1 // SPDX-License-Identifier: GPL-2.0 2 #include <string.h> 3 #include <linux/tcp.h> 4 #include <linux/bpf.h> 5 #include <netinet/in.h> 6 #include <bpf/bpf_helpers.h> 7 8 char _license[] SEC("license") = "GPL"; 9 10 int page_size = 0; /* userspace should set it */ 11 12 #ifndef SOL_TCP 13 #define SOL_TCP IPPROTO_TCP 14 #endif 15 16 #define SOL_CUSTOM 0xdeadbeef 17 18 struct sockopt_sk { 19 __u8 val; 20 }; 21 22 struct { 23 __uint(type, BPF_MAP_TYPE_SK_STORAGE); 24 __uint(map_flags, BPF_F_NO_PREALLOC); 25 __type(key, int); 26 __type(value, struct sockopt_sk); 27 } socket_storage_map SEC(".maps"); 28 29 SEC("cgroup/getsockopt") 30 int _getsockopt(struct bpf_sockopt *ctx) 31 { 32 __u8 *optval_end = ctx->optval_end; 33 __u8 *optval = ctx->optval; 34 struct sockopt_sk *storage; 35 36 /* Make sure bpf_get_netns_cookie is callable. 37 */ 38 if (bpf_get_netns_cookie(NULL) == 0) 39 return 0; 40 41 if (bpf_get_netns_cookie(ctx) == 0) 42 return 0; 43 44 if (ctx->level == SOL_IP && ctx->optname == IP_TOS) { 45 /* Not interested in SOL_IP:IP_TOS; 46 * let next BPF program in the cgroup chain or kernel 47 * handle it. 48 */ 49 ctx->optlen = 0; /* bypass optval>PAGE_SIZE */ 50 return 1; 51 } 52 53 if (ctx->level == SOL_SOCKET && ctx->optname == SO_SNDBUF) { 54 /* Not interested in SOL_SOCKET:SO_SNDBUF; 55 * let next BPF program in the cgroup chain or kernel 56 * handle it. 57 */ 58 return 1; 59 } 60 61 if (ctx->level == SOL_TCP && ctx->optname == TCP_CONGESTION) { 62 /* Not interested in SOL_TCP:TCP_CONGESTION; 63 * let next BPF program in the cgroup chain or kernel 64 * handle it. 65 */ 66 return 1; 67 } 68 69 if (ctx->level == SOL_TCP && ctx->optname == TCP_ZEROCOPY_RECEIVE) { 70 /* Verify that TCP_ZEROCOPY_RECEIVE triggers. 71 * It has a custom implementation for performance 72 * reasons. 73 */ 74 75 /* Check that optval contains address (__u64) */ 76 if (optval + sizeof(__u64) > optval_end) 77 return 0; /* bounds check */ 78 79 if (((struct tcp_zerocopy_receive *)optval)->address != 0) 80 return 0; /* unexpected data */ 81 82 return 1; 83 } 84 85 if (ctx->level == SOL_IP && ctx->optname == IP_FREEBIND) { 86 if (optval + 1 > optval_end) 87 return 0; /* bounds check */ 88 89 ctx->retval = 0; /* Reset system call return value to zero */ 90 91 /* Always export 0x55 */ 92 optval[0] = 0x55; 93 ctx->optlen = 1; 94 95 /* Userspace buffer is PAGE_SIZE * 2, but BPF 96 * program can only see the first PAGE_SIZE 97 * bytes of data. 98 */ 99 if (optval_end - optval != page_size) 100 return 0; /* unexpected data size */ 101 102 return 1; 103 } 104 105 if (ctx->level != SOL_CUSTOM) 106 return 0; /* deny everything except custom level */ 107 108 if (optval + 1 > optval_end) 109 return 0; /* bounds check */ 110 111 storage = bpf_sk_storage_get(&socket_storage_map, ctx->sk, 0, 112 BPF_SK_STORAGE_GET_F_CREATE); 113 if (!storage) 114 return 0; /* couldn't get sk storage */ 115 116 if (!ctx->retval) 117 return 0; /* kernel should not have handled 118 * SOL_CUSTOM, something is wrong! 119 */ 120 ctx->retval = 0; /* Reset system call return value to zero */ 121 122 optval[0] = storage->val; 123 ctx->optlen = 1; 124 125 return 1; 126 } 127 128 SEC("cgroup/setsockopt") 129 int _setsockopt(struct bpf_sockopt *ctx) 130 { 131 __u8 *optval_end = ctx->optval_end; 132 __u8 *optval = ctx->optval; 133 struct sockopt_sk *storage; 134 135 /* Make sure bpf_get_netns_cookie is callable. 136 */ 137 if (bpf_get_netns_cookie(NULL) == 0) 138 return 0; 139 140 if (bpf_get_netns_cookie(ctx) == 0) 141 return 0; 142 143 if (ctx->level == SOL_IP && ctx->optname == IP_TOS) { 144 /* Not interested in SOL_IP:IP_TOS; 145 * let next BPF program in the cgroup chain or kernel 146 * handle it. 147 */ 148 ctx->optlen = 0; /* bypass optval>PAGE_SIZE */ 149 return 1; 150 } 151 152 if (ctx->level == SOL_SOCKET && ctx->optname == SO_SNDBUF) { 153 /* Overwrite SO_SNDBUF value */ 154 155 if (optval + sizeof(__u32) > optval_end) 156 return 0; /* bounds check */ 157 158 *(__u32 *)optval = 0x55AA; 159 ctx->optlen = 4; 160 161 return 1; 162 } 163 164 if (ctx->level == SOL_TCP && ctx->optname == TCP_CONGESTION) { 165 /* Always use cubic */ 166 167 if (optval + 5 > optval_end) 168 return 0; /* bounds check */ 169 170 memcpy(optval, "cubic", 5); 171 ctx->optlen = 5; 172 173 return 1; 174 } 175 176 if (ctx->level == SOL_IP && ctx->optname == IP_FREEBIND) { 177 /* Original optlen is larger than PAGE_SIZE. */ 178 if (ctx->optlen != page_size * 2) 179 return 0; /* unexpected data size */ 180 181 if (optval + 1 > optval_end) 182 return 0; /* bounds check */ 183 184 /* Make sure we can trim the buffer. */ 185 optval[0] = 0; 186 ctx->optlen = 1; 187 188 /* Usepace buffer is PAGE_SIZE * 2, but BPF 189 * program can only see the first PAGE_SIZE 190 * bytes of data. 191 */ 192 if (optval_end - optval != page_size) 193 return 0; /* unexpected data size */ 194 195 return 1; 196 } 197 198 if (ctx->level != SOL_CUSTOM) 199 return 0; /* deny everything except custom level */ 200 201 if (optval + 1 > optval_end) 202 return 0; /* bounds check */ 203 204 storage = bpf_sk_storage_get(&socket_storage_map, ctx->sk, 0, 205 BPF_SK_STORAGE_GET_F_CREATE); 206 if (!storage) 207 return 0; /* couldn't get sk storage */ 208 209 storage->val = optval[0]; 210 ctx->optlen = -1; /* BPF has consumed this option, don't call kernel 211 * setsockopt handler. 212 */ 213 214 return 1; 215 } 216