xref: /openbmc/linux/net/ipv4/bpf_tcp_ca.c (revision 9123e3a7)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019 Facebook  */
3 
4 #include <linux/types.h>
5 #include <linux/bpf_verifier.h>
6 #include <linux/bpf.h>
7 #include <linux/btf.h>
8 #include <linux/filter.h>
9 #include <net/tcp.h>
10 #include <net/bpf_sk_storage.h>
11 
12 static u32 optional_ops[] = {
13 	offsetof(struct tcp_congestion_ops, init),
14 	offsetof(struct tcp_congestion_ops, release),
15 	offsetof(struct tcp_congestion_ops, set_state),
16 	offsetof(struct tcp_congestion_ops, cwnd_event),
17 	offsetof(struct tcp_congestion_ops, in_ack_event),
18 	offsetof(struct tcp_congestion_ops, pkts_acked),
19 	offsetof(struct tcp_congestion_ops, min_tso_segs),
20 	offsetof(struct tcp_congestion_ops, sndbuf_expand),
21 	offsetof(struct tcp_congestion_ops, cong_control),
22 };
23 
24 static u32 unsupported_ops[] = {
25 	offsetof(struct tcp_congestion_ops, get_info),
26 };
27 
28 static const struct btf_type *tcp_sock_type;
29 static u32 tcp_sock_id, sock_id;
30 
31 static int btf_sk_storage_get_ids[5];
32 static struct bpf_func_proto btf_sk_storage_get_proto __read_mostly;
33 
34 static int btf_sk_storage_delete_ids[5];
35 static struct bpf_func_proto btf_sk_storage_delete_proto __read_mostly;
36 
37 static void convert_sk_func_proto(struct bpf_func_proto *to, int *to_btf_ids,
38 				  const struct bpf_func_proto *from)
39 {
40 	int i;
41 
42 	*to = *from;
43 	to->btf_id = to_btf_ids;
44 	for (i = 0; i < ARRAY_SIZE(to->arg_type); i++) {
45 		if (to->arg_type[i] == ARG_PTR_TO_SOCKET) {
46 			to->arg_type[i] = ARG_PTR_TO_BTF_ID;
47 			to->btf_id[i] = tcp_sock_id;
48 		}
49 	}
50 }
51 
52 static int bpf_tcp_ca_init(struct btf *btf)
53 {
54 	s32 type_id;
55 
56 	type_id = btf_find_by_name_kind(btf, "sock", BTF_KIND_STRUCT);
57 	if (type_id < 0)
58 		return -EINVAL;
59 	sock_id = type_id;
60 
61 	type_id = btf_find_by_name_kind(btf, "tcp_sock", BTF_KIND_STRUCT);
62 	if (type_id < 0)
63 		return -EINVAL;
64 	tcp_sock_id = type_id;
65 	tcp_sock_type = btf_type_by_id(btf, tcp_sock_id);
66 
67 	convert_sk_func_proto(&btf_sk_storage_get_proto,
68 			      btf_sk_storage_get_ids,
69 			      &bpf_sk_storage_get_proto);
70 	convert_sk_func_proto(&btf_sk_storage_delete_proto,
71 			      btf_sk_storage_delete_ids,
72 			      &bpf_sk_storage_delete_proto);
73 
74 	return 0;
75 }
76 
77 static bool is_optional(u32 member_offset)
78 {
79 	unsigned int i;
80 
81 	for (i = 0; i < ARRAY_SIZE(optional_ops); i++) {
82 		if (member_offset == optional_ops[i])
83 			return true;
84 	}
85 
86 	return false;
87 }
88 
89 static bool is_unsupported(u32 member_offset)
90 {
91 	unsigned int i;
92 
93 	for (i = 0; i < ARRAY_SIZE(unsupported_ops); i++) {
94 		if (member_offset == unsupported_ops[i])
95 			return true;
96 	}
97 
98 	return false;
99 }
100 
101 extern struct btf *btf_vmlinux;
102 
103 static bool bpf_tcp_ca_is_valid_access(int off, int size,
104 				       enum bpf_access_type type,
105 				       const struct bpf_prog *prog,
106 				       struct bpf_insn_access_aux *info)
107 {
108 	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
109 		return false;
110 	if (type != BPF_READ)
111 		return false;
112 	if (off % size != 0)
113 		return false;
114 
115 	if (!btf_ctx_access(off, size, type, prog, info))
116 		return false;
117 
118 	if (info->reg_type == PTR_TO_BTF_ID && info->btf_id == sock_id)
119 		/* promote it to tcp_sock */
120 		info->btf_id = tcp_sock_id;
121 
122 	return true;
123 }
124 
125 static int bpf_tcp_ca_btf_struct_access(struct bpf_verifier_log *log,
126 					const struct btf_type *t, int off,
127 					int size, enum bpf_access_type atype,
128 					u32 *next_btf_id)
129 {
130 	size_t end;
131 
132 	if (atype == BPF_READ)
133 		return btf_struct_access(log, t, off, size, atype, next_btf_id);
134 
135 	if (t != tcp_sock_type) {
136 		bpf_log(log, "only read is supported\n");
137 		return -EACCES;
138 	}
139 
140 	switch (off) {
141 	case bpf_ctx_range(struct inet_connection_sock, icsk_ca_priv):
142 		end = offsetofend(struct inet_connection_sock, icsk_ca_priv);
143 		break;
144 	case offsetof(struct inet_connection_sock, icsk_ack.pending):
145 		end = offsetofend(struct inet_connection_sock,
