1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019 Facebook */
3
4 #include <linux/init.h>
5 #include <linux/types.h>
6 #include <linux/bpf_verifier.h>
7 #include <linux/bpf.h>
8 #include <linux/btf.h>
9 #include <linux/btf_ids.h>
10 #include <linux/filter.h>
11 #include <net/tcp.h>
12 #include <net/bpf_sk_storage.h>
13
14 /* "extern" is to avoid sparse warning. It is only used in bpf_struct_ops.c. */
15 static struct bpf_struct_ops bpf_tcp_congestion_ops;
16
17 static u32 unsupported_ops[] = {
18 offsetof(struct tcp_congestion_ops, get_info),
19 };
20
21 static const struct btf_type *tcp_sock_type;
22 static u32 tcp_sock_id, sock_id;
23 static const struct btf_type *tcp_congestion_ops_type;
24
25 static int bpf_tcp_ca_init(struct btf *btf)
26 {
27 s32 type_id;
28
29 type_id = btf_find_by_name_kind(btf, "sock", BTF_KIND_STRUCT);
30 if (type_id < 0)
31 return -EINVAL;
32 sock_id = type_id;
33
34 type_id = btf_find_by_name_kind(btf, "tcp_sock", BTF_KIND_STRUCT);
35 if (type_id < 0)
36 return -EINVAL;
37 tcp_sock_id = type_id;
38 tcp_sock_type = btf_type_by_id(btf, tcp_sock_id);
39
40 type_id = btf_find_by_name_kind(btf, "tcp_congestion_ops", BTF_KIND_STRUCT);
41 if (type_id < 0)
42 return -EINVAL;
43 tcp_congestion_ops_type = btf_type_by_id(btf, type_id);
44
45 return 0;
46 }
47
48 static bool is_unsupported(u32 member_offset)
49 {
50 unsigned int i;
51
52 for (i = 0; i < ARRAY_SIZE(unsupported_ops); i++) {
53 if (member_offset == unsupported_ops[i])
54 return true;
55 }
56
57 return false;
58 }
59
60 static bool bpf_tcp_ca_is_valid_access(int off, int size,
61 enum bpf_access_type type,
62 const struct bpf_prog *prog,
63 struct bpf_insn_access_aux *info)
64 {
65 if (!bpf_tracing_btf_ctx_access(off, size, type, prog, info))
66 return false;
67
68 if (base_type(info->reg_type) == PTR_TO_BTF_ID &&
69 !bpf_type_has_unsafe_modifiers(info->reg_type) &&
70 info->btf_id == sock_id)
71 /* promote it to tcp_sock */
72 info->btf_id = tcp_sock_id;
73
74 return true;
75 }
76
77 static int bpf_tcp_ca_btf_struct_access(struct bpf_verifier_log *log,
78 const struct bpf_reg_state *reg,
79 int off, int size)
80 {
81 const struct btf_type *t;
82 size_t end;
83
84 t = btf_type_by_id(reg->btf, reg->btf_id);
85 if (t != tcp_sock_type) {
86 bpf_log(log, "only read is supported\n");
87 return -EACCES;
88 }
89
90 switch (off) {
91 case offsetof(struct sock, sk_pacing_rate):
92 end = offsetofend(struct sock, sk_pacing_rate);
93 break;
94 case offsetof(struct sock, sk_pacing_status):
95 end = offsetofend(struct sock, sk_pacing_status);
96 break;
97 case bpf_ctx_range(struct inet_connection_sock, icsk_ca_priv):
98 end = offsetofend(struct inet_connection_sock, icsk_ca_priv);
99 break;
100 case offsetof(struct inet_connection_sock, icsk_ack.pending):
101 end = offsetofend(struct inet_connection_sock,
102 icsk_ack.pending);
103 break;
104 case offsetof(struct tcp_sock, snd_cwnd):
105 end = offsetofend(struct tcp_sock, snd_cwnd);
106 break;
107 case offsetof(struct tcp_sock, snd_cwnd_cnt):
108 end = offsetofend(struct tcp_sock, snd_cwnd_cnt);
109 break;
110 case offsetof(struct tcp_sock, snd_cwnd_stamp):
111 end = offsetofend(struct tcp_sock, snd_cwnd_stamp);
112 break;
113 case offsetof(struct tcp_sock, snd_ssthresh):
114 end = offsetofend(struct tcp_sock, snd_ssthresh);
115 break;
116 case offsetof(struct tcp_sock, ecn_flags):
117 end = offsetofend(struct tcp_sock, ecn_flags);
118 break;
119 case offsetof(struct tcp_sock, app_limited):
120 end = offsetofend(struct tcp_sock, app_limited);
121 break;
122 default:
123 bpf_log(log, "no write support to tcp_sock at off %d\n", off);
