1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019 Facebook */
3 
4 #include <linux/bpf.h>
5 #include <netinet/in.h>
6 #include <stdbool.h>
7 
8 #include <bpf/bpf_helpers.h>
9 #include <bpf/bpf_endian.h>
10 
11 enum bpf_linum_array_idx {
12 	EGRESS_LINUM_IDX,
13 	INGRESS_LINUM_IDX,
14 	READ_SK_DST_PORT_LINUM_IDX,
15 	__NR_BPF_LINUM_ARRAY_IDX,
16 };
17 
18 struct {
19 	__uint(type, BPF_MAP_TYPE_ARRAY);
20 	__uint(max_entries, __NR_BPF_LINUM_ARRAY_IDX);
21 	__type(key, __u32);
22 	__type(value, __u32);
23 } linum_map SEC(".maps");
24 
25 struct bpf_spinlock_cnt {
26 	struct bpf_spin_lock lock;
27 	__u32 cnt;
28 };
29 
30 struct {
31 	__uint(type, BPF_MAP_TYPE_SK_STORAGE);
32 	__uint(map_flags, BPF_F_NO_PREALLOC);
33 	__type(key, int);
34 	__type(value, struct bpf_spinlock_cnt);
35 } sk_pkt_out_cnt SEC(".maps");
36 
37 struct {
38 	__uint(type, BPF_MAP_TYPE_SK_STORAGE);
39 	__uint(map_flags, BPF_F_NO_PREALLOC);
40 	__type(key, int);
41 	__type(value, struct bpf_spinlock_cnt);
42 } sk_pkt_out_cnt10 SEC(".maps");
43 
44 struct tcp_sock {
45 	__u32	lsndtime;
46 } __attribute__((preserve_access_index));
47 
48 struct bpf_tcp_sock listen_tp = {};
49 struct sockaddr_in6 srv_sa6 = {};
50 struct bpf_tcp_sock cli_tp = {};
51 struct bpf_tcp_sock srv_tp = {};
52 struct bpf_sock listen_sk = {};
53 struct bpf_sock srv_sk = {};
54 struct bpf_sock cli_sk = {};
55 __u64 parent_cg_id = 0;
56 __u64 child_cg_id = 0;
57 __u64 lsndtime = 0;
58 
is_loopback6(__u32 * a6)59 static bool is_loopback6(__u32 *a6)
60 {
61 	return !a6[0] && !a6[1] && !a6[2] && a6[3] == bpf_htonl(1);
62 }
63 
skcpy(struct bpf_sock * dst,const struct bpf_sock * src)64 static void skcpy(struct bpf_sock *dst,
65 		  const struct bpf_sock *src)
66 {
67 	dst->bound_dev_if = src->bound_dev_if;
68 	dst->family = src->family;
69 	dst->type = src->type;
70 	dst->protocol = src->protocol;
71 	dst->mark = src->mark;
72 	dst->priority = src->priority;
73 	dst->src_ip4 = src->src_ip4;
74 	dst->src_ip6[0] = src->src_ip6[0];
75 	dst->src_ip6[1] = src->src_ip6[1];
76 	dst->src_ip6[2] = src->src_ip6[2];
77 	dst->src_ip6[3] = src->src_ip6[3];
78 	dst->src_port = src->src_port;
79 	dst->dst_ip4 = src->dst_ip4;
80 	dst->dst_ip6[0] = src->dst_ip6[0];
81 	dst->dst_ip6[1] = src->dst_ip6[1];
82 	dst->dst_ip6[2] = src->dst_ip6[2];
83 	dst->dst_ip6[3] = src->dst_ip6[3];
84 	dst->dst_port = src->dst_port;
85 	dst->state = src->state;
86 }
87 
tpcpy(struct bpf_tcp_sock * dst,const struct bpf_tcp_sock * src)88 static void tpcpy(struct bpf_tcp_sock *dst,
89 		  const struct bpf_tcp_sock *src)
90 {
91 	dst->snd_cwnd = src->snd_cwnd;
92 	dst->srtt_us = src->srtt_us;
93 	dst->rtt_min = src->rtt_min;
94 	dst->snd_ssthresh = src->snd_ssthresh;
95 	dst->rcv_nxt = src->rcv_nxt;
96 	dst->snd_nxt = src->snd_nxt;
97 	dst->snd_una = src->snd_una;
98 	dst->mss_cache = src->mss_cache;
99 	dst->ecn_flags = src->ecn_flags;
100 	dst->rate_delivered = src->rate_delivered;
101 	dst->rate_interval_us = src->rate_interval_us;
102 	dst->packets_out = src->packets_out;
103 	dst->retrans_out = src->retrans_out;
104 	dst->total_retrans = src->total_retrans;
105 	dst->segs_in = src->segs_in;
