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
bpf_tcp_ca_init(struct btf * btf)31 static int bpf_tcp_ca_init(struct btf *btf)
32 {
33 s32 type_id;
34
35 type_id = btf_find_by_name_kind(btf, "sock", BTF_KIND_STRUCT);
36 if (type_id < 0)
37 return -EINVAL;
38 sock_id = type_id;
39
40 type_id = btf_find_by_name_kind(btf, "tcp_sock", BTF_KIND_STRUCT);
41 if (type_id < 0)
42 return -EINVAL;
43 tcp_sock_id = type_id;
44 tcp_sock_type = btf_type_by_id(btf, tcp_sock_id);
45
46 return 0;
47 }
48
is_optional(u32 member_offset)49 static bool is_optional(u32 member_offset)
50 {
51 unsigned int i;
52
53 for (i = 0; i < ARRAY_SIZE(optional_ops); i++) {
54 if (member_offset == optional_ops[i])
55 return true;
56 }
57
58 return false;
59 }
60
is_unsupported(u32 member_offset)61 static bool is_unsupported(u32 member_offset)
62 {
63 unsigned int i;
64
65 for (i = 0; i < ARRAY_SIZE(unsupported_ops); i++) {
66 if (member_offset == unsupported_ops[i])
67 return true;
68 }
69
70 return false;
71 }
72
73 extern struct btf *btf_vmlinux;
74
bpf_tcp_ca_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)75 static bool bpf_tcp_ca_is_valid_access(int off, int size,
76 enum bpf_access_type type,
77 const struct bpf_prog *prog,
78 struct bpf_insn_access_aux *info)
79 {
80 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
81 return false;
82 if (type != BPF_READ)
83 return false;
84 if (off % size != 0)
85 return false;
86
87 if (!btf_ctx_access(off, size, type, prog, info))
88 return false;
89
90 if (info->reg_type == PTR_TO_BTF_ID && info->btf_id == sock_id)
91 /* promote it to tcp_sock */
92 info->btf_id = tcp_sock_id;
93
94 return true;
95 }
96
bpf_tcp_ca_btf_struct_access(struct bpf_verifier_log * log,const struct btf_type * t,int off,int size,enum bpf_access_type atype,u32 * next_btf_id)97 static int bpf_tcp_ca_btf_struct_access(struct bpf_verifier_log *log,
98 const struct btf_type *t, int off,
99 int size, enum bpf_access_type atype,
100 u32 *next_btf_id)
101 {
102 size_t end;
103
104 if (atype == BPF_READ)
105 return btf_struct_access(log, t, off, size, atype, next_btf_id);
106
107 if (t != tcp_sock_type) {
108 bpf_log(log, "only read is supported\n");
109 return -EACCES;
110 }
111
112 switch (off) {
113 case bpf_ctx_range(struct inet_connection_sock, icsk_ca_priv):
114 end = offsetofend(struct inet_connection_sock, icsk_ca_priv);
115 break;
116 case offsetof(struct inet_connection_sock, icsk_ack.pending):
117 end = offsetofend(struct inet_connection_sock,
118 icsk_ack.pending);
119 break;
120 case offsetof(struct tcp_sock, snd_cwnd):
121 end = offsetofend(struct tcp_sock, snd_cwnd);
122 break;
123 case offsetof(struct tcp_sock, snd_cwnd_cnt):
124 end = offsetofend(struct tcp_sock, snd_cwnd_cnt);
125 break;
126 case offsetof(struct tcp_sock, snd_ssthresh):
127 end = offsetofend(struct tcp_sock, snd_ssthresh);
128 break;
129 case offsetof(struct tcp_sock, ecn_flags):
130 end = offsetofend(struct tcp_sock, ecn_flags);
131 break;
132 default:
133 bpf_log(log, "no write support to tcp_sock at off %d\n", off);
134 return -EACCES;
135 }
136
137 if (off + size > end) {
138 bpf_log(log,
139 "write access at off %d with size %d beyond the member of tcp_sock ended at %zu\n",
140 off, size, end);
141 return -EACCES;
142 }
143
144 return NOT_INIT;
145 }
146
BPF_CALL_2(bpf_tcp_send_ack,struct tcp_sock *,tp,u32,rcv_nxt)147 BPF_CALL_2(bpf_tcp_send_ack, struct tcp_sock *, tp, u32, rcv_nxt)
148 {
