| 1 |
|
| 2 | |
| 3 | * syscall_wrapper.h - x86 specific wrappers to syscall definitions
|
| 4 |
|
| 5 |
|
| 6 | #ifndef _ASM_X86_SYSCALL_WRAPPER_H
|
| 7 | #define _ASM_X86_SYSCALL_WRAPPER_H
|
| 8 |
|
| 9 | #include <asm/ptrace.h>
|
| 10 |
|
| 11 | extern long __x64_sys_ni_syscall(const struct pt_regs *regs);
|
| 12 | extern long __ia32_sys_ni_syscall(const struct pt_regs *regs);
|
| 13 |
|
| 14 | |
| 15 | * Instead of the generic __SYSCALL_DEFINEx() definition, the x86 version takes
|
| 16 | * struct pt_regs *regs as the only argument of the syscall stub(s) named as:
|
| 17 | * __x64_sys_*() - 64-bit native syscall
|
| 18 | * __ia32_sys_*() - 32-bit native syscall or common compat syscall
|
| 19 | * __ia32_compat_sys_*() - 32-bit compat syscall
|
| 20 | * __x64_compat_sys_*() - 64-bit X32 compat syscall
|
| 21 | *
|
| 22 | * The registers are decoded according to the ABI:
|
| 23 | * 64-bit: RDI, RSI, RDX, R10, R8, R9
|
| 24 | * 32-bit: EBX, ECX, EDX, ESI, EDI, EBP
|
| 25 | *
|
| 26 | * The stub then passes the decoded arguments to the __se_sys_*() wrapper to
|
| 27 | * perform sign-extension (omitted for zero-argument syscalls). Finally the
|
| 28 | * arguments are passed to the __do_sys_*() function which is the actual
|
| 29 | * syscall. These wrappers are marked as inline so the compiler can optimize
|
| 30 | * the functions where appropriate.
|
| 31 | *
|
| 32 | * Example assembly (slightly re-ordered for better readability):
|
| 33 | *
|
| 34 | * <__x64_sys_recv>: <-- syscall with 4 parameters
|
| 35 | * callq <__fentry__>
|
| 36 | *
|
| 37 | * mov 0x70(%rdi),%rdi <-- decode regs->di
|
| 38 | * mov 0x68(%rdi),%rsi <-- decode regs->si
|
| 39 | * mov 0x60(%rdi),%rdx <-- decode regs->dx
|
| 40 | * mov 0x38(%rdi),%rcx <-- decode regs->r10
|
| 41 | *
|
| 42 | * xor %r9d,%r9d <-- clear %r9
|
| 43 | * xor %r8d,%r8d <-- clear %r8
|
| 44 | *
|
| 45 | * callq __sys_recvfrom <-- do the actual work in __sys_recvfrom()
|
| 46 | * which takes 6 arguments
|
| 47 | *
|
| 48 | * cltq <-- extend return value to 64-bit
|
| 49 | * retq <-- return
|
| 50 | *
|
| 51 | * This approach avoids leaking random user-provided register content down
|
| 52 | * the call chain.
|
| 53 |
|
| 54 |
|
| 55 |
|
| 56 | #define SC_X86_64_REGS_TO_ARGS(x, ...) \
|
| 57 | __MAP(x,__SC_ARGS \
|
| 58 | ,,regs->di,,regs->si,,regs->dx \
|
| 59 | ,,regs->r10,,regs->r8,,regs->r9) \
|
| 60 |
|
| 61 |
|
| 62 |
|
| 63 | #define SYSCALL_PT_ARG6(m, t1, t2, t3, t4, t5, t6) \
|
| 64 | SYSCALL_PT_ARG5(m, t1, t2, t3, t4, t5), m(t6, (regs->bp))
|
| 65 | #define SYSCALL_PT_ARG5(m, t1, t2, t3, t4, t5) \
|
| 66 | SYSCALL_PT_ARG4(m, t1, t2, t3, t4), m(t5, (regs->di))
|
| 67 | #define SYSCALL_PT_ARG4(m, t1, t2, t3, t4) \
|
| 68 | SYSCALL_PT_ARG3(m, t1, t2, t3), m(t4, (regs->si))
|
| 69 | #define SYSCALL_PT_ARG3(m, t1, t2, t3) \
|
| 70 | SYSCALL_PT_ARG2(m, t1, t2), m(t3, (regs->dx))
|
| 71 | #define SYSCALL_PT_ARG2(m, t1, t2) \
|
| 72 | SYSCALL_PT_ARG1(m, t1), m(t2, (regs->cx))
|
| 73 | #define SYSCALL_PT_ARG1(m, t1) m(t1, (regs->bx))
|
| 74 | #define SYSCALL_PT_ARGS(x, ...) SYSCALL_PT_ARG##x(__VA_ARGS__)
|
| 75 |
|
| 76 | #define __SC_COMPAT_CAST(t, a) \
|
| 77 | (__typeof(__builtin_choose_expr(__TYPE_IS_L(t), 0, 0U))) \
|
| 78 | (unsigned int)a
|
| 79 |
|
| 80 |
|
| 81 | #define SC_IA32_REGS_TO_ARGS(x, ...) \
|
| 82 | SYSCALL_PT_ARGS(x, __SC_COMPAT_CAST, \
|
