]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
0b452006 | 2 | |
4f5dd394 | 3 | #include <ctype.h> |
0b452006 | 4 | #include <errno.h> |
11c3a366 TA |
5 | #include <limits.h> |
6 | #include <linux/oom.h> | |
4f5dd394 LP |
7 | #include <stdbool.h> |
8 | #include <stdio.h> | |
11c3a366 | 9 | #include <stdlib.h> |
9bfaffd5 | 10 | #include <sys/mman.h> |
e2047ba9 | 11 | #include <sys/mount.h> |
7b3e062c | 12 | #include <sys/personality.h> |
405f8907 | 13 | #include <sys/prctl.h> |
4f5dd394 LP |
14 | #include <sys/types.h> |
15 | #include <sys/wait.h> | |
11c3a366 | 16 | #include <syslog.h> |
4f5dd394 | 17 | #include <unistd.h> |
349cc4a5 | 18 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
19 | #include <valgrind/valgrind.h> |
20 | #endif | |
0b452006 | 21 | |
b5efdb8a | 22 | #include "alloc-util.h" |
6e5f1b57 | 23 | #include "architecture.h" |
21c491e1 | 24 | #include "env-util.h" |
39090201 | 25 | #include "errno-util.h" |
aad3a64d | 26 | #include "escape.h" |
3ffd4af2 | 27 | #include "fd-util.h" |
0b452006 | 28 | #include "fileio.h" |
f4f15635 | 29 | #include "fs-util.h" |
7b3e062c | 30 | #include "ioprio.h" |
e3b4efd2 | 31 | #include "locale-util.h" |
0b452006 | 32 | #include "log.h" |
11c3a366 | 33 | #include "macro.h" |
0a970718 | 34 | #include "memory-util.h" |
f5947a5e YW |
35 | #include "missing_sched.h" |
36 | #include "missing_syscall.h" | |
0cb8e3d1 | 37 | #include "namespace-util.h" |
aad3a64d | 38 | #include "path-util.h" |
93cc7779 | 39 | #include "process-util.h" |
8869a0b4 | 40 | #include "raw-clone.h" |
909106eb | 41 | #include "rlimit-util.h" |
93cc7779 | 42 | #include "signal-util.h" |
1359fffa | 43 | #include "stat-util.h" |
298f466f | 44 | #include "stdio-util.h" |
7b3e062c | 45 | #include "string-table.h" |
07630cea | 46 | #include "string-util.h" |
4c253ed1 | 47 | #include "terminal-util.h" |
b1d4f8e1 | 48 | #include "user-util.h" |
bc28751e | 49 | #include "utf8.h" |
0b452006 | 50 | |
0e85cbcf ZJS |
51 | /* The kernel limits userspace processes to TASK_COMM_LEN (16 bytes), but allows higher values for its own |
52 | * workers, e.g. "kworker/u9:3-kcryptd/253:0". Let's pick a fixed smallish limit that will work for the kernel. | |
53 | */ | |
54 | #define COMM_MAX_LEN 128 | |
55 | ||
0a51b45c | 56 | static int get_process_state(pid_t pid) { |
5c7b9974 | 57 | _cleanup_free_ char *line = NULL; |
0b452006 RC |
58 | const char *p; |
59 | char state; | |
60 | int r; | |
0b452006 RC |
61 | |
62 | assert(pid >= 0); | |
63 | ||
5c7b9974 LP |
64 | /* Shortcut: if we are enquired about our own state, we are obviously running */ |
65 | if (pid == 0 || pid == getpid_cached()) | |
66 | return (unsigned char) 'R'; | |
67 | ||
0b452006 | 68 | p = procfs_file_alloca(pid, "stat"); |
a644184a | 69 | |
0b452006 | 70 | r = read_one_line_file(p, &line); |
a644184a LP |
71 | if (r == -ENOENT) |
72 | return -ESRCH; | |
0b452006 RC |
73 | if (r < 0) |
74 | return r; | |
75 | ||
76 | p = strrchr(line, ')'); | |
77 | if (!p) | |
78 | return -EIO; | |
79 | ||
80 | p++; | |
81 | ||
82 | if (sscanf(p, " %c", &state) != 1) | |
83 | return -EIO; | |
84 | ||
85 | return (unsigned char) state; | |
86 | } | |
87 | ||
ce268825 LP |
88 | int get_process_comm(pid_t pid, char **ret) { |
89 | _cleanup_free_ char *escaped = NULL, *comm = NULL; | |
0b452006 RC |
90 | int r; |
91 | ||
ce268825 | 92 | assert(ret); |
0b452006 RC |
93 | assert(pid >= 0); |
94 | ||
cde93ba2 LP |
95 | if (pid == 0 || pid == getpid_cached()) { |
96 | comm = new0(char, TASK_COMM_LEN + 1); /* Must fit in 16 byte according to prctl(2) */ | |
97 | if (!comm) | |
98 | return -ENOMEM; | |
99 | ||
100 | if (prctl(PR_GET_NAME, comm) < 0) | |
101 | return -errno; | |
102 | } else { | |
103 | const char *p; | |
104 | ||
105 | p = procfs_file_alloca(pid, "comm"); | |
106 | ||
107 | /* Note that process names of kernel threads can be much longer than TASK_COMM_LEN */ | |
108 | r = read_one_line_file(p, &comm); | |
109 | if (r == -ENOENT) | |
110 | return -ESRCH; | |
111 | if (r < 0) | |
112 | return r; | |
113 | } | |
114 | ||
0e85cbcf | 115 | escaped = new(char, COMM_MAX_LEN); |
ce268825 LP |
116 | if (!escaped) |
117 | return -ENOMEM; | |
118 | ||
ce268825 | 119 | /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */ |
0e85cbcf | 120 | cellescape(escaped, COMM_MAX_LEN, comm); |
0b452006 | 121 | |
ce268825 LP |
122 | *ret = TAKE_PTR(escaped); |
123 | return 0; | |
0b452006 RC |
124 | } |
125 | ||
61977664 ZJS |
126 | static int get_process_cmdline_nulstr( |
127 | pid_t pid, | |
128 | size_t max_size, | |
129 | ProcessCmdlineFlags flags, | |
130 | char **ret, | |
131 | size_t *ret_size) { | |
132 | ||
0b452006 | 133 | const char *p; |
61977664 | 134 | char *t; |
bc28751e | 135 | size_t k; |
7b7a060e | 136 | int r; |
0b452006 | 137 | |
61977664 ZJS |
138 | /* Retrieves a process' command line as a "sized nulstr", i.e. possibly without the last NUL, but |
139 | * with a specified size. | |
69281c49 | 140 | * |
61977664 ZJS |
141 | * If PROCESS_CMDLINE_COMM_FALLBACK is specified in flags and the process has no command line set |
142 | * (the case for kernel threads), or has a command line that resolves to the empty string, will | |
143 | * return the "comm" name of the process instead. This will use at most _SC_ARG_MAX bytes of input | |
144 | * data. | |
145 | * | |
146 | * Returns an error, 0 if output was read but is truncated, 1 otherwise. | |
147 | */ | |
69281c49 | 148 | |
0b452006 | 149 | p = procfs_file_alloca(pid, "cmdline"); |
61977664 ZJS |
150 | r = read_virtual_file(p, max_size, &t, &k); /* Let's assume that each input byte results in >= 1 |
151 | * columns of output. We ignore zero-width codepoints. */ | |
fdeea3f4 ZJS |
152 | if (r == -ENOENT) |
153 | return -ESRCH; | |
154 | if (r < 0) | |
155 | return r; | |
35bbbf85 | 156 | |
61977664 ZJS |
157 | if (k == 0) { |
158 | t = mfree(t); | |
159 | ||
09c1dcee | 160 | if (!(flags & PROCESS_CMDLINE_COMM_FALLBACK)) |
0b452006 RC |
161 | return -ENOENT; |
162 | ||
bc28751e | 163 | /* Kernel threads have no argv[] */ |
61977664 | 164 | _cleanup_free_ char *comm = NULL; |
69281c49 | 165 | |
61977664 | 166 | r = get_process_comm(pid, &comm); |
bc28751e ZJS |
167 | if (r < 0) |
168 | return r; | |
69281c49 | 169 | |
61977664 | 170 | t = strjoin("[", comm, "]"); |
bc28751e ZJS |
171 | if (!t) |
172 | return -ENOMEM; | |
61977664 ZJS |
173 | |
174 | k = strlen(t); | |
175 | r = k <= max_size; | |
176 | if (r == 0) /* truncation */ | |
177 | t[max_size] = '\0'; | |
0b452006 RC |
178 | } |
179 | ||
61977664 ZJS |
180 | *ret = t; |
181 | *ret_size = k; | |
182 | return r; | |
183 | } | |
eb1ec489 | 184 | |
61977664 ZJS |
185 | int get_process_cmdline(pid_t pid, size_t max_columns, ProcessCmdlineFlags flags, char **line) { |
186 | _cleanup_free_ char *t = NULL; | |
187 | size_t k; | |
188 | char *ans; | |
e3b4efd2 | 189 | |
61977664 ZJS |
190 | assert(line); |
191 | assert(pid >= 0); | |
192 | ||
19cc6d5e | 193 | /* Retrieve and format a commandline. See above for discussion of retrieval options. |
61977664 ZJS |
194 | * |
195 | * There are two main formatting modes: | |
196 | * | |
