|
| 1 | +// +build linux |
| 2 | + |
| 3 | +package native |
| 4 | + |
| 5 | +import "github.com/opencontainers/runc/libcontainer/configs" |
| 6 | + |
| 7 | +var defaultSeccompProfile = &configs.Seccomp{ |
| 8 | + DefaultAction: configs.Allow, |
| 9 | + Syscalls: []*configs.Syscall{ |
| 10 | + { |
| 11 | + // Quota and Accounting syscalls which could let containers |
| 12 | + // disable their own resource limits or process accounting |
| 13 | + Name: "acct", |
| 14 | + Action: configs.Errno, |
| 15 | + Args: []*configs.Arg{}, |
| 16 | + }, |
| 17 | + { |
| 18 | + // Prevent containers from using the kernel keyring, |
| 19 | + // which is not namespaced |
| 20 | + Name: "add_key", |
| 21 | + Action: configs.Errno, |
| 22 | + Args: []*configs.Arg{}, |
| 23 | + }, |
| 24 | + { |
| 25 | + // Similar to clock_settime and settimeofday |
| 26 | + // Time/Date is not namespaced |
| 27 | + Name: "adjtimex", |
| 28 | + Action: configs.Errno, |
| 29 | + Args: []*configs.Arg{}, |
| 30 | + }, |
| 31 | + { |
| 32 | + // Time/Date is not namespaced |
| 33 | + Name: "clock_settime", |
| 34 | + Action: configs.Errno, |
| 35 | + Args: []*configs.Arg{}, |
| 36 | + }, |
| 37 | + { |
| 38 | + // Deny cloning new namespaces |
| 39 | + Name: "clone", |
| 40 | + Action: configs.Errno, |
| 41 | + Args: []*configs.Arg{ |
| 42 | + { |
| 43 | + // flags from sched.h |
| 44 | + // CLONE_NEWUTS 0x04000000 |
| 45 | + // CLONE_NEWIPC 0x08000000 |
| 46 | + // CLONE_NEWUSER 0x10000000 |
| 47 | + // CLONE_NEWPID 0x20000000 |
| 48 | + // CLONE_NEWNET 0x40000000 |
| 49 | + Index: 0, |
| 50 | + Value: uint64(0x04000000), |
| 51 | + Op: configs.GreaterThanOrEqualTo, |
| 52 | + }, |
| 53 | + { |
| 54 | + // flags from sched.h |
| 55 | + // CLONE_NEWNS 0x00020000 |
| 56 | + Index: 0, |
| 57 | + Value: uint64(0x00020000), |
| 58 | + Op: configs.EqualTo, |
| 59 | + }, |
| 60 | + }, |
| 61 | + }, |
| 62 | + { |
| 63 | + // Deny manipulation and functions on kernel modules. |
| 64 | + Name: "create_module", |
| 65 | + Action: configs.Errno, |
| 66 | + Args: []*configs.Arg{}, |
| 67 | + }, |
| 68 | + { |
| 69 | + // Deny manipulation and functions on kernel modules. |
| 70 | + Name: "delete_module", |
| 71 | + Action: configs.Errno, |
| 72 | + Args: []*configs.Arg{}, |
| 73 | + }, |
| 74 | + { |
| 75 | + // Deny retrieval of exported kernel and module symbols |
| 76 | + Name: "get_kernel_syms", |
| 77 | + Action: configs.Errno, |
| 78 | + Args: []*configs.Arg{}, |
| 79 | + }, |
| 80 | + { |
| 81 | + // Terrifying syscalls that modify kernel memory and NUMA settings. |
| 82 | + // They're gated by CAP_SYS_NICE, |
| 83 | + // which we do not retain by default in containers. |
| 84 | + Name: "get_mempolicy", |
| 85 | + Action: configs.Errno, |
| 86 | + Args: []*configs.Arg{}, |
| 87 | + }, |
| 88 | + { |
| 89 | + // Deny getting the list of robust futexes |
| 90 | + Name: "get_robust_list", |
| 91 | + Action: configs.Errno, |
| 92 | + Args: []*configs.Arg{}, |
| 93 | + }, |
| 94 | + { |
| 95 | + // Deny manipulation and functions on kernel modules. |
| 96 | + Name: "init_module", |
| 97 | + Action: configs.Errno, |
| 98 | + Args: []*configs.Arg{}, |
| 99 | + }, |
| 100 | + { |
| 101 | + // Prevent containers from modifying kernel I/O privilege levels. |
| 102 | + // Already restricted as containers drop CAP_SYS_RAWIO by default. |
| 103 | + Name: "ioperm", |
| 104 | + Action: configs.Errno, |
| 105 | + Args: []*configs.Arg{}, |
| 106 | + }, |
| 107 | + { |
| 108 | + // Prevent containers from modifying kernel I/O privilege levels. |