| 83 | __MAP(x, __SC_TYPE, __VA_ARGS__)) \
|
| 84 |
|
| 85 | #define __SYS_STUB0(abi, name) \
|
| 86 | long __##abi##_##name(const struct pt_regs *regs); \
|
| 87 | ALLOW_ERROR_INJECTION(__##abi##_##name, ERRNO); \
|
| 88 | long __##abi##_##name(const struct pt_regs *regs) \
|
| 89 | __alias(__do_##name);
|
| 90 |
|
| 91 | #define __SYS_STUBx(abi, name, ...) \
|
| 92 | long __##abi##_##name(const struct pt_regs *regs); \
|
| 93 | ALLOW_ERROR_INJECTION(__##abi##_##name, ERRNO); \
|
| 94 | long __##abi##_##name(const struct pt_regs *regs) \
|
| 95 | { \
|
| 96 | return __se_##name(__VA_ARGS__); \
|
| 97 | }
|
| 98 |
|
| 99 | #define __COND_SYSCALL(abi, name) \
|
| 100 | __weak long __##abi##_##name(const struct pt_regs *__unused); \
|
| 101 | __weak long __##abi##_##name(const struct pt_regs *__unused) \
|
| 102 | { \
|
| 103 | return sys_ni_syscall(); \
|
| 104 | }
|
| 105 |
|
| 106 | #ifdef CONFIG_X86_64
|
| 107 | #define __X64_SYS_STUB0(name) \
|
| 108 | __SYS_STUB0(x64, sys_##name)
|
| 109 |
|
| 110 | #define __X64_SYS_STUBx(x, name, ...) \
|
| 111 | __SYS_STUBx(x64, sys##name, \
|
| 112 | SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__))
|
| 113 |
|
| 114 | #define __X64_COND_SYSCALL(name) \
|
| 115 | __COND_SYSCALL(x64, sys_##name)
|
| 116 |
|
| 117 | #else
|
| 118 | #define __X64_SYS_STUB0(name)
|
| 119 | #define __X64_SYS_STUBx(x, name, ...)
|
| 120 | #define __X64_COND_SYSCALL(name)
|
| 121 | #endif
|
| 122 |
|
| 123 | #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
|
| 124 | #define __IA32_SYS_STUB0(name) \
|
| 125 | __SYS_STUB0(ia32, sys_##name)
|
| 126 |
|
| 127 | #define __IA32_SYS_STUBx(x, name, ...) \
|
| 128 | __SYS_STUBx(ia32, sys##name, \
|
| 129 | SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__))
|
| 130 |
|
| 131 | #define __IA32_COND_SYSCALL(name) \
|
| 132 | __COND_SYSCALL(ia32, sys_##name)
|
| 133 |
|
| 134 | #else
|
| 135 | #define __IA32_SYS_STUB0(name)
|
| 136 | #define __IA32_SYS_STUBx(x, name, ...)
|
| 137 | #define __IA32_COND_SYSCALL(name)
|
| 138 | #endif
|
| 139 |
|
| 140 | #ifdef CONFIG_IA32_EMULATION
|
| 141 | |
| 142 | * For IA32 emulation, we need to handle "compat" syscalls *and* create
|
| 143 | * additional wrappers (aptly named __ia32_sys_xyzzy) which decode the
|
| 144 | * ia32 regs in the proper order for shared or "common" syscalls. As some
|
| 145 | * syscalls may not be implemented, we need to expand COND_SYSCALL in
|
| 146 | * kernel/sys_ni.c to cover this case as well.
|
| 147 |
|
| 148 | #define __IA32_COMPAT_SYS_STUB0(name) \
|
| 149 | __SYS_STUB0(ia32, compat_sys_##name)
|
| 150 |
|
| 151 | #define __IA32_COMPAT_SYS_STUBx(x, name, ...) \
|
| 152 | __SYS_STUBx(ia32, compat_sys##name, \
|
| 153 | SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__))
|
| 154 |
|
| 155 | #define __IA32_COMPAT_COND_SYSCALL(name) \
|
| 156 | __COND_SYSCALL(ia32, compat_sys_##name)
|
| 157 |
|
| 158 | #else
|
| 159 | #define __IA32_COMPAT_SYS_STUB0(name)
|
| 160 | #define __IA32_COMPAT_SYS_STUBx(x, name, ...)
|
| 161 | #define __IA32_COMPAT_COND_SYSCALL(name)
|
| 162 | #endif
|
| 163 |
|
| 164 |
|
| 165 | #ifdef CONFIG_X86_X32_ABI
|
| 166 | |
| 167 | * For the x32 ABI, we need to create a stub for compat_sys_*() which is aware
|
| 168 | * of the x86-64-style parameter ordering of x32 syscalls. The syscalls common
|
| 169 | * with x86_64 obviously do not need such care.