197 | * - when PROCESS_CMDLINE_QUOTE is specified, output is quoted in C/Python style. If no shell special | |
198 | * characters are present, this output can be copy-pasted into the terminal to execute. UTF-8 | |
199 | * output is assumed. | |
200 | * | |
201 | * - otherwise, a compact non-roundtrippable form is returned. Non-UTF8 bytes are replaced by �. The | |
202 | * returned string is of the specified console width at most, abbreviated with an ellipsis. | |
203 | * | |
204 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
205 | * PROCESS_CMDLINE_COMM_FALLBACK is not specified). Returns 0 and sets *line otherwise. */ | |
206 | ||
207 | int full = get_process_cmdline_nulstr(pid, max_columns, flags, &t, &k); | |
208 | if (full < 0) | |
209 | return full; | |
210 | ||
99009ed0 ZJS |
211 | if (flags & (PROCESS_CMDLINE_QUOTE | PROCESS_CMDLINE_QUOTE_POSIX)) { |
212 | ShellEscapeFlags shflags = SHELL_ESCAPE_EMPTY | | |
213 | FLAGS_SET(flags, PROCESS_CMDLINE_QUOTE_POSIX) * SHELL_ESCAPE_POSIX; | |
214 | ||
61977664 ZJS |
215 | assert(!(flags & PROCESS_CMDLINE_USE_LOCALE)); |
216 | ||
217 | _cleanup_strv_free_ char **args = NULL; | |
218 | ||
219 | args = strv_parse_nulstr(t, k); | |
220 | if (!args) | |
221 | return -ENOMEM; | |
222 | ||
223 | for (size_t i = 0; args[i]; i++) { | |
224 | char *e; | |
225 | ||
99009ed0 | 226 | e = shell_maybe_quote(args[i], shflags); |
61977664 ZJS |
227 | if (!e) |
228 | return -ENOMEM; | |
229 | ||
230 | free_and_replace(args[i], e); | |
231 | } | |
232 | ||
233 | ans = strv_join(args, " "); | |
234 | if (!ans) | |
235 | return -ENOMEM; | |
236 | ||
237 | } else { | |
238 | /* Arguments are separated by NULs. Let's replace those with spaces. */ | |
239 | for (size_t i = 0; i < k - 1; i++) | |
240 | if (t[i] == '\0') | |
241 | t[i] = ' '; | |
242 | ||
243 | delete_trailing_chars(t, WHITESPACE); | |
244 | ||
245 | bool eight_bit = (flags & PROCESS_CMDLINE_USE_LOCALE) && !is_locale_utf8(); | |
246 | ||
247 | ans = escape_non_printable_full(t, max_columns, | |
248 | eight_bit * XESCAPE_8_BIT | !full * XESCAPE_FORCE_ELLIPSIS); | |
249 | if (!ans) | |
250 | return -ENOMEM; | |
251 | ||
252 | ans = str_realloc(ans); | |
253 | } | |
eb1ec489 | 254 | |
61977664 | 255 | *line = ans; |
0b452006 RC |
256 | return 0; |
257 | } | |
258 | ||
ad4f7f67 | 259 | static int update_argv(const char name[], size_t l) { |
c55104ce ZJS |
260 | static int can_do = -1; |
261 | ||
262 | if (can_do == 0) | |
263 | return 0; | |
264 | can_do = false; /* We'll set it to true only if the whole process works */ | |
265 | ||
01f989c6 JW |
266 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the |
267 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
268 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
269 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
270 | * mmap() is not. */ | |
271 | if (geteuid() != 0) | |
ad4f7f67 ZJS |
272 | return log_debug_errno(SYNTHETIC_ERRNO(EPERM), |
273 | "Skipping PR_SET_MM, as we don't have privileges."); | |
274 | ||
275 | static size_t mm_size = 0; | |
276 | static char *mm = NULL; | |
277 | int r; | |
278 | ||
279 | if (mm_size < l+1) { | |
9bfaffd5 LP |
280 | size_t nn_size; |
281 | char *nn; | |
282 | ||
9bfaffd5 LP |
283 | nn_size = PAGE_ALIGN(l+1); |
284 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
ad4f7f67 ZJS |
285 | if (nn == MAP_FAILED) |
286 | return log_debug_errno(errno, "mmap() failed: %m"); | |
405f8907 | 287 | |
9bfaffd5 LP |
288 | strncpy(nn, name, nn_size); |
289 | ||
290 | /* Now, let's tell the kernel about this new memory */ | |
291 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
c55104ce ZJS |
292 | if (ERRNO_IS_PRIVILEGE(errno)) |
293 | return log_debug_errno(errno, "PR_SET_MM_ARG_START failed: %m"); | |
294 | ||
14ee72b7 FS |
295 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
296 | * below the desired start address, in which case the kernel may have kicked this back due | |
297 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
298 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
299 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
300 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
301 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
302 | * resizing it to what we want. */ | |
303 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
304 | ||
305 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
ad4f7f67 | 306 | r = log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); |
14ee72b7 | 307 | (void) munmap(nn, nn_size); |
ad4f7f67 | 308 | return r; |
14ee72b7 | 309 | } |
9bfaffd5 | 310 | |
ad4f7f67 ZJS |
311 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) |
312 | return log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
14ee72b7 FS |
313 | } else { |
314 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
315 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
316 | * and continue. */ | |
317 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
318 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
319 | } | |
9bfaffd5 LP |
320 | |
321 | if (mm) | |
322 | (void) munmap(mm, mm_size); | |
323 | ||
324 | mm = nn; | |
325 | mm_size = nn_size; | |
01f989c6 | 326 | } else { |
9bfaffd5 LP |
327 | strncpy(mm, name, mm_size); |
328 | ||
01f989c6 JW |
329 | /* Update the end pointer, continuing regardless of any failure. */ |
330 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
331 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
332 | } | |
333 | ||
c55104ce | 334 | can_do = true; |
ad4f7f67 ZJS |
335 | return 0; |
336 | } | |
337 | ||
338 | int rename_process(const char name[]) { | |
339 | bool truncated = false; | |
340 | ||
341 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
342 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
343 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
344 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
345 | * truncated. | |
346 | * | |
347 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
348 | ||
349 | if (isempty(name)) | |
350 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
351 | ||
352 | if (!is_main_thread()) | |
353 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
354 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
355 | * process name that isn't correct on any other threads */ | |
356 | ||
357 | size_t l = strlen(name); | |
358 | ||
359 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we | |
360 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
361 | if (prctl(PR_SET_NAME, name) < 0) | |
362 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
363 | if (l >= TASK_COMM_LEN) /* Linux userspace process names can be 15 chars at max */ | |
364 | truncated = true; | |
365 | ||
366 | /* Second step, change glibc's ID of the process name. */ | |
367 | if (program_invocation_name) { | |
368 | size_t k; | |
369 | ||
370 | k = strlen(program_invocation_name); | |
371 | strncpy(program_invocation_name, name, k); | |
372 | if (l > k) | |
373 | truncated = true; | |
374 | } | |
375 | ||
376 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
