syscall_wrapper.h (9348B)
1/* SPDX-License-Identifier: GPL-2.0 */ 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 9struct pt_regs; 10 11extern long __x64_sys_ni_syscall(const struct pt_regs *regs); 12extern 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/* Mapping of registers to parameters for syscalls on x86-64 and x32 */ 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/* Mapping of registers to parameters for syscalls on i386 */ 62#define SC_IA32_REGS_TO_ARGS(x, ...) \ 63 __MAP(x,__SC_ARGS \ 64 ,,(unsigned int)regs->bx,,(unsigned int)regs->cx \ 65 ,,(unsigned int)regs->dx,,(unsigned int)regs->si \ 66 ,,(unsigned int)regs->di,,(unsigned int)regs->bp) 67 68#define __SYS_STUB0(abi, name) \ 69 long __##abi##_##name(const struct pt_regs *regs); \ 70 ALLOW_ERROR_INJECTION(__##abi##_##name, ERRNO); \ 71 long __##abi##_##name(const struct pt_regs *regs) \ 72 __alias(__do_##name); 73 74#define __SYS_STUBx(abi, name, ...) \ 75 long __##abi##_##name(const struct pt_regs *regs); \ 76 ALLOW_ERROR_INJECTION(__##abi##_##name, ERRNO); \ 77 long __##abi##_##name(const struct pt_regs *regs) \ 78 { \ 79 return __se_##name(__VA_ARGS__); \ 80 } 81 82#define __COND_SYSCALL(abi, name) \ 83 __weak long __##abi##_##name(const struct pt_regs *__unused); \ 84 __weak long __##abi##_##name(const struct pt_regs *__unused) \ 85 { \ 86 return sys_ni_syscall(); \ 87 } 88 89#define __SYS_NI(abi, name) \ 90 SYSCALL_ALIAS(__##abi##_##name, sys_ni_posix_timers); 91 92#ifdef CONFIG_X86_64 93#define __X64_SYS_STUB0(name) \ 94 __SYS_STUB0(x64, sys_##name) 95 96#define __X64_SYS_STUBx(x, name, ...) \ 97 __SYS_STUBx(x64, sys##name, \ 98 SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__)) 99 100#define __X64_COND_SYSCALL(name) \ 101 __COND_SYSCALL(x64, sys_##name) 102 103#define __X64_SYS_NI(name) \ 104 __SYS_NI(x64, sys_##name) 105#else /* CONFIG_X86_64 */ 106#define __X64_SYS_STUB0(name) 107#define __X64_SYS_STUBx(x, name, ...) 108#define __X64_COND_SYSCALL(name) 109#define __X64_SYS_NI(name) 110#endif /* CONFIG_X86_64 */ 111 112#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 113#define __IA32_SYS_STUB0(name) \ 114 __SYS_STUB0(ia32, sys_##name) 115 116#define __IA32_SYS_STUBx(x, name, ...) \ 117 __SYS_STUBx(ia32, sys##name, \ 118 SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__)) 119 120#define __IA32_COND_SYSCALL(name) \ 121 __COND_SYSCALL(ia32, sys_##name) 122 123#define __IA32_SYS_NI(name) \ 124 __SYS_NI(ia32, sys_##name) 125#else /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ 126#define __IA32_SYS_STUB0(name) 127#define __IA32_SYS_STUBx(x, name, ...) 128#define __IA32_COND_SYSCALL(name) 129#define __IA32_SYS_NI(name) 130#endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ 131 132#ifdef CONFIG_IA32_EMULATION 133/* 134 * For IA32 emulation, we need to handle "compat" syscalls *and* create 135 * additional wrappers (aptly named __ia32_sys_xyzzy) which decode the 136 * ia32 regs in the proper order for shared or "common" syscalls. As some 137 * syscalls may not be implemented, we need to expand COND_SYSCALL in 138 * kernel/sys_ni.c and SYS_NI in kernel/time/posix-stubs.c to cover this 139 * case as well. 140 */ 141#define __IA32_COMPAT_SYS_STUB0(name) \ 142 __SYS_STUB0(ia32, compat_sys_##name) 143 144#define __IA32_COMPAT_SYS_STUBx(x, name, ...) \ 145 __SYS_STUBx(ia32, compat_sys##name, \ 146 SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__)) 147 148#define __IA32_COMPAT_COND_SYSCALL(name) \ 149 __COND_SYSCALL(ia32, compat_sys_##name) 150 151#define __IA32_COMPAT_SYS_NI(name) \ 152 __SYS_NI(ia32, compat_sys_##name) 153 154#else /* CONFIG_IA32_EMULATION */ 155#define __IA32_COMPAT_SYS_STUB0(name) 156#define __IA32_COMPAT_SYS_STUBx(x, name, ...) 