]>
Commit | Line | Data |
---|---|---|
1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ | |
2 | ||
3 | #include <sys/wait.h> | |
4 | #include <unistd.h> | |
5 | ||
6 | #include "alloc-util.h" | |
7 | #include "errno-util.h" | |
8 | #include "fd-util.h" | |
9 | #include "format-util.h" | |
10 | #include "hash-funcs.h" | |
11 | #include "log.h" | |
12 | #include "parse-util.h" | |
13 | #include "pidfd-util.h" | |
14 | #include "pidref.h" | |
15 | #include "process-util.h" | |
16 | #include "siphash24.h" | |
17 | ||
18 | int pidref_acquire_pidfd_id(PidRef *pidref) { | |
19 | int r; | |
20 | ||
21 | assert(pidref); | |
22 | ||
23 | if (!pidref_is_set(pidref)) | |
24 | return -ESRCH; | |
25 | ||
26 | if (pidref_is_remote(pidref)) | |
27 | return -EREMOTE; | |
28 | ||
29 | if (pidref->fd < 0) | |
30 | return -ENOMEDIUM; | |
31 | ||
32 | if (pidref->fd_id > 0) | |
33 | return 0; | |
34 | ||
35 | r = pidfd_get_inode_id(pidref->fd, &pidref->fd_id); | |
36 | if (r < 0) { | |
37 | if (!ERRNO_IS_NEG_NOT_SUPPORTED(r)) | |
38 | log_debug_errno(r, "Failed to get inode number of pidfd for pid " PID_FMT ": %m", | |
39 | pidref->pid); | |
40 | return r; | |
41 | } | |
42 | ||
43 | return 0; | |
44 | } | |
45 | ||
46 | bool pidref_equal(PidRef *a, PidRef *b) { | |
47 | ||
48 | /* If this is the very same structure, it definitely refers to the same process */ | |
49 | if (a == b) | |
50 | return true; | |
51 | ||
52 | if (!pidref_is_set(a)) | |
53 | return !pidref_is_set(b); | |
54 | ||
55 | if (!pidref_is_set(b)) | |
56 | return false; | |
57 | ||
58 | if (a->pid != b->pid) | |
59 | return false; | |
60 | ||
61 | if (pidref_is_remote(a)) { | |
62 | /* If one is remote and the other isn't, they are not the same */ | |
63 | if (!pidref_is_remote(b)) | |
64 | return false; | |
65 | ||
66 | /* If both are remote, compare fd IDs if we have both, otherwise don't bother, and cut things short */ | |
67 | if (a->fd_id == 0 || b->fd_id == 0) | |
68 | return true; | |
69 | } else { | |
70 | /* If the other side is remote, then this is not the same */ | |
71 | if (pidref_is_remote(b)) | |
72 | return false; | |
73 | ||
74 | /* PID1 cannot exit, hence it cannot change pidfs ids, hence no point in comparing them, we | |
75 | * can shortcut things */ | |
76 | if (a->pid == 1) | |
77 | return true; | |
78 | ||
79 | /* Try to compare pidfds using their inode numbers. This way we can ensure that we | |
80 | * don't spuriously consider two PidRefs equal if the pid has been reused once. Note | |
81 | * that we ignore all errors here, not only EOPNOTSUPP, as fstat() might fail due to | |
82 | * many reasons. */ | |
83 | if (pidref_acquire_pidfd_id(a) < 0 || pidref_acquire_pidfd_id(b) < 0) | |
84 | return true; | |
85 | } | |
86 | ||
87 | return a->fd_id == b->fd_id; | |
88 | } | |
89 | ||
90 | int pidref_set_pid(PidRef *pidref, pid_t pid) { | |
91 | uint64_t pidfdid = 0; | |
92 | int fd; | |
93 | ||
94 | assert(pidref); | |