377 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
378 | * the end. This is the best option for changing /proc/self/cmdline. */ | |
379 | (void) update_argv(name, l); | |
380 | ||
9bfaffd5 LP |
381 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if |
382 | * it still looks here */ | |
405f8907 | 383 | if (saved_argc > 0) { |
9bfaffd5 LP |
384 | if (saved_argv[0]) { |
385 | size_t k; | |
386 | ||
387 | k = strlen(saved_argv[0]); | |
388 | strncpy(saved_argv[0], name, k); | |
389 | if (l > k) | |
390 | truncated = true; | |
391 | } | |
405f8907 | 392 | |
ad4f7f67 | 393 | for (int i = 1; i < saved_argc; i++) { |
405f8907 LP |
394 | if (!saved_argv[i]) |
395 | break; | |
396 | ||
397 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
398 | } | |
399 | } | |
9bfaffd5 LP |
400 | |
401 | return !truncated; | |
405f8907 LP |
402 | } |
403 | ||
0b452006 | 404 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
405 | _cleanup_free_ char *line = NULL; |
406 | unsigned long long flags; | |
407 | size_t l, i; | |
0b452006 | 408 | const char *p; |
36b5119a LP |
409 | char *q; |
410 | int r; | |
0b452006 | 411 | |
4c701096 | 412 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 413 | return 0; |
36b5119a LP |
414 | if (!pid_is_valid(pid)) |
415 | return -EINVAL; | |
0b452006 | 416 | |
36b5119a LP |
417 | p = procfs_file_alloca(pid, "stat"); |
418 | r = read_one_line_file(p, &line); | |
419 | if (r == -ENOENT) | |
420 | return -ESRCH; | |
421 | if (r < 0) | |
422 | return r; | |
0b452006 | 423 | |
36b5119a LP |
424 | /* Skip past the comm field */ |
425 | q = strrchr(line, ')'); | |
426 | if (!q) | |
427 | return -EINVAL; | |
428 | q++; | |
429 | ||
430 | /* Skip 6 fields to reach the flags field */ | |
431 | for (i = 0; i < 6; i++) { | |
432 | l = strspn(q, WHITESPACE); | |
433 | if (l < 1) | |
434 | return -EINVAL; | |
435 | q += l; | |
436 | ||
437 | l = strcspn(q, WHITESPACE); | |
438 | if (l < 1) | |
439 | return -EINVAL; | |
440 | q += l; | |
a644184a | 441 | } |
0b452006 | 442 | |
f21f31b2 | 443 | /* Skip preceding whitespace */ |
36b5119a LP |
444 | l = strspn(q, WHITESPACE); |
445 | if (l < 1) | |
446 | return -EINVAL; | |
447 | q += l; | |
35bbbf85 | 448 | |
36b5119a LP |
449 | /* Truncate the rest */ |
450 | l = strcspn(q, WHITESPACE); | |
451 | if (l < 1) | |
452 | return -EINVAL; | |
453 | q[l] = 0; | |
0b452006 | 454 | |
36b5119a LP |
455 | r = safe_atollu(q, &flags); |
456 | if (r < 0) | |
457 | return r; | |
0b452006 | 458 | |
36b5119a | 459 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
460 | } |
461 | ||
462 | int get_process_capeff(pid_t pid, char **capeff) { | |
463 | const char *p; | |
a644184a | 464 | int r; |
0b452006 RC |
465 | |
466 | assert(capeff); | |
467 | assert(pid >= 0); | |
468 | ||
469 | p = procfs_file_alloca(pid, "status"); | |
470 | ||
c4cd1d4d | 471 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
472 | if (r == -ENOENT) |
473 | return -ESRCH; | |
474 | ||
475 | return r; | |
0b452006 RC |
476 | } |
477 | ||
478 | static int get_process_link_contents(const char *proc_file, char **name) { | |
479 | int r; | |
480 | ||
481 | assert(proc_file); | |
482 | assert(name); | |
483 | ||
484 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
485 | if (r == -ENOENT) |
486 | return -ESRCH; | |
0b452006 | 487 | if (r < 0) |
a644184a | 488 | return r; |
0b452006 RC |
489 | |
490 | return 0; | |
491 | } | |
492 | ||
493 | int get_process_exe(pid_t pid, char **name) { | |
494 | const char *p; | |
495 | char *d; | |
496 | int r; | |
497 | ||
498 | assert(pid >= 0); | |
499 | ||
500 | p = procfs_file_alloca(pid, "exe"); | |
501 | r = get_process_link_contents(p, name); | |
502 | if (r < 0) | |
503 | return r; | |
504 | ||
505 | d = endswith(*name, " (deleted)"); | |
506 | if (d) | |
507 | *d = '\0'; | |
508 | ||
509 | return 0; | |
510 | } | |
511 | ||
512 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
513 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 514 | const char *p; |
7e7a16a0 | 515 | int r; |
0b452006 RC |
516 | |
517 | assert(field); | |
518 | assert(uid); | |
519 | ||
07b38ba5 | 520 | if (pid < 0) |
6f8cbcdb LP |
521 | return -EINVAL; |
522 | ||
0b452006 | 523 | p = procfs_file_alloca(pid, "status"); |
fdeea3f4 ZJS |
524 | r = fopen_unlocked(p, "re", &f); |
525 | if (r == -ENOENT) | |
526 | return -ESRCH; | |
527 | if (r < 0) | |
528 | return r; | |
35bbbf85 | 529 | |
7e7a16a0 LP |
530 | for (;;) { |
531 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
532 | char *l; |
533 | ||
7e7a16a0 LP |
534 | r = read_line(f, LONG_LINE_MAX, &line); |
535 | if (r < 0) | |
536 | return r; | |
537 | if (r == 0) | |
538 | break; | |
539 | ||
0b452006 RC |
540 | l = strstrip(line); |
541 | ||
542 | if (startswith(l, field)) { | |
543 | l += strlen(field); | |
544 | l += strspn(l, WHITESPACE); | |
545 | ||
546 | l[strcspn(l, WHITESPACE)] = 0; | |
547 | ||
548 | return parse_uid(l, uid); | |
549 | } | |
550 | } | |
551 | ||
552 | return -EIO; | |
553 | } | |
554 | ||
555 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
556 | |
557 | if (pid == 0 || pid == getpid_cached()) { | |
558 | *uid = getuid(); | |
559 | return 0; | |
560 | } | |
561 | ||
0b452006 RC |
562 | return get_process_id(pid, "Uid:", uid); |
563 | } | |
564 | ||
565 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
566 | |
567 | if (pid == 0 || pid == getpid_cached()) { | |
568 | *gid = getgid(); | |
569 | return 0; | |
570 | } | |
571 | ||
0b452006 RC |
572 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
573 | return get_process_id(pid, "Gid:", gid); | |
574 | } | |
575 | ||
576 | int get_process_cwd(pid_t pid, char **cwd) { | |
577 | const char *p; | |
578 | ||
579 | assert(pid >= 0); | |
580 | ||
aad3a64d LP |
581 | if (pid == 0 || pid == getpid_cached()) |
582 | return safe_getcwd(cwd); | |
583 | ||
0b452006 RC |
584 | p = procfs_file_alloca(pid, "cwd"); |
585 | ||
586 | return get_process_link_contents(p, cwd); | |
587 | } | |
588 | ||
589 | int get_process_root(pid_t pid, char **root) { | |
590 | const char *p; | |
591 | ||
592 | assert(pid >= 0); | |
593 | ||
594 | p = procfs_file_alloca(pid, "root"); | |
595 | ||
596 | return get_process_link_contents(p, root); | |
597 | } | |
598 | ||
2a7797e9 LP |
599 | #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U) |
600 | ||
0b452006 RC |
601 | int get_process_environ(pid_t pid, char **env) { |
602 | _cleanup_fclose_ FILE *f = NULL; | |
603 | _cleanup_free_ char *outcome = NULL; | |
319a4f4b | 604 | size_t sz = 0; |
2a7797e9 LP |
605 | const char *p; |
606 | int r; | |
0b452006 RC |
607 | |
608 | assert(pid >= 0); | |
609 | assert(env); | |
610 | ||
611 | p = procfs_file_alloca(pid, "environ"); | |
612 | ||
fdeea3f4 ZJS |
613 | r = fopen_unlocked(p, "re", &f); |
614 | if (r == -ENOENT) | |
615 | return -ESRCH; | |
616 | if (r < 0) | |
617 | return r; | |
35bbbf85 | 618 | |
2a7797e9 LP |
619 | for (;;) { |
620 | char c; | |
621 | ||
622 | if (sz >= ENVIRONMENT_BLOCK_MAX) | |
623 | return -ENOBUFS; | |
624 | ||
319a4f4b | 625 | if (!GREEDY_REALLOC(outcome, sz + 5)) |
0b452006 RC |
626 | return -ENOMEM; |
627 | ||
