src/basic/process-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <linux/oom.h>
   4 #include <pthread.h>
   5 #include <spawn.h>
   6 #include <stdio.h>
   7 #include <sys/mount.h>
   8 #include <sys/personality.h>
   9 #include <sys/prctl.h>
  10 #include <sys/wait.h>
  11 #include <syslog.h>
  12 #include <threads.h>
  13 #include <unistd.h>
  14 #if HAVE_VALGRIND_VALGRIND_H
  15 #include <valgrind/valgrind.h>
  16 #endif
  17
  18 #include "sd-messages.h"
  19
  20 #include "alloc-util.h"
  21 #include "architecture.h"
  22 #include "argv-util.h"
  23 #include "cgroup-util.h"
  24 #include "dirent-util.h"
  25 #include "env-file.h"
  26 #include "errno-util.h"
  27 #include "escape.h"
  28 #include "fd-util.h"
  29 #include "fileio.h"
  30 #include "fs-util.h"
  31 #include "hostname-util.h"
  32 #include "io-util.h"
  33 #include "iovec-util.h"
  34 #include "locale-util.h"
  35 #include "log.h"
  36 #include "memory-util.h"
  37 #include "mountpoint-util.h"
  38 #include "namespace-util.h"
  39 #include "nulstr-util.h"
  40 #include "parse-util.h"
  41 #include "path-util.h"
  42 #include "pidfd-util.h"
  43 #include "pidref.h"
  44 #include "process-util.h"
  45 #include "raw-clone.h"
  46 #include "rlimit-util.h"
  47 #include "signal-util.h"
  48 #include "socket-util.h"
  49 #include "stat-util.h"
  50 #include "stdio-util.h"
  51 #include "string-table.h"
  52 #include "string-util.h"
  53 #include "time-util.h"
  54 #include "user-util.h"
  55
  56 /* The kernel limits userspace processes to TASK_COMM_LEN (16 bytes), but allows higher values for its own
  57  * workers, e.g. "kworker/u9:3-kcryptd/253:0". Let's pick a fixed smallish limit that will work for the kernel.
  58  */
  59 #define COMM_MAX_LEN 128
  60
  61 static int get_process_state(pid_t pid) {
  62         _cleanup_free_ char *line = NULL;
  63         const char *p;
  64         char state;
  65         int r;
  66
  67         assert(pid >= 0);
  68
  69         /* Shortcut: if we are enquired about our own state, we are obviously running */
  70         if (pid == 0 || pid == getpid_cached())
  71                 return (unsigned char) 'R';
  72
  73         p = procfs_file_alloca(pid, "stat");
  74
  75         r = read_one_line_file(p, &line);
  76         if (r == -ENOENT)
  77                 return -ESRCH;
  78         if (r < 0)
  79                 return r;
  80
  81         p = strrchr(line, ')');
  82         if (!p)
  83                 return -EIO;
  84
  85         p++;
  86
  87         if (sscanf(p, " %c", &state) != 1)
  88                 return -EIO;
  89
  90         return (unsigned char) state;
  91 }
  92
  93 int pid_get_comm(pid_t pid, char **ret) {
  94         _cleanup_free_ char *escaped = NULL, *comm = NULL;
  95         int r;
  96
  97         assert(pid >= 0);
  98         assert(ret);
  99
 100         if (pid == 0 || pid == getpid_cached()) {
 101                 comm = new0(char, TASK_COMM_LEN + 1); /* Must fit in 16 byte according to prctl(2) */
 102                 if (!comm)
 103                         return -ENOMEM;
 104
 105                 if (prctl(PR_GET_NAME, comm) < 0)
 106                         return -errno;
 107         } else {
 108                 const char *p;
 109
 110                 p = procfs_file_alloca(pid, "comm");
 111
 112                 /* Note that process names of kernel threads can be much longer than TASK_COMM_LEN */
 113                 r = read_one_line_file(p, &comm);
 114                 if (r == -ENOENT)
 115                         return -ESRCH;
 116                 if (r < 0)
 117                         return r;
 118         }
 119
 120         escaped = new(char, COMM_MAX_LEN);
 121         if (!escaped)
 122                 return -ENOMEM;
 123
 124         /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */
 125         cellescape(escaped, COMM_MAX_LEN, comm);
 126
 127         *ret = TAKE_PTR(escaped);
 128         return 0;
 129 }
 130
 131 int pidref_get_comm(const PidRef *pid, char **ret) {
 132         _cleanup_free_ char *comm = NULL;
 133         int r;
 134
 135         if (!pidref_is_set(pid))
 136                 return -ESRCH;
 137
 138         if (pidref_is_remote(pid))
 139                 return -EREMOTE;
 140
 141         r = pid_get_comm(pid->pid, &comm);
 142         if (r < 0)
 143                 return r;
 144
 145         r = pidref_verify(pid);
 146         if (r < 0)
 147                 return r;
 148
 149         if (ret)
 150                 *ret = TAKE_PTR(comm);
 151         return 0;
 152 }
 153
 154 static int pid_get_cmdline_nulstr(
 155                 pid_t pid,
 156                 size_t max_size,
 157                 ProcessCmdlineFlags flags,
 158                 char **ret,
 159                 size_t *ret_size) {
 160
 161         _cleanup_free_ char *t = NULL;
 162         const char *p;
 163         size_t k;
 164         int r;
 165
 166         /* Retrieves a process' command line as a "sized nulstr", i.e. possibly without the last NUL, but
 167          * with a specified size.
 168          *
 169          * If PROCESS_CMDLINE_COMM_FALLBACK is specified in flags and the process has no command line set
 170          * (the case for kernel threads), or has a command line that resolves to the empty string, will
 171          * return the "comm" name of the process instead. This will use at most _SC_ARG_MAX bytes of input
 172          * data.
 173          *
 174          * Returns an error, 0 if output was read but is truncated, 1 otherwise.
 175          */
 176
 177         p = procfs_file_alloca(pid, "cmdline");
 178         r = read_virtual_file(p, max_size, &t, &k); /* Let's assume that each input byte results in >= 1
 179                                                      * columns of output. We ignore zero-width codepoints. */
 180         if (r == -ENOENT)
 181                 return -ESRCH;
 182         if (r < 0)
 183                 return r;
 184
 185         if (k == 0) {
 186                 if (!(flags & PROCESS_CMDLINE_COMM_FALLBACK))
 187                         return -ENOENT;
 188
 189                 /* Kernel threads have no argv[] */
 190                 _cleanup_free_ char *comm = NULL;
 191
 192                 r = pid_get_comm(pid, &comm);
 193                 if (r < 0)
 194                         return r;
 195
 196                 free(t);
 197                 t = strjoin("[", comm, "]");
 198                 if (!t)
 199                         return -ENOMEM;
 200
 201                 k = strlen(t);
 202                 r = k <= max_size;
 203                 if (r == 0) /* truncation */
 204                         t[max_size] = '\0';
 205         }
 206
 207         if (ret)
 208                 *ret = TAKE_PTR(t);
 209         if (ret_size)
 210                 *ret_size = k;
 211
 212         return r;
 213 }
 214
 215 int pid_get_cmdline(pid_t pid, size_t max_columns, ProcessCmdlineFlags flags, char **ret) {
 216         _cleanup_free_ char *t = NULL;
 217         size_t k;
 218         char *ans;
 219
 220         assert(pid >= 0);
 221         assert(ret);
 222
 223         /* Retrieve and format a command line. See above for discussion of retrieval options.
