Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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
9#include <asm/ptrace.h>
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#ifdef CONFIG_X86_64
90#define __X64_SYS_STUB0(name) \
91 __SYS_STUB0(x64, sys_##name)
92
93#define __X64_SYS_STUBx(x, name, ...) \
94 __SYS_STUBx(x64, sys##name, \
95 SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__))
96
97#define __X64_COND_SYSCALL(name) \
98 __COND_SYSCALL(x64, sys_##name)
99
100#else /* CONFIG_X86_64 */
101#define __X64_SYS_STUB0(name)
102#define __X64_SYS_STUBx(x, name, ...)
103#define __X64_COND_SYSCALL(name)
104#endif /* CONFIG_X86_64 */
105
106#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
107#define __IA32_SYS_STUB0(name) \
108 __SYS_STUB0(ia32, sys_##name)
109
110#define __IA32_SYS_STUBx(x, name, ...) \
111 __SYS_STUBx(ia32, sys##name, \
112 SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__))
113
114#define __IA32_COND_SYSCALL(name) \
115 __COND_SYSCALL(ia32, sys_##name)
116
117#else /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
118#define __IA32_SYS_STUB0(name)
119#define __IA32_SYS_STUBx(x, name, ...)
120#define __IA32_COND_SYSCALL(name)
121#endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
122
123#ifdef CONFIG_IA32_EMULATION
124/*
125 * For IA32 emulation, we need to handle "compat" syscalls *and* create
126 * additional wrappers (aptly named __ia32_sys_xyzzy) which decode the
127 * ia32 regs in the proper order for shared or "common" syscalls. As some
128 * syscalls may not be implemented, we need to expand COND_SYSCALL in
129 * kernel/sys_ni.c to cover this case as well.
130 */
131#define __IA32_COMPAT_SYS_STUB0(name) \
132 __SYS_STUB0(ia32, compat_sys_##name)
133
134#define __IA32_COMPAT_SYS_STUBx(x, name, ...) \
135 __SYS_STUBx(ia32, compat_sys##name, \
136 SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__))
137
138#define __IA32_COMPAT_COND_SYSCALL(name) \
139 __COND_SYSCALL(ia32, compat_sys_##name)
140
141#else /* CONFIG_IA32_EMULATION */
142#define __IA32_COMPAT_SYS_STUB0(name)
143#define __IA32_COMPAT_SYS_STUBx(x, name, ...)
144#define __IA32_COMPAT_COND_SYSCALL(name)
145#endif /* CONFIG_IA32_EMULATION */
146
147
148#ifdef CONFIG_X86_X32_ABI
149/*
150 * For the x32 ABI, we need to create a stub for compat_sys_*() which is aware
151 * of the x86-64-style parameter ordering of x32 syscalls. The syscalls common
152 * with x86_64 obviously do not need such care.
153 */
154#define __X32_COMPAT_SYS_STUB0(name) \
155 __SYS_STUB0(x64, compat_sys_##name)
156
157#define __X32_COMPAT_SYS_STUBx(x, name, ...) \
158 __SYS_STUBx(x64, compat_sys##name, \
159 SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__))
160
161#define __X32_COMPAT_COND_SYSCALL(name) \
162 __COND_SYSCALL(x64, compat_sys_##name)
163
164#else /* CONFIG_X86_X32_ABI */
165#define __X32_COMPAT_SYS_STUB0(name)
166#define __X32_COMPAT_SYS_STUBx(x, name, ...)
167#define __X32_COMPAT_COND_SYSCALL(name)
168#endif /* CONFIG_X86_X32_ABI */
169
170
171#ifdef CONFIG_COMPAT
172/*
173 * Compat means IA32_EMULATION and/or X86_X32. As they use a different
174 * mapping of registers to parameters, we need to generate stubs for each
175 * of them.
176 */
177#define COMPAT_SYSCALL_DEFINE0(name) \
178 static long \
179 __do_compat_sys_##name(const struct pt_regs *__unused); \
180 __IA32_COMPAT_SYS_STUB0(name) \
181 __X32_COMPAT_SYS_STUB0(name) \
182 static long \
183 __do_compat_sys_##name(const struct pt_regs *__unused)
184
185#define COMPAT_SYSCALL_DEFINEx(x, name, ...) \
186 static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
187 static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
188 __IA32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \
189 __X32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \
190 static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \
191 { \
192 return __do_compat_sys##name(__MAP(x,__SC_DELOUSE,__VA_ARGS__));\
193 } \
194 static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))
195
196/*
197 * As some compat syscalls may not be implemented, we need to expand
198 * COND_SYSCALL_COMPAT in kernel/sys_ni.c to cover this case as well.
199 */
200#define COND_SYSCALL_COMPAT(name) \
201 __IA32_COMPAT_COND_SYSCALL(name) \
202 __X32_COMPAT_COND_SYSCALL(name)
203
204#endif /* CONFIG_COMPAT */
205
206#define __SYSCALL_DEFINEx(x, name, ...) \
207 static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
208 static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
209 __X64_SYS_STUBx(x, name, __VA_ARGS__) \
210 __IA32_SYS_STUBx(x, name, __VA_ARGS__) \
211 static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \
212 { \
213 long ret = __do_sys##name(__MAP(x,__SC_CAST,__VA_ARGS__));\
214 __MAP(x,__SC_TEST,__VA_ARGS__); \
215 __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__)); \
216 return ret; \
217 } \
218 static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))
219
220/*
221 * As the generic SYSCALL_DEFINE0() macro does not decode any parameters for
222 * obvious reasons, and passing struct pt_regs *regs to it in %rdi does not
223 * hurt, we only need to re-define it here to keep the naming congruent to
224 * SYSCALL_DEFINEx() -- which is essential for the COND_SYSCALL() macro
225 * to work correctly.
226 */
227#define SYSCALL_DEFINE0(sname) \
228 SYSCALL_METADATA(_##sname, 0); \
229 static long __do_sys_##sname(const struct pt_regs *__unused); \
230 __X64_SYS_STUB0(sname) \
231 __IA32_SYS_STUB0(sname) \
232 static long __do_sys_##sname(const struct pt_regs *__unused)
233
234#define COND_SYSCALL(name) \
235 __X64_COND_SYSCALL(name) \
236 __IA32_COND_SYSCALL(name)
237
238
239/*
240 * For VSYSCALLS, we need to declare these three syscalls with the new
241 * pt_regs-based calling convention for in-kernel use.
242 */
243long __x64_sys_getcpu(const struct pt_regs *regs);
244long __x64_sys_gettimeofday(const struct pt_regs *regs);
245long __x64_sys_time(const struct pt_regs *regs);
246
247#endif /* _ASM_X86_SYSCALL_WRAPPER_H */