2a7797e9 LP |
628 | r = safe_fgetc(f, &c); |
629 | if (r < 0) | |
630 | return r; | |
631 | if (r == 0) | |
632 | break; | |
633 | ||
0b452006 RC |
634 | if (c == '\0') |
635 | outcome[sz++] = '\n'; | |
636 | else | |
637 | sz += cescape_char(c, outcome + sz); | |
638 | } | |
639 | ||
2a7797e9 | 640 | outcome[sz] = '\0'; |
ae2a15bc | 641 | *env = TAKE_PTR(outcome); |
0b452006 RC |
642 | |
643 | return 0; | |
644 | } | |
645 | ||
6bc73acb | 646 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
647 | int r; |
648 | _cleanup_free_ char *line = NULL; | |
649 | long unsigned ppid; | |
650 | const char *p; | |
651 | ||
652 | assert(pid >= 0); | |
653 | assert(_ppid); | |
654 | ||
6f8cbcdb | 655 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
656 | *_ppid = getppid(); |
657 | return 0; | |
658 | } | |
659 | ||
660 | p = procfs_file_alloca(pid, "stat"); | |
661 | r = read_one_line_file(p, &line); | |
a644184a LP |
662 | if (r == -ENOENT) |
663 | return -ESRCH; | |
0b452006 RC |
664 | if (r < 0) |
665 | return r; | |
666 | ||
667 | /* Let's skip the pid and comm fields. The latter is enclosed | |
668 | * in () but does not escape any () in its value, so let's | |
669 | * skip over it manually */ | |
670 | ||
671 | p = strrchr(line, ')'); | |
672 | if (!p) | |
673 | return -EIO; | |
674 | ||
675 | p++; | |
676 | ||
677 | if (sscanf(p, " " | |
678 | "%*c " /* state */ | |
679 | "%lu ", /* ppid */ | |
680 | &ppid) != 1) | |
681 | return -EIO; | |
682 | ||
683 | if ((long unsigned) (pid_t) ppid != ppid) | |
684 | return -ERANGE; | |
685 | ||
686 | *_ppid = (pid_t) ppid; | |
687 | ||
688 | return 0; | |
689 | } | |
690 | ||
5e37d193 FB |
691 | int get_process_umask(pid_t pid, mode_t *umask) { |
692 | _cleanup_free_ char *m = NULL; | |
693 | const char *p; | |
694 | int r; | |
695 | ||
696 | assert(umask); | |
697 | assert(pid >= 0); | |
698 | ||
699 | p = procfs_file_alloca(pid, "status"); | |
700 | ||
701 | r = get_proc_field(p, "Umask", WHITESPACE, &m); | |
702 | if (r == -ENOENT) | |
703 | return -ESRCH; | |
704 | ||
705 | return parse_mode(m, umask); | |
706 | } | |
707 | ||
0b452006 RC |
708 | int wait_for_terminate(pid_t pid, siginfo_t *status) { |
709 | siginfo_t dummy; | |
710 | ||
711 | assert(pid >= 1); | |
712 | ||
713 | if (!status) | |
714 | status = &dummy; | |
715 | ||
716 | for (;;) { | |
717 | zero(*status); | |
718 | ||
719 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
720 | ||
721 | if (errno == EINTR) | |
722 | continue; | |
723 | ||
3f0083a2 | 724 | return negative_errno(); |
0b452006 RC |
725 | } |
726 | ||
727 | return 0; | |
728 | } | |
729 | } | |
730 | ||
731 | /* | |
732 | * Return values: | |
733 | * < 0 : wait_for_terminate() failed to get the state of the | |
734 | * process, the process was terminated by a signal, or | |
735 | * failed for an unknown reason. | |
736 | * >=0 : The process terminated normally, and its exit code is | |
737 | * returned. | |
738 | * | |
739 | * That is, success is indicated by a return value of zero, and an | |
740 | * error is indicated by a non-zero value. | |
741 | * | |
742 | * A warning is emitted if the process terminates abnormally, | |
743 | * and also if it returns non-zero unless check_exit_code is true. | |
744 | */ | |
7d4904fe LP |
745 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
746 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 747 | siginfo_t status; |
7d4904fe | 748 | int r, prio; |
0b452006 | 749 | |
0b452006 RC |
750 | assert(pid > 1); |
751 | ||
7d4904fe LP |
752 | if (!name) { |
753 | r = get_process_comm(pid, &buffer); | |
754 | if (r < 0) | |
755 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
756 | else | |
757 | name = buffer; | |
758 | } | |
759 | ||
760 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
761 | ||
0b452006 RC |
762 | r = wait_for_terminate(pid, &status); |
763 | if (r < 0) | |
7d4904fe | 764 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
765 | |
766 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
767 | if (status.si_status != EXIT_SUCCESS) |
768 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
769 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
770 | else |
771 | log_debug("%s succeeded.", name); | |
772 | ||
773 | return status.si_status; | |
7d4904fe | 774 | |
3742095b | 775 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 776 | |
7d4904fe | 777 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
778 | return -EPROTO; |
779 | } | |
780 | ||
7d4904fe | 781 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
782 | return -EPROTO; |
783 | } | |
784 | ||
d5641e0d KW |
785 | /* |
786 | * Return values: | |
e225e5c3 LP |
787 | * |
788 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
789 | * was terminated by a signal, or failed for an unknown reason. | |
790 | * | |
d5641e0d KW |
791 | * >=0 : The process terminated normally with no failures. |
792 | * | |
e225e5c3 LP |
793 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
794 | * states are indicated by error is indicated by a non-zero value. | |
795 | * | |
796 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
797 | * to remain entirely race-free. | |
d5641e0d KW |
798 | */ |
799 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
800 | sigset_t mask; | |
801 | int r; | |
802 | usec_t until; | |
803 | ||
804 | assert_se(sigemptyset(&mask) == 0); | |
805 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
806 | ||
807 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
808 | * pid to exit. */ | |
496db330 | 809 | until = usec_add(now(CLOCK_MONOTONIC), timeout); |
d5641e0d KW |
810 | for (;;) { |
811 | usec_t n; | |
812 | siginfo_t status = {}; | |
813 | struct timespec ts; | |
814 | ||
815 | n = now(CLOCK_MONOTONIC); | |
816 | if (n >= until) | |
817 | break; | |
818 | ||
819 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
820 | /* Assuming we woke due to the child exiting. */ | |
821 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
822 | if (status.si_pid == pid) { | |
823 | /* This is the correct child.*/ | |
824 | if (status.si_code == CLD_EXITED) | |
825 | return (status.si_status == 0) ? 0 : -EPROTO; | |
826 | else | |
827 | return -EPROTO; | |
828 | } | |
829 | } | |
830 | /* Not the child, check for errors and proceed appropriately */ | |
831 | if (r < 0) { | |
832 | switch (r) { | |
833 | case -EAGAIN: | |
834 | /* Timed out, child is likely hung. */ | |
835 | return -ETIMEDOUT; | |
836 | case -EINTR: | |
837 | /* Received a different signal and should retry */ | |
838 | continue; | |
839 | default: | |
840 | /* Return any unexpected errors */ | |
841 | return r; | |
842 | } | |
843 | } | |
844 | } | |
845 | ||
846 | return -EPROTO; | |
847 | } | |
848 | ||
89c9030d LP |
849 | void sigkill_wait(pid_t pid) { |
850 | assert(pid > 1); | |
851 | ||
53640e6f | 852 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
853 | (void) wait_for_terminate(pid, NULL); |
854 | } | |
855 | ||
856 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
857 | PROTECT_ERRNO; |
858 | ||
4d0d3d41 LP |
859 | if (!pid) |
860 | return; | |
861 | if (*pid <= 1) | |
