1 | // SPDX-License-Identifier: GPL-2.0-or-later |
2 | /* |
3 | * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us> |
4 | */ |
5 | |
6 | #include <linux/module.h> |
7 | #include <linux/init.h> |
8 | #include <linux/kernel.h> |
9 | #include <linux/skbuff.h> |
10 | #include <linux/rtnetlink.h> |
11 | #include <linux/filter.h> |
12 | #include <linux/bpf.h> |
13 | |
14 | #include <net/netlink.h> |
15 | #include <net/sock.h> |
16 | #include <net/pkt_sched.h> |
17 | #include <net/pkt_cls.h> |
18 | |
19 | #include <linux/tc_act/tc_bpf.h> |
20 | #include <net/tc_act/tc_bpf.h> |
21 | #include <net/tc_wrapper.h> |
22 | |
23 | #define ACT_BPF_NAME_LEN 256 |
24 | |
25 | struct tcf_bpf_cfg { |
26 | struct bpf_prog *filter; |
27 | struct sock_filter *bpf_ops; |
28 | const char *bpf_name; |
29 | u16 bpf_num_ops; |
30 | bool is_ebpf; |
31 | }; |
32 | |
33 | static struct tc_action_ops act_bpf_ops; |
34 | |
35 | TC_INDIRECT_SCOPE int tcf_bpf_act(struct sk_buff *skb, |
36 | const struct tc_action *act, |
37 | struct tcf_result *res) |
38 | { |
39 | bool at_ingress = skb_at_tc_ingress(skb); |
40 | struct tcf_bpf *prog = to_bpf(act); |
41 | struct bpf_prog *filter; |
42 | int action, filter_res; |
43 | |
44 | tcf_lastuse_update(tm: &prog->tcf_tm); |
45 | bstats_update(this_cpu_ptr(prog->common.cpu_bstats), skb); |
46 | |
47 | filter = rcu_dereference(prog->filter); |
48 | if (at_ingress) { |
49 | __skb_push(skb, len: skb->mac_len); |
50 | bpf_compute_data_pointers(skb); |
51 | filter_res = bpf_prog_run(prog: filter, ctx: skb); |
52 | __skb_pull(skb, len: skb->mac_len); |
53 | } else { |
54 | bpf_compute_data_pointers(skb); |
55 | filter_res = bpf_prog_run(prog: filter, ctx: skb); |
56 | } |
57 | if (unlikely(!skb->tstamp && skb->mono_delivery_time)) |
58 | skb->mono_delivery_time = 0; |
59 | if (skb_sk_is_prefetched(skb) && filter_res != TC_ACT_OK) |
60 | skb_orphan(skb); |
61 | |
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. |
65 | * |
66 | * In case a different well-known TC_ACT opcode has been |
67 | * returned, it will overwrite the default one. |
68 | * |
69 | * For everything else that is unknown, TC_ACT_UNSPEC is |
70 | * returned. |
71 | */ |
72 | switch (filter_res) { |
73 | case TC_ACT_PIPE: |
74 | case TC_ACT_RECLASSIFY: |
75 | case TC_ACT_OK: |
76 | case TC_ACT_REDIRECT: |
77 | action = filter_res; |
78 | break; |
79 | case TC_ACT_SHOT: |
80 | action = filter_res; |
81 | qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats)); |
82 | break; |
83 | case TC_ACT_UNSPEC: |
84 | action = prog->tcf_action; |
85 | break; |
86 | default: |
87 | action = TC_ACT_UNSPEC; |
88 | break; |
89 | } |
90 | |
91 | return action; |
92 | } |
93 | |
94 | static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog) |
95 | { |
96 | return !prog->bpf_ops; |
97 | } |
98 | |
99 | static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog, |
100 | struct sk_buff *skb) |
101 | { |
102 | struct nlattr *nla; |
103 | |
104 | if (nla_put_u16(skb, attrtype: TCA_ACT_BPF_OPS_LEN, value: prog->bpf_num_ops)) |
105 | return -EMSGSIZE; |
106 | |
107 | nla = nla_reserve(skb, attrtype: TCA_ACT_BPF_OPS, attrlen: prog->bpf_num_ops * |
108 | sizeof(struct sock_filter)); |
109 | if (nla == NULL) |
110 | return -EMSGSIZE; |
111 | |
112 | memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla)); |