106 	dst->data_segs_in = src->data_segs_in;
107 	dst->segs_out = src->segs_out;
108 	dst->data_segs_out = src->data_segs_out;
109 	dst->lost_out = src->lost_out;
110 	dst->sacked_out = src->sacked_out;
111 	dst->bytes_received = src->bytes_received;
112 	dst->bytes_acked = src->bytes_acked;
113 }
114 
115 /* Always return CG_OK so that no pkt will be filtered out */
116 #define CG_OK 1
117 
118 #define RET_LOG() ({						\
119 	linum = __LINE__;					\
120 	bpf_map_update_elem(&linum_map, &linum_idx, &linum, BPF_ANY);	\
121 	return CG_OK;						\
122 })
123 
124 SEC("cgroup_skb/egress")
egress_read_sock_fields(struct __sk_buff * skb)125 int egress_read_sock_fields(struct __sk_buff *skb)
126 {
127 	struct bpf_spinlock_cnt cli_cnt_init = { .lock = {}, .cnt = 0xeB9F };
128 	struct bpf_spinlock_cnt *pkt_out_cnt, *pkt_out_cnt10;
129 	struct bpf_tcp_sock *tp, *tp_ret;
130 	struct bpf_sock *sk, *sk_ret;
131 	__u32 linum, linum_idx;
132 	struct tcp_sock *ktp;
133 
134 	linum_idx = EGRESS_LINUM_IDX;
135 
136 	sk = skb->sk;
137 	if (!sk)
138 		RET_LOG();
139 
140 	/* Not testing the egress traffic or the listening socket,
141 	 * which are covered by the cgroup_skb/ingress test program.
142 	 */
143 	if (sk->family != AF_INET6 || !is_loopback6(sk->src_ip6) ||
144 	    sk->state == BPF_TCP_LISTEN)
145 		return CG_OK;
146 
147 	if (sk->src_port == bpf_ntohs(srv_sa6.sin6_port)) {
148 		/* Server socket */
149 		sk_ret = &srv_sk;
150 		tp_ret = &srv_tp;
151 	} else if (sk->dst_port == srv_sa6.sin6_port) {
152 		/* Client socket */
153 		sk_ret = &cli_sk;
154 		tp_ret = &cli_tp;
155 	} else {
156 		/* Not the testing egress traffic */
157 		return CG_OK;
158 	}
159 
160 	/* It must be a fullsock for cgroup_skb/egress prog */
161 	sk = bpf_sk_fullsock(sk);
162 	if (!sk)
163 		RET_LOG();
164 
165 	/* Not the testing egress traffic */
166 	if (sk->protocol != IPPROTO_TCP)
167 		return CG_OK;
168 
169 	tp = bpf_tcp_sock(sk);
170 	if (!tp)
171 		RET_LOG();
172 
173 	skcpy(sk_ret, sk);
174 	tpcpy(tp_ret, tp);
175 
176 	if (sk_ret == &srv_sk) {
177 		ktp = bpf_skc_to_tcp_sock(sk);
178 
179 		if (!ktp)
180 			RET_LOG();
181 
182 		lsndtime = ktp->lsndtime;
183 
184 		child_cg_id = bpf_sk_cgroup_id(ktp);
185 		if (!child_cg_id)
186 			RET_LOG();
187 
188 		parent_cg_id = bpf_sk_ancestor_cgroup_id(ktp, 2);
189 		if (!parent_cg_id)
190 			RET_LOG();
191 
192 		/* The userspace has created it for srv sk */
193 		pkt_out_cnt = bpf_sk_storage_get(&sk_pkt_out_cnt, ktp, 0, 0);
194 		pkt_out_cnt10 = bpf_sk_storage_get(&sk_pkt_out_cnt10, ktp,
195 						   0, 0);
196 	} else {
197 		pkt_out_cnt = bpf_sk_storage_get(&sk_pkt_out_cnt, sk,
198 						 &cli_cnt_init,
199 						 BPF_SK_STORAGE_GET_F_CREATE);
200 		pkt_out_cnt10 = bpf_sk_storage_get(&sk_pkt_out_cnt10,
201 						   sk, &cli_cnt_init,
202 						   BPF_SK_STORAGE_GET_F_CREATE);
203 	}
204 
205 	if (!pkt_out_cnt || !pkt_out_cnt10)
206 		RET_LOG();
207 
208 	/* Even both cnt and cnt10 have lock defined in their BTF,
209 	 * intentionally one cnt takes lock while one does not
210 	 * as a test for the spinlock support in BPF_MAP_TYPE_SK_STORAGE.
