| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | |
| 3 | #ifndef __PIDFD_H |
| 4 | #define __PIDFD_H |
| 5 | |
| 6 | #define _GNU_SOURCE |
| 7 | #include <errno.h> |
| 8 | #include <fcntl.h> |
| 9 | #include <sched.h> |
| 10 | #include <signal.h> |
| 11 | #include <stdio.h> |
| 12 | #include <stdlib.h> |
| 13 | #include <string.h> |
| 14 | #include <syscall.h> |
| 15 | #include <sys/ioctl.h> |
| 16 | #include <sys/types.h> |
| 17 | #include <sys/wait.h> |
| 18 | |
| 19 | /* |
| 20 | * Remove the userspace definitions of the following preprocessor symbols |
| 21 | * to avoid duplicate-definition warnings from the subsequent in-kernel |
| 22 | * definitions. |
| 23 | */ |
| 24 | #undef SCHED_NORMAL |
| 25 | #undef SCHED_FLAG_KEEP_ALL |
| 26 | #undef SCHED_FLAG_UTIL_CLAMP |
| 27 | |
| 28 | #include "kselftest.h" |
| 29 | #include "../clone3/clone3_selftests.h" |
| 30 | |
| 31 | #ifndef FD_PIDFS_ROOT |
| 32 | #define FD_PIDFS_ROOT -10002 |
| 33 | #endif |
| 34 | |
| 35 | #ifndef P_PIDFD |
| 36 | #define P_PIDFD 3 |
| 37 | #endif |
| 38 | |
| 39 | #ifndef CLONE_NEWTIME |
| 40 | #define CLONE_NEWTIME 0x00000080 |
| 41 | #endif |
| 42 | |
| 43 | #ifndef CLONE_PIDFD |
| 44 | #define CLONE_PIDFD 0x00001000 |
| 45 | #endif |
| 46 | |
| 47 | #ifndef __NR_pidfd_open |
| 48 | #define __NR_pidfd_open 434 |
| 49 | #endif |
| 50 | |
| 51 | #ifndef __NR_pidfd_send_signal |
| 52 | #define __NR_pidfd_send_signal 424 |
| 53 | #endif |
| 54 | |
| 55 | #ifndef __NR_clone3 |
| 56 | #define __NR_clone3 435 |
| 57 | #endif |
| 58 | |
| 59 | #ifndef __NR_pidfd_getfd |
| 60 | #define __NR_pidfd_getfd 438 |
| 61 | #endif |
| 62 | |
| 63 | #ifndef PIDFD_NONBLOCK |
| 64 | #define PIDFD_NONBLOCK O_NONBLOCK |
| 65 | #endif |
| 66 | |
| 67 | #ifndef PIDFD_SELF_THREAD |
| 68 | #define PIDFD_SELF_THREAD -10000 /* Current thread. */ |
| 69 | #endif |
| 70 | |
| 71 | #ifndef PIDFD_SELF_THREAD_GROUP |
| 72 | #define PIDFD_SELF_THREAD_GROUP -10001 /* Current thread group leader. */ |
| 73 | #endif |
| 74 | |
| 75 | #ifndef PIDFD_SELF |
| 76 | #define PIDFD_SELF PIDFD_SELF_THREAD |
| 77 | #endif |
| 78 | |
| 79 | #ifndef PIDFD_SELF_PROCESS |
| 80 | #define PIDFD_SELF_PROCESS PIDFD_SELF_THREAD_GROUP |
| 81 | #endif |
| 82 | |
| 83 | #ifndef PIDFS_IOCTL_MAGIC |
| 84 | #define PIDFS_IOCTL_MAGIC 0xFF |
| 85 | #endif |
| 86 | |
| 87 | #ifndef PIDFD_GET_CGROUP_NAMESPACE |
| 88 | #define PIDFD_GET_CGROUP_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 1) |
| 89 | #endif |
| 90 | |
| 91 | #ifndef PIDFD_GET_IPC_NAMESPACE |
| 92 | #define PIDFD_GET_IPC_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 2) |
| 93 | #endif |
| 94 | |
| 95 | #ifndef PIDFD_GET_MNT_NAMESPACE |
| 96 | #define PIDFD_GET_MNT_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 3) |
| 97 | #endif |
| 98 | |
| 99 | #ifndef PIDFD_GET_NET_NAMESPACE |
| 100 | #define PIDFD_GET_NET_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 4) |
| 101 | #endif |
| 102 | |
| 103 | #ifndef PIDFD_GET_PID_NAMESPACE |
| 104 | #define PIDFD_GET_PID_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 5) |
| 105 | #endif |
| 106 | |
| 107 | #ifndef PIDFD_GET_PID_FOR_CHILDREN_NAMESPACE |
| 108 | #define PIDFD_GET_PID_FOR_CHILDREN_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 6) |
