0; | |
220 | } | |
221 | ||
222 | static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg) | |
223 | { | |
224 | struct bpf_prog *fp; | |
225 | char *name = NULL; | |
226 | u32 bpf_fd; | |
227 | ||
228 | bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]); | |
229 | ||
230 | fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT); | |
231 | if (IS_ERR(fp)) | |
232 | return PTR_ERR(fp); | |
233 | ||
234 | if (tb[TCA_ACT_BPF_NAME]) { | |
235 | name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL); | |
236 | if (!name) { | |
237 | bpf_prog_put(fp); | |
238 | return -ENOMEM; | |
239 | } | |
240 | } | |
241 | ||
242 | cfg->bpf_name = name; | |
243 | cfg->filter = fp; | |
244 | cfg->is_ebpf = true; | |
245 | ||
246 | return 0; | |
247 | } | |
248 | ||
249 | static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg) | |
250 | { | |
251 | if (cfg->is_ebpf) | |
252 | bpf_prog_put(cfg->filter); | |
253 | else | |
254 | bpf_prog_destroy(cfg->filter); | |
255 | ||
256 | kfree(cfg->bpf_ops); | |
257 | kfree(cfg->bpf_name); | |
258 | } | |
259 | ||
260 | static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog, | |
261 | struct tcf_bpf_cfg *cfg) | |
262 | { | |
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 | |
266 | */ | |
267 | cfg->filter = rcu_dereference_protected(prog->filter, 1); | |
268 | ||
269 | cfg->bpf_ops = prog->bpf_ops; | |
270 | cfg->bpf_name = prog->bpf_name; | |
271 | } | |
272 | ||
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) | |
276 | { | |
277 | struct tc_action_net *tn = net_generic(net, bpf_net_id); | |
278 | struct nlattr *tb[TCA_ACT_BPF_MAX + 1]; | |
279 | struct tcf_bpf_cfg cfg, old; | |
280 | struct tc_act_bpf *parm; | |
281 | struct tcf_bpf *prog; | |
282 | bool is_bpf, is_ebpf; | |
283 | int ret, res = 0; | |
284 | ||
285 | if (!nla) | |
286 | return -EINVAL; | |
287 | ||
288 | ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy, NULL); | |
289 | if (ret < 0) | |
290 | return ret; | |
291 | ||
292 | if (!tb[TCA_ACT_BPF_PARMS]) | |
293 | return -EINVAL; | |
294 | ||
295 | parm = nla_data(tb[TCA_ACT_BPF_PARMS]); | |
296 | ||
297 | if (!tcf_idr_check(tn, parm->index, act, bind)) { | |
298 | ret = tcf_idr_create(tn, parm->index, est, act, | |
299 | &act_bpf_ops, bind, true); | |
300 | if (ret < 0) | |
301 | return ret; | |
302 | ||
303 | res = ACT_P_CREATED; | |
304 | } else { | |
305 | /* Don't override defaults. */ | |
306 | if (bind) | |
307 | return 0; | |
308 | ||
309 | tcf_idr_release(*act, bind); | |
310 | if (!replace) | |
311 | return -EEXIST; | |
312 | } | |
313 | ||
314 | is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS]; | |
315 | is_ebpf = tb[TCA_ACT_BPF_FD]; | |
316 | ||
317 | if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) { | |
318 | ret = -EINVAL; | |
319 | goto out; | |
320 | } | |
321 | ||
322 | memset(&cfg, 0, sizeof(cfg)); | |
323 | ||
324 | ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) : | |
325 | tcf_bpf_init_from_efd(tb, &cfg); | |
326 | if (ret < 0) | |
327 | goto out; | |
328 | ||
329 | prog = to_bpf(*act); | |
330 | ASSERT_RTNL(); | |
331 | ||
332 | if (res != ACT_P_CREATED) | |
333 | tcf_bpf_prog_fill_cfg(prog, &old); | |