113 | |
114 | return 0; |
115 | } |
116 | |
117 | static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog, |
118 | struct sk_buff *skb) |
119 | { |
120 | struct nlattr *nla; |
121 | |
122 | if (prog->bpf_name && |
123 | nla_put_string(skb, attrtype: TCA_ACT_BPF_NAME, str: prog->bpf_name)) |
124 | return -EMSGSIZE; |
125 | |
126 | if (nla_put_u32(skb, attrtype: TCA_ACT_BPF_ID, value: prog->filter->aux->id)) |
127 | return -EMSGSIZE; |
128 | |
129 | nla = nla_reserve(skb, attrtype: TCA_ACT_BPF_TAG, attrlen: sizeof(prog->filter->tag)); |
130 | if (nla == NULL) |
131 | return -EMSGSIZE; |
132 | |
133 | memcpy(nla_data(nla), prog->filter->tag, nla_len(nla)); |
134 | |
135 | return 0; |
136 | } |
137 | |
138 | static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act, |
139 | int bind, int ref) |
140 | { |
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 = refcount_read(r: &prog->tcf_refcnt) - ref, |
146 | .bindcnt = atomic_read(v: &prog->tcf_bindcnt) - bind, |
147 | }; |
148 | struct tcf_t tm; |
149 | int ret; |
150 | |
151 | spin_lock_bh(lock: &prog->tcf_lock); |
152 | opt.action = prog->tcf_action; |
153 | if (nla_put(skb, attrtype: TCA_ACT_BPF_PARMS, attrlen: sizeof(opt), data: &opt)) |
154 | goto nla_put_failure; |
155 | |
156 | if (tcf_bpf_is_ebpf(prog)) |
157 | ret = tcf_bpf_dump_ebpf_info(prog, skb); |
158 | else |
159 | ret = tcf_bpf_dump_bpf_info(prog, skb); |
160 | if (ret) |
161 | goto nla_put_failure; |
162 | |
163 | tcf_tm_dump(dtm: &tm, stm: &prog->tcf_tm); |
164 | if (nla_put_64bit(skb, attrtype: TCA_ACT_BPF_TM, attrlen: sizeof(tm), data: &tm, |
165 | padattr: TCA_ACT_BPF_PAD)) |
166 | goto nla_put_failure; |
167 | |
168 | spin_unlock_bh(lock: &prog->tcf_lock); |
169 | return skb->len; |
170 | |
171 | nla_put_failure: |
172 | spin_unlock_bh(lock: &prog->tcf_lock); |
173 | nlmsg_trim(skb, mark: tp); |
174 | return -1; |
175 | } |
176 | |
177 | static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = { |
178 | [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) }, |
179 | [TCA_ACT_BPF_FD] = { .type = NLA_U32 }, |
180 | [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, |
181 | .len = ACT_BPF_NAME_LEN }, |
182 | [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 }, |
183 | [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY, |
184 | .len = sizeof(struct sock_filter) * BPF_MAXINSNS }, |
185 | }; |
186 | |
187 | static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg) |
188 | { |
189 | struct sock_filter *bpf_ops; |
190 | struct sock_fprog_kern fprog_tmp; |
191 | struct bpf_prog *fp; |
192 | u16 bpf_size, bpf_num_ops; |
193 | int ret; |
194 | |
195 | bpf_num_ops = nla_get_u16(nla: tb[TCA_ACT_BPF_OPS_LEN]); |
196 | if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0) |
197 | return -EINVAL; |
198 | |
199 | bpf_size = bpf_num_ops * sizeof(*bpf_ops); |
200 | if (bpf_size != nla_len(nla: tb[TCA_ACT_BPF_OPS])) |
201 | return -EINVAL; |
202 | |
203 | bpf_ops = kmemdup(p: nla_data(nla: tb[TCA_ACT_BPF_OPS]), size: bpf_size, GFP_KERNEL); |
204 | if (bpf_ops == NULL) |
205 | return -ENOMEM; |
206 | |
207 | fprog_tmp.len = bpf_num_ops; |
208 | fprog_tmp.filter = bpf_ops; |
209 | |
210 | ret = bpf_prog_create(pfp: &fp, fprog: &fprog_tmp); |
211 | if (ret < 0) { |
212 | kfree(objp: bpf_ops); |
213 | return ret; |
214 | } |