124 return -EACCES;
125 }
126
127 if (off + size > end) {
128 bpf_log(log,
129 "write access at off %d with size %d beyond the member of tcp_sock ended at %zu\n",
130 off, size, end);
131 return -EACCES;
132 }
133
134 return 0;
135 }
136
137 BPF_CALL_2(bpf_tcp_send_ack, struct tcp_sock *, tp, u32, rcv_nxt)
138 {
139 /* bpf_tcp_ca prog cannot have NULL tp */
140 __tcp_send_ack((struct sock *)tp, rcv_nxt);
141 return 0;
142 }
143
144 static const struct bpf_func_proto bpf_tcp_send_ack_proto = {
145 .func = bpf_tcp_send_ack,
146 .gpl_only = false,
147 /* In case we want to report error later */
148 .ret_type = RET_INTEGER,
149 .arg1_type = ARG_PTR_TO_BTF_ID,
150 .arg1_btf_id = &tcp_sock_id,
151 .arg2_type = ARG_ANYTHING,
152 };
153
154 static u32 prog_ops_moff(const struct bpf_prog *prog)
155 {
156 const struct btf_member *m;
157 const struct btf_type *t;
158 u32 midx;
159
160 midx = prog->expected_attach_type;
161 t = tcp_congestion_ops_type;
162 m = &btf_type_member(t)[midx];
163
164 return __btf_member_bit_offset(t, m) / 8;
165 }
166
167 static const struct bpf_func_proto *
168 bpf_tcp_ca_get_func_proto(enum bpf_func_id func_id,
169 const struct bpf_prog *prog)
170 {
171 switch (func_id) {
172 case BPF_FUNC_tcp_send_ack:
173 return &bpf_tcp_send_ack_proto;
174 case BPF_FUNC_sk_storage_get:
175 return &bpf_sk_storage_get_proto;
176 case BPF_FUNC_sk_storage_delete:
177 return &bpf_sk_storage_delete_proto;
178 case BPF_FUNC_setsockopt:
179 /* Does not allow release() to call setsockopt.
180 * release() is called when the current bpf-tcp-cc
181 * is retiring. It is not allowed to call
182 * setsockopt() to make further changes which
183 * may potentially allocate new resources.
184 */
185 if (prog_ops_moff(prog) !=
186 offsetof(struct tcp_congestion_ops, release))
187 return &bpf_sk_setsockopt_proto;
188 return NULL;
189 case BPF_FUNC_getsockopt:
190 /* Since get/setsockopt is usually expected to
191 * be available together, disable getsockopt for
192 * release also to avoid usage surprise.
193 * The bpf-tcp-cc already has a more powerful way
194 * to read tcp_sock from the PTR_TO_BTF_ID.
195 */
196 if (prog_ops_moff(prog) !=
197 offsetof(struct tcp_congestion_ops, release))
198 return &bpf_sk_getsockopt_proto;
199 return NULL;
200 case BPF_FUNC_ktime_get_coarse_ns:
201 return &bpf_ktime_get_coarse_ns_proto;
202 default:
203 return bpf_base_func_proto(func_id, prog);
204 }
205 }
206
207 BTF_KFUNCS_START(bpf_tcp_ca_check_kfunc_ids)
208 BTF_ID_FLAGS(func, tcp_reno_ssthresh)
209 BTF_ID_FLAGS(func, tcp_reno_cong_avoid)
210 BTF_ID_FLAGS(func, tcp_reno_undo_cwnd)
211 BTF_ID_FLAGS(func, tcp_slow_start)
212 BTF_ID_FLAGS(func, tcp_cong_avoid_ai)
213 BTF_KFUNCS_END(bpf_tcp_ca_check_kfunc_ids)
214
215 static const struct btf_kfunc_id_set bpf_tcp_ca_kfunc_set = {
216 .owner = THIS_MODULE,
217 .set = &bpf_tcp_ca_check_kfunc_ids,
218 };
219
220 static const struct bpf_verifier_ops bpf_tcp_ca_verifier_ops = {
221 .get_func_proto = bpf_tcp_ca_get_func_proto,
222 .is_valid_access = bpf_tcp_ca_is_valid_access,
223 .btf_struct_access = bpf_tcp_ca_btf_struct_access,
224 };
225
226 static int bpf_tcp_ca_init_member(const struct btf_type *t,
227 const struct btf_member *member,
228 void *kdata, const void *udata)
229 {
230 const struct tcp_congestion_ops *utcp_ca;
231 struct tcp_congestion_ops *tcp_ca;
232 u32 moff;
233
234 utcp_ca = (const struct tcp_congestion_ops *)udata;
235 tcp_ca = (struct tcp_congestion_ops *)kdata;
236
237 moff = __btf_member_bit_offset(t, member) / 8;
238 switch (moff) {
239 case offsetof(struct tcp_congestion_ops, flags):