 224          *
 225          * There are two main formatting modes:
 226          *
 227          * - when PROCESS_CMDLINE_QUOTE is specified, output is quoted in C/Python style. If no shell special
 228          *   characters are present, this output can be copy-pasted into the terminal to execute. UTF-8
 229          *   output is assumed.
 230          *
 231          * - otherwise, a compact non-roundtrippable form is returned. Non-UTF8 bytes are replaced by �. The
 232          *   returned string is of the specified console width at most, abbreviated with an ellipsis.
 233          *
 234          * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
 235          * PROCESS_CMDLINE_COMM_FALLBACK is not specified). Returns 0 and sets *line otherwise. */
 236
 237         int full = pid_get_cmdline_nulstr(pid, max_columns, flags, &t, &k);
 238         if (full < 0)
 239                 return full;
 240
 241         if (flags & (PROCESS_CMDLINE_QUOTE | PROCESS_CMDLINE_QUOTE_POSIX)) {
 242                 ShellEscapeFlags shflags = SHELL_ESCAPE_EMPTY |
 243                         FLAGS_SET(flags, PROCESS_CMDLINE_QUOTE_POSIX) * SHELL_ESCAPE_POSIX;
 244
 245                 assert(!(flags & PROCESS_CMDLINE_USE_LOCALE));
 246
 247                 _cleanup_strv_free_ char **args = NULL;
 248
 249                 /* Drop trailing NULs, otherwise strv_parse_nulstr() adds additional empty strings at the end.
 250                  * See also issue #21186. */
 251                 args = strv_parse_nulstr_full(t, k, /* drop_trailing_nuls = */ true);
 252                 if (!args)
 253                         return -ENOMEM;
 254
 255                 ans = quote_command_line(args, shflags);
 256                 if (!ans)
 257                         return -ENOMEM;
 258         } else {
 259                 /* Arguments are separated by NULs. Let's replace those with spaces. */
 260                 for (size_t i = 0; i < k - 1; i++)
 261                         if (t[i] == '\0')
 262                                 t[i] = ' ';
 263
 264                 delete_trailing_chars(t, WHITESPACE);
 265
 266                 bool eight_bit = (flags & PROCESS_CMDLINE_USE_LOCALE) && !is_locale_utf8();
 267
 268                 ans = escape_non_printable_full(t, max_columns,
 269                                                 eight_bit * XESCAPE_8_BIT | !full * XESCAPE_FORCE_ELLIPSIS);
 270                 if (!ans)
 271                         return -ENOMEM;
 272
 273                 ans = str_realloc(ans);
 274         }
 275
 276         *ret = ans;
 277         return 0;
 278 }
 279
 280 int pidref_get_cmdline(const PidRef *pid, size_t max_columns, ProcessCmdlineFlags flags, char **ret) {
 281         _cleanup_free_ char *s = NULL;
 282         int r;
 283
 284         if (!pidref_is_set(pid))
 285                 return -ESRCH;
 286
 287         if (pidref_is_remote(pid))
 288                 return -EREMOTE;
 289
 290         r = pid_get_cmdline(pid->pid, max_columns, flags, &s);
 291         if (r < 0)
 292                 return r;
 293
 294         r = pidref_verify(pid);
 295         if (r < 0)
 296                 return r;
 297
 298         if (ret)
 299                 *ret = TAKE_PTR(s);
 300         return 0;
 301 }
 302
 303 int pid_get_cmdline_strv(pid_t pid, ProcessCmdlineFlags flags, char ***ret) {
 304         _cleanup_free_ char *t = NULL;
 305         char **args;
 306         size_t k;
 307         int r;
 308
 309         assert(pid >= 0);
 310         assert((flags & ~PROCESS_CMDLINE_COMM_FALLBACK) == 0);
 311         assert(ret);
 312
 313         r = pid_get_cmdline_nulstr(pid, SIZE_MAX, flags, &t, &k);
 314         if (r < 0)
 315                 return r;
 316
 317         args = strv_parse_nulstr_full(t, k, /* drop_trailing_nuls = */ true);
 318         if (!args)
 319                 return -ENOMEM;
 320
 321         *ret = args;
 322         return 0;
 323 }
 324
 325 int pidref_get_cmdline_strv(const PidRef *pid, ProcessCmdlineFlags flags, char ***ret) {
 326         _cleanup_strv_free_ char **args = NULL;
 327         int r;
 328
 329         if (!pidref_is_set(pid))
 330                 return -ESRCH;
 331
 332         if (pidref_is_remote(pid))
 333                 return -EREMOTE;
 334
 335         r = pid_get_cmdline_strv(pid->pid, flags, &args);
 336         if (r < 0)
 337                 return r;
 338
 339         r = pidref_verify(pid);
 340         if (r < 0)
 341                 return r;
 342
 343         if (ret)
 344                 *ret = TAKE_PTR(args);
 345
 346         return 0;
 347 }
 348
 349 int container_get_leader(const char *machine, pid_t *pid) {
 350         _cleanup_free_ char *s = NULL, *class = NULL;
 351         const char *p;
 352         pid_t leader;
 353         int r;
 354
 355         assert(machine);
 356         assert(pid);
 357
 358         if (streq(machine, ".host")) {
 359                 *pid = 1;
 360                 return 0;
 361         }
 362
 363         if (!hostname_is_valid(machine, 0))
 364                 return -EINVAL;
 365
 366         p = strjoina("/run/systemd/machines/", machine);
 367         r = parse_env_file(NULL, p,
 368                            "LEADER", &s,
 369                            "CLASS", &class);
 370         if (r == -ENOENT)
 371                 return -EHOSTDOWN;
 372         if (r < 0)
 373                 return r;
 374         if (!s)
 375                 return -EIO;
 376
 377         if (!streq_ptr(class, "container"))
 378                 return -EIO;
 379
 380         r = parse_pid(s, &leader);
 381         if (r < 0)
 382                 return r;
 383         if (leader <= 1)
 384                 return -EIO;
 385
 386         *pid = leader;
 387         return 0;
 388 }
 389
 390 int pid_is_kernel_thread(pid_t pid) {
 391         _cleanup_free_ char *line = NULL;
 392         unsigned long long flags;
 393         size_t l, i;
 394         const char *p;
 395         char *q;
 396         int r;
 397
 398         if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
 399                 return 0;
 400         if (!pid_is_valid(pid))
 401                 return -EINVAL;
 402
 403         p = procfs_file_alloca(pid, "stat");
 404         r = read_one_line_file(p, &line);
 405         if (r == -ENOENT)
 406                 return -ESRCH;
 407         if (r < 0)
 408                 return r;
 409
 410         /* Skip past the comm field */
 411         q = strrchr(line, ')');
 412         if (!q)
 413                 return -EINVAL;
 414         q++;
 415
 416         /* Skip 6 fields to reach the flags field */
 417         for (i = 0; i < 6; i++) {
 418                 l = strspn(q, WHITESPACE);
 419                 if (l < 1)
 420                         return -EINVAL;
 421                 q += l;
 422
 423                 l = strcspn(q, WHITESPACE);
 424                 if (l < 1)
 425                         return -EINVAL;
 426                 q += l;
 427         }
 428