215 | |
216 | cfg->bpf_ops = bpf_ops; |
217 | cfg->bpf_num_ops = bpf_num_ops; |
218 | cfg->filter = fp; |
219 | cfg->is_ebpf = false; |
220 | |
221 | return 0; |
222 | } |
223 | |
224 | static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg) |
225 | { |
226 | struct bpf_prog *fp; |
227 | char *name = NULL; |
228 | u32 bpf_fd; |
229 | |
230 | bpf_fd = nla_get_u32(nla: tb[TCA_ACT_BPF_FD]); |
231 | |
232 | fp = bpf_prog_get_type(ufd: bpf_fd, type: BPF_PROG_TYPE_SCHED_ACT); |
233 | if (IS_ERR(ptr: fp)) |
234 | return PTR_ERR(ptr: fp); |
235 | |
236 | if (tb[TCA_ACT_BPF_NAME]) { |
237 | name = nla_memdup(src: tb[TCA_ACT_BPF_NAME], GFP_KERNEL); |
238 | if (!name) { |
239 | bpf_prog_put(prog: fp); |
240 | return -ENOMEM; |
241 | } |
242 | } |
243 | |
244 | cfg->bpf_name = name; |
245 | cfg->filter = fp; |
246 | cfg->is_ebpf = true; |
247 | |
248 | return 0; |
249 | } |
250 | |
251 | static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg) |
252 | { |
253 | struct bpf_prog *filter = cfg->filter; |
254 | |
255 | if (filter) { |
256 | if (cfg->is_ebpf) |
257 | bpf_prog_put(prog: filter); |
258 | else |
259 | bpf_prog_destroy(fp: filter); |
260 | } |
261 | |
262 | kfree(objp: cfg->bpf_ops); |
263 | kfree(objp: cfg->bpf_name); |
264 | } |
265 | |
266 | static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog, |
267 | struct tcf_bpf_cfg *cfg) |
268 | { |
269 | cfg->is_ebpf = tcf_bpf_is_ebpf(prog); |
270 | /* updates to prog->filter are prevented, since it's called either |
271 | * with tcf lock or during final cleanup in rcu callback |
272 | */ |
273 | cfg->filter = rcu_dereference_protected(prog->filter, 1); |
274 | |
275 | cfg->bpf_ops = prog->bpf_ops; |
276 | cfg->bpf_name = prog->bpf_name; |
277 | } |
278 | |
279 | static int tcf_bpf_init(struct net *net, struct nlattr *nla, |
280 | struct nlattr *est, struct tc_action **act, |
281 | struct tcf_proto *tp, u32 flags, |
282 | struct netlink_ext_ack *extack) |
283 | { |
284 | struct tc_action_net *tn = net_generic(net, id: act_bpf_ops.net_id); |
285 | bool bind = flags & TCA_ACT_FLAGS_BIND; |
286 | struct nlattr *tb[TCA_ACT_BPF_MAX + 1]; |
287 | struct tcf_chain *goto_ch = NULL; |
288 | struct tcf_bpf_cfg cfg, old; |
289 | struct tc_act_bpf *parm; |
290 | struct tcf_bpf *prog; |
291 | bool is_bpf, is_ebpf; |
292 | int ret, res = 0; |
293 | u32 index; |
294 | |
295 | if (!nla) |
296 | return -EINVAL; |
297 | |
298 | ret = nla_parse_nested_deprecated(tb, TCA_ACT_BPF_MAX, nla, |
299 | policy: act_bpf_policy, NULL); |
300 | if (ret < 0) |
301 | return ret; |
302 | |
303 | if (!tb[TCA_ACT_BPF_PARMS]) |
304 | return -EINVAL; |
305 | |
306 | parm = nla_data(nla: tb[TCA_ACT_BPF_PARMS]); |
307 | index = parm->index; |
308 | ret = tcf_idr_check_alloc(tn, index: &index, a: act, bind); |
309 | if (!ret) { |
310 | ret = tcf_idr_create(tn, index, est, a: act, |
311 | ops: &act_bpf_ops, bind, cpustats: true, flags); |
312 | if (ret < 0) { |
313 | tcf_idr_cleanup(tn, index); |
314 | return ret; |
315 | } |
316 | |
317 | res = ACT_P_CREATED; |
318 | } else if (ret > 0) { |
319 | /* Don't override defaults. */ |
320 | if (bind) |
321 | return 0; |
322 | |
323 | if (!(flags & TCA_ACT_FLAGS_REPLACE)) { |
324 | tcf_idr_release(a: *act, bind); |
325 | return -EEXIST; |
326 | } |
327 | } else { |
328 | return ret; |
329 | } |