240 if (utcp_ca->flags & ~TCP_CONG_MASK)
241 return -EINVAL;
242 tcp_ca->flags = utcp_ca->flags;
243 return 1;
244 case offsetof(struct tcp_congestion_ops, name):
245 if (bpf_obj_name_cpy(tcp_ca->name, utcp_ca->name,
246 sizeof(tcp_ca->name)) <= 0)
247 return -EINVAL;
248 return 1;
249 }
250
251 return 0;
252 }
253
254 static int bpf_tcp_ca_check_member(const struct btf_type *t,
255 const struct btf_member *member,
256 const struct bpf_prog *prog)
257 {
258 if (is_unsupported(__btf_member_bit_offset(t, member) / 8))
259 return -ENOTSUPP;
260 return 0;
261 }
262
263 static int bpf_tcp_ca_reg(void *kdata, struct bpf_link *link)
264 {
265 return tcp_register_congestion_control(kdata);
266 }
267
268 static void bpf_tcp_ca_unreg(void *kdata, struct bpf_link *link)
269 {
270 tcp_unregister_congestion_control(kdata);
271 }
272
273 static int bpf_tcp_ca_update(void *kdata, void *old_kdata, struct bpf_link *link)
274 {
275 return tcp_update_congestion_control(kdata, old_kdata);
276 }
277
278 static int bpf_tcp_ca_validate(void *kdata)
279 {
280 return tcp_validate_congestion_control(kdata);
281 }
282
283 static u32 bpf_tcp_ca_ssthresh(struct sock *sk)
284 {
285 return 0;
286 }
287
288 static void bpf_tcp_ca_cong_avoid(struct sock *sk, u32 ack, u32 acked)
289 {
290 }
291
292 static void bpf_tcp_ca_set_state(struct sock *sk, u8 new_state)
293 {
294 }
295
296 static void bpf_tcp_ca_cwnd_event(struct sock *sk, enum tcp_ca_event ev)
297 {
298 }
299
300 static void bpf_tcp_ca_in_ack_event(struct sock *sk, u32 flags)
301 {
302 }
303
304 static void bpf_tcp_ca_pkts_acked(struct sock *sk, const struct ack_sample *sample)
305 {
306 }
307
308 static u32 bpf_tcp_ca_min_tso_segs(struct sock *sk)
309 {
310 return 0;
311 }
312
313 static void bpf_tcp_ca_cong_control(struct sock *sk, u32 ack, int flag,
314 const struct rate_sample *rs)
315 {
316 }
317
318 static u32 bpf_tcp_ca_undo_cwnd(struct sock *sk)
319 {
320 return 0;
321 }
322
323 static u32 bpf_tcp_ca_sndbuf_expand(struct sock *sk)
324 {
325 return 0;
326 }
327
328 static void __bpf_tcp_ca_init(struct sock *sk)
329 {
330 }
331
332 static void __bpf_tcp_ca_release(struct sock *sk)
333 {
334 }
335
336 static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = {
337 .ssthresh = bpf_tcp_ca_ssthresh,
338 .cong_avoid = bpf_tcp_ca_cong_avoid,
339 .set_state = bpf_tcp_ca_set_state,
340 .cwnd_event = bpf_tcp_ca_cwnd_event,
341 .in_ack_event = bpf_tcp_ca_in_ack_event,
342 .pkts_acked = bpf_tcp_ca_pkts_acked,
343 .min_tso_segs = bpf_tcp_ca_min_tso_segs,
344 .cong_control = bpf_tcp_ca_cong_control,
345 .undo_cwnd = bpf_tcp_ca_undo_cwnd,
346 .sndbuf_expand = bpf_tcp_ca_sndbuf_expand,
347
348 .init = __bpf_tcp_ca_init,
349 .release = __bpf_tcp_ca_release,
350 };
351
352 static struct bpf_struct_ops bpf_tcp_congestion_ops = {
353 .verifier_ops = &bpf_tcp_ca_verifier_ops,
354 .reg = bpf_tcp_ca_reg,
355 .unreg = bpf_tcp_ca_unreg,
356 .update = bpf_tcp_ca_update,
357 .check_member = bpf_tcp_ca_check_member,
358 .init_member = bpf_tcp_ca_init_member,
359 .init = bpf_tcp_ca_init,
360 .validate = bpf_tcp_ca_validate,
361 .name = "tcp_congestion_ops",
362 .cfi_stubs = &__bpf_ops_tcp_congestion_ops,
363 .owner = THIS_MODULE,
364 };
365
366 static int __init bpf_tcp_ca_kfunc_init(void)
367 {
368 int ret;
369
370 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &bpf_tcp_ca_kfunc_set);
371 ret = ret ?: register_bpf_struct_ops(&bpf_tcp_congestion_ops, tcp_congestion_ops);
372
373 return ret;
374 }
375 late_initcall(bpf_tcp_ca_kfunc_init);
376
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.