 429         /* Skip preceding whitespace */
 430         l = strspn(q, WHITESPACE);
 431         if (l < 1)
 432                 return -EINVAL;
 433         q += l;
 434
 435         /* Truncate the rest */
 436         l = strcspn(q, WHITESPACE);
 437         if (l < 1)
 438                 return -EINVAL;
 439         q[l] = 0;
 440
 441         r = safe_atollu(q, &flags);
 442         if (r < 0)
 443                 return r;
 444
 445         return !!(flags & PF_KTHREAD);
 446 }
 447
 448 int pidref_is_kernel_thread(const PidRef *pid) {
 449         int result, r;
 450
 451         if (!pidref_is_set(pid))
 452                 return -ESRCH;
 453
 454         if (pidref_is_remote(pid))
 455                 return -EREMOTE;
 456
 457         result = pid_is_kernel_thread(pid->pid);
 458         if (result < 0)
 459                 return result;
 460
 461         r = pidref_verify(pid); /* Verify that the PID wasn't reused since */
 462         if (r < 0)
 463                 return r;
 464
 465         return result;
 466 }
 467
 468 static int get_process_link_contents(pid_t pid, const char *proc_file, char **ret) {
 469         const char *p;
 470         int r;
 471
 472         assert(proc_file);
 473
 474         p = procfs_file_alloca(pid, proc_file);
 475
 476         r = readlink_malloc(p, ret);
 477         return (r == -ENOENT && proc_mounted() > 0) ? -ESRCH : r;
 478 }
 479
 480 int get_process_exe(pid_t pid, char **ret) {
 481         char *d;
 482         int r;
 483
 484         assert(pid >= 0);
 485
 486         r = get_process_link_contents(pid, "exe", ret);
 487         if (r < 0)
 488                 return r;
 489
 490         if (ret) {
 491                 d = endswith(*ret, " (deleted)");
 492                 if (d)
 493                         *d = '\0';
 494         }
 495
 496         return 0;
 497 }
 498
 499 int pid_get_uid(pid_t pid, uid_t *ret) {
 500         int r;
 501
 502         assert(pid >= 0);
 503         assert(ret);
 504
 505         if (pid == 0 || pid == getpid_cached()) {
 506                 *ret = getuid();
 507                 return 0;
 508         }
 509
 510         _cleanup_free_ char *v = NULL;
 511         r = procfs_file_get_field(pid, "status", "Uid", &v);
 512         if (r == -ENOENT)
 513                 return -ESRCH;
 514         if (r < 0)
 515                 return r;
 516
 517         return parse_uid(v, ret);
 518 }
 519
 520 int pidref_get_uid(const PidRef *pid, uid_t *ret) {
 521         int r;
 522
 523         if (!pidref_is_set(pid))
 524                 return -ESRCH;
 525
 526         if (pidref_is_remote(pid))
 527                 return -EREMOTE;
 528
 529         if (pid->fd >= 0) {
 530                 r = pidfd_get_uid(pid->fd, ret);
 531                 if (!ERRNO_IS_NEG_NOT_SUPPORTED(r))
 532                         return r;
 533         }
 534
 535         uid_t uid;
 536         r = pid_get_uid(pid->pid, &uid);
 537         if (r < 0)
 538                 return r;
 539
 540         r = pidref_verify(pid);
 541         if (r < 0)
 542                 return r;
 543
 544         if (ret)
 545                 *ret = uid;
 546         return 0;
 547 }
 548
 549 int get_process_gid(pid_t pid, gid_t *ret) {
 550         int r;
 551
 552         assert(pid >= 0);
 553         assert(ret);
 554
 555         if (pid == 0 || pid == getpid_cached()) {
 556                 *ret = getgid();
 557                 return 0;
 558         }
 559
 560         _cleanup_free_ char *v = NULL;
 561         r = procfs_file_get_field(pid, "status", "Gid", &v);
 562         if (r == -ENOENT)
 563                 return -ESRCH;
 564         if (r < 0)
 565                 return r;
 566
 567         return parse_gid(v, ret);
 568 }
 569
 570 int get_process_cwd(pid_t pid, char **ret) {
 571         assert(pid >= 0);
 572
 573         if (pid == 0 || pid == getpid_cached())
 574                 return safe_getcwd(ret);
 575
 576         return get_process_link_contents(pid, "cwd", ret);
 577 }
 578
 579 int get_process_root(pid_t pid, char **ret) {
 580         assert(pid >= 0);
 581         return get_process_link_contents(pid, "root", ret);
 582 }
 583
 584 #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U)
 585
 586 int get_process_environ(pid_t pid, char **ret) {
 587         _cleanup_fclose_ FILE *f = NULL;
 588         _cleanup_free_ char *outcome = NULL;
 589         size_t sz = 0;
 590         const char *p;
 591         int r;
 592
 593         assert(pid >= 0);
 594         assert(ret);
 595
 596         p = procfs_file_alloca(pid, "environ");
 597
 598         r = fopen_unlocked(p, "re", &f);
 599         if (r == -ENOENT)
 600                 return -ESRCH;
 601         if (r < 0)
 602                 return r;
 603
 604         for (;;) {
 605                 char c;
 606
 607                 if (sz >= ENVIRONMENT_BLOCK_MAX)
 608                         return -ENOBUFS;
 609
 610                 if (!GREEDY_REALLOC(outcome, sz + 5))
 611                         return -ENOMEM;
 612
 613                 r = safe_fgetc(f, &c);
 614                 if (r < 0)
 615                         return r;
 616                 if (r == 0)
 617                         break;
 618
 619                 if (c == '\0')
 620                         outcome[sz++] = '\n';
 621                 else
 622                         sz += cescape_char(c, outcome + sz);
 623         }
 624
 625         outcome[sz] = '\0';
 626         *ret = TAKE_PTR(outcome);
 627
 628         return 0;
 629 }
 630
 631 int pid_get_ppid(pid_t pid, pid_t *ret) {
 632         _cleanup_free_ char *line = NULL;
 633         unsigned long ppid;
 634         const char *p;
 635         int r;
 636
 637         assert(pid >= 0);
 638
 639         if (pid == 0)
 640                 pid = getpid_cached();
 641         if (pid == 1) /* PID 1 has no parent, shortcut this case */
 642                 return -EADDRNOTAVAIL;
 643
 644         if (pid == getpid_cached()) {
 645                 if (ret)
 646                         *ret = getppid();
 647                 return 0;
 648         }
 649
 650         p = procfs_file_alloca(pid, "stat");
 651         r = read_one_line_file(p, &line);
 652         if (r == -ENOENT)
 653                 return -ESRCH;
 654         if (r < 0)
 655                 return r;
 656
 657         /* Let's skip the pid and comm fields. The latter is enclosed in () but does not escape any () in its
 658          * value, so let's skip over it manually */
 659
 660         p = strrchr(line, ')');
 661         if (!p)
 662                 return -EIO;
 663         p++;
 664
 665         if (sscanf(p, " "
 666                    "%*c "  /* state */
 667                    "%lu ", /* ppid */
 668                    &ppid) != 1)
 669                 return -EIO;
 670
 671         /* If ppid is zero the process has no parent. Which might be the case for PID 1 (caught above)
 672          * but also for processes originating in other namespaces that are inserted into a pidns.