211 	 */
212 	pkt_out_cnt->cnt += 1;
213 	bpf_spin_lock(&pkt_out_cnt10->lock);
214 	pkt_out_cnt10->cnt += 10;
215 	bpf_spin_unlock(&pkt_out_cnt10->lock);
216 
217 	return CG_OK;
218 }
219 
220 SEC("cgroup_skb/ingress")
ingress_read_sock_fields(struct __sk_buff * skb)221 int ingress_read_sock_fields(struct __sk_buff *skb)
222 {
223 	struct bpf_tcp_sock *tp;
224 	__u32 linum, linum_idx;
225 	struct bpf_sock *sk;
226 
227 	linum_idx = INGRESS_LINUM_IDX;
228 
229 	sk = skb->sk;
230 	if (!sk)
231 		RET_LOG();
232 
233 	/* Not the testing ingress traffic to the server */
234 	if (sk->family != AF_INET6 || !is_loopback6(sk->src_ip6) ||
235 	    sk->src_port != bpf_ntohs(srv_sa6.sin6_port))
236 		return CG_OK;
237 
238 	/* Only interested in the listening socket */
239 	if (sk->state != BPF_TCP_LISTEN)
240 		return CG_OK;
241 
242 	/* It must be a fullsock for cgroup_skb/ingress prog */
243 	sk = bpf_sk_fullsock(sk);
244 	if (!sk)
245 		RET_LOG();
246 
247 	tp = bpf_tcp_sock(sk);
248 	if (!tp)
249 		RET_LOG();
250 
251 	skcpy(&listen_sk, sk);
252 	tpcpy(&listen_tp, tp);
253 
254 	return CG_OK;
255 }
256 
257 /*
258  * NOTE: 4-byte load from bpf_sock at dst_port offset is quirky. It
259  * gets rewritten by the access converter to a 2-byte load for
260  * backward compatibility. Treating the load result as a be16 value
261  * makes the code portable across little- and big-endian platforms.
262  */
sk_dst_port__load_word(struct bpf_sock * sk)263 static __noinline bool sk_dst_port__load_word(struct bpf_sock *sk)
264 {
265 	__u32 *word = (__u32 *)&sk->dst_port;
266 	return word[0] == bpf_htons(0xcafe);
267 }
268 
sk_dst_port__load_half(struct bpf_sock * sk)269 static __noinline bool sk_dst_port__load_half(struct bpf_sock *sk)
270 {
271 	__u16 *half;
272 
273 	asm volatile ("");
274 	half = (__u16 *)&sk->dst_port;
275 	return half[0] == bpf_htons(0xcafe);
276 }
277 
sk_dst_port__load_byte(struct bpf_sock * sk)278 static __noinline bool sk_dst_port__load_byte(struct bpf_sock *sk)
279 {
280 	__u8 *byte = (__u8 *)&sk->dst_port;
281 	return byte[0] == 0xca && byte[1] == 0xfe;
282 }
283 
284 SEC("cgroup_skb/egress")
read_sk_dst_port(struct __sk_buff * skb)285 int read_sk_dst_port(struct __sk_buff *skb)
286 {
287 	__u32 linum, linum_idx;
288 	struct bpf_sock *sk;
289 
290 	linum_idx = READ_SK_DST_PORT_LINUM_IDX;
291 
292 	sk = skb->sk;
293 	if (!sk)
294 		RET_LOG();
295 
296 	/* Ignore everything but the SYN from the client socket */
297 	if (sk->state != BPF_TCP_SYN_SENT)
298 		return CG_OK;
299 
300 	if (!sk_dst_port__load_word(sk))
301 		RET_LOG();
302 	if (!sk_dst_port__load_half(sk))
303 		RET_LOG();
304 	if (!sk_dst_port__load_byte(sk))
305 		RET_LOG();
306 
307 	return CG_OK;
308 }
309 
310 char _license[] SEC("license") = "GPL";
311