95 | ||
96 | if (pid < 0) | |
97 | return -ESRCH; | |
98 | if (pid == 0) { | |
99 | pid = getpid_cached(); | |
100 | (void) pidfd_get_inode_id_self_cached(&pidfdid); | |
101 | } | |
102 | ||
103 | fd = pidfd_open(pid, 0); | |
104 | if (fd < 0) { | |
105 | /* Graceful fallback in case the kernel is out of fds */ | |
106 | if (!ERRNO_IS_RESOURCE(errno)) | |
107 | return log_debug_errno(errno, "Failed to open pidfd for pid " PID_FMT ": %m", pid); | |
108 | ||
109 | fd = -EBADF; | |
110 | } | |
111 | ||
112 | *pidref = (PidRef) { | |
113 | .fd = fd, | |
114 | .pid = pid, | |
115 | .fd_id = pidfdid, | |
116 | }; | |
117 | ||
118 | return 0; | |
119 | } | |
120 | ||
121 | int pidref_set_pidstr(PidRef *pidref, const char *pid) { | |
122 | pid_t nr; | |
123 | int r; | |
124 | ||
125 | assert(pidref); | |
126 | ||
127 | r = parse_pid(pid, &nr); | |
128 | if (r < 0) | |
129 | return r; | |
130 | ||
131 | return pidref_set_pid(pidref, nr); | |
132 | } | |
133 | ||
134 | int pidref_set_pidfd(PidRef *pidref, int fd) { | |
135 | int r; | |
136 | ||
137 | assert(pidref); | |
138 | ||
139 | if (fd < 0) | |
140 | return -EBADF; | |
141 | ||
142 | int fd_copy = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
143 | if (fd_copy < 0) { | |
144 | pid_t pid; | |
145 | ||
146 | if (!ERRNO_IS_RESOURCE(errno)) | |
147 | return -errno; | |
148 | ||
149 | /* Graceful fallback if we are out of fds */ | |
150 | r = pidfd_get_pid(fd, &pid); | |
151 | if (r < 0) | |
152 | return r; | |
153 | ||
154 | *pidref = PIDREF_MAKE_FROM_PID(pid); | |
155 | return 0; | |
156 | } | |
157 | ||
158 | return pidref_set_pidfd_consume(pidref, fd_copy); | |
159 | } | |
160 | ||
161 | int pidref_set_pidfd_take(PidRef *pidref, int fd) { | |
162 | pid_t pid; | |
163 | int r; | |
164 | ||
165 | assert(pidref); | |
166 | ||
167 | if (fd < 0) | |
168 | return -EBADF; | |
169 | ||
170 | r = pidfd_get_pid(fd, &pid); | |
171 | if (r < 0) | |
172 | return r; | |
173 | ||
174 | *pidref = (PidRef) { | |
175 | .fd = fd, | |
176 | .pid = pid, | |
177 | }; | |
178 | ||
179 | return 0; | |
180 | } | |
181 | ||
182 | int pidref_set_pidfd_consume(PidRef *pidref, int fd) { | |
183 | int r; | |
184 | ||
185 | r = pidref_set_pidfd_take(pidref, fd); | |
186 | if (r < 0) | |
187 | safe_close(fd); | |
188 | ||
189 | return r; | |
190 | } | |
191 | ||
192 | int pidref_set_parent(PidRef *ret) { | |
193 | _cleanup_(pidref_done) PidRef parent = PIDREF_NULL; | |
194 | pid_t ppid; | |
195 | int r; | |
196 | ||
197 | assert(ret); | |
198 | ||
199 | /* Acquires a pidref to our parent process. Deals with the fact that parent processes might exit, and | |
200 | * we get reparented to other processes, with our old parent's PID already being recycled. */ | |
201 | ||
202 | ppid = getppid(); | |
203 | for (;;) { | |
204 | r = pidref_set_pid(&parent, ppid); | |
205 | if (r < 0) | |
206 | return r; | |
207 | ||