 673          * Return a recognizable error in this case. */
 674         if (ppid == 0)
 675                 return -EADDRNOTAVAIL;
 676
 677         if ((pid_t) ppid < 0 || (unsigned long) (pid_t) ppid != ppid)
 678                 return -ERANGE;
 679
 680         if (ret)
 681                 *ret = (pid_t) ppid;
 682
 683         return 0;
 684 }
 685
 686 int pidref_get_ppid(const PidRef *pidref, pid_t *ret) {
 687         int r;
 688
 689         if (!pidref_is_set(pidref))
 690                 return -ESRCH;
 691
 692         if (pidref_is_remote(pidref))
 693                 return -EREMOTE;
 694
 695         if (pidref->fd >= 0) {
 696                 r = pidfd_get_ppid(pidref->fd, ret);
 697                 if (!ERRNO_IS_NEG_NOT_SUPPORTED(r))
 698                         return r;
 699         }
 700
 701         pid_t ppid;
 702         r = pid_get_ppid(pidref->pid, ret ? &ppid : NULL);
 703         if (r < 0)
 704                 return r;
 705
 706         r = pidref_verify(pidref);
 707         if (r < 0)
 708                 return r;
 709
 710         if (ret)
 711                 *ret = ppid;
 712         return 0;
 713 }
 714
 715 int pidref_get_ppid_as_pidref(const PidRef *pidref, PidRef *ret) {
 716         pid_t ppid;
 717         int r;
 718
 719         assert(ret);
 720
 721         r = pidref_get_ppid(pidref, &ppid);
 722         if (r < 0)
 723                 return r;
 724
 725         for (unsigned attempt = 0; attempt < 16; attempt++) {
 726                 _cleanup_(pidref_done) PidRef parent = PIDREF_NULL;
 727
 728                 r = pidref_set_pid(&parent, ppid);
 729                 if (r < 0)
 730                         return r;
 731
 732                 /* If we have a pidfd of the original PID, let's verify that the process we acquired really
 733                  * is the parent still */
 734                 if (pidref->fd >= 0) {
 735                         r = pidref_get_ppid(pidref, &ppid);
 736                         if (r < 0)
 737                                 return r;
 738
 739                         /* Did the PPID change since we queried it? if so we might have pinned the wrong
 740                          * process, if its PID got reused by now. Let's try again */
 741                         if (parent.pid != ppid)
 742                                 continue;
 743                 }
 744
 745                 *ret = TAKE_PIDREF(parent);
 746                 return 0;
 747         }
 748
 749         /* Give up after 16 tries */
 750         return -ENOTRECOVERABLE;
 751 }
 752
 753 int pid_get_start_time(pid_t pid, usec_t *ret) {
 754         _cleanup_free_ char *line = NULL;
 755         const char *p;
 756         int r;
 757
 758         assert(pid >= 0);
 759
 760         p = procfs_file_alloca(pid, "stat");
 761         r = read_one_line_file(p, &line);
 762         if (r == -ENOENT)
 763                 return -ESRCH;
 764         if (r < 0)
 765                 return r;
 766
 767         /* Let's skip the pid and comm fields. The latter is enclosed in () but does not escape any () in its
 768          * value, so let's skip over it manually */
 769
 770         p = strrchr(line, ')');
 771         if (!p)
 772                 return -EIO;
 773         p++;
 774
 775         unsigned long llu;
 776
 777         if (sscanf(p, " "
 778                    "%*c " /* state */
 779                    "%*u " /* ppid */
 780                    "%*u " /* pgrp */
 781                    "%*u " /* session */
 782                    "%*u " /* tty_nr */
 783                    "%*u " /* tpgid */
 784                    "%*u " /* flags */
 785                    "%*u " /* minflt */
 786                    "%*u " /* cminflt */
 787                    "%*u " /* majflt */
 788                    "%*u " /* cmajflt */
 789                    "%*u " /* utime */
 790                    "%*u " /* stime */
 791                    "%*u " /* cutime */
 792                    "%*u " /* cstime */
 793                    "%*i " /* priority */
 794                    "%*i " /* nice */
 795                    "%*u " /* num_threads */
 796                    "%*u " /* itrealvalue */
 797                    "%lu ", /* starttime */
 798                    &llu) != 1)
 799                 return -EIO;
 800
 801         if (ret)
 802                 *ret = jiffies_to_usec(llu); /* CLOCK_BOOTTIME */
 803
 804         return 0;
 805 }
 806
 807 int pidref_get_start_time(const PidRef *pid, usec_t *ret) {
 808         usec_t t;
 809         int r;
 810
 811         if (!pidref_is_set(pid))
 812                 return -ESRCH;
 813
 814         if (pidref_is_remote(pid))
 815                 return -EREMOTE;
 816
 817         r = pid_get_start_time(pid->pid, ret ? &t : NULL);
 818         if (r < 0)
 819                 return r;
 820
 821         r = pidref_verify(pid);
 822         if (r < 0)
 823                 return r;
 824
 825         if (ret)
 826                 *ret = t;
 827
 828         return 0;
 829 }
 830
 831 int get_process_umask(pid_t pid, mode_t *ret) {
 832         _cleanup_free_ char *m = NULL;
 833         int r;
 834
 835         assert(pid >= 0);
 836         assert(ret);
 837
 838         r = procfs_file_get_field(pid, "status", "Umask", &m);
 839         if (r == -ENOENT)
 840                 return -ESRCH;
 841         if (r < 0)
 842                 return r;
 843
 844         return parse_mode(m, ret);
 845 }
 846
 847 int wait_for_terminate(pid_t pid, siginfo_t *ret) {
 848         return pidref_wait_for_terminate(&PIDREF_MAKE_FROM_PID(pid), ret);
 849 }
 850
 851 /*
 852  * Return values:
 853  * < 0 : wait_for_terminate() failed to get the state of the
 854  *       process, the process was terminated by a signal, or
 855  *       failed for an unknown reason.
