1 | /* Set flags signalling availability of kernel features based on given |
2 | kernel version number. |
3 | Copyright (C) 1999-2021 Free Software Foundation, Inc. |
4 | This file is part of the GNU C Library. |
5 | |
6 | The GNU C Library is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either |
9 | version 2.1 of the License, or (at your option) any later version. |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | Lesser General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Lesser General Public |
17 | License along with the GNU C Library; if not, see |
18 | <https://www.gnu.org/licenses/>. */ |
19 | |
20 | /* This file must not contain any C code. At least it must be protected |
21 | to allow using the file also in assembler files. */ |
22 | |
23 | #ifndef _LINUX_KERNEL_FEATURES_H |
24 | #define _LINUX_KERNEL_FEATURES_H 1 |
25 | |
26 | #include <bits/wordsize.h> |
27 | |
28 | #ifndef __LINUX_KERNEL_VERSION |
29 | /* We assume the worst; all kernels should be supported. */ |
30 | # define __LINUX_KERNEL_VERSION 0 |
31 | #endif |
32 | |
33 | /* We assume for __LINUX_KERNEL_VERSION the same encoding used in |
34 | linux/version.h. I.e., the major, minor, and subminor all get a |
35 | byte with the major number being in the highest byte. This means |
36 | we can do numeric comparisons. |
37 | |
38 | In the following we will define certain symbols depending on |
39 | whether the describes kernel feature is available in the kernel |
40 | version given by __LINUX_KERNEL_VERSION. We are not always exactly |
41 | recording the correct versions in which the features were |
42 | introduced. If somebody cares these values can afterwards be |
43 | corrected. */ |
44 | |
45 | /* The statfs64 syscalls are available in 2.5.74 (but not for alpha). */ |
46 | #define __ASSUME_STATFS64 1 |
47 | |
48 | /* pselect/ppoll were introduced just after 2.6.16-rc1. On x86_64 and |
49 | SH this appeared first in 2.6.19-rc1, on ia64 in 2.6.22-rc1. */ |
50 | #define __ASSUME_PSELECT 1 |
51 | |
52 | /* Support for inter-process robust mutexes was added in 2.6.17 (but |
53 | some architectures lack futex_atomic_cmpxchg_inatomic in some |
54 | configurations). */ |
55 | #define __ASSUME_SET_ROBUST_LIST 1 |
56 | |
57 | /* Support for various CLOEXEC and NONBLOCK flags was added in |
58 | 2.6.27. */ |
59 | #define __ASSUME_IN_NONBLOCK 1 |
60 | |
61 | /* Support for preadv and pwritev was added in 2.6.30. */ |
62 | #define __ASSUME_PREADV 1 |
63 | #define __ASSUME_PWRITEV 1 |
64 | |
65 | /* Support for sendmmsg functionality was added in 3.0. */ |
66 | #define __ASSUME_SENDMMSG 1 |
67 | |
68 | /* On most architectures, most socket syscalls are supported for all |
69 | supported kernel versions, but on some socketcall architectures |
70 | separate syscalls were only added later. */ |
71 | #define __ASSUME_SENDMSG_SYSCALL 1 |
72 | #define __ASSUME_RECVMSG_SYSCALL 1 |
73 | #define __ASSUME_ACCEPT_SYSCALL 1 |
74 | #define __ASSUME_CONNECT_SYSCALL 1 |
75 | #define __ASSUME_RECVFROM_SYSCALL 1 |
76 | #define __ASSUME_SENDTO_SYSCALL 1 |
77 | #define __ASSUME_ACCEPT4_SYSCALL 1 |
78 | #define __ASSUME_RECVMMSG_SYSCALL 1 |
79 | #define __ASSUME_SENDMMSG_SYSCALL 1 |
80 | |
81 | /* Support for SysV IPC through wired syscalls. All supported architectures |
82 | either support ipc syscall and/or all the ipc correspondent syscalls. */ |
83 | #define __ASSUME_DIRECT_SYSVIPC_SYSCALLS 1 |
84 | /* The generic default __IPC_64 value is 0x0, however some architectures |
85 | require a different value of 0x100. */ |
86 | #define __ASSUME_SYSVIPC_DEFAULT_IPC_64 1 |
87 | |
88 | /* All supported architectures reserve a 32-bit for MODE field in sysvipc |
89 | ipc_perm. However, some kernel ABI interfaces still expect a 16-bit |
90 | field. This is only an issue if arch-defined IPC_PERM padding is on a |
91 | wrong position regarding endianness. In this case, the IPC control |
92 | routines (msgctl, semctl, and semtctl) requires to shift the value to |
93 | correct place. |
94 | The ABIs that requires it define __ASSUME_SYSVIPC_BROKEN_MODE_T. */ |
95 | |
96 | /* Support for p{read,write}v2 was added in 4.6. However Linux default |
97 | implementation does not assume the __ASSUME_* and instead use a fallback |
98 | implementation based on p{read,write}v and returning an error for |
99 | non supported flags. */ |
100 | |
101 | /* Support for the renameat2 system call was added in kernel 3.15. */ |
102 | #if __LINUX_KERNEL_VERSION >= 0x030F00 |
103 | # define __ASSUME_RENAMEAT2 |
104 | #endif |