862 | return; | |
863 | ||
89c9030d | 864 | sigkill_wait(*pid); |
4d0d3d41 LP |
865 | } |
866 | ||
392cf1d0 SL |
867 | void sigterm_wait(pid_t pid) { |
868 | assert(pid > 1); | |
869 | ||
53640e6f | 870 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
871 | (void) wait_for_terminate(pid, NULL); |
872 | } | |
873 | ||
0b452006 RC |
874 | int kill_and_sigcont(pid_t pid, int sig) { |
875 | int r; | |
876 | ||
877 | r = kill(pid, sig) < 0 ? -errno : 0; | |
878 | ||
26f417d3 LP |
879 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
880 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 881 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 882 | (void) kill(pid, SIGCONT); |
0b452006 RC |
883 | |
884 | return r; | |
885 | } | |
886 | ||
e70f4453 | 887 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
888 | _cleanup_fclose_ FILE *f = NULL; |
889 | char *value = NULL; | |
0b452006 | 890 | const char *path; |
0d90bd92 LP |
891 | size_t l, sum = 0; |
892 | int r; | |
0b452006 RC |
893 | |
894 | assert(pid >= 0); | |
895 | assert(field); | |
e70f4453 LP |
896 | assert(ret); |
897 | ||
898 | if (pid == 0 || pid == getpid_cached()) { | |
899 | const char *e; | |
900 | ||
901 | e = getenv(field); | |
902 | if (!e) { | |
903 | *ret = NULL; | |
904 | return 0; | |
905 | } | |
906 | ||
907 | value = strdup(e); | |
908 | if (!value) | |
909 | return -ENOMEM; | |
910 | ||
911 | *ret = value; | |
912 | return 1; | |
913 | } | |
0b452006 | 914 | |
0d90bd92 LP |
915 | if (!pid_is_valid(pid)) |
916 | return -EINVAL; | |
917 | ||
0b452006 RC |
918 | path = procfs_file_alloca(pid, "environ"); |
919 | ||
fdeea3f4 ZJS |
920 | r = fopen_unlocked(path, "re", &f); |
921 | if (r == -ENOENT) | |
922 | return -ESRCH; | |
923 | if (r < 0) | |
924 | return r; | |
35bbbf85 | 925 | |
0b452006 | 926 | l = strlen(field); |
0d90bd92 LP |
927 | for (;;) { |
928 | _cleanup_free_ char *line = NULL; | |
0b452006 | 929 | |
0d90bd92 LP |
930 | if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */ |
931 | return -ENOBUFS; | |
0b452006 | 932 | |
0d90bd92 LP |
933 | r = read_nul_string(f, LONG_LINE_MAX, &line); |
934 | if (r < 0) | |
935 | return r; | |
936 | if (r == 0) /* EOF */ | |
937 | break; | |
0b452006 | 938 | |
0d90bd92 | 939 | sum += r; |
0b452006 | 940 | |
041b5ae1 | 941 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
942 | value = strdup(line + l + 1); |
943 | if (!value) | |
944 | return -ENOMEM; | |
945 | ||
e70f4453 LP |
946 | *ret = value; |
947 | return 1; | |
0b452006 | 948 | } |
0d90bd92 | 949 | } |
0b452006 | 950 | |
e70f4453 LP |
951 | *ret = NULL; |
952 | return 0; | |
0b452006 RC |
953 | } |
954 | ||
4d051546 FB |
955 | int pid_is_my_child(pid_t pid) { |
956 | pid_t ppid; | |
957 | int r; | |
958 | ||
959 | if (pid <= 1) | |
960 | return false; | |
961 | ||
962 | r = get_process_ppid(pid, &ppid); | |
963 | if (r < 0) | |
964 | return r; | |
965 | ||
966 | return ppid == getpid_cached(); | |
967 | } | |
968 | ||
0b452006 RC |
969 | bool pid_is_unwaited(pid_t pid) { |
970 | /* Checks whether a PID is still valid at all, including a zombie */ | |
971 | ||
07b38ba5 | 972 | if (pid < 0) |
0b452006 RC |
973 | return false; |
974 | ||
5fd9b2c5 LP |
975 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
976 | return true; | |
977 | ||
6f8cbcdb LP |
978 | if (pid == getpid_cached()) |
979 | return true; | |
980 | ||
0b452006 RC |
981 | if (kill(pid, 0) >= 0) |
982 | return true; | |
983 | ||
984 | return errno != ESRCH; | |
985 | } | |
986 | ||
987 | bool pid_is_alive(pid_t pid) { | |
988 | int r; | |
989 | ||
990 | /* Checks whether a PID is still valid and not a zombie */ | |
991 | ||
07b38ba5 | 992 | if (pid < 0) |
0b452006 RC |
993 | return false; |
994 | ||
5fd9b2c5 LP |
995 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
996 | return true; | |
997 | ||
6f8cbcdb LP |
998 | if (pid == getpid_cached()) |
999 | return true; | |
1000 | ||
0b452006 | 1001 | r = get_process_state(pid); |
4c701096 | 1002 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
1003 | return false; |
1004 | ||
1005 | return true; | |
1006 | } | |
d4510856 | 1007 | |
1359fffa MS |
1008 | int pid_from_same_root_fs(pid_t pid) { |
1009 | const char *root; | |
1010 | ||
07b38ba5 | 1011 | if (pid < 0) |
6f8cbcdb LP |
1012 | return false; |
1013 | ||
1014 | if (pid == 0 || pid == getpid_cached()) | |
1015 | return true; | |
1359fffa MS |
1016 | |
1017 | root = procfs_file_alloca(pid, "root"); | |
1018 | ||
e3f791a2 | 1019 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
1020 | } |
1021 | ||
d4510856 LP |
1022 | bool is_main_thread(void) { |
1023 | static thread_local int cached = 0; | |
1024 | ||
1025 | if (_unlikely_(cached == 0)) | |
df0ff127 | 1026 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
1027 | |
1028 | return cached > 0; | |
1029 | } | |
7b3e062c | 1030 | |
848e863a | 1031 | _noreturn_ void freeze(void) { |
7b3e062c | 1032 | |
3da48d7a EV |
1033 | log_close(); |
1034 | ||
7b3e062c | 1035 | /* Make sure nobody waits for us on a socket anymore */ |
7acf581a | 1036 | (void) close_all_fds(NULL, 0); |
7b3e062c LP |
1037 | |
1038 | sync(); | |
1039 | ||
8647283e MS |
1040 | /* Let's not freeze right away, but keep reaping zombies. */ |
1041 | for (;;) { | |
1042 | int r; | |
1043 | siginfo_t si = {}; | |
1044 | ||
1045 | r = waitid(P_ALL, 0, &si, WEXITED); | |
1046 | if (r < 0 && errno != EINTR) | |
1047 | break; | |
1048 | } | |
1049 | ||
1050 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
1051 | for (;;) |
1052 | pause(); | |
1053 | } | |
1054 | ||
1055 | bool oom_score_adjust_is_valid(int oa) { | |
1056 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
1057 | } | |
1058 | ||
1059 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1060 | int architecture; |
7b3e062c | 1061 | |
0c0fea07 LP |
1062 | if (!p) |
1063 | return PERSONALITY_INVALID; | |
1064 | ||
6e5f1b57 LP |
1065 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1066 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1067 | * the same register size. */ | |
1068 | ||
1069 | architecture = architecture_from_string(p); | |
1070 | if (architecture < 0) | |
1071 | return PERSONALITY_INVALID; | |
7b3e062c | 1072 | |
0c0fea07 | 1073 | if (architecture == native_architecture()) |
7b3e062c | 1074 | return PER_LINUX; |
0c0fea07 LP |
1075 | #ifdef SECONDARY_ARCHITECTURE |
1076 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1077 | return PER_LINUX32; |
7b3e062c LP |
1078 | #endif |
1079 | ||
1080 | return PERSONALITY_INVALID; | |
1081 | } | |
1082 | ||
1083 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1084 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1085 | |
7b3e062c | 1086 | if (p == PER_LINUX) |
0c0fea07 LP |
1087 | architecture = native_architecture(); |
1088 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1089 | else if (p == PER_LINUX32) |
0c0fea07 | 1090 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1091 | #endif |
1092 | ||
6e5f1b57 LP |
1093 | if (architecture < 0) |
1094 | return NULL; | |
1095 | ||
1096 | return architecture_to_string(architecture); | |
7b3e062c LP |