 856  * >=0 : The process terminated normally, and its exit code is
 857  *       returned.
 858  *
 859  * That is, success is indicated by a return value of zero, and an
 860  * error is indicated by a non-zero value.
 861  *
 862  * A warning is emitted if the process terminates abnormally,
 863  * and also if it returns non-zero unless check_exit_code is true.
 864  */
 865 int pidref_wait_for_terminate_and_check(const char *name, PidRef *pidref, WaitFlags flags) {
 866         int r;
 867
 868         if (!pidref_is_set(pidref))
 869                 return -ESRCH;
 870         if (pidref_is_remote(pidref))
 871                 return -EREMOTE;
 872         if (pidref->pid == 1 || pidref_is_self(pidref))
 873                 return -ECHILD;
 874
 875         _cleanup_free_ char *buffer = NULL;
 876         if (!name) {
 877                 r = pidref_get_comm(pidref, &buffer);
 878                 if (r < 0)
 879                         log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pidref->pid);
 880                 else
 881                         name = buffer;
 882         }
 883
 884         int prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
 885
 886         siginfo_t status;
 887         r = pidref_wait_for_terminate(pidref, &status);
 888         if (r < 0)
 889                 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
 890
 891         if (status.si_code == CLD_EXITED) {
 892                 if (status.si_status != EXIT_SUCCESS)
 893                         log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
 894                                  "%s failed with exit status %i.", strna(name), status.si_status);
 895                 else
 896                         log_debug("%s succeeded.", name);
 897
 898                 return status.si_status;
 899
 900         } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
 901
 902                 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
 903                 return -EPROTO;
 904         }
 905
 906         log_full(prio, "%s failed due to unknown reason.", strna(name));
 907         return -EPROTO;
 908 }
 909
 910 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
 911         return pidref_wait_for_terminate_and_check(name, &PIDREF_MAKE_FROM_PID(pid), flags);
 912 }
 913
 914 /*
 915  * Return values:
 916  *
 917  * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process
 918  *       was terminated by a signal, or failed for an unknown reason.
 919  *
 920  * >=0 : The process terminated normally with no failures.
 921  *
 922  * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure
 923  * states are indicated by error is indicated by a non-zero value.
 924  *
 925  * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off
 926  * to remain entirely race-free.
 927  */
 928 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
 929         sigset_t mask;
 930         int r;
 931         usec_t until;
 932
 933         assert_se(sigemptyset(&mask) == 0);
 934         assert_se(sigaddset(&mask, SIGCHLD) == 0);
 935
 936         /* Drop into a sigtimewait-based timeout. Waiting for the
 937          * pid to exit. */
 938         until = usec_add(now(CLOCK_MONOTONIC), timeout);
 939         for (;;) {
 940                 usec_t n;
 941                 siginfo_t status = {};
 942
 943                 n = now(CLOCK_MONOTONIC);
 944                 if (n >= until)
 945                         break;
 946
 947                 r = RET_NERRNO(sigtimedwait(&mask, NULL, TIMESPEC_STORE(until - n)));
 948                 /* Assuming we woke due to the child exiting. */
 949                 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
 950                         if (status.si_pid == pid) {
 951                                 /* This is the correct child. */
 952                                 if (status.si_code == CLD_EXITED)
 953                                         return status.si_status == 0 ? 0 : -EPROTO;
 954                                 else
 955                                         return -EPROTO;
 956                         }
 957                 }
 958                 /* Not the child, check for errors and proceed appropriately */
 959                 if (r < 0) {
 960                         switch (r) {
 961                         case -EAGAIN:
 962                                 /* Timed out, child is likely hung. */
 963                                 return -ETIMEDOUT;
 964                         case -EINTR:
 965                                 /* Received a different signal and should retry */
 966                                 continue;
 967                         default:
 968                                 /* Return any unexpected errors */
 969                                 return r;
 970                         }
 971                 }
 972         }
 973
 974         return -EPROTO;
 975 }
 976
 977 void sigkill_wait(pid_t pid) {
 978         assert(pid > 1);
 979
 980         (void) kill(pid, SIGKILL);
 981         (void) wait_for_terminate(pid, NULL);
 982 }
 983
 984 void sigkill_waitp(pid_t *pid) {
 985         PROTECT_ERRNO;
 986
 987         if (!pid)
 988                 return;
 989         if (*pid <= 1)
 990                 return;
 991
 992         sigkill_wait(*pid);
 993 }
 994
 995 void sigterm_wait(pid_t pid) {
 996         assert(pid > 1);
 997
 998         (void) kill_and_sigcont(pid, SIGTERM);
 999         (void) wait_for_terminate(pid, NULL);
1000 }
1001
1002 void sigkill_nowait(pid_t pid) {
1003         assert(pid > 1);
1004
1005         (void) kill(pid, SIGKILL);
1006 }
1007
1008 void sigkill_nowaitp(pid_t *pid) {
1009         PROTECT_ERRNO;
1010
1011         if (!pid)
1012                 return;
1013         if (*pid <= 1)
1014                 return;
1015
1016         sigkill_nowait(*pid);
1017 }
1018
1019 int kill_and_sigcont(pid_t pid, int sig) {
1020         int r;
1021
1022         r = RET_NERRNO(kill(pid, sig));
1023
1024         /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
1025          * affected by a process being suspended anyway. */
1026         if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
1027                 (void) kill(pid, SIGCONT);
1028
1029         return r;
1030 }
1031
1032 int getenv_for_pid(pid_t pid, const char *field, char **ret) {
1033         _cleanup_fclose_ FILE *f = NULL;
1034         const char *path;
1035         size_t sum = 0;
1036         int r;
1037
1038         assert(pid >= 0);
1039         assert(field);
1040         assert(ret);
1041
1042         if (pid == 0 || pid == getpid_cached())
1043                 return strdup_to_full(ret, getenv(field));
1044
1045         if (!pid_is_valid(pid))
1046                 return -EINVAL;
1047
1048         path = procfs_file_alloca(pid, "environ");
1049
1050         r = fopen_unlocked(path, "re", &f);
1051         if (r == -ENOENT)
1052                 return -ESRCH;
1053         if (r < 0)
1054                 return r;
1055
1056         for (;;) {
1057                 _cleanup_free_ char *line = NULL;
1058                 const char *match;
1059
1060                 if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */
1061                         return -ENOBUFS;
1062
1063                 r = read_nul_string(f, LONG_LINE_MAX, &line);
1064                 if (r < 0)
1065                         return r;
1066                 if (r == 0)  /* EOF */
1067                         break;
1068
1069                 sum += r;
1070
1071                 match = startswith(line, field);
1072                 if (match && *match == '=')
1073                         return strdup_to_full(ret, match + 1);
1074         }
1075
1076         *ret = NULL;
1077         return 0;
1078 }
1079
1080 int pidref_is_my_child(PidRef *pid) {
1081         int r;
1082
1083         if (!pidref_is_set(pid))
1084                 return -ESRCH;
1085
1086         if (pidref_is_remote(pid))
1087                 return -EREMOTE;
1088
1089         if (pid->pid == 1 || pidref_is_self(pid))
1090                 return false;
1091
1092         pid_t ppid;
1093         r = pidref_get_ppid(pid, &ppid);
1094         if (r == -EADDRNOTAVAIL) /* if this process is outside of our pidns, it is definitely not our child */
1095                 return false;
1096         if (r < 0)
1097                 return r;
1098
1099         return ppid == getpid_cached();
1100 }
1101
1102 int pid_is_my_child(pid_t pid) {
1103
1104         if (pid == 0)
1105                 return false;
1106
1107         return pidref_is_my_child(&PIDREF_MAKE_FROM_PID(pid));
1108 }
1109
1110 int pidref_is_unwaited(PidRef *pid) {
1111         int r;
1112
1113         /* Checks whether a PID is still valid at all, including a zombie */
1114
1115         if (!pidref_is_set(pid))
1116                 return -ESRCH;
1117
1118         if (pidref_is_remote(pid))
1119                 return -EREMOTE;
1120
1121         if (pid->pid == 1 || pidref_is_self(pid))
1122                 return true;
1123
1124         r = pidref_kill(pid, 0);
1125         if (r == -ESRCH)
1126                 return false;
1127         if (r < 0)
1128                 return r;
1129
1130         return true;
1131 }
1132
1133 int pid_is_unwaited(pid_t pid) {
1134
1135         if (pid == 0)
1136                 return true;
1137
1138         return pidref_is_unwaited(&PIDREF_MAKE_FROM_PID(pid));
1139 }
1140
1141 int pid_is_alive(pid_t pid) {
1142         int r;
1143
1144         /* Checks whether a PID is still valid and not a zombie */
1145
1146         if (pid < 0)
1147                 return -ESRCH;
1148
1149         if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
1150                 return true;
1151
1152         if (pid == getpid_cached())
1153                 return true;
1154
1155         r = get_process_state(pid);
1156         if (r == -ESRCH)
1157                 return false;
1158         if (r < 0)
1159                 return r;
1160
1161         return r != 'Z';
1162 }
1163
1164 int pidref_is_alive(const PidRef *pidref) {
1165         int r, result;
1166
1167         if (!pidref_is_set(pidref))
1168                 return -ESRCH;
1169
1170         if (pidref_is_remote(pidref))
1171                 return -EREMOTE;
1172
1173         result = pid_is_alive(pidref->pid);
1174         if (result < 0) {
1175                 assert(result != -ESRCH);
1176                 return result;
1177         }
1178
1179         r = pidref_verify(pidref);
1180         if (r == -ESRCH)
1181                 return false;
1182         if (r < 0)
1183                 return r;
1184
1185         return result;
1186 }
1187
1188 int pidref_from_same_root_fs(PidRef *a, PidRef *b) {
1189         _cleanup_(pidref_done) PidRef self = PIDREF_NULL;
1190         int r;
1191
1192         /* Checks if the two specified processes have the same root fs. Either can be specified as NULL in
1193          * which case we'll check against ourselves. */
1194
1195         if (!a || !b) {
1196                 r = pidref_set_self(&self);
1197                 if (r < 0)
1198                         return r;
1199                 if (!a)
1200                         a = &self;
1201                 if (!b)
1202                         b = &self;
1203         }
1204
1205         if (!pidref_is_set(a) || !pidref_is_set(b))
1206                 return -ESRCH;
1207
1208         /* If one of the two processes have the same root they cannot have the same root fs, but if both of
1209          * them do we don't know */
1210         if (pidref_is_remote(a) && pidref_is_remote(b))
1211                 return -EREMOTE;
1212         if (pidref_is_remote(a) || pidref_is_remote(b))
1213                 return false;
1214
1215         if (pidref_equal(a, b))
1216                 return true;
1217
1218         const char *roota = procfs_file_alloca(a->pid, "root");
1219         const char *rootb = procfs_file_alloca(b->pid, "root");
1220
1221         int result = inode_same(roota, rootb, 0);
1222         if (result == -ENOENT)
1223                 return proc_mounted() == 0 ? -ENOSYS : -ESRCH;
1224         if (result < 0)
1225                 return result;
1226
1227         r = pidref_verify(a);
1228         if (r < 0)
1229                 return r;
1230         r = pidref_verify(b);
1231         if (r < 0)
1232                 return r;
1233
1234         return result;
1235 }
1236
1237 bool is_main_thread(void) {
1238         static thread_local int cached = -1;
1239
1240         if (cached < 0)
1241                 cached = getpid_cached() == gettid();
1242
1243         return cached;
1244 }
1245
1246 bool oom_score_adjust_is_valid(int oa) {
1247         return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
1248 }
1249
1250 unsigned long personality_from_string(const char *p) {
1251         Architecture architecture;
1252
1253         if (!p)
1254                 return PERSONALITY_INVALID;
1255
1256         /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
1257          * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
1258          * the same register size. */
1259
1260         architecture = architecture_from_string(p);
1261         if (architecture < 0)
1262                 return PERSONALITY_INVALID;
1263
1264         if (architecture == native_architecture())
1265                 return PER_LINUX;
1266 #ifdef ARCHITECTURE_SECONDARY
1267         if (architecture == ARCHITECTURE_SECONDARY)
1268                 return PER_LINUX32;
1269 #endif
1270
1271         return PERSONALITY_INVALID;
1272 }
1273
1274 const char* personality_to_string(unsigned long p) {
1275         Architecture architecture = _ARCHITECTURE_INVALID;
1276
1277         if (p == PER_LINUX)
1278                 architecture = native_architecture();
1279 #ifdef ARCHITECTURE_SECONDARY
1280         else if (p == PER_LINUX32)
1281                 architecture = ARCHITECTURE_SECONDARY;
1282 #endif
1283
1284         if (architecture < 0)
1285                 return NULL;
1286
1287         return architecture_to_string(architecture);
1288 }
1289
1290 int safe_personality(unsigned long p) {
1291         int ret;
1292
1293         /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1294          * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1295          * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1296          * the return value indicating the same issue, so that we are definitely on the safe side.