| 109 | #endif |
| 110 | |
| 111 | #ifndef PIDFD_GET_TIME_NAMESPACE |
| 112 | #define PIDFD_GET_TIME_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 7) |
| 113 | #endif |
| 114 | |
| 115 | #ifndef PIDFD_GET_TIME_FOR_CHILDREN_NAMESPACE |
| 116 | #define PIDFD_GET_TIME_FOR_CHILDREN_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 8) |
| 117 | #endif |
| 118 | |
| 119 | #ifndef PIDFD_GET_USER_NAMESPACE |
| 120 | #define PIDFD_GET_USER_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 9) |
| 121 | #endif |
| 122 | |
| 123 | #ifndef PIDFD_GET_UTS_NAMESPACE |
| 124 | #define PIDFD_GET_UTS_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 10) |
| 125 | #endif |
| 126 | |
| 127 | #ifndef PIDFD_GET_INFO |
| 128 | #define PIDFD_GET_INFO _IOWR(PIDFS_IOCTL_MAGIC, 11, struct pidfd_info) |
| 129 | #endif |
| 130 | |
| 131 | #ifndef PIDFD_INFO_PID |
| 132 | #define PIDFD_INFO_PID (1UL << 0) /* Always returned, even if not requested */ |
| 133 | #endif |
| 134 | |
| 135 | #ifndef PIDFD_INFO_CREDS |
| 136 | #define PIDFD_INFO_CREDS (1UL << 1) /* Always returned, even if not requested */ |
| 137 | #endif |
| 138 | |
| 139 | #ifndef PIDFD_INFO_CGROUPID |
| 140 | #define PIDFD_INFO_CGROUPID (1UL << 2) /* Always returned if available, even if not requested */ |
| 141 | #endif |
| 142 | |
| 143 | #ifndef PIDFD_INFO_EXIT |
| 144 | #define PIDFD_INFO_EXIT (1UL << 3) /* Always returned if available, even if not requested */ |
| 145 | #endif |
| 146 | |
| 147 | #ifndef PIDFD_INFO_COREDUMP |
| 148 | #define PIDFD_INFO_COREDUMP (1UL << 4) |
| 149 | #endif |
| 150 | |
| 151 | #ifndef PIDFD_INFO_SUPPORTED_MASK |
| 152 | #define PIDFD_INFO_SUPPORTED_MASK (1UL << 5) |
| 153 | #endif |
| 154 | |
| 155 | #ifndef PIDFD_INFO_COREDUMP_SIGNAL |
| 156 | #define PIDFD_INFO_COREDUMP_SIGNAL (1UL << 6) |
| 157 | #endif |
| 158 | |
| 159 | #ifndef PIDFD_COREDUMPED |
| 160 | #define PIDFD_COREDUMPED (1U << 0) /* Did crash and... */ |
| 161 | #endif |
| 162 | |
| 163 | #ifndef PIDFD_COREDUMP_SKIP |
| 164 | #define PIDFD_COREDUMP_SKIP (1U << 1) /* coredumping generation was skipped. */ |
| 165 | #endif |
| 166 | |
| 167 | #ifndef PIDFD_COREDUMP_USER |
| 168 | #define PIDFD_COREDUMP_USER (1U << 2) /* coredump was done as the user. */ |
| 169 | #endif |
| 170 | |
| 171 | #ifndef PIDFD_COREDUMP_ROOT |
| 172 | #define PIDFD_COREDUMP_ROOT (1U << 3) /* coredump was done as root. */ |
| 173 | #endif |
| 174 | |
| 175 | #ifndef PIDFD_THREAD |
| 176 | #define PIDFD_THREAD O_EXCL |
| 177 | #endif |
| 178 | |
| 179 | struct pidfd_info { |
| 180 | __u64 mask; |
| 181 | __u64 cgroupid; |
| 182 | __u32 pid; |
| 183 | __u32 tgid; |
| 184 | __u32 ppid; |
| 185 | __u32 ruid; |
| 186 | __u32 rgid; |
| 187 | __u32 euid; |
| 188 | __u32 egid; |
| 189 | __u32 suid; |
| 190 | __u32 sgid; |
| 191 | __u32 fsuid; |
| 192 | __u32 fsgid; |
| 193 | __s32 exit_code; |
| 194 | struct { |
| 195 | __u32 coredump_mask; |
| 196 | __u32 coredump_signal; |
| 197 | }; |
| 198 | __u64 supported_mask; |
| 199 | }; |
| 200 | |
| 201 | /* |
| 202 | * The kernel reserves 300 pids via RESERVED_PIDS in kernel/pid.c |
| 203 | * That means, when it wraps around any pid < 300 will be skipped. |
| 204 | * So we need to use a pid > 300 in order to test recycling. |