1097 | } |
1098 | ||
21022b9d LP |
1099 | int safe_personality(unsigned long p) { |
1100 | int ret; | |
1101 | ||
1102 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1103 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1104 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1105 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1106 | * | |
1107 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1108 | ||
1109 | errno = 0; | |
1110 | ret = personality(p); | |
1111 | if (ret < 0) { | |
1112 | if (errno != 0) | |
1113 | return -errno; | |
1114 | ||
1115 | errno = -ret; | |
1116 | } | |
1117 | ||
1118 | return ret; | |
1119 | } | |
1120 | ||
e8132d63 LP |
1121 | int opinionated_personality(unsigned long *ret) { |
1122 | int current; | |
1123 | ||
1124 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1125 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1126 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1127 | ||
21022b9d | 1128 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1129 | if (current < 0) |
21022b9d | 1130 | return current; |
e8132d63 LP |
1131 | |
1132 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1133 | *ret = PER_LINUX32; | |
1134 | else | |
1135 | *ret = PER_LINUX; | |
1136 | ||
1137 | return 0; | |
1138 | } | |
1139 | ||
dcadc967 | 1140 | void valgrind_summary_hack(void) { |
349cc4a5 | 1141 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1142 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1143 | pid_t pid; |
8869a0b4 | 1144 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1145 | if (pid < 0) |
1146 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1147 | else if (pid == 0) | |
1148 | exit(EXIT_SUCCESS); | |
1149 | else { | |
1150 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1151 | (void) wait_for_terminate(pid, NULL); | |
1152 | } | |
1153 | } | |
1154 | #endif | |
1155 | } | |
1156 | ||
93bab288 | 1157 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1158 | /* Suitable for usage in qsort() */ |
93bab288 | 1159 | return CMP(*a, *b); |
291d565a LP |
1160 | } |
1161 | ||
7f452159 LP |
1162 | int ioprio_parse_priority(const char *s, int *ret) { |
1163 | int i, r; | |
1164 | ||
1165 | assert(s); | |
1166 | assert(ret); | |
1167 | ||
1168 | r = safe_atoi(s, &i); | |
1169 | if (r < 0) | |
1170 | return r; | |
1171 | ||
1172 | if (!ioprio_priority_is_valid(i)) | |
1173 | return -EINVAL; | |
1174 | ||
1175 | *ret = i; | |
1176 | return 0; | |
1177 | } | |
1178 | ||
5c30a6d2 LP |
1179 | /* The cached PID, possible values: |
1180 | * | |
1181 | * == UNSET [0] → cache not initialized yet | |
1182 | * == BUSY [-1] → some thread is initializing it at the moment | |
1183 | * any other → the cached PID | |
1184 | */ | |
1185 | ||
1186 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1187 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1188 | ||
1189 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1190 | ||
799a960d | 1191 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1192 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1193 | cached_pid = CACHED_PID_UNSET; | |
1194 | } | |
1195 | ||
1196 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1197 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1198 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1199 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
d34dae18 | 1200 | extern void* __dso_handle _weak_; |
5c30a6d2 LP |
1201 | |
1202 | pid_t getpid_cached(void) { | |
5d71bac3 | 1203 | static bool installed = false; |
5c30a6d2 LP |
1204 | pid_t current_value; |
1205 | ||
1206 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1207 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1208 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1209 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1210 | * | |
1211 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1212 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1213 | */ |
1214 | ||
1215 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1216 | ||
1217 | switch (current_value) { | |
1218 | ||
1219 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1220 | pid_t new_pid; | |
1221 | ||
996def17 | 1222 | new_pid = raw_getpid(); |
5c30a6d2 | 1223 | |
5d71bac3 LP |
1224 | if (!installed) { |
1225 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1226 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1227 | * we'll check for errors only in the most generic fashion possible. */ | |
1228 | ||
1229 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1230 | /* OOM? Let's try again later */ | |
1231 | cached_pid = CACHED_PID_UNSET; | |
1232 | return new_pid; | |
1233 | } | |
1234 | ||
1235 | installed = true; | |
5c30a6d2 LP |
1236 | } |
1237 | ||
1238 | cached_pid = new_pid; | |
1239 | return new_pid; | |
1240 | } | |
1241 | ||
1242 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1243 | return raw_getpid(); |
5c30a6d2 LP |
1244 | |
1245 | default: /* Properly initialized */ | |
1246 | return current_value; | |
1247 | } | |
1248 | } | |
1249 | ||
fba868fa LP |
1250 | int must_be_root(void) { |
1251 | ||
1252 | if (geteuid() == 0) | |
1253 | return 0; | |
1254 | ||
baaa35ad | 1255 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1256 | } |
1257 | ||
4b3abcd0 MG |
1258 | static void restore_sigsetp(sigset_t **ssp) { |
1259 | if (*ssp) | |
1260 | (void) sigprocmask(SIG_SETMASK, *ssp, NULL); | |
1261 | } | |
1262 | ||
4c253ed1 LP |
1263 | int safe_fork_full( |
1264 | const char *name, | |
1265 | const int except_fds[], | |
1266 | size_t n_except_fds, | |
1267 | ForkFlags flags, | |
1268 | pid_t *ret_pid) { | |
1269 | ||
1270 | pid_t original_pid, pid; | |
1f5d1e02 | 1271 | sigset_t saved_ss, ss; |
4b3abcd0 | 1272 | _cleanup_(restore_sigsetp) sigset_t *saved_ssp = NULL; |
d7e38115 | 1273 | bool block_signals = false, block_all = false; |
b6e1fff1 | 1274 | int prio, r; |
4c253ed1 LP |
1275 | |
1276 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1277 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1278 | ||
b6e1fff1 LP |
1279 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1280 | ||
4c253ed1 LP |
1281 | original_pid = getpid_cached(); |
1282 | ||
48f813c4 LP |
1283 | if (flags & FORK_FLUSH_STDIO) { |
1284 | fflush(stdout); | |
1285 | fflush(stderr); /* This one shouldn't be necessary, stderr should be unbuffered anyway, but let's better be safe than sorry */ | |
1286 | } | |
1287 | ||
1f5d1e02 | 1288 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
1f5d1e02 LP |
1289 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1290 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1291 | |
cd2a429e | 1292 | assert_se(sigfillset(&ss) >= 0); |
d7e38115 | 1293 | block_signals = block_all = true; |
1f5d1e02 LP |
1294 | |
1295 | } else if (flags & FORK_WAIT) { | |
1f5d1e02 LP |
1296 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ |
1297 | ||
cd2a429e ZJS |
1298 | assert_se(sigemptyset(&ss) >= 0); |
1299 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1f5d1e02 | 1300 | block_signals = true; |
4c253ed1 LP |
1301 | } |
1302 | ||
4b3abcd0 | 1303 | if (block_signals) { |