1297          *
1298          * See https://github.com/systemd/systemd/issues/6737 */
1299
1300         errno = 0;
1301         ret = personality(p);
1302         if (ret < 0) {
1303                 if (errno != 0)
1304                         return -errno;
1305
1306                 errno = -ret;
1307         }
1308
1309         return ret;
1310 }
1311
1312 int opinionated_personality(unsigned long *ret) {
1313         int current;
1314
1315         /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1316          * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1317          * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1318
1319         current = safe_personality(PERSONALITY_INVALID);
1320         if (current < 0)
1321                 return current;
1322
1323         if (((unsigned long) current & OPINIONATED_PERSONALITY_MASK) == PER_LINUX32)
1324                 *ret = PER_LINUX32;
1325         else
1326                 *ret = PER_LINUX;
1327
1328         return 0;
1329 }
1330
1331 void valgrind_summary_hack(void) {
1332 #if HAVE_VALGRIND_VALGRIND_H
1333         if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1334                 pid_t pid;
1335                 pid = raw_clone(SIGCHLD);
1336                 if (pid < 0)
1337                         log_struct_errno(
1338                                 LOG_EMERG, errno,
1339                                 LOG_MESSAGE_ID(SD_MESSAGE_VALGRIND_HELPER_FORK_STR),
1340                                 LOG_MESSAGE("Failed to fork off valgrind helper: %m"));
1341                 else if (pid == 0)
1342                         exit(EXIT_SUCCESS);
1343                 else {
1344                         log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1345                         (void) wait_for_terminate(pid, NULL);
1346                 }
1347         }
1348 #endif
1349 }
1350
1351 int pid_compare_func(const pid_t *a, const pid_t *b) {
1352         /* Suitable for usage in qsort() */
1353         return CMP(*a, *b);
1354 }
1355
1356 bool nice_is_valid(int n) {
1357         return n >= PRIO_MIN && n < PRIO_MAX;
1358 }
1359
1360 bool sched_policy_is_valid(int i) {
1361         return IN_SET(i, SCHED_OTHER, SCHED_BATCH, SCHED_IDLE, SCHED_FIFO, SCHED_RR);
1362 }
1363
1364 bool sched_priority_is_valid(int i) {
1365         return i >= 0 && i <= sched_get_priority_max(SCHED_RR);
1366 }
1367
1368 /* The cached PID, possible values:
1369  *
1370  *     == UNSET [0]  → cache not initialized yet
1371  *     == BUSY [-1]  → some thread is initializing it at the moment
1372  *     any other     → the cached PID
1373  */
1374
1375 #define CACHED_PID_UNSET ((pid_t) 0)
1376 #define CACHED_PID_BUSY ((pid_t) -1)
1377
1378 static pid_t cached_pid = CACHED_PID_UNSET;
1379
1380 void reset_cached_pid(void) {
1381         /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1382         cached_pid = CACHED_PID_UNSET;
1383 }
1384
1385 pid_t getpid_cached(void) {
1386         static bool installed = false;
1387         pid_t current_value = CACHED_PID_UNSET;
1388
1389         /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1390          * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1391          * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1392          * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1393          *
1394          * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1395          * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1396          */
1397
1398         (void) __atomic_compare_exchange_n(
1399                         &cached_pid,
1400                         &current_value,
1401                         CACHED_PID_BUSY,
1402                         false,
1403                         __ATOMIC_SEQ_CST,
1404                         __ATOMIC_SEQ_CST);
1405
1406         switch (current_value) {
1407
1408         case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1409                 pid_t new_pid;
1410
1411                 new_pid = getpid();
1412
1413                 if (!installed) {
1414                         /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
1415                          * only half-documented (glibc doesn't document it but LSB does — though only superficially)
1416                          * we'll check for errors only in the most generic fashion possible. */
1417
1418                         if (pthread_atfork(NULL, NULL, reset_cached_pid) != 0) {
1419                                 /* OOM? Let's try again later */
1420                                 cached_pid = CACHED_PID_UNSET;
1421                                 return new_pid;
1422                         }
1423
1424                         installed = true;
1425                 }
1426
1427                 cached_pid = new_pid;
1428                 return new_pid;
1429         }
1430
1431         case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1432                 return getpid();
1433
1434         default: /* Properly initialized */
1435                 return current_value;
1436         }
1437 }
1438
1439 int must_be_root(void) {
1440
1441         if (geteuid() == 0)
1442                 return 0;
1443
1444         return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root.");
1445 }
1446
1447 pid_t clone_with_nested_stack(int (*fn)(void *), int flags, void *userdata) {
1448         size_t ps;
1449         pid_t pid;
1450         void *mystack;
1451
1452         /* A wrapper around glibc's clone() call that automatically sets up a "nested" stack. Only supports
1453          * invocations without CLONE_VM, so that we can continue to use the parent's stack mapping.