| 205 | */ |
| 206 | #define PID_RECYCLE 1000 |
| 207 | |
| 208 | /* |
| 209 | * Define a few custom error codes for the child process to clearly indicate |
| 210 | * what is happening. This way we can tell the difference between a system |
| 211 | * error, a test error, etc. |
| 212 | */ |
| 213 | #define PIDFD_PASS 0 |
| 214 | #define PIDFD_FAIL 1 |
| 215 | #define PIDFD_ERROR 2 |
| 216 | #define PIDFD_SKIP 3 |
| 217 | #define PIDFD_XFAIL 4 |
| 218 | |
| 219 | static inline int sys_waitid(int which, pid_t pid, siginfo_t *info, int options) |
| 220 | { |
| 221 | return syscall(__NR_waitid, which, pid, info, options, NULL); |
| 222 | } |
| 223 | |
| 224 | static inline int wait_for_pid(pid_t pid) |
| 225 | { |
| 226 | int status, ret; |
| 227 | |
| 228 | again: |
| 229 | ret = waitpid(pid, &status, 0); |
| 230 | if (ret == -1) { |
| 231 | if (errno == EINTR) |
| 232 | goto again; |
| 233 | |
| 234 | ksft_print_msg("waitpid returned -1, errno=%d\n" , errno); |
| 235 | return -1; |
| 236 | } |
| 237 | |
| 238 | if (!WIFEXITED(status)) { |
| 239 | ksft_print_msg( |
| 240 | "waitpid !WIFEXITED, WIFSIGNALED=%d, WTERMSIG=%d\n" , |
| 241 | WIFSIGNALED(status), WTERMSIG(status)); |
| 242 | return -1; |
| 243 | } |
| 244 | |
| 245 | ret = WEXITSTATUS(status); |
| 246 | return ret; |
| 247 | } |
| 248 | |
| 249 | static inline int sys_pidfd_open(pid_t pid, unsigned int flags) |
| 250 | { |
| 251 | return syscall(__NR_pidfd_open, pid, flags); |
| 252 | } |
| 253 | |
| 254 | static inline int sys_pidfd_send_signal(int pidfd, int sig, siginfo_t *info, |
| 255 | unsigned int flags) |
| 256 | { |
| 257 | return syscall(__NR_pidfd_send_signal, pidfd, sig, info, flags); |
| 258 | } |
| 259 | |
| 260 | static inline int sys_pidfd_getfd(int pidfd, int fd, int flags) |
| 261 | { |
| 262 | return syscall(__NR_pidfd_getfd, pidfd, fd, flags); |
| 263 | } |
| 264 | |
| 265 | static inline int sys_memfd_create(const char *name, unsigned int flags) |
| 266 | { |
| 267 | return syscall(__NR_memfd_create, name, flags); |
| 268 | } |
| 269 | |
| 270 | static inline pid_t create_child(int *pidfd, unsigned flags) |
| 271 | { |
| 272 | struct __clone_args args = { |
| 273 | .flags = CLONE_PIDFD | flags, |
| 274 | .exit_signal = SIGCHLD, |
| 275 | .pidfd = ptr_to_u64(pidfd), |
| 276 | }; |
| 277 | |
| 278 | return sys_clone3(args: &args, size: sizeof(struct __clone_args)); |
| 279 | } |
| 280 | |
| 281 | static inline ssize_t read_nointr(int fd, void *buf, size_t count) |
| 282 | { |
| 283 | ssize_t ret; |
| 284 | |
| 285 | do { |
| 286 | ret = read(fd, buf, count); |
| 287 | } while (ret < 0 && errno == EINTR); |
| 288 | |
| 289 | return ret; |
| 290 | } |
| 291 | |
| 292 | static inline ssize_t write_nointr(int fd, const void *buf, size_t count) |
| 293 | { |
| 294 | ssize_t ret; |
| 295 | |
| 296 | do { |
| 297 | ret = write(fd, buf, count); |
| 298 | } while (ret < 0 && errno == EINTR); |
| 299 | |
| 300 | return ret; |
| 301 | } |
| 302 | |
| 303 | static inline int sys_execveat(int dirfd, const char *pathname, |
| 304 | char *const argv[], char *const envp[], |
| 305 | int flags) |
| 306 | { |
| 307 | return syscall(__NR_execveat, dirfd, pathname, argv, envp, flags); |
| 308 | } |
| 309 | |
| 310 | #endif /* __PIDFD_H */ |
| 311 | |