|
| 170 |
|
| 171 | #define __X32_COMPAT_SYS_STUB0(name) \
|
| 172 | __SYS_STUB0(x64, compat_sys_##name)
|
| 173 |
|
| 174 | #define __X32_COMPAT_SYS_STUBx(x, name, ...) \
|
| 175 | __SYS_STUBx(x64, compat_sys##name, \
|
| 176 | SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__))
|
| 177 |
|
| 178 | #define __X32_COMPAT_COND_SYSCALL(name) \
|
| 179 | __COND_SYSCALL(x64, compat_sys_##name)
|
| 180 |
|
| 181 | #else
|
| 182 | #define __X32_COMPAT_SYS_STUB0(name)
|
| 183 | #define __X32_COMPAT_SYS_STUBx(x, name, ...)
|
| 184 | #define __X32_COMPAT_COND_SYSCALL(name)
|
| 185 | #endif
|
| 186 |
|
| 187 |
|
| 188 | #ifdef CONFIG_COMPAT
|
| 189 | |
| 190 | * Compat means IA32_EMULATION and/or X86_X32. As they use a different
|
| 191 | * mapping of registers to parameters, we need to generate stubs for each
|
| 192 | * of them.
|
| 193 |
|
| 194 | #define COMPAT_SYSCALL_DEFINE0(name) \
|
| 195 | static long \
|
| 196 | __do_compat_sys_##name(const struct pt_regs *__unused); \
|
| 197 | __IA32_COMPAT_SYS_STUB0(name) \
|
| 198 | __X32_COMPAT_SYS_STUB0(name) \
|
| 199 | static long \
|
| 200 | __do_compat_sys_##name(const struct pt_regs *__unused)
|
| 201 |
|
| 202 | #define COMPAT_SYSCALL_DEFINEx(x, name, ...) \
|
| 203 | static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
|
| 204 | static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
|
| 205 | __IA32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \
|
| 206 | __X32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \
|
| 207 | static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \
|
| 208 | { \
|
| 209 | return __do_compat_sys##name(__MAP(x,__SC_DELOUSE,__VA_ARGS__));\
|
| 210 | } \
|
| 211 | static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))
|
| 212 |
|
| 213 | |
| 214 | * As some compat syscalls may not be implemented, we need to expand
|
| 215 | * COND_SYSCALL_COMPAT in kernel/sys_ni.c to cover this case as well.
|
| 216 |
|
| 217 | #define COND_SYSCALL_COMPAT(name) \
|
| 218 | __IA32_COMPAT_COND_SYSCALL(name) \
|
| 219 | __X32_COMPAT_COND_SYSCALL(name)
|
| 220 |
|
| 221 | #endif
|
| 222 |
|
| 223 | #define __SYSCALL_DEFINEx(x, name, ...) \
|
| 224 | static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
|
| 225 | static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
|
| 226 | __X64_SYS_STUBx(x, name, __VA_ARGS__) \
|
| 227 | __IA32_SYS_STUBx(x, name, __VA_ARGS__) \
|
| 228 | static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \
|
| 229 | { \
|
| 230 | long ret = __do_sys##name(__MAP(x,__SC_CAST,__VA_ARGS__));\
|
| 231 | __MAP(x,__SC_TEST,__VA_ARGS__); \
|
| 232 | __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__)); \
|
| 233 | return ret; \
|
| 234 | } \
|
| 235 | static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))
|
| 236 |
|
| 237 | |
| 238 | * As the generic SYSCALL_DEFINE0() macro does not decode any parameters for
|
| 239 | * obvious reasons, and passing struct pt_regs *regs to it in %rdi does not
|
| 240 | * hurt, we only need to re-define it here to keep the naming congruent to
|
| 241 | * SYSCALL_DEFINEx() -- which is essential for the COND_SYSCALL() macro
|
| 242 | * to work correctly.
|
| 243 |
|
| 244 | #define SYSCALL_DEFINE0(sname) \
|
| 245 | SYSCALL_METADATA(_##sname, 0); \
|
| 246 | static long __do_sys_##sname(const struct pt_regs *__unused); \
|
| 247 | __X64_SYS_STUB0(sname) \
|
| 248 | __IA32_SYS_STUB0(sname) \
|
| 249 | static long __do_sys_##sname(const struct pt_regs *__unused)
|
| 250 |
|
| 251 | #define COND_SYSCALL(name) \
|
| 252 | __X64_COND_SYSCALL(name) \
|
| 253 | __IA32_COND_SYSCALL(name)
|
| 254 |
|
| 255 |
|
| 256 | |
| 257 | * For VSYSCALLS, we need to declare these three syscalls with the new
|
| 258 | * pt_regs-based calling convention for in-kernel use.
|
| 259 |
|
| 260 | long __x64_sys_getcpu(const struct pt_regs *regs);
|
| 261 | long __x64_sys_gettimeofday(const struct pt_regs *regs);
|
| 262 | long __x64_sys_time(const struct pt_regs *regs);
|
| 263 |
|
| 264 | #endif
|