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
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 * __x32_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 { \
85 return sys_ni_syscall(); \
86 }
87
88#define __SYS_NI(abi, name) \
89 SYSCALL_ALIAS(__##abi##_##name, sys_ni_posix_timers);
90
91#ifdef CONFIG_X86_64
92#define __X64_SYS_STUB0(name) \
93 __SYS_STUB0(x64, sys_##name)
94
95#define __X64_SYS_STUBx(x, name, ...) \
96 __SYS_STUBx(x64, sys##name, \
97 SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__))
98
99#define __X64_COND_SYSCALL(name) \
100 __COND_SYSCALL(x64, sys_##name)
101
102#define __X64_SYS_NI(name) \
103 __SYS_NI(x64, sys_##name)
104#else /* CONFIG_X86_64 */
105#define __X64_SYS_STUB0(name)
106#define __X64_SYS_STUBx(x, name, ...)
107#define __X64_COND_SYSCALL(name)
108#define __X64_SYS_NI(name)
109#endif /* CONFIG_X86_64 */
110
111#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
112#define __IA32_SYS_STUB0(name) \
113 __SYS_STUB0(ia32, sys_##name)
114
115#define __IA32_SYS_STUBx(x, name, ...) \
116 __SYS_STUBx(ia32, sys##name, \
117 SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__))
118
119#define __IA32_COND_SYSCALL(name) \
120 __COND_SYSCALL(ia32, sys_##name)
121
122#define __IA32_SYS_NI(name) \
123 __SYS_NI(ia32, sys_##name)
124#else /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
125#define __IA32_SYS_STUB0(name)
126#define __IA32_SYS_STUBx(x, name, ...)
127#define __IA32_COND_SYSCALL(name)
128#define __IA32_SYS_NI(name)
129#endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
130
131#ifdef CONFIG_IA32_EMULATION
132/*
133 * For IA32 emulation, we need to handle "compat" syscalls *and* create
134 * additional wrappers (aptly named __ia32_sys_xyzzy) which decode the
135 * ia32 regs in the proper order for shared or "common" syscalls. As some
136 * syscalls may not be implemented, we need to expand COND_SYSCALL in
137 * kernel/sys_ni.c and SYS_NI in kernel/time/posix-stubs.c to cover this
138 * case as well.
139 */
140#define __IA32_COMPAT_SYS_STUB0(name) \
141 __SYS_STUB0(ia32, compat_sys_##name)
142
143#define __IA32_COMPAT_SYS_STUBx(x, name, ...) \
144 __SYS_STUBx(ia32, compat_sys##name, \
145 SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__))
146
147#define __IA32_COMPAT_COND_SYSCALL(name) \
148 __COND_SYSCALL(ia32, compat_sys_##name)
149
150#define __IA32_COMPAT_SYS_NI(name) \
151 __SYS_NI(ia32, compat_sys_##name)
152
153#else /* CONFIG_IA32_EMULATION */
154#define __IA32_COMPAT_SYS_STUB0(name)
155#define __IA32_COMPAT_SYS_STUBx(x, name, ...)
156#define __IA32_COMPAT_COND_SYSCALL(name)
157#define __IA32_COMPAT_SYS_NI(name)
158#endif /* CONFIG_IA32_EMULATION */
159
160
161#ifdef CONFIG_X86_X32
162/*
163 * For the x32 ABI, we need to create a stub for compat_sys_*() which is aware
164 * of the x86-64-style parameter ordering of x32 syscalls. The syscalls common
165 * with x86_64 obviously do not need such care.
166 */
167#define __X32_COMPAT_SYS_STUB0(name) \
168 __SYS_STUB0(x32, compat_sys_##name)
169
170#define __X32_COMPAT_SYS_STUBx(x, name, ...) \
171 __SYS_STUBx(x32, compat_sys##name, \
172 SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__))
173
174#define __X32_COMPAT_COND_SYSCALL(name) \
175 __COND_SYSCALL(x32, compat_sys_##name)
176
177#define __X32_COMPAT_SYS_NI(name) \
178 __SYS_NI(x32, compat_sys_##name)
179#else /* CONFIG_X86_X32 */
180#define __X32_COMPAT_SYS_STUB0(name)
181#define __X32_COMPAT_SYS_STUBx(x, name, ...)