208 | if (parent.fd < 0) /* If pidfds are not available, then we are done */ | |
209 | break; | |
210 | ||
211 | pid_t now_ppid = getppid(); | |
212 | if (now_ppid == ppid) /* If our ppid is still the same, then we are done */ | |
213 | break; | |
214 | ||
215 | /* Otherwise let's try again with the new ppid */ | |
216 | ppid = now_ppid; | |
217 | pidref_done(&parent); | |
218 | } | |
219 | ||
220 | *ret = TAKE_PIDREF(parent); | |
221 | return 0; | |
222 | } | |
223 | ||
224 | void pidref_done(PidRef *pidref) { | |
225 | assert(pidref); | |
226 | ||
227 | *pidref = (PidRef) { | |
228 | .fd = safe_close(pidref->fd), | |
229 | }; | |
230 | } | |
231 | ||
232 | PidRef* pidref_free(PidRef *pidref) { | |
233 | /* Regularly, this is an embedded structure. But sometimes we want it on the heap too */ | |
234 | if (!pidref) | |
235 | return NULL; | |
236 | ||
237 | pidref_done(pidref); | |
238 | return mfree(pidref); | |
239 | } | |
240 | ||
241 | int pidref_copy(const PidRef *pidref, PidRef *ret) { | |
242 | _cleanup_(pidref_done) PidRef copy = PIDREF_NULL; | |
243 | ||
244 | /* If NULL is passed we'll generate a PidRef that refers to no process. This makes it easy to | |
245 | * copy pidref fields that might or might not reference a process yet. */ | |
246 | ||
247 | assert(ret); | |
248 | ||
249 | if (pidref) { | |
250 | if (pidref_is_remote(pidref)) /* Propagate remote flag */ | |
251 | copy.fd = -EREMOTE; | |
252 | else if (pidref->fd >= 0) { | |
253 | copy.fd = fcntl(pidref->fd, F_DUPFD_CLOEXEC, 3); | |
254 | if (copy.fd < 0) { | |
255 | if (!ERRNO_IS_RESOURCE(errno)) | |
256 | return -errno; | |
257 | ||
258 | copy.fd = -EBADF; | |
259 | } | |
260 | } | |
261 | ||
262 | copy.pid = pidref->pid; | |
263 | copy.fd_id = pidref->fd_id; | |
264 | } | |
265 | ||
266 | *ret = TAKE_PIDREF(copy); | |
267 | return 0; | |
268 | } | |
269 | ||
270 | int pidref_dup(const PidRef *pidref, PidRef **ret) { | |
271 | _cleanup_(pidref_freep) PidRef *dup_pidref = NULL; | |
272 | int r; | |
273 | ||
274 | /* Allocates a new PidRef on the heap, making it a copy of the specified pidref. This does not try to | |
275 | * acquire a pidfd if we don't have one yet! */ | |
276 | ||
277 | assert(ret); | |
278 | ||
279 | dup_pidref = newdup(PidRef, &PIDREF_NULL, 1); | |
280 | if (!dup_pidref) | |
281 | return -ENOMEM; | |
282 | ||
283 | r = pidref_copy(pidref, dup_pidref); | |
284 | if (r < 0) | |
285 | return r; | |
286 | ||
287 | *ret = TAKE_PTR(dup_pidref); | |
288 | return 0; | |
289 | } | |
290 | ||
291 | int pidref_new_from_pid(pid_t pid, PidRef **ret) { | |
292 | _cleanup_(pidref_freep) PidRef *n = NULL; | |
293 | int r; | |
294 | ||
295 | assert(ret); | |
296 | ||
297 | if (pid < 0) | |
298 | return -ESRCH; | |
299 | ||
300 | n = new(PidRef, 1); | |
301 | if (!n) | |
302 | return -ENOMEM; | |
303 | ||
304 | *n = PIDREF_NULL; | |
305 | ||
306 | r = pidref_set_pid(n, pid); | |
307 | if (r < 0) | |
308 | return r; | |
309 | ||
310 | *ret = TAKE_PTR(n); | |
311 | return 0; | |
312 | } | |
313 | ||
314 | int pidref_kill(const PidRef *pidref, int sig) { | |
315 | ||
316 | if (!pidref) | |
317 | return -ESRCH; | |
318 | ||
319 | if (pidref_is_remote(pidref)) | |
320 | return -EREMOTE; | |
321 | ||
322 | if (pidref->fd >= 0) | |
323 | return RET_NERRNO(pidfd_send_signal(pidref->fd, sig, NULL, 0)); | |
324 | ||
325 | if (pidref->pid > 0) | |
326 | return RET_NERRNO(kill(pidref->pid, sig)); | |
327 | ||
328 | return -ESRCH; | |
329 | } | |
330 | ||
331 | int pidref_kill_and_sigcont(const PidRef *pidref, int sig) { | |
332 | int r; | |
333 | ||
334 | r = pidref_kill(pidref, sig); | |
335 | if (r < 0) | |
336 | return r; | |
337 | ||
338 | if (!IN_SET(sig, SIGCONT, SIGKILL)) | |
339 | (void) pidref_kill(pidref, SIGCONT); | |
340 | ||
341 | return 0; | |
342 | } | |
343 | ||
344 | int pidref_sigqueue(const PidRef *pidref, int sig, int value) { | |
345 | ||
346 | if (!pidref) | |
347 | return -ESRCH; | |
348 | ||
349 | if (pidref_is_remote(pidref)) | |
350 | return -EREMOTE; | |
351 | ||
352 | if (pidref->fd >= 0) { | |
353 | siginfo_t si; | |
354 | ||
355 | /* We can't use structured initialization here, since the structure contains various unions | |
356 | * and these fields lie in overlapping (carefully aligned) unions that LLVM is allergic to | |
357 | * allow assignments to */ | |
358 | zero(si); | |
359 | si.si_signo = sig; | |
360 | si.si_code = SI_QUEUE; | |
361 | si.si_pid = getpid_cached(); | |
362 | si.si_uid = getuid(); | |
363 | si.si_value.sival_int = value; | |
364 | ||
365 | return RET_NERRNO(pidfd_send_signal(pidref->fd, sig, &si, 0)); | |
366 | } | |
367 | ||
368 | if (pidref->pid > 0) | |
369 | return RET_NERRNO(sigqueue(pidref->pid, sig, (const union sigval) { .sival_int = value })); | |
370 | ||
371 | return -ESRCH; | |
372 | } | |
373 | ||
374 | int pidref_verify(const PidRef *pidref) { | |
375 | int r; | |
376 | ||
377 | /* This is a helper that is supposed to be called after reading information from procfs via a | |
378 | * PidRef. It ensures that the PID we track still matches the PIDFD we pin. If this value differs | |
379 | * after a procfs read, we might have read the data from a recycled PID. */ | |
380 | ||
381 | if (!pidref_is_set(pidref)) | |
382 | return -ESRCH; | |
383 | ||
384 | if (pidref_is_remote(pidref)) | |
385 | return -EREMOTE; | |
386 | ||
387 | if (pidref->pid == 1) | |
388 | return 1; /* PID 1 can never go away, hence never be recycled to a different process → return 1 */ | |
389 | ||
390 | if (pidref->fd < 0) | |
391 | return 0; /* If we don't have a pidfd we cannot validate it, hence we assume it's all OK → return 0 */ | |
392 | ||
393 | r = pidfd_verify_pid(pidref->fd, pidref->pid); | |
394 | if (r < 0) | |
395 | return r; | |
396 | ||
397 | return 1; /* We have a pidfd and it still points to the PID we have, hence all is *really* OK → return 1 */ | |
398 | } | |
399 | ||
400 | bool pidref_is_self(PidRef *pidref) { | |
401 | if (!pidref_is_set(pidref)) | |
402 | return false; | |
403 | ||
404 | if (pidref_is_remote(pidref)) | |
405 | return false; | |
406 | ||
407 | if (pidref->pid != getpid_cached()) | |
408 | return false; | |
409 | ||
410 | /* PID1 cannot exit, hence no point in comparing pidfd IDs, they can never change */ | |
411 | if (pidref->pid == 1) | |
412 | return true; | |
413 | ||
414 | /* Also compare pidfd ID if we can get it */ | |
415 | if (pidref_acquire_pidfd_id(pidref) < 0) | |
416 | return true; | |
417 | ||
418 | uint64_t self_id; | |
419 | if (pidfd_get_inode_id_self_cached(&self_id) < 0) | |
420 | return true; | |
421 | ||
422 | return pidref->fd_id == self_id; | |
423 | } | |
424 | ||
425 | int pidref_wait(PidRef *pidref, siginfo_t *ret, int options) { | |
426 | int r; | |
427 | ||
428 | if (!pidref_is_set(pidref)) | |
429 | return -ESRCH; | |
430 | ||
431 | if (pidref_is_remote(pidref)) | |
432 | return -EREMOTE; | |
433 | ||
434 | if (pidref->pid == 1 || pidref_is_self(pidref)) | |
435 | return -ECHILD; | |
436 | ||
437 | siginfo_t si = {}; | |
438 | if (pidref->fd >= 0) | |
439 | r = RET_NERRNO(waitid(P_PIDFD, pidref->fd, &si, options)); | |
440 | else | |
441 | r = RET_NERRNO(waitid(P_PID, pidref->pid, &si, options)); | |
442 | if (r < 0) | |
443 | return r; | |
444 | ||
445 | if (ret) | |
446 | *ret = si; | |
447 | ||
448 | return 0; | |
449 | } | |
450 | ||
451 | int pidref_wait_for_terminate(PidRef *pidref, siginfo_t *ret) { | |
452 | int r; | |
453 | ||
454 | for (;;) { | |
455 | r = pidref_wait(pidref, ret, WEXITED); | |
456 | if (r != -EINTR) | |
457 | return r; | |
458 | } | |
459 | } | |
460 | ||
461 | bool pidref_is_automatic(const PidRef *pidref) { | |
462 | return pidref && pid_is_automatic(pidref->pid); | |
463 | } | |
464 | ||
465 | void pidref_hash_func(const PidRef *pidref, struct siphash *state) { | |
466 | siphash24_compress_typesafe(pidref->pid, state); | |
467 | } | |
468 | ||
469 | int pidref_compare_func(const PidRef *a, const PidRef *b) { | |
470 | int r; | |
471 | ||
472 | assert(a); | |
473 | assert(b); | |
474 | ||
475 | r = CMP(pidref_is_set(a), pidref_is_set(b)); | |
476 | if (r != 0) | |
477 | return r; | |
478 | ||
479 | r = CMP(pidref_is_automatic(a), pidref_is_automatic(b)); | |
480 | if (r != 0) | |
481 | return r; | |
482 | ||
483 | r = CMP(pidref_is_remote(a), pidref_is_remote(b)); | |
484 | if (r != 0) | |
485 | return r; | |
486 | ||
487 | r = CMP(a->pid, b->pid); | |
488 | if (r != 0) | |
489 | return r; | |
490 | ||
491 | if (a->fd_id != 0 && b->fd_id != 0) | |
492 | return CMP(a->fd_id, b->fd_id); | |
493 | ||
494 | return 0; | |
495 | } | |
496 | ||
497 | DEFINE_HASH_OPS(pidref_hash_ops, PidRef, pidref_hash_func, pidref_compare_func); | |
498 | ||
499 | DEFINE_HASH_OPS_WITH_KEY_DESTRUCTOR(pidref_hash_ops_free, | |
500 | PidRef, pidref_hash_func, pidref_compare_func, | |
501 | pidref_free); |