330 | |
331 | ret = tcf_action_check_ctrlact(action: parm->action, tp, handle: &goto_ch, newchain: extack); |
332 | if (ret < 0) |
333 | goto release_idr; |
334 | |
335 | is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS]; |
336 | is_ebpf = tb[TCA_ACT_BPF_FD]; |
337 | |
338 | if (is_bpf == is_ebpf) { |
339 | ret = -EINVAL; |
340 | goto put_chain; |
341 | } |
342 | |
343 | memset(&cfg, 0, sizeof(cfg)); |
344 | |
345 | ret = is_bpf ? tcf_bpf_init_from_ops(tb, cfg: &cfg) : |
346 | tcf_bpf_init_from_efd(tb, cfg: &cfg); |
347 | if (ret < 0) |
348 | goto put_chain; |
349 | |
350 | prog = to_bpf(*act); |
351 | |
352 | spin_lock_bh(lock: &prog->tcf_lock); |
353 | if (res != ACT_P_CREATED) |
354 | tcf_bpf_prog_fill_cfg(prog, cfg: &old); |
355 | |
356 | prog->bpf_ops = cfg.bpf_ops; |
357 | prog->bpf_name = cfg.bpf_name; |
358 | |
359 | if (cfg.bpf_num_ops) |
360 | prog->bpf_num_ops = cfg.bpf_num_ops; |
361 | |
362 | goto_ch = tcf_action_set_ctrlact(a: *act, action: parm->action, newchain: goto_ch); |
363 | rcu_assign_pointer(prog->filter, cfg.filter); |
364 | spin_unlock_bh(lock: &prog->tcf_lock); |
365 | |
366 | if (goto_ch) |
367 | tcf_chain_put_by_act(chain: goto_ch); |
368 | |
369 | if (res != ACT_P_CREATED) { |
370 | /* make sure the program being replaced is no longer executing */ |
371 | synchronize_rcu(); |
372 | tcf_bpf_cfg_cleanup(cfg: &old); |
373 | } |
374 | |
375 | return res; |
376 | |
377 | put_chain: |
378 | if (goto_ch) |
379 | tcf_chain_put_by_act(chain: goto_ch); |
380 | |
381 | release_idr: |
382 | tcf_idr_release(a: *act, bind); |
383 | return ret; |
384 | } |
385 | |
386 | static void tcf_bpf_cleanup(struct tc_action *act) |
387 | { |
388 | struct tcf_bpf_cfg tmp; |
389 | |
390 | tcf_bpf_prog_fill_cfg(to_bpf(act), cfg: &tmp); |
391 | tcf_bpf_cfg_cleanup(cfg: &tmp); |
392 | } |
393 | |
394 | static struct tc_action_ops act_bpf_ops __read_mostly = { |
395 | .kind = "bpf" , |
396 | .id = TCA_ID_BPF, |
397 | .owner = THIS_MODULE, |
398 | .act = tcf_bpf_act, |
399 | .dump = tcf_bpf_dump, |
400 | .cleanup = tcf_bpf_cleanup, |
401 | .init = tcf_bpf_init, |
402 | .size = sizeof(struct tcf_bpf), |
403 | }; |
404 | |
405 | static __net_init int bpf_init_net(struct net *net) |
406 | { |
407 | struct tc_action_net *tn = net_generic(net, id: act_bpf_ops.net_id); |
408 | |
409 | return tc_action_net_init(net, tn, ops: &act_bpf_ops); |
410 | } |
411 | |
412 | static void __net_exit bpf_exit_net(struct list_head *net_list) |
413 | { |
414 | tc_action_net_exit(net_list, id: act_bpf_ops.net_id); |
415 | } |
416 | |
417 | static struct pernet_operations bpf_net_ops = { |
418 | .init = bpf_init_net, |
419 | .exit_batch = bpf_exit_net, |
420 | .id = &act_bpf_ops.net_id, |
421 | .size = sizeof(struct tc_action_net), |
422 | }; |
423 | |
424 | static int __init bpf_init_module(void) |
425 | { |
426 | return tcf_register_action(a: &act_bpf_ops, ops: &bpf_net_ops); |
427 | } |
428 | |
429 | static void __exit bpf_cleanup_module(void) |
430 | { |
431 | tcf_unregister_action(a: &act_bpf_ops, ops: &bpf_net_ops); |
432 | } |
433 | |
434 | module_init(bpf_init_module); |
435 | module_exit(bpf_cleanup_module); |
436 | |
437 | MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>" ); |
438 | MODULE_DESCRIPTION("TC BPF based action" ); |
439 | MODULE_LICENSE("GPL v2" ); |
440 | |