157#define __IA32_COMPAT_COND_SYSCALL(name) 158#define __IA32_COMPAT_SYS_NI(name) 159#endif /* CONFIG_IA32_EMULATION */ 160 161 162#ifdef CONFIG_X86_X32_ABI 163/* 164 * For the x32 ABI, we need to create a stub for compat_sys_*() which is aware 165 * of the x86-64-style parameter ordering of x32 syscalls. The syscalls common 166 * with x86_64 obviously do not need such care. 167 */ 168#define __X32_COMPAT_SYS_STUB0(name) \ 169 __SYS_STUB0(x64, compat_sys_##name) 170 171#define __X32_COMPAT_SYS_STUBx(x, name, ...) \ 172 __SYS_STUBx(x64, compat_sys##name, \ 173 SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__)) 174 175#define __X32_COMPAT_COND_SYSCALL(name) \ 176 __COND_SYSCALL(x64, compat_sys_##name) 177 178#define __X32_COMPAT_SYS_NI(name) \ 179 __SYS_NI(x64, compat_sys_##name) 180#else /* CONFIG_X86_X32_ABI */ 181#define __X32_COMPAT_SYS_STUB0(name) 182#define __X32_COMPAT_SYS_STUBx(x, name, ...) 183#define __X32_COMPAT_COND_SYSCALL(name) 184#define __X32_COMPAT_SYS_NI(name) 185#endif /* CONFIG_X86_X32_ABI */ 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 and COMPAT_SYS_NI in 216 * kernel/time/posix-stubs.c to cover this case as well. 217 */ 218#define COND_SYSCALL_COMPAT(name) \ 219 __IA32_COMPAT_COND_SYSCALL(name) \ 220 __X32_COMPAT_COND_SYSCALL(name) 221 222#define COMPAT_SYS_NI(name) \ 223 __IA32_COMPAT_SYS_NI(name) \ 224 __X32_COMPAT_SYS_NI(name) 225 226#endif /* CONFIG_COMPAT */ 227 228#define __SYSCALL_DEFINEx(x, name, ...) \ 229 static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \ 230 static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\ 231 __X64_SYS_STUBx(x, name, __VA_ARGS__) \ 232 __IA32_SYS_STUBx(x, name, __VA_ARGS__) \ 233 static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \ 234 { \ 235 long ret = __do_sys##name(__MAP(x,__SC_CAST,__VA_ARGS__));\ 236 __MAP(x,__SC_TEST,__VA_ARGS__); \ 237 __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__)); \ 238 return ret; \ 239 } \ 240 static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__)) 241 242/* 243 * As the generic SYSCALL_DEFINE0() macro does not decode any parameters for 244 * obvious reasons, and passing struct pt_regs *regs to it in %rdi does not 245 * hurt, we only need to re-define it here to keep the naming congruent to 246 * SYSCALL_DEFINEx() -- which is essential for the COND_SYSCALL() and SYS_NI() 247 * macros to work correctly. 248 */ 249#define SYSCALL_DEFINE0(sname) \ 250 SYSCALL_METADATA(_##sname, 0); \ 251 static long __do_sys_##sname(const struct pt_regs *__unused); \ 252 __X64_SYS_STUB0(sname) \ 253 __IA32_SYS_STUB0(sname) \ 254 static long __do_sys_##sname(const struct pt_regs *__unused) 255 256#define COND_SYSCALL(name) \ 257 __X64_COND_SYSCALL(name) \ 258 __IA32_COND_SYSCALL(name) 259 260#define SYS_NI(name) \ 261 __X64_SYS_NI(name) \ 262 __IA32_SYS_NI(name) 263 264 265/* 266 * For VSYSCALLS, we need to declare these three syscalls with the new 267 * pt_regs-based calling convention for in-kernel use. 268 */ 269long __x64_sys_getcpu(const struct pt_regs *regs); 270long __x64_sys_gettimeofday(const struct pt_regs *regs); 271long __x64_sys_time(const struct pt_regs *regs); 272 273#endif /* _ASM_X86_SYSCALL_WRAPPER_H */