334 | ||
335 | prog->bpf_ops = cfg.bpf_ops; | |
336 | prog->bpf_name = cfg.bpf_name; | |
337 | ||
338 | if (cfg.bpf_num_ops) | |
339 | prog->bpf_num_ops = cfg.bpf_num_ops; | |
340 | ||
341 | prog->tcf_action = parm->action; | |
342 | rcu_assign_pointer(prog->filter, cfg.filter); | |
343 | ||
344 | if (res == ACT_P_CREATED) { | |
345 | tcf_idr_insert(tn, *act); | |
346 | } else { | |
347 | /* make sure the program being replaced is no longer executing */ | |
348 | synchronize_rcu(); | |
349 | tcf_bpf_cfg_cleanup(&old); | |
350 | } | |
351 | ||
352 | return res; | |
353 | out: | |
354 | if (res == ACT_P_CREATED) | |
355 | tcf_idr_cleanup(*act, est); | |
356 | ||
357 | return ret; | |
358 | } | |
359 | ||
360 | static void tcf_bpf_cleanup(struct tc_action *act, int bind) | |
361 | { | |
362 | struct tcf_bpf_cfg tmp; | |
363 | ||
364 | tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp); | |
365 | tcf_bpf_cfg_cleanup(&tmp); | |
366 | } | |
367 | ||
368 | static int tcf_bpf_walker(struct net *net, struct sk_buff *skb, | |
369 | struct netlink_callback *cb, int type, | |
370 | const struct tc_action_ops *ops) | |
371 | { | |
372 | struct tc_action_net *tn = net_generic(net, bpf_net_id); | |
373 | ||
374 | return tcf_generic_walker(tn, skb, cb, type, ops); | |
375 | } | |
376 | ||
377 | static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index) | |
378 | { | |
379 | struct tc_action_net *tn = net_generic(net, bpf_net_id); | |
380 | ||
381 | return tcf_idr_search(tn, a, index); | |
382 | } | |
383 | ||
384 | static struct tc_action_ops act_bpf_ops __read_mostly = { | |
385 | .kind = "bpf", | |
386 | .type = TCA_ACT_BPF, | |
387 | .owner = THIS_MODULE, | |
388 | .act = tcf_bpf, | |
389 | .dump = tcf_bpf_dump, | |
390 | .cleanup = tcf_bpf_cleanup, | |
391 | .init = tcf_bpf_init, | |
392 | .walk = tcf_bpf_walker, | |
393 | .lookup = tcf_bpf_search, | |
394 | .size = sizeof(struct tcf_bpf), | |
395 | }; | |
396 | ||
397 | static __net_init int bpf_init_net(struct net *net) | |
398 | { | |
399 | struct tc_action_net *tn = net_generic(net, bpf_net_id); | |
400 | ||
401 | return tc_action_net_init(tn, &act_bpf_ops); | |
402 | } | |
403 | ||
404 | static void __net_exit bpf_exit_net(struct net *net) | |
405 | { | |
406 | struct tc_action_net *tn = net_generic(net, bpf_net_id); | |
407 | ||
408 | tc_action_net_exit(tn); | |
409 | } | |
410 | ||
411 | static struct pernet_operations bpf_net_ops = { | |
412 | .init = bpf_init_net, | |
413 | .exit = bpf_exit_net, | |
414 | .id = &bpf_net_id, | |
415 | .size = sizeof(struct tc_action_net), | |
416 | }; | |
417 | ||
418 | static int __init bpf_init_module(void) | |
419 | { | |
420 | return tcf_register_action(&act_bpf_ops, &bpf_net_ops); | |
421 | } | |
422 | ||
423 | static void __exit bpf_cleanup_module(void) | |
424 | { | |
425 | tcf_unregister_action(&act_bpf_ops, &bpf_net_ops); | |
426 | } | |
427 | ||
428 | module_init(bpf_init_module); | |
429 | module_exit(bpf_cleanup_module); | |
430 | ||
431 | MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>"); | |
432 | MODULE_DESCRIPTION("TC BPF based action"); | |
433 | MODULE_LICENSE("GPL v2"); |