149 /* bpf_tcp_ca prog cannot have NULL tp */
150 __tcp_send_ack((struct sock *)tp, rcv_nxt);
151 return 0;
152 }
153
154 static const struct bpf_func_proto bpf_tcp_send_ack_proto = {
155 .func = bpf_tcp_send_ack,
156 .gpl_only = false,
157 /* In case we want to report error later */
158 .ret_type = RET_INTEGER,
159 .arg1_type = ARG_PTR_TO_BTF_ID,
160 .arg1_btf_id = &tcp_sock_id,
161 .arg2_type = ARG_ANYTHING,
162 };
163
164 static const struct bpf_func_proto *
bpf_tcp_ca_get_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)165 bpf_tcp_ca_get_func_proto(enum bpf_func_id func_id,
166 const struct bpf_prog *prog)
167 {
168 switch (func_id) {
169 case BPF_FUNC_tcp_send_ack:
170 return &bpf_tcp_send_ack_proto;
171 case BPF_FUNC_sk_storage_get:
172 return &bpf_sk_storage_get_proto;
173 case BPF_FUNC_sk_storage_delete:
174 return &bpf_sk_storage_delete_proto;
175 default:
176 return bpf_base_func_proto(func_id);
177 }
178 }
179
180 static const struct bpf_verifier_ops bpf_tcp_ca_verifier_ops = {
181 .get_func_proto = bpf_tcp_ca_get_func_proto,
182 .is_valid_access = bpf_tcp_ca_is_valid_access,
183 .btf_struct_access = bpf_tcp_ca_btf_struct_access,
184 };
185
bpf_tcp_ca_init_member(const struct btf_type * t,const struct btf_member * member,void * kdata,const void * udata)186 static int bpf_tcp_ca_init_member(const struct btf_type *t,
187 const struct btf_member *member,
188 void *kdata, const void *udata)
189 {
190 const struct tcp_congestion_ops *utcp_ca;
191 struct tcp_congestion_ops *tcp_ca;
192 int prog_fd;
193 u32 moff;
194
195 utcp_ca = (const struct tcp_congestion_ops *)udata;
196 tcp_ca = (struct tcp_congestion_ops *)kdata;
197
198 moff = btf_member_bit_offset(t, member) / 8;
199 switch (moff) {
200 case offsetof(struct tcp_congestion_ops, flags):
201 if (utcp_ca->flags & ~TCP_CONG_MASK)
202 return -EINVAL;
203 tcp_ca->flags = utcp_ca->flags;
204 return 1;
205 case offsetof(struct tcp_congestion_ops, name):
206 if (bpf_obj_name_cpy(tcp_ca->name, utcp_ca->name,
207 sizeof(tcp_ca->name)) <= 0)
208 return -EINVAL;
209 if (tcp_ca_find(utcp_ca->name))
210 return -EEXIST;
211 return 1;
212 }
213
214 if (!btf_type_resolve_func_ptr(btf_vmlinux, member->type, NULL))
215 return 0;
216
217 /* Ensure bpf_prog is provided for compulsory func ptr */
218 prog_fd = (int)(*(unsigned long *)(udata + moff));
219 if (!prog_fd && !is_optional(moff) && !is_unsupported(moff))
220 return -EINVAL;
221
222 return 0;
223 }
224
bpf_tcp_ca_check_member(const struct btf_type * t,const struct btf_member * member)225 static int bpf_tcp_ca_check_member(const struct btf_type *t,
226 const struct btf_member *member)
227 {
228 if (is_unsupported(btf_member_bit_offset(t, member) / 8))
229 return -ENOTSUPP;
230 return 0;
231 }
232
bpf_tcp_ca_reg(void * kdata)233 static int bpf_tcp_ca_reg(void *kdata)
234 {
235 return tcp_register_congestion_control(kdata);
236 }
237
bpf_tcp_ca_unreg(void * kdata)238 static void bpf_tcp_ca_unreg(void *kdata)
239 {
240 tcp_unregister_congestion_control(kdata);
241 }
242
243 /* Avoid sparse warning. It is only used in bpf_struct_ops.c. */
244 extern struct bpf_struct_ops bpf_tcp_congestion_ops;
245
246 struct bpf_struct_ops bpf_tcp_congestion_ops = {
247 .verifier_ops = &bpf_tcp_ca_verifier_ops,
248 .reg = bpf_tcp_ca_reg,
249 .unreg = bpf_tcp_ca_unreg,
250 .check_member = bpf_tcp_ca_check_member,
251 .init_member = bpf_tcp_ca_init_member,
252 .init = bpf_tcp_ca_init,
253 .name = "tcp_congestion_ops",
254 };
255