1f5d1e02 LP |
1304 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) |
1305 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
4b3abcd0 MG |
1306 | saved_ssp = &saved_ss; |
1307 | } | |
1f5d1e02 | 1308 | |
5f968736 LP |
1309 | if ((flags & (FORK_NEW_MOUNTNS|FORK_NEW_USERNS)) != 0) |
1310 | pid = raw_clone(SIGCHLD| | |
1311 | (FLAGS_SET(flags, FORK_NEW_MOUNTNS) ? CLONE_NEWNS : 0) | | |
1312 | (FLAGS_SET(flags, FORK_NEW_USERNS) ? CLONE_NEWUSER : 0)); | |
be39f6ee LP |
1313 | else |
1314 | pid = fork(); | |
4b3abcd0 MG |
1315 | if (pid < 0) |
1316 | return log_full_errno(prio, errno, "Failed to fork: %m"); | |
4c253ed1 LP |
1317 | if (pid > 0) { |
1318 | /* We are in the parent process */ | |
1319 | ||
1f5d1e02 LP |
1320 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1321 | ||
1322 | if (flags & FORK_WAIT) { | |
d7e38115 MG |
1323 | if (block_all) { |
1324 | /* undo everything except SIGCHLD */ | |
1325 | ss = saved_ss; | |
1326 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1327 | (void) sigprocmask(SIG_SETMASK, &ss, NULL); | |
1328 | } | |
1329 | ||
1f5d1e02 LP |
1330 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); |
1331 | if (r < 0) | |
1332 | return r; | |
1333 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1334 | return -EPROTO; | |
1335 | } | |
1336 | ||
4c253ed1 LP |
1337 | if (ret_pid) |
1338 | *ret_pid = pid; | |
1339 | ||
1340 | return 1; | |
1341 | } | |
1342 | ||
1343 | /* We are in the child process */ | |
1344 | ||
4b3abcd0 MG |
1345 | /* Restore signal mask manually */ |
1346 | saved_ssp = NULL; | |
1347 | ||
4c253ed1 LP |
1348 | if (flags & FORK_REOPEN_LOG) { |
1349 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1350 | log_close(); | |
1351 | log_set_open_when_needed(true); | |
1352 | } | |
1353 | ||
1354 | if (name) { | |
1355 | r = rename_process(name); | |
1356 | if (r < 0) | |
b6e1fff1 LP |
1357 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1358 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1359 | } |
1360 | ||
97033ba4 LP |
1361 | if (flags & (FORK_DEATHSIG|FORK_DEATHSIG_SIGINT)) |
1362 | if (prctl(PR_SET_PDEATHSIG, (flags & FORK_DEATHSIG_SIGINT) ? SIGINT : SIGTERM) < 0) { | |
b6e1fff1 | 1363 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1364 | _exit(EXIT_FAILURE); |
1365 | } | |
1366 | ||
1367 | if (flags & FORK_RESET_SIGNALS) { | |
1368 | r = reset_all_signal_handlers(); | |
1369 | if (r < 0) { | |
b6e1fff1 | 1370 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1371 | _exit(EXIT_FAILURE); |
1372 | } | |
1373 | ||
1374 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1375 | r = reset_signal_mask(); | |
1376 | if (r < 0) { | |
b6e1fff1 | 1377 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1378 | _exit(EXIT_FAILURE); |
1379 | } | |
1380 | } else if (block_signals) { /* undo what we did above */ | |
1381 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1382 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1383 | _exit(EXIT_FAILURE); |
1384 | } | |
1385 | } | |
1386 | ||
1387 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1388 | pid_t ppid; |
4c253ed1 LP |
1389 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1390 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1391 | ||
7ddc2dc5 SL |
1392 | ppid = getppid(); |
1393 | if (ppid == 0) | |
d7b34e38 | 1394 | /* Parent is in a different PID namespace. */; |
7ddc2dc5 | 1395 | else if (ppid != original_pid) { |
4c253ed1 LP |
1396 | log_debug("Parent died early, raising SIGTERM."); |
1397 | (void) raise(SIGTERM); | |
1398 | _exit(EXIT_FAILURE); | |
1399 | } | |
1400 | } | |
1401 | ||
d94a24ca | 1402 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1403 | |
1404 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1405 | ||
1406 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1407 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1408 | _exit(EXIT_FAILURE); | |
1409 | } | |
1410 | } | |
1411 | ||
4c253ed1 LP |
1412 | if (flags & FORK_CLOSE_ALL_FDS) { |
1413 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1414 | log_close(); | |
1415 | ||
1416 | r = close_all_fds(except_fds, n_except_fds); | |
1417 | if (r < 0) { | |
b6e1fff1 | 1418 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1419 | _exit(EXIT_FAILURE); |
1420 | } | |
1421 | } | |
1422 | ||
1423 | /* When we were asked to reopen the logs, do so again now */ | |
1424 | if (flags & FORK_REOPEN_LOG) { | |
1425 | log_open(); | |
1426 | log_set_open_when_needed(false); | |
1427 | } | |
1428 | ||
1429 | if (flags & FORK_NULL_STDIO) { | |
1430 | r = make_null_stdio(); | |
1431 | if (r < 0) { | |
b6e1fff1 | 1432 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1433 | _exit(EXIT_FAILURE); |
1434 | } | |
8987afc4 LP |
1435 | |
1436 | } else if (flags & FORK_STDOUT_TO_STDERR) { | |
8987afc4 | 1437 | if (dup2(STDERR_FILENO, STDOUT_FILENO) < 0) { |
bddeb54c | 1438 | log_full_errno(prio, errno, "Failed to connect stdout to stderr: %m"); |
8987afc4 LP |
1439 | _exit(EXIT_FAILURE); |
1440 | } | |
4c253ed1 LP |
1441 | } |
1442 | ||
909106eb LP |
1443 | if (flags & FORK_RLIMIT_NOFILE_SAFE) { |
1444 | r = rlimit_nofile_safe(); | |
1445 | if (r < 0) { | |
1446 | log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m"); | |
1447 | _exit(EXIT_FAILURE); | |
1448 | } | |
1449 | } | |
1450 | ||
4c253ed1 LP |
1451 | if (ret_pid) |
1452 | *ret_pid = getpid_cached(); | |
1453 | ||
1454 | return 0; | |
1455 | } | |
1456 | ||
27096982 LP |
1457 | int namespace_fork( |
1458 | const char *outer_name, | |
1459 | const char *inner_name, | |
1460 | const int except_fds[], | |
1461 | size_t n_except_fds, | |
1462 | ForkFlags flags, | |
1463 | int pidns_fd, | |
1464 | int mntns_fd, | |
1465 | int netns_fd, | |
1466 | int userns_fd, | |
1467 | int root_fd, | |
1468 | pid_t *ret_pid) { | |
1469 | ||
1470 | int r; | |
1471 | ||
1472 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1473 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1474 | * /proc/self/fd works correctly. */ | |
1475 | ||
1476 | r = safe_fork_full(outer_name, except_fds, n_except_fds, (flags|FORK_DEATHSIG) & ~(FORK_REOPEN_LOG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE), ret_pid); | |
1477 | if (r < 0) | |
1478 | return r; | |
1479 | if (r == 0) { | |
1480 | pid_t pid; | |
1481 | ||
1482 | /* Child */ | |
1483 | ||
1484 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1485 | if (r < 0) { | |
1486 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1487 | _exit(EXIT_FAILURE); | |
1488 | } | |
1489 | ||
1490 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1491 | r = safe_fork_full(inner_name, except_fds, n_except_fds, flags & ~(FORK_WAIT|FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_NULL_STDIO), &pid); | |
1492 | if (r < 0) | |
1493 | _exit(EXIT_FAILURE); | |
1494 | if (r == 0) { | |
1495 | /* Child */ | |
1496 | if (ret_pid) | |
1497 | *ret_pid = pid; | |
1498 | return 0; | |
1499 | } | |
1500 | ||
1501 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1502 | if (r < 0) | |
1503 | _exit(EXIT_FAILURE); | |
1504 | ||
1505 | _exit(r); | |
1506 | } | |
1507 | ||