| 109 | + // Already restricted as containers drop CAP_SYS_RAWIO by default. |
| 110 | + Name: "iopl", |
| 111 | + Action: configs.Errno, |
| 112 | + Args: []*configs.Arg{}, |
| 113 | + }, |
| 114 | + { |
| 115 | + // Sister syscall of kexec_load that does the same thing, |
| 116 | + // slightly different arguments |
| 117 | + Name: "kexec_file_load", |
| 118 | + Action: configs.Errno, |
| 119 | + Args: []*configs.Arg{}, |
| 120 | + }, |
| 121 | + { |
| 122 | + // Deny loading a new kernel for later execution |
| 123 | + Name: "kexec_load", |
| 124 | + Action: configs.Errno, |
| 125 | + Args: []*configs.Arg{}, |
| 126 | + }, |
| 127 | + { |
| 128 | + // Prevent containers from using the kernel keyring, |
| 129 | + // which is not namespaced |
| 130 | + Name: "keyctl", |
| 131 | + Action: configs.Errno, |
| 132 | + Args: []*configs.Arg{}, |
| 133 | + }, |
| 134 | + { |
| 135 | + // Tracing/profiling syscalls, |
| 136 | + // which could leak a lot of information on the host |
| 137 | + Name: "lookup_dcookie", |
| 138 | + Action: configs.Errno, |
| 139 | + Args: []*configs.Arg{}, |
| 140 | + }, |
| 141 | + { |
| 142 | + // Terrifying syscalls that modify kernel memory and NUMA settings. |
| 143 | + // They're gated by CAP_SYS_NICE, |
| 144 | + // which we do not retain by default in containers. |
| 145 | + Name: "mbind", |
| 146 | + Action: configs.Errno, |
| 147 | + Args: []*configs.Arg{}, |
| 148 | + }, |
| 149 | + { |
| 150 | + // Terrifying syscalls that modify kernel memory and NUMA settings. |
| 151 | + // They're gated by CAP_SYS_NICE, |
| 152 | + // which we do not retain by default in containers. |
| 153 | + Name: "migrate_pages", |
| 154 | + Action: configs.Errno, |
| 155 | + Args: []*configs.Arg{}, |
| 156 | + }, |
| 157 | + { |
| 158 | + // Old syscall only used in 16-bit code, |
| 159 | + // and a potential information leak |
| 160 | + Name: "modify_ldt", |
| 161 | + Action: configs.Errno, |
| 162 | + Args: []*configs.Arg{}, |
| 163 | + }, |
| 164 | + { |
| 165 | + // Deny mount |
| 166 | + Name: "mount", |
| 167 | + Action: configs.Errno, |
| 168 | + Args: []*configs.Arg{}, |
| 169 | + }, |
| 170 | + { |
| 171 | + // Terrifying syscalls that modify kernel memory and NUMA settings. |
| 172 | + // They're gated by CAP_SYS_NICE, |
| 173 | + // which we do not retain by default in containers. |
| 174 | + Name: "move_pages", |
| 175 | + Action: configs.Errno, |
| 176 | + Args: []*configs.Arg{}, |
| 177 | + }, |
| 178 | + { |
| 179 | + // Deny interaction with the kernel nfs daemon |
| 180 | + Name: "nfsservctl", |
| 181 | + Action: configs.Errno, |
| 182 | + Args: []*configs.Arg{}, |
| 183 | + }, |
| 184 | + { |
| 185 | + // Cause of an old container breakout, |
| 186 | + // might as well restrict it to be on the safe side |
| 187 | + Name: "open_by_handle_at", |
| 188 | + Action: configs.Errno, |
| 189 | + Args: []*configs.Arg{}, |
| 190 | + }, |
| 191 | + { |
| 192 | + // Tracing/profiling syscalls, |
| 193 | + // which could leak a lot of information on the host |
| 194 | + Name: "perf_event_open", |
| 195 | + Action: configs.Errno, |
| 196 | + Args: []*configs.Arg{}, |
| 197 | + }, |
| 198 | + { |
| 199 | + // Prevent container from enabling BSD emulation. |
| 200 | + // Not inherently dangerous, but poorly tested, |
| 201 | + // potential for a lot of kernel vulns in this. |
| 202 | + Name: "personality", |
| 203 | + Action: configs.Errno, |
| 204 | + Args: []*configs.Arg{}, |
| 205 | + }, |
| 206 | + { |
| 207 | + // Deny pivot_root |
| 208 | + Name: "pivot_root", |
| 209 | + Action: configs.Errno, |
| 210 | + Args: []*configs.Arg{}, |
| 211 | + }, |
| 212 | + { |
| 213 | + // Already blocked by dropping CAP_PTRACE |
| 214 | + Name: "ptrace", |
| 215 | + Action: configs.Errno, |
| 216 | + Args: []*configs.Arg{}, |
| 217 | + }, |
| 218 | + { |
| 219 | + // Deny manipulation and functions on kernel modules. |
| 220 | + Name: "query_module", |
| 221 | + Action: configs.Errno, |
| 222 | + Args: []*configs.Arg{}, |
| 223 | + }, |
| 224 | + { |
| 225 | + // Quota and Accounting syscalls which could let containers |
| 226 | + // disable their own resource limits or process accounting |
| 227 | + Name: "quotactl", |
| 228 | + Action: configs.Errno, |
| 229 | + Args: []*configs.Arg{}, |
| 230 | + }, |
| 231 | + { |
| 232 | + // Probably a bad idea to let containers reboot the host |
| 233 | + Name: "reboot", |
| 234 | + Action: configs.Errno, |
| 235 | + Args: []*configs.Arg{}, |
| 236 | + }, |
| 237 | + { |
| 238 | + // Probably a bad idea to let containers restart |
| 239 | + Name: "restart_syscall", |
| 240 | + Action: configs.Errno, |
| 241 | + Args: []*configs.Arg{}, |
| 242 | + }, |
| 243 | + { |
| 244 | + // Prevent containers from using the kernel keyring, |
| 245 | + // which is not namespaced |
| 246 | + Name: "request_key", |
| 247 | + Action: configs.Errno, |
| 248 | + Args: []*configs.Arg{}, |
| 249 | + }, |
| 250 | + { |
| 251 | + // meta, deny seccomp |
| 252 | + Name: "seccomp", |
| 253 | + Action: configs.Errno, |
| 254 | + Args: []*configs.Arg{}, |
| 255 | + }, |
| 256 | + { |
| 257 | + // Terrifying syscalls that modify kernel memory and NUMA settings. |
| 258 | + // They're gated by CAP_SYS_NICE, |
| 259 | + // which we do not retain by default in containers. |
| 260 | + Name: "set_mempolicy", |
| 261 | + Action: configs.Errno, |
| 262 | + Args: []*configs.Arg{}, |
| 263 | + }, |
| 264 | + { |
| 265 | + // deny associating a thread with a namespace |
| 266 | + Name: "setns", |
| 267 | + Action: configs.Errno, |
| 268 | + Args: []*configs.Arg{}, |
| 269 | + }, |
| 270 | + { |
| 271 | + // Deny setting the list of robust futexes |
| 272 | + Name: "set_robust_list", |
| 273 | + Action: configs.Errno, |
| 274 | + Args: []*configs.Arg{}, |
| 275 | + }, |
| 276 | + { |
| 277 | + // Time/Date is not namespaced |
| 278 | + Name: "settimeofday", |
| 279 | + Action: configs.Errno, |
| 280 | + Args: []*configs.Arg{}, |
| 281 | + }, |
| 282 | + { |
| 283 | + // Deny start/stop swapping to file/device |
| 284 | + Name: "swapon", |
| 285 | + Action: configs.Errno, |
| 286 | + Args: []*configs.Arg{}, |
| 287 | + }, |
| 288 | + { |
| 289 | + // Deny start/stop swapping to file/device |
| 290 | + Name: "swapoff", |
| 291 | + Action: configs.Errno, |
| 292 | + Args: []*configs.Arg{}, |
| 293 | + }, |
| 294 | + { |
| 295 | + // Deny read/write system parameters |
| 296 | + Name: "_sysctl", |
| 297 | + Action: configs.Errno, |
| 298 | + Args: []*configs.Arg{}, |
| 299 | + }, |
| 300 | + { |
| 301 | + // Deny umount |
| 302 | + Name: "umount2", |
| 303 | + Action: configs.Errno, |
| 304 | + Args: []*configs.Arg{}, |
| 305 | + }, |
| 306 | + { |
| 307 | + // Same as clone |
| 308 | + Name: "unshare", |
| 309 | + Action: configs.Errno, |
| 310 | + Args: []*configs.Arg{}, |
| 311 | + }, |
| 312 | + { |
| 313 | + // Older syscall related to shared libraries, unused for a long time |
| 314 | + Name: "uselib", |
| 315 | + Action: configs.Errno, |
| 316 | + Args: []*configs.Arg{}, |
| 317 | + }, |
| 318 | + }, |
| 319 | +} |
0 commit comments