105 | |
106 | /* Support for the execveat syscall was added in 3.19. */ |
107 | #if __LINUX_KERNEL_VERSION >= 0x031300 |
108 | # define __ASSUME_EXECVEAT 1 |
109 | #endif |
110 | |
111 | #if __LINUX_KERNEL_VERSION >= 0x040400 |
112 | # define __ASSUME_MLOCK2 1 |
113 | #endif |
114 | |
115 | /* Support for statx was added in kernel 4.11. */ |
116 | #if __LINUX_KERNEL_VERSION >= 0x040B00 |
117 | # define __ASSUME_STATX 1 |
118 | #endif |
119 | |
120 | /* Support for clone call used on fork. The signature varies across the |
121 | architectures with current 4 different variants: |
122 | |
123 | 1. long int clone (unsigned long flags, unsigned long newsp, |
124 | int *parent_tidptr, unsigned long tls, |
125 | int *child_tidptr) |
126 | |
127 | 2. long int clone (unsigned long newsp, unsigned long clone_flags, |
128 | int *parent_tidptr, int * child_tidptr, |
129 | unsigned long tls) |
130 | |
131 | 3. long int clone (unsigned long flags, unsigned long newsp, |
132 | int stack_size, int *parent_tidptr, |
133 | int *child_tidptr, unsigned long tls) |
134 | |
135 | 4. long int clone (unsigned long flags, unsigned long newsp, |
136 | int *parent_tidptr, int *child_tidptr, |
137 | unsigned long tls) |
138 | |
139 | The fourth variant is intended to be used as the default for newer ports, |
140 | Also IA64 uses the third variant but with __NR_clone2 instead of |
141 | __NR_clone. |
142 | |
143 | The macros names to define the variant used for the architecture is |
144 | similar to kernel: |
145 | |
146 | - __ASSUME_CLONE_BACKWARDS: for variant 1. |
147 | - __ASSUME_CLONE_BACKWARDS2: for variant 2 (s390). |
148 | - __ASSUME_CLONE_BACKWARDS3: for variant 3 (microblaze). |
149 | - __ASSUME_CLONE_DEFAULT: for variant 4. |
150 | - __ASSUME_CLONE2: for clone2 with variant 3 (ia64). |
151 | */ |
152 | |
153 | #define __ASSUME_CLONE_DEFAULT 1 |
154 | |
155 | /* Support for 64-bit time_t in the system call interface. When this |
156 | flag is set, the kernel provides a version of each of these system |
157 | calls that accepts 64-bit time_t: |
158 | |
159 | clock_adjtime(64) |
160 | clock_gettime(64) |
161 | clock_settime(64) |
162 | clock_getres(_time64) |
163 | clock_nanosleep(_time64) |
164 | futex(_time64) |
165 | mq_timedreceive(_time64) |
166 | mq_timedsend(_time64) |
167 | ppoll(_time64) |
168 | pselect6(_time64) |
169 | rt_sigtimedwait(_time64) |
170 | sched_rr_get_interval(_time64) |
171 | timer_gettime(64) |
172 | timer_settime(64) |
173 | timerfd_gettime(64) |
174 | timerfd_settime(64) |
175 | utimensat(_time64) |
176 | |
177 | On architectures where time_t has historically been 64 bits, |
178 | only the 64-bit version of each system call exists, and there |
179 | are no suffixes on the __NR_ constants. |
180 | |
181 | On architectures where time_t has historically been 32 bits, |
182 | both 32-bit and 64-bit versions of each system call may exist, |
183 | depending on the kernel version. When the 64-bit version exists, |
184 | there is a '64' or '_time64' suffix on the name of its __NR_ |
185 | constant, as shown above. |
186 | |
187 | This flag is always set for Linux 5.1 and later. Prior to that |
188 | version, it is set only for some CPU architectures and ABIs: |
189 | |
190 | - __WORDSIZE == 64 - all supported architectures where pointers |
191 | are 64 bits also have always had 64-bit time_t. |
192 | |
193 | - __WORDSIZE == 32 && __SYSCALL_WORDSIZE == 64 - this describes |
194 | only one supported configuration, x86's 'x32' subarchitecture, |
195 | where pointers are 32 bits but time_t has always been 64 bits. |
196 | |
197 | __ASSUME_TIME64_SYSCALLS being set does not mean __TIMESIZE is 64, |
198 | and __TIMESIZE equal to 64 does not mean __ASSUME_TIME64_SYSCALLS |
199 | is set. All four cases are possible. */ |
200 | |
201 | #if __LINUX_KERNEL_VERSION >= 0x050100 \ |
202 | || __WORDSIZE == 64 \ |
203 | || (defined __SYSCALL_WORDSIZE && __SYSCALL_WORDSIZE == 64) |
204 | # define __ASSUME_TIME64_SYSCALLS 1 |
205 | #endif |
206 | |
207 | /* Linux waitid prior kernel 5.4 does not support waiting for the current |
208 | process group. */ |
209 | #if __LINUX_KERNEL_VERSION >= 0x050400 |
210 | # define __ASSUME_WAITID_PID0_P_PGID |
211 | #endif |
212 | |
213 | /* The faccessat2 system call was introduced across all architectures |
214 | in Linux 5.8. */ |
215 | #if __LINUX_KERNEL_VERSION >= 0x050800 |
216 | # define __ASSUME_FACCESSAT2 1 |
217 | #else |
218 | # define __ASSUME_FACCESSAT2 0 |
219 | #endif |
220 | |
221 | #endif /* kernel-features.h */ |
222 | |