1508 | return 1; | |
1509 | } | |
1510 | ||
da6053d0 | 1511 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1512 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1513 | size_t n, i; |
78752f2e LP |
1514 | va_list ap; |
1515 | char **l; | |
1516 | int r; | |
1517 | ||
1518 | assert(path); | |
1519 | ||
1520 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1521 | ||
7e0ed2e9 SB |
1522 | r = safe_fork_full(name, |
1523 | except, | |
1524 | n_except, | |
1525 | FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS|FORK_REOPEN_LOG, | |
1526 | ret_pid); | |
78752f2e LP |
1527 | if (r < 0) |
1528 | return r; | |
1529 | if (r > 0) | |
1530 | return 0; | |
1531 | ||
1532 | /* In the child: */ | |
1533 | ||
1534 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1535 | stderr_is_tty = isatty(STDERR_FILENO); | |
1536 | ||
1537 | if (!stdout_is_tty || !stderr_is_tty) { | |
1538 | int fd; | |
1539 | ||
1540 | /* Detach from stdout/stderr. and reopen | |
1541 | * /dev/tty for them. This is important to | |
1542 | * ensure that when systemctl is started via | |
1543 | * popen() or a similar call that expects to | |
1544 | * read EOF we actually do generate EOF and | |
1545 | * not delay this indefinitely by because we | |
1546 | * keep an unused copy of stdin around. */ | |
1547 | fd = open("/dev/tty", O_WRONLY); | |
1548 | if (fd < 0) { | |
1549 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1550 | _exit(EXIT_FAILURE); | |
1551 | } | |
1552 | ||
1553 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1554 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1555 | _exit(EXIT_FAILURE); | |
1556 | } | |
1557 | ||
1558 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1559 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1560 | _exit(EXIT_FAILURE); | |
1561 | } | |
1562 | ||
e7685a77 | 1563 | safe_close_above_stdio(fd); |
78752f2e LP |
1564 | } |
1565 | ||
595225af LP |
1566 | (void) rlimit_nofile_safe(); |
1567 | ||
78752f2e LP |
1568 | /* Count arguments */ |
1569 | va_start(ap, path); | |
1570 | for (n = 0; va_arg(ap, char*); n++) | |
1571 | ; | |
1572 | va_end(ap); | |
1573 | ||
1574 | /* Allocate strv */ | |
cf409d15 | 1575 | l = newa(char*, n + 1); |
78752f2e LP |
1576 | |
1577 | /* Fill in arguments */ | |
1578 | va_start(ap, path); | |
1579 | for (i = 0; i <= n; i++) | |
1580 | l[i] = va_arg(ap, char*); | |
1581 | va_end(ap); | |
1582 | ||
1583 | execv(path, l); | |
1584 | _exit(EXIT_FAILURE); | |
1585 | } | |
1586 | ||
9f8168eb LP |
1587 | int set_oom_score_adjust(int value) { |
1588 | char t[DECIMAL_STR_MAX(int)]; | |
1589 | ||
1590 | sprintf(t, "%i", value); | |
1591 | ||
1592 | return write_string_file("/proc/self/oom_score_adj", t, | |
1593 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1594 | } | |
1595 | ||
298f466f LP |
1596 | int pidfd_get_pid(int fd, pid_t *ret) { |
1597 | char path[STRLEN("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)]; | |
1598 | _cleanup_free_ char *fdinfo = NULL; | |
1599 | char *p; | |
1600 | int r; | |
1601 | ||
1602 | if (fd < 0) | |
1603 | return -EBADF; | |
1604 | ||
1605 | xsprintf(path, "/proc/self/fdinfo/%i", fd); | |
1606 | ||
627055ce | 1607 | r = read_full_virtual_file(path, &fdinfo, NULL); |
298f466f LP |
1608 | if (r == -ENOENT) /* if fdinfo doesn't exist we assume the process does not exist */ |
1609 | return -ESRCH; | |
1610 | if (r < 0) | |
1611 | return r; | |
1612 | ||
1613 | p = startswith(fdinfo, "Pid:"); | |
1614 | if (!p) { | |
1615 | p = strstr(fdinfo, "\nPid:"); | |
1616 | if (!p) | |
1617 | return -ENOTTY; /* not a pidfd? */ | |
1618 | ||
1619 | p += 5; | |
1620 | } | |
1621 | ||
1622 | p += strspn(p, WHITESPACE); | |
1623 | p[strcspn(p, WHITESPACE)] = 0; | |
1624 | ||
1625 | return parse_pid(p, ret); | |
1626 | } | |
1627 | ||
39090201 DJL |
1628 | static int rlimit_to_nice(rlim_t limit) { |
1629 | if (limit <= 1) | |
1630 | return PRIO_MAX-1; /* i.e. 19 */ | |
1631 | ||
1632 | if (limit >= -PRIO_MIN + PRIO_MAX) | |
1633 | return PRIO_MIN; /* i.e. -20 */ | |
1634 | ||
1635 | return PRIO_MAX - (int) limit; | |
1636 | } | |
1637 | ||
1638 | int setpriority_closest(int priority) { | |
1639 | int current, limit, saved_errno; | |
1640 | struct rlimit highest; | |
1641 | ||
1642 | /* Try to set requested nice level */ | |
1643 | if (setpriority(PRIO_PROCESS, 0, priority) >= 0) | |
1644 | return 1; | |
1645 | ||
1646 | /* Permission failed */ | |
1647 | saved_errno = -errno; | |
1648 | if (!ERRNO_IS_PRIVILEGE(saved_errno)) | |
1649 | return saved_errno; | |
1650 | ||
1651 | errno = 0; | |
1652 | current = getpriority(PRIO_PROCESS, 0); | |
1653 | if (errno != 0) | |
1654 | return -errno; | |
1655 | ||
1656 | if (priority == current) | |
1657 | return 1; | |
1658 | ||
1659 | /* Hmm, we'd expect that raising the nice level from our status quo would always work. If it doesn't, | |
1660 | * then the whole setpriority() system call is blocked to us, hence let's propagate the error | |
1661 | * right-away */ | |
1662 | if (priority > current) | |
1663 | return saved_errno; | |
1664 | ||
1665 | if (getrlimit(RLIMIT_NICE, &highest) < 0) | |
1666 | return -errno; | |
1667 | ||
1668 | limit = rlimit_to_nice(highest.rlim_cur); | |
1669 | ||
1670 | /* We are already less nice than limit allows us */ | |
1671 | if (current < limit) { | |
1672 | log_debug("Cannot raise nice level, permissions and the resource limit do not allow it."); | |
1673 | return 0; | |
1674 | } | |
1675 | ||
1676 | /* Push to the allowed limit */ | |
1677 | if (setpriority(PRIO_PROCESS, 0, limit) < 0) | |
1678 | return -errno; | |
1679 | ||
1680 | log_debug("Cannot set requested nice level (%i), used next best (%i).", priority, limit); | |
1681 | return 0; | |
1682 | } | |
1683 | ||
2306d177 LP |
1684 | bool invoked_as(char *argv[], const char *token) { |
1685 | if (!argv || isempty(argv[0])) | |
1686 | return false; | |
1687 | ||
1688 | if (isempty(token)) | |
1689 | return false; | |
1690 | ||
1691 | return strstr(last_path_component(argv[0]), token); | |
1692 | } | |
1693 | ||
7b3e062c LP |
1694 | static const char *const ioprio_class_table[] = { |
1695 | [IOPRIO_CLASS_NONE] = "none", | |
1696 | [IOPRIO_CLASS_RT] = "realtime", | |
1697 | [IOPRIO_CLASS_BE] = "best-effort", | |
f44b3035 | 1698 | [IOPRIO_CLASS_IDLE] = "idle", |
7b3e062c LP |
1699 | }; |
1700 | ||
10062bbc | 1701 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1702 | |
1703 | static const char *const sigchld_code_table[] = { | |
1704 | [CLD_EXITED] = "exited", | |
1705 | [CLD_KILLED] = "killed", | |
1706 | [CLD_DUMPED] = "dumped", | |
1707 | [CLD_TRAPPED] = "trapped", | |
1708 | [CLD_STOPPED] = "stopped", | |
1709 | [CLD_CONTINUED] = "continued", | |
1710 | }; | |
1711 | ||
1712 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1713 | ||
1714 | static const char* const sched_policy_table[] = { | |
1715 | [SCHED_OTHER] = "other", | |
1716 | [SCHED_BATCH] = "batch", | |
1717 | [SCHED_IDLE] = "idle", | |
1718 | [SCHED_FIFO] = "fifo", | |
f44b3035 | 1719 | [SCHED_RR] = "rr", |
7b3e062c LP |
1720 | }; |
1721 | ||
1722 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |