| 1 |
/* SPDX-License-Identifier: LGPL-2.1-or-later */ |
| 2 |
|
| 3 |
#include <poll.h> |
| 4 |
#include <sys/wait.h> |
| 5 |
#include <unistd.h> |
| 6 |
|
| 7 |
#include "alloc-util.h" |
| 8 |
#include "errno-util.h" |
| 9 |
#include "fd-util.h" |
| 10 |
#include "fiber-ops.h" |
| 11 |
#include "format-util.h" |
| 12 |
#include "hash-funcs.h" |
| 13 |
#include "io-util.h" |
| 14 |
#include "log.h" |
| 15 |
#include "parse-util.h" |
| 16 |
#include "pidfd-util.h" |
| 17 |
#include "pidref.h" |
| 18 |
#include "process-util.h" |
| 19 |
#include "siphash24.h" |
| 20 |
#include "time-util.h" |
| 21 |
|
| 22 |
int pidref_acquire_pidfd_id(PidRef *pidref) { |
| 23 |
int r; |
| 24 |
|
| 25 |
assert(pidref); |
| 26 |
|
| 27 |
if (!pidref_is_set(pidref)) |
| 28 |
return -ESRCH; |
| 29 |
|
| 30 |
if (pidref_is_remote(pidref)) |
| 31 |
return -EREMOTE; |
| 32 |
|
| 33 |
if (pidref->fd < 0) |
| 34 |
return -ENOMEDIUM; |
| 35 |
|
| 36 |
if (pidref->fd_id > 0) |
| 37 |
return 0; |
| 38 |
|
| 39 |
r = pidfd_get_inode_id(pidref->fd, &pidref->fd_id); |
| 40 |
if (r < 0) { |
| 41 |
if (!ERRNO_IS_NEG_NOT_SUPPORTED(r)) |
| 42 |
log_debug_errno(r, "Failed to get inode number of pidfd for pid " PID_FMT ": %m", |
| 43 |
pidref->pid); |
| 44 |
return r; |
| 45 |
} |
| 46 |
|
| 47 |
return 0; |
| 48 |
} |
| 49 |
|
| 50 |
bool pidref_equal(PidRef *a, PidRef *b) { |
| 51 |
|
| 52 |
/* If this is the very same structure, it definitely refers to the same process */ |
| 53 |
if (a == b) |
| 54 |
return true; |
| 55 |
|
| 56 |
if (!pidref_is_set(a)) |
| 57 |
return !pidref_is_set(b); |
| 58 |
|
| 59 |
if (!pidref_is_set(b)) |
| 60 |
return false; |
| 61 |
|
| 62 |
if (a->pid != b->pid) |
| 63 |
return false; |
| 64 |
|
| 65 |
if (pidref_is_remote(a)) { |
| 66 |
/* If one is remote and the other isn't, they are not the same */ |
| 67 |
if (!pidref_is_remote(b)) |
| 68 |
return false; |
| 69 |
|
| 70 |
/* If both are remote, compare fd IDs if we have both, otherwise don't bother, and cut things short */ |
| 71 |
if (a->fd_id == 0 || b->fd_id == 0) |
| 72 |
return true; |
| 73 |
} else { |
| 74 |
/* If the other side is remote, then this is not the same */ |
| 75 |
if (pidref_is_remote(b)) |
| 76 |
return false; |
| 77 |
|
| 78 |
/* PID1 cannot exit, hence it cannot change pidfs ids, hence no point in comparing them, we |
| 79 |
* can shortcut things */ |
| 80 |
if (a->pid == 1) |
| 81 |
return true; |
| 82 |
|
| 83 |
/* Try to compare pidfds using their inode numbers. This way we can ensure that we |
| 84 |
* don't spuriously consider two PidRefs equal if the pid has been reused once. Note |
| 85 |
* that we ignore all errors here, not only EOPNOTSUPP, as fstat() might fail due to |
| 86 |
* many reasons. */ |
| 87 |
if (pidref_acquire_pidfd_id(a) < 0 || pidref_acquire_pidfd_id(b) < 0) |
| 88 |
return true; |
| 89 |
} |
| 90 |
|
| 91 |
return a->fd_id == b->fd_id; |
| 92 |
} |
| 93 |
|
| 94 |
int pidref_set_pid(PidRef *pidref, pid_t pid) { |
| 95 |
uint64_t pidfdid = 0; |
| 96 |
int fd; |
| 97 |
|
| 98 |
assert(pidref); |
| 99 |
|
| 100 |
if (pid < 0) |
| 101 |
return -ESRCH; |
| 102 |
if (pid == 0) { |
| 103 |
pid = getpid_cached(); |
| 104 |
(void) pidfd_get_inode_id_self_cached(&pidfdid); |
| 105 |
} |
| 106 |
|
| 107 |
fd = pidfd_open(pid, 0); |
| 108 |
if (fd < 0) { |
| 109 |
/* Graceful fallback in case the kernel is out of fds */ |
| 110 |
if (!ERRNO_IS_RESOURCE(errno)) |
| 111 |
return log_debug_errno(errno, "Failed to open pidfd for pid " PID_FMT ": %m", pid); |
| 112 |
|
| 113 |
fd = -EBADF; |
| 114 |
} |
| 115 |
|
| 116 |
*pidref = (PidRef) { |
| 117 |
.fd = fd, |
| 118 |
.pid = pid, |
| 119 |
.fd_id = pidfdid, |
| 120 |
}; |
| 121 |
|
| 122 |
return 0; |
| 123 |
} |
| 124 |
|
| 125 |
int pidref_set_pid_and_pidfd_id( |
| 126 |
PidRef *pidref, |
| 127 |
pid_t pid, |
| 128 |
uint64_t pidfd_id) { |
| 129 |
|
| 130 |
int r; |
| 131 |
|
| 132 |
assert(pidref); |
| 133 |
|
| 134 |
_cleanup_(pidref_done) PidRef n = PIDREF_NULL; |
| 135 |
r = pidref_set_pid(&n, pid); |
| 136 |
if (r < 0) |
| 137 |
return r; |
| 138 |
|
| 139 |
if (pidfd_id > 0) { |
| 140 |
r = pidref_acquire_pidfd_id(&n); |
| 141 |
if (r < 0) |
| 142 |
return r; |
| 143 |
|
| 144 |
if (n.fd_id != pidfd_id) |
| 145 |
return -ESRCH; |
| 146 |
} |
| 147 |
|
| 148 |
*pidref = TAKE_PIDREF(n); |
| 149 |
return 0; |
| 150 |
} |
| 151 |
|
| 152 |
int pidref_set_pidstr(PidRef *pidref, const char *pid) { |
| 153 |
pid_t nr; |
| 154 |
int r; |
| 155 |
|
| 156 |
assert(pidref); |
| 157 |
|
| 158 |
r = parse_pid(pid, &nr); |
| 159 |
if (r < 0) |
| 160 |
return r; |
| 161 |
|
| 162 |
return pidref_set_pid(pidref, nr); |
| 163 |
} |
| 164 |
|
| 165 |
int pidref_set_pidfd(PidRef *pidref, int fd) { |
| 166 |
int r; |
| 167 |
|
| 168 |
assert(pidref); |
| 169 |
|
| 170 |
if (fd < 0) |
| 171 |
return -EBADF; |
| 172 |
|
| 173 |
int fd_copy = fcntl(fd, F_DUPFD_CLOEXEC, 3); |
| 174 |
if (fd_copy < 0) { |
| 175 |
pid_t pid; |
| 176 |
|
| 177 |
if (!ERRNO_IS_RESOURCE(errno)) |
| 178 |
return -errno; |
| 179 |
|
| 180 |
/* Graceful fallback if we are out of fds */ |
| 181 |
r = pidfd_get_pid(fd, &pid); |
| 182 |
if (r < 0) |
| 183 |
return r; |
| 184 |
|
| 185 |
*pidref = PIDREF_MAKE_FROM_PID(pid); |
| 186 |
return 0; |
| 187 |
} |
| 188 |
|
| 189 |
return pidref_set_pidfd_consume(pidref, fd_copy); |
| 190 |
} |
| 191 |
|
| 192 |
int pidref_set_pidfd_take(PidRef *pidref, int fd) { |
| 193 |
pid_t pid; |
| 194 |
int r; |
| 195 |
|
| 196 |
assert(pidref); |
| 197 |
|
| 198 |
if (fd < 0) |
| 199 |
return -EBADF; |
| 200 |
|
| 201 |
r = pidfd_get_pid(fd, &pid); |
| 202 |
if (r < 0) |
| 203 |
return r; |
| 204 |
|
| 205 |
*pidref = (PidRef) { |
| 206 |
.fd = fd, |
| 207 |
.pid = pid, |
| 208 |
}; |
| 209 |
|
| 210 |
return 0; |
| 211 |
} |
| 212 |
|
| 213 |
int pidref_set_pidfd_consume(PidRef *pidref, int fd) { |
| 214 |
int r; |
| 215 |
|
| 216 |
r = pidref_set_pidfd_take(pidref, fd); |
| 217 |
if (r < 0) |
| 218 |
safe_close(fd); |
| 219 |
|
| 220 |
return r; |
| 221 |
} |
| 222 |
|
| 223 |
int pidref_set_parent(PidRef *ret) { |
| 224 |
_cleanup_(pidref_done) PidRef parent = PIDREF_NULL; |
| 225 |
pid_t ppid; |
| 226 |
int r; |
| 227 |
|
| 228 |
assert(ret); |
| 229 |
|
| 230 |
/* Acquires a pidref to our parent process. Deals with the fact that parent processes might exit, and |
| 231 |
* we get reparented to other processes, with our old parent's PID already being recycled. */ |
| 232 |
|
| 233 |
ppid = getppid(); |
| 234 |
for (;;) { |
| 235 |
r = pidref_set_pid(&parent, ppid); |
| 236 |
if (r < 0) |
| 237 |
return r; |
| 238 |
|
| 239 |
if (parent.fd < 0) /* If pidfds are not available, then we are done */ |
| 240 |
break; |
| 241 |
|
| 242 |
pid_t now_ppid = getppid(); |
| 243 |
if (now_ppid == ppid) /* If our ppid is still the same, then we are done */ |
| 244 |
break; |
| 245 |
|
| 246 |
/* Otherwise let's try again with the new ppid */ |
| 247 |
ppid = now_ppid; |
| 248 |
pidref_done(&parent); |
| 249 |
} |
| 250 |
|
| 251 |
*ret = TAKE_PIDREF(parent); |
| 252 |
return 0; |
| 253 |
} |
| 254 |
|
| 255 |
void pidref_done(PidRef *pidref) { |
| 256 |
assert(pidref); |
| 257 |
|
| 258 |
*pidref = (PidRef) { |
| 259 |
.fd = safe_close(pidref->fd), |
| 260 |
}; |
| 261 |
} |
| 262 |
|
| 263 |
PidRef* pidref_free(PidRef *pidref) { |
| 264 |
/* Regularly, this is an embedded structure. But sometimes we want it on the heap too */ |
| 265 |
if (!pidref) |
| 266 |
return NULL; |
| 267 |
|
| 268 |
pidref_done(pidref); |
| 269 |
return mfree(pidref); |
| 270 |
} |
| 271 |
|
| 272 |
int pidref_copy(const PidRef *pidref, PidRef *ret) { |
| 273 |
_cleanup_(pidref_done) PidRef copy = PIDREF_NULL; |
| 274 |
|
| 275 |
/* If NULL is passed we'll generate a PidRef that refers to no process. This makes it easy to |
| 276 |
* copy pidref fields that might or might not reference a process yet. */ |
| 277 |
|
| 278 |
assert(ret); |
| 279 |
|
| 280 |
if (pidref) { |
| 281 |
if (pidref_is_remote(pidref)) /* Propagate remote flag */ |
| 282 |
copy.fd = -EREMOTE; |
| 283 |
else if (pidref->fd >= 0) { |
| 284 |
copy.fd = fcntl(pidref->fd, F_DUPFD_CLOEXEC, 3); |
| 285 |
if (copy.fd < 0) { |
| 286 |
if (!ERRNO_IS_RESOURCE(errno)) |
| 287 |
return -errno; |
| 288 |
|
| 289 |
copy.fd = -EBADF; |
| 290 |
} |
| 291 |
} |
| 292 |
|
| 293 |
copy.pid = pidref->pid; |
| 294 |
copy.fd_id = pidref->fd_id; |
| 295 |
} |
| 296 |
|
| 297 |
*ret = TAKE_PIDREF(copy); |
| 298 |
return 0; |
| 299 |
} |
| 300 |
|
| 301 |
int pidref_dup(const PidRef *pidref, PidRef **ret) { |
| 302 |
_cleanup_(pidref_freep) PidRef *dup_pidref = NULL; |
| 303 |
int r; |
| 304 |
|
| 305 |
/* Allocates a new PidRef on the heap, making it a copy of the specified pidref. This does not try to |
| 306 |
* acquire a pidfd if we don't have one yet! */ |
| 307 |
|
| 308 |
assert(ret); |
| 309 |
|
| 310 |
dup_pidref = newdup(PidRef, &PIDREF_NULL, 1); |
| 311 |
if (!dup_pidref) |
| 312 |
return -ENOMEM; |
| 313 |
|
| 314 |
r = pidref_copy(pidref, dup_pidref); |
| 315 |
if (r < 0) |
| 316 |
return r; |
| 317 |
|
| 318 |
*ret = TAKE_PTR(dup_pidref); |
| 319 |
return 0; |
| 320 |
} |
| 321 |
|
| 322 |
int pidref_new_from_pid(pid_t pid, PidRef **ret) { |
| 323 |
_cleanup_(pidref_freep) PidRef *n = NULL; |
| 324 |
int r; |
| 325 |
|
| 326 |
assert(ret); |
| 327 |
|
| 328 |
if (pid < 0) |
| 329 |
return -ESRCH; |
| 330 |
|
| 331 |
n = new(PidRef, 1); |
| 332 |
if (!n) |
| 333 |
return -ENOMEM; |
| 334 |
|
| 335 |
*n = PIDREF_NULL; |
| 336 |
|
| 337 |
r = pidref_set_pid(n, pid); |
| 338 |
if (r < 0) |
| 339 |
return r; |
| 340 |
|
| 341 |
*ret = TAKE_PTR(n); |
| 342 |
return 0; |
| 343 |
} |
| 344 |
|
| 345 |
int pidref_kill(const PidRef *pidref, int sig) { |
| 346 |
|
| 347 |
if (!pidref) |
| 348 |
return -ESRCH; |
| 349 |
|
| 350 |
if (pidref_is_remote(pidref)) |
| 351 |
return -EREMOTE; |
| 352 |
|
| 353 |
if (pidref->fd >= 0) |
| 354 |
return RET_NERRNO(pidfd_send_signal(pidref->fd, sig, NULL, 0)); |
| 355 |
|
| 356 |
if (pidref->pid > 0) |
| 357 |
return RET_NERRNO(kill(pidref->pid, sig)); |
| 358 |
|
| 359 |
return -ESRCH; |
| 360 |
} |
| 361 |
|
| 362 |
int pidref_kill_and_sigcont(const PidRef *pidref, int sig) { |
| 363 |
int r; |
| 364 |
|
| 365 |
r = pidref_kill(pidref, sig); |
| 366 |
if (r < 0) |
| 367 |
return r; |
| 368 |
|
| 369 |
if (!IN_SET(sig, SIGCONT, SIGKILL)) |
| 370 |
(void) pidref_kill(pidref, SIGCONT); |
| 371 |
|
| 372 |
return 0; |
| 373 |
} |
| 374 |
|
| 375 |
int pidref_sigqueue(const PidRef *pidref, int sig, int value) { |
| 376 |
|
| 377 |
if (!pidref) |
| 378 |
return -ESRCH; |
| 379 |
|
| 380 |
if (pidref_is_remote(pidref)) |
| 381 |
return -EREMOTE; |
| 382 |
|
| 383 |
if (pidref->fd >= 0) { |
| 384 |
siginfo_t si; |
| 385 |
|
| 386 |
/* We can't use structured initialization here, since the structure contains various unions |
| 387 |
* and these fields lie in overlapping (carefully aligned) unions that LLVM is allergic to |
| 388 |
* allow assignments to */ |
| 389 |
zero(si); |
| 390 |
si.si_signo = sig; |
| 391 |
si.si_code = SI_QUEUE; |
| 392 |
si.si_pid = getpid_cached(); |
| 393 |
si.si_uid = getuid(); |
| 394 |
si.si_value.sival_int = value; |
| 395 |
|
| 396 |
return RET_NERRNO(pidfd_send_signal(pidref->fd, sig, &si, 0)); |
| 397 |
} |
| 398 |
|
| 399 |
if (pidref->pid > 0) |
| 400 |
return RET_NERRNO(sigqueue(pidref->pid, sig, (const union sigval) { .sival_int = value })); |
| 401 |
|
| 402 |
return -ESRCH; |
| 403 |
} |
| 404 |
|
| 405 |
int pidref_verify(const PidRef *pidref) { |
| 406 |
int r; |
| 407 |
|
| 408 |
/* This is a helper that is supposed to be called after reading information from procfs via a |
| 409 |
* PidRef. It ensures that the PID we track still matches the PIDFD we pin. If this value differs |
| 410 |
* after a procfs read, we might have read the data from a recycled PID. */ |
| 411 |
|
| 412 |
if (!pidref_is_set(pidref)) |
| 413 |
return -ESRCH; |
| 414 |
|
| 415 |
if (pidref_is_remote(pidref)) |
| 416 |
return -EREMOTE; |
| 417 |
|
| 418 |
if (pidref->pid == 1) |
| 419 |
return 1; /* PID 1 can never go away, hence never be recycled to a different process → return 1 */ |
| 420 |
|
| 421 |
if (pidref->fd < 0) |
| 422 |
return 0; /* If we don't have a pidfd we cannot validate it, hence we assume it's all OK → return 0 */ |
| 423 |
|
| 424 |
r = pidfd_verify_pid(pidref->fd, pidref->pid); |
| 425 |
if (r < 0) |
| 426 |
return r; |
| 427 |
|
| 428 |
return 1; /* We have a pidfd and it still points to the PID we have, hence all is *really* OK → return 1 */ |
| 429 |
} |
| 430 |
|
| 431 |
bool pidref_is_self(PidRef *pidref) { |
| 432 |
if (!pidref_is_set(pidref)) |
| 433 |
return false; |
| 434 |
|
| 435 |
if (pidref_is_remote(pidref)) |
| 436 |
return false; |
| 437 |
|
| 438 |
if (pidref->pid != getpid_cached()) |
| 439 |
return false; |
| 440 |
|
| 441 |
/* PID1 cannot exit, hence no point in comparing pidfd IDs, they can never change */ |
| 442 |
if (pidref->pid == 1) |
| 443 |
return true; |
| 444 |
|
| 445 |
/* Also compare pidfd ID if we can get it */ |
| 446 |
if (pidref_acquire_pidfd_id(pidref) < 0) |
| 447 |
return true; |
| 448 |
|
| 449 |
uint64_t self_id; |
| 450 |
if (pidfd_get_inode_id_self_cached(&self_id) < 0) |
| 451 |
return true; |
| 452 |
|
| 453 |
return pidref->fd_id == self_id; |
| 454 |
} |
| 455 |
|
| 456 |
int pidref_wait_for_terminate_full(PidRef *pidref, usec_t timeout, siginfo_t *ret_si) { |
| 457 |
int r; |
| 458 |
|
| 459 |
assert(timeout > 0); |
| 460 |
|
| 461 |
if (!pidref_is_set(pidref)) |
| 462 |
return -ESRCH; |
| 463 |
|
| 464 |
if (pidref_is_remote(pidref)) |
| 465 |
return -EREMOTE; |
| 466 |
|
| 467 |
if (pidref->pid == 1 || pidref_is_self(pidref)) |
| 468 |
return -ECHILD; |
| 469 |
|
| 470 |
if (pidref->fd < 0 && (timeout != USEC_INFINITY || fiber_ops_is_set())) |
| 471 |
return -ENOMEDIUM; |
| 472 |
|
| 473 |
usec_t ts = timeout == USEC_INFINITY ? USEC_INFINITY : usec_add(now(CLOCK_MONOTONIC), timeout); |
| 474 |
|
| 475 |
/* Poll the pidfd before waitid() if either there's a finite timeout (so we can honor it) or |
| 476 |
* we're on a fiber (so fd_wait_for_event() can suspend us instead of blocking the event loop |
| 477 |
* inside waitid()). Otherwise let waitid() block directly. The precondition above guarantees |
| 478 |
* pidref->fd >= 0 in both cases. */ |
| 479 |
bool poll_first = ts != USEC_INFINITY || fiber_ops_is_set(); |
| 480 |
|
| 481 |
for (;;) { |
| 482 |
if (poll_first) { |
| 483 |
usec_t left; |
| 484 |
|
| 485 |
if (ts == USEC_INFINITY) |
| 486 |
left = USEC_INFINITY; |
| 487 |
else { |
| 488 |
left = usec_sub_unsigned(ts, now(CLOCK_MONOTONIC)); |
| 489 |
if (left == 0) |
| 490 |
return -ETIMEDOUT; |
| 491 |
} |
| 492 |
|
| 493 |
r = fd_wait_for_event(pidref->fd, POLLIN, left); |
| 494 |
if (r == 0) |
| 495 |
return -ETIMEDOUT; |
| 496 |
if (r == -EINTR) |
| 497 |
continue; |
| 498 |
if (r < 0) |
| 499 |
return r; |
| 500 |
} |
| 501 |
|
| 502 |
siginfo_t si = {}; |
| 503 |
|
| 504 |
if (pidref->fd >= 0) |
| 505 |
r = RET_NERRNO(waitid(P_PIDFD, pidref->fd, &si, WEXITED)); |
| 506 |
else |
| 507 |
r = RET_NERRNO(waitid(P_PID, pidref->pid, &si, WEXITED)); |
| 508 |
if (r >= 0) { |
| 509 |
if (ret_si) |
| 510 |
*ret_si = si; |
| 511 |
return 0; |
| 512 |
} |
| 513 |
if (r != -EINTR) |
| 514 |
return r; |
| 515 |
} |
| 516 |
} |
| 517 |
|
| 518 |
bool pidref_is_automatic(const PidRef *pidref) { |
| 519 |
return pidref && pid_is_automatic(pidref->pid); |
| 520 |
} |
| 521 |
|
| 522 |
void pidref_hash_func(const PidRef *pidref, struct siphash *state) { |
| 523 |
siphash24_compress_typesafe(pidref->pid, state); |
| 524 |
} |
| 525 |
|
| 526 |
int pidref_compare_func(const PidRef *a, const PidRef *b) { |
| 527 |
int r; |
| 528 |
|
| 529 |
assert(a); |
| 530 |
assert(b); |
| 531 |
|
| 532 |
r = CMP(pidref_is_set(a), pidref_is_set(b)); |
| 533 |
if (r != 0) |
| 534 |
return r; |
| 535 |
|
| 536 |
r = CMP(pidref_is_automatic(a), pidref_is_automatic(b)); |
| 537 |
if (r != 0) |
| 538 |
return r; |
| 539 |
|
| 540 |
r = CMP(pidref_is_remote(a), pidref_is_remote(b)); |
| 541 |
if (r != 0) |
| 542 |
return r; |
| 543 |
|
| 544 |
r = CMP(a->pid, b->pid); |
| 545 |
if (r != 0) |
| 546 |
return r; |
| 547 |
|
| 548 |
if (a->fd_id != 0 && b->fd_id != 0) |
| 549 |
return CMP(a->fd_id, b->fd_id); |
| 550 |
|
| 551 |
return 0; |
| 552 |
} |
| 553 |
|
| 554 |
DEFINE_HASH_OPS(pidref_hash_ops, PidRef, pidref_hash_func, pidref_compare_func); |
| 555 |
|
| 556 |
DEFINE_HASH_OPS_WITH_KEY_DESTRUCTOR(pidref_hash_ops_free, |
| 557 |
PidRef, pidref_hash_func, pidref_compare_func, |
| 558 |
pidref_free); |