146 				  icsk_ack.pending);
147 		break;
148 	case offsetof(struct tcp_sock, snd_cwnd):
149 		end = offsetofend(struct tcp_sock, snd_cwnd);
150 		break;
151 	case offsetof(struct tcp_sock, snd_cwnd_cnt):
152 		end = offsetofend(struct tcp_sock, snd_cwnd_cnt);
153 		break;
154 	case offsetof(struct tcp_sock, snd_ssthresh):
155 		end = offsetofend(struct tcp_sock, snd_ssthresh);
156 		break;
157 	case offsetof(struct tcp_sock, ecn_flags):
158 		end = offsetofend(struct tcp_sock, ecn_flags);
159 		break;
160 	default:
161 		bpf_log(log, "no write support to tcp_sock at off %d\n", off);
162 		return -EACCES;
163 	}
164 
165 	if (off + size > end) {
166 		bpf_log(log,
167 			"write access at off %d with size %d beyond the member of tcp_sock ended at %zu\n",
168 			off, size, end);
169 		return -EACCES;
170 	}
171 
172 	return NOT_INIT;
173 }
174 
175 BPF_CALL_2(bpf_tcp_send_ack, struct tcp_sock *, tp, u32, rcv_nxt)
176 {
177 	/* bpf_tcp_ca prog cannot have NULL tp */
178 	__tcp_send_ack((struct sock *)tp, rcv_nxt);
179 	return 0;
180 }
181 
182 static const struct bpf_func_proto bpf_tcp_send_ack_proto = {
183 	.func		= bpf_tcp_send_ack,
184 	.gpl_only	= false,
185 	/* In case we want to report error later */
186 	.ret_type	= RET_INTEGER,
187 	.arg1_type	= ARG_PTR_TO_BTF_ID,
188 	.arg2_type	= ARG_ANYTHING,
189 	.btf_id		= &tcp_sock_id,
190 };
191 
192 static const struct bpf_func_proto *
193 bpf_tcp_ca_get_func_proto(enum bpf_func_id func_id,
194 			  const struct bpf_prog *prog)
195 {
196 	switch (func_id) {
197 	case BPF_FUNC_tcp_send_ack:
198 		return &bpf_tcp_send_ack_proto;
199 	case BPF_FUNC_sk_storage_get:
200 		return &btf_sk_storage_get_proto;
201 	case BPF_FUNC_sk_storage_delete:
202 		return &btf_sk_storage_delete_proto;
203 	default:
204 		return bpf_base_func_proto(func_id);
205 	}
206 }
207 
208 static const struct bpf_verifier_ops bpf_tcp_ca_verifier_ops = {
209 	.get_func_proto		= bpf_tcp_ca_get_func_proto,
210 	.is_valid_access	= bpf_tcp_ca_is_valid_access,
211 	.btf_struct_access	= bpf_tcp_ca_btf_struct_access,
212 };
213 
214 static int bpf_tcp_ca_init_member(const struct btf_type *t,
215 				  const struct btf_member *member,
216 				  void *kdata, const void *udata)
217 {
218 	const struct tcp_congestion_ops *utcp_ca;
219 	struct tcp_congestion_ops *tcp_ca;
220 	int prog_fd;
221 	u32 moff;
222 
223 	utcp_ca = (const struct tcp_congestion_ops *)udata;
224 	tcp_ca = (struct tcp_congestion_ops *)kdata;
225 
226 	moff = btf_member_bit_offset(t, member) / 8;
227 	switch (moff) {
228 	case offsetof(struct tcp_congestion_ops, flags):
229 		if (utcp_ca->flags & ~TCP_CONG_MASK)
230 			return -EINVAL;
231 		tcp_ca->flags = utcp_ca->flags;
232 		return 1;
233 	case offsetof(struct tcp_congestion_ops, name):
234 		if (bpf_obj_name_cpy(tcp_ca->name, utcp_ca->name,
235 				     sizeof(tcp_ca->name)) <= 0)
236 			return -EINVAL;
237 		if (tcp_ca_find(utcp_ca->name))
238 			return -EEXIST;
239 		return 1;
240 	}
241 
242 	if (!btf_type_resolve_func_ptr(btf_vmlinux, member->type, NULL))
243 		return 0;
244 
245 	/* Ensure bpf_prog is provided for compulsory func ptr */
246 	prog_fd = (int)(*(unsigned long *)(udata + moff));
247 	if (!prog_fd && !is_optional(moff) && !is_unsupported(moff))
248 		return -EINVAL;
249 
250 	return 0;
251 }
252 
253 static int bpf_tcp_ca_check_member(const struct btf_type *t,
254 				   const struct btf_member *member)
255 {
256 	if (is_unsupported(btf_member_bit_offset(t, member) / 8))
257 		return -ENOTSUPP;
258 	return 0;
259 }
260 
261 static int bpf_tcp_ca_reg(void *kdata)
262 {
263 	return tcp_register_congestion_control(kdata);
264 }
265 
266 static void bpf_tcp_ca_unreg(void *kdata)
267 {
268 	tcp_unregister_congestion_control(kdata);
269 }
270 
271 /* Avoid sparse warning.  It is only used in bpf_struct_ops.c. */
272 extern struct bpf_struct_ops bpf_tcp_congestion_ops;
273 
274 struct bpf_struct_ops bpf_tcp_congestion_ops = {
275 	.verifier_ops = &bpf_tcp_ca_verifier_ops,
276 	.reg = bpf_tcp_ca_reg,
277 	.unreg = bpf_tcp_ca_unreg,
278 	.check_member = bpf_tcp_ca_check_member,
279 	.init_member = bpf_tcp_ca_init_member,
280 	.init = bpf_tcp_ca_init,
281 	.name = "tcp_congestion_ops",
282 };
283