182#define __X32_COMPAT_COND_SYSCALL(name)
183#define __X32_COMPAT_SYS_NI(name)
184#endif /* CONFIG_X86_X32 */
185
186
187#ifdef CONFIG_COMPAT
188/*
189 * Compat means IA32_EMULATION and/or X86_X32. As they use a different
190 * mapping of registers to parameters, we need to generate stubs for each
191 * of them.
192 */
193#define COMPAT_SYSCALL_DEFINE0(name) \
194 static long \
195 __do_compat_sys_##name(const struct pt_regs *__unused); \
196 __IA32_COMPAT_SYS_STUB0(name) \
197 __X32_COMPAT_SYS_STUB0(name) \
198 static long \
199 __do_compat_sys_##name(const struct pt_regs *__unused)
200
201#define COMPAT_SYSCALL_DEFINEx(x, name, ...) \
202 static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
203 static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
204 __IA32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \
205 __X32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \
206 static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \
207 { \
208 return __do_compat_sys##name(__MAP(x,__SC_DELOUSE,__VA_ARGS__));\
209 } \
210 static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))
211
212/*
213 * As some compat syscalls may not be implemented, we need to expand
214 * COND_SYSCALL_COMPAT in kernel/sys_ni.c and COMPAT_SYS_NI in
215 * kernel/time/posix-stubs.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#define COMPAT_SYS_NI(name) \
222 __IA32_COMPAT_SYS_NI(name) \
223 __X32_COMPAT_SYS_NI(name)
224
225#endif /* CONFIG_COMPAT */
226
227#define __SYSCALL_DEFINEx(x, name, ...) \
228 static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \
229 static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\
230 __X64_SYS_STUBx(x, name, __VA_ARGS__) \
231 __IA32_SYS_STUBx(x, name, __VA_ARGS__) \
232 static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \
233 { \
234 long ret = __do_sys##name(__MAP(x,__SC_CAST,__VA_ARGS__));\
235 __MAP(x,__SC_TEST,__VA_ARGS__); \
236 __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__)); \
237 return ret; \
238 } \
239 static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))
240
241/*
242 * As the generic SYSCALL_DEFINE0() macro does not decode any parameters for
243 * obvious reasons, and passing struct pt_regs *regs to it in %rdi does not
244 * hurt, we only need to re-define it here to keep the naming congruent to
245 * SYSCALL_DEFINEx() -- which is essential for the COND_SYSCALL() and SYS_NI()
246 * macros to work correctly.
247 */
248#define SYSCALL_DEFINE0(sname) \
249 SYSCALL_METADATA(_##sname, 0); \
250 static long __do_sys_##sname(const struct pt_regs *__unused); \
251 __X64_SYS_STUB0(sname) \
252 __IA32_SYS_STUB0(sname) \
253 static long __do_sys_##sname(const struct pt_regs *__unused)
254
255#define COND_SYSCALL(name) \
256 __X64_COND_SYSCALL(name) \
257 __IA32_COND_SYSCALL(name)
258
259#define SYS_NI(name) \
260 __X64_SYS_NI(name) \
261 __IA32_SYS_NI(name)
262
263
264/*
265 * For VSYSCALLS, we need to declare these three syscalls with the new
266 * pt_regs-based calling convention for in-kernel use.
267 */
268long __x64_sys_getcpu(const struct pt_regs *regs);
269long __x64_sys_gettimeofday(const struct pt_regs *regs);
270long __x64_sys_time(const struct pt_regs *regs);
271
272#endif /* _ASM_X86_SYSCALL_WRAPPER_H */