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 if (ctx->level == SOL_IP && ctx->optname == IP_TOS) { 37 /* Not interested in SOL_IP:IP_TOS; 38 * let next BPF program in the cgroup chain or kernel 39 * handle it. 40 */ 41 ctx->optlen = 0; /* bypass optval>PAGE_SIZE */ 42 return 1; 43 } 44 45 if (ctx->level == SOL_SOCKET && ctx->optname == SO_SNDBUF) { 46 /* Not interested in SOL_SOCKET:SO_SNDBUF; 47 * let next BPF program in the cgroup chain or kernel 48 * handle it. 49 */ 50 return 1; 51 } 52 53 if (ctx->level == SOL_TCP && ctx->optname == TCP_CONGESTION) { 54 /* Not interested in SOL_TCP:TCP_CONGESTION; 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_ZEROCOPY_RECEIVE) { 62 /* Verify that TCP_ZEROCOPY_RECEIVE triggers. 63 * It has a custom implementation for performance 64 * reasons. 65 */ 66 67 if (optval + sizeof(struct tcp_zerocopy_receive) > optval_end) 68 return 0; /* EPERM, bounds check */ 69 70 if (((struct tcp_zerocopy_receive *)optval)->address != 0) 71 return 0; /* EPERM, unexpected data */ 72 73 return 1; 74 } 75 76 if (ctx->level == SOL_IP && ctx->optname == IP_FREEBIND) { 77 if (optval + 1 > optval_end) 78 return 0; /* EPERM, bounds check */ 79 80 ctx->retval = 0; /* Reset system call return value to zero */ 81 82 /* Always export 0x55 */ 83 optval[0] = 0x55; 84 ctx->optlen = 1; 85 86 /* Userspace buffer is PAGE_SIZE * 2, but BPF 87 * program can only see the first PAGE_SIZE 88 * bytes of data. 89 */ 90 if (optval_end - optval != page_size) 91 return 0; /* EPERM, unexpected data size */ 92 93 return 1; 94 } 95 96 if (ctx->level != SOL_CUSTOM) 97 return 0; /* EPERM, deny everything except custom level */ 98 99 if (optval + 1 > optval_end) 100 return 0; /* EPERM, bounds check */ 101 102 storage = bpf_sk_storage_get(&socket_storage_map, ctx->sk, 0, 103 BPF_SK_STORAGE_GET_F_CREATE); 104 if (!storage) 105 return 0; /* EPERM, couldn't get sk storage */ 106 107 if (!ctx->retval) 108 return 0; /* EPERM, kernel should not have handled 109 * SOL_CUSTOM, something is wrong! 110 */ 111 ctx->retval = 0; /* Reset system call return value to zero */ 112 113 optval[0] = storage->val; 114 ctx->optlen = 1; 115 116 return 1; 117 } 118 119 SEC("cgroup/setsockopt") 120 int _setsockopt(struct bpf_sockopt *ctx) 121 { 122 __u8 *optval_end = ctx->optval_end; 123 __u8 *optval = ctx->optval; 124 struct sockopt_sk *storage; 125 126 if (ctx->level == SOL_IP && ctx->optname == IP_TOS) { 127 /* Not interested in SOL_IP:IP_TOS; 128 * let next BPF program in the cgroup chain or kernel 129 * handle it. 130 */ 131 ctx->optlen = 0; /* bypass optval>PAGE_SIZE */ 132 return 1; 133 } 134 135 if (ctx->level == SOL_SOCKET && ctx->optname == SO_SNDBUF) { 136 /* Overwrite SO_SNDBUF value */ 137 138 if (optval + sizeof(__u32) > optval_end) 139 return 0; /* EPERM, bounds check */ 140 141 *(__u32 *)optval = 0x55AA; 142 ctx->optlen = 4; 143 144 return 1; 145 } 146 147 if (ctx->level == SOL_TCP && ctx->optname == TCP_CONGESTION) { 148 /* Always use cubic */ 149 150 if (optval + 5 > optval_end) 151 return 0; /* EPERM, bounds check */ 152 153 memcpy(optval, "cubic", 5); 154 ctx->optlen = 5; 155 156 return 1; 157 } 158 159 if (ctx->level == SOL_IP && ctx->optname == IP_FREEBIND) { 160 /* Original optlen is larger than PAGE_SIZE. */ 161 if (ctx->optlen != page_size * 2) 162 return 0; /* EPERM, unexpected data size */ 163 164 if (optval + 1 > optval_end) 165 return 0; /* EPERM, bounds check */ 166 167 /* Make sure we can trim the buffer. */ 168 optval[0] = 0; 169 ctx->optlen = 1; 170 171 /* Usepace buffer is PAGE_SIZE * 2, but BPF 172 * program can only see the first PAGE_SIZE 173 * bytes of data. 174 */ 175 if (optval_end - optval != page_size) 176 return 0; /* EPERM, unexpected data size */ 177 178 return 1; 179 } 180 181 if (ctx->level != SOL_CUSTOM) 182 return 0; /* EPERM, deny everything except custom level */ 183 184 if (optval + 1 > optval_end) 185 return 0; /* EPERM, bounds check */ 186 187 storage = bpf_sk_storage_get(&socket_storage_map, ctx->sk, 0, 188 BPF_SK_STORAGE_GET_F_CREATE); 189 if (!storage) 190 return 0; /* EPERM, couldn't get sk storage */ 191 192 storage->val = optval[0]; 193 ctx->optlen = -1; /* BPF has consumed this option, don't call kernel 194 * setsockopt handler. 195 */ 196 197 return 1; 198 } 199