1454          *
1455          * Note: glibc's clone() wrapper does not synchronize malloc() locks. This means that if the parent
1456          * is threaded these locks will be in an undefined state in the child, and hence memory allocations
1457          * are likely going to run into deadlocks. Hence: if you use this function make sure your parent is
1458          * strictly single-threaded or your child never calls malloc(). */
1459
1460         assert((flags & (CLONE_VM|CLONE_PARENT_SETTID|CLONE_CHILD_SETTID|
1461                          CLONE_CHILD_CLEARTID|CLONE_SETTLS)) == 0);
1462
1463         /* We allocate some space on the stack to use as the stack for the child (hence "nested"). Note that
1464          * the net effect is that the child will have the start of its stack inside the stack of the parent,
1465          * but since they are a CoW copy of each other that's fine. We allocate one page-aligned page. But
1466          * since we don't want to deal with differences between systems where the stack grows backwards or
1467          * forwards we'll allocate one more and place the stack address in the middle. Except that we also
1468          * want it page aligned, hence we'll allocate one page more. Makes 3. */
1469
1470         ps = page_size();
1471         mystack = alloca(ps*3);
1472         mystack = (uint8_t*) mystack + ps; /* move pointer one page ahead since stacks usually grow backwards */
1473         mystack = (void*) ALIGN_TO((uintptr_t) mystack, ps); /* align to page size (moving things further ahead) */
1474
1475 #if HAVE_CLONE
1476         pid = clone(fn, mystack, flags, userdata);
1477 #else
1478         pid = __clone2(fn, mystack, ps, flags, userdata);
1479 #endif
1480         if (pid < 0)
1481                 return -errno;
1482
1483         return pid;
1484 }
1485
1486 static void restore_sigsetp(sigset_t **ssp) {
1487         if (*ssp)
1488                 (void) sigprocmask(SIG_SETMASK, *ssp, NULL);
1489 }
1490
1491 static int fork_flags_to_signal(ForkFlags flags) {
1492         return (flags & FORK_DEATHSIG_SIGTERM) ? SIGTERM :
1493                 (flags & FORK_DEATHSIG_SIGINT) ? SIGINT :
1494                                                  SIGKILL;
1495 }
1496
1497 int pidref_safe_fork_full(
1498                 const char *name,
1499                 const int stdio_fds[3],
1500                 int except_fds[],
1501                 size_t n_except_fds,
1502                 ForkFlags flags,
1503                 PidRef *ret_pid) {
1504
1505         pid_t original_pid, pid;
1506         sigset_t saved_ss, ss;
1507         _unused_ _cleanup_(restore_sigsetp) sigset_t *saved_ssp = NULL;
1508         bool block_signals = false, block_all = false, intermediary = false;
1509         _cleanup_close_pair_ int pidref_transport_fds[2] = EBADF_PAIR;
1510         int prio, r;
1511
1512         assert(!FLAGS_SET(flags, FORK_WAIT|FORK_FREEZE));
1513         assert(!FLAGS_SET(flags, FORK_DETACH) ||
1514                (flags & (FORK_WAIT|FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGINT|FORK_DEATHSIG_SIGKILL)) == 0);
1515
1516         /* A wrapper around fork(), that does a couple of important initializations in addition to mere
1517          * forking. If provided, ret_pid is initialized in both the parent and the child process, both times
1518          * referencing the child process. Returns == 0 in the child and > 0 in the parent. */
1519
1520         prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1521
1522         original_pid = getpid_cached();
1523
1524         if (flags & FORK_FLUSH_STDIO) {
1525                 fflush(stdout);
1526                 fflush(stderr); /* This one shouldn't be necessary, stderr should be unbuffered anyway, but let's better be safe than sorry */
1527         }
1528
1529         if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGINT)) {
1530                 /* We temporarily block all signals, so that the new child has them blocked initially. This
1531                  * way, we can be sure that SIGTERMs are not lost we might send to the child. (Note that for
1532                  * FORK_DEATHSIG_SIGKILL we don't bother, since it cannot be blocked anyway.) */
1533
1534                 assert_se(sigfillset(&ss) >= 0);
1535                 block_signals = block_all = true;
1536
1537         } else if (flags & FORK_WAIT) {
1538                 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1539
1540                 assert_se(sigemptyset(&ss) >= 0);
1541                 assert_se(sigaddset(&ss, SIGCHLD) >= 0);
1542                 block_signals = true;
1543         }
1544
1545         if (block_signals) {
1546                 if (sigprocmask(SIG_BLOCK, &ss, &saved_ss) < 0)
1547                         return log_full_errno(prio, errno, "Failed to block signal mask: %m");
1548                 saved_ssp = &saved_ss;
1549         }
1550
1551         if (FLAGS_SET(flags, FORK_DETACH)) {
1552                 /* Fork off intermediary child if needed */
1553
1554                 r = is_reaper_process();
1555                 if (r < 0)
1556                         return log_full_errno(prio, r, "Failed to determine if we are a reaper process: %m");
1557
1558                 if (!r) {
1559                         /* Not a reaper process, hence do a double fork() so we are reparented to one */
1560
1561                         if (ret_pid && socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, pidref_transport_fds) < 0)
1562                                 return log_full_errno(prio, errno, "Failed to allocate pidref socket: %m");
1563
1564                         pid = fork();
1565                         if (pid < 0)
1566                                 return log_full_errno(prio, errno, "Failed to fork off '%s': %m", strna(name));
1567                         if (pid > 0) {
1568                                 log_debug("Successfully forked off intermediary '%s' as PID " PID_FMT ".", strna(name), pid);
1569
1570                                 pidref_transport_fds[1] = safe_close(pidref_transport_fds[1]);
1571
1572                                 if (pidref_transport_fds[0] >= 0) {
1573                                         /* Wait for the intermediary child to exit so the caller can be certain the actual child
1574                                          * process has been reparented by the time this function returns. */
1575                                         r = wait_for_terminate_and_check(name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0);
1576                                         if (r < 0)
1577                                                 return log_full_errno(prio, r, "Failed to wait for intermediary process: %m");
1578                                         if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1579                                                 return -EPROTO;
1580
1581                                         int pidfd;
1582                                         ssize_t n = receive_one_fd_iov(
1583                                                         pidref_transport_fds[0],
1584                                                         &IOVEC_MAKE(&pid, sizeof(pid)),
1585                                                         /* iovlen= */ 1,
1586                                                         /* flags= */ 0,
1587                                                         &pidfd);
1588                                         if (n < 0)
1589                                                 return log_full_errno(prio, n, "Failed to receive child pidref: %m");
1590
1591                                         *ret_pid = (PidRef) { .pid = pid, .fd = pidfd };
1592                                 }
1593
1594                                 return 1; /* return in the parent */
1595                         }
1596
1597                         pidref_transport_fds[0] = safe_close(pidref_transport_fds[0]);
1598                         intermediary = true;
1599                 }
1600         }
1601
1602         if ((flags & (FORK_NEW_MOUNTNS|FORK_NEW_USERNS|FORK_NEW_NETNS|FORK_NEW_PIDNS)) != 0)
1603                 pid = raw_clone(SIGCHLD|
1604                                 (FLAGS_SET(flags, FORK_NEW_MOUNTNS) ? CLONE_NEWNS : 0) |
1605                                 (FLAGS_SET(flags, FORK_NEW_USERNS) ? CLONE_NEWUSER : 0) |
1606                                 (FLAGS_SET(flags, FORK_NEW_NETNS) ? CLONE_NEWNET : 0) |
1607                                 (FLAGS_SET(flags, FORK_NEW_PIDNS) ? CLONE_NEWPID : 0));
1608         else
1609                 pid = fork();
1610         if (pid < 0)
1611                 return log_full_errno(prio, errno, "Failed to fork off '%s': %m", strna(name));
1612         if (pid > 0) {
1613
1614                 /* If we are in the intermediary process, exit now */
1615                 if (intermediary) {
1616                         if (pidref_transport_fds[1] >= 0) {
1617                                 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
1618
1619                                 r = pidref_set_pid(&pidref, pid);
1620                                 if (r < 0) {
1621                                         log_full_errno(prio, r, "Failed to open reference to PID "PID_FMT": %m", pid);
1622                                         _exit(EXIT_FAILURE);
1623                                 }
1624
1625                                 r = send_one_fd_iov(
1626                                                 pidref_transport_fds[1],
1627                                                 pidref.fd,
1628                                                 &IOVEC_MAKE(&pidref.pid, sizeof(pidref.pid)),
1629                                                 /* iovlen= */ 1,
1630                                                 /* flags= */ 0);
1631                                 if (r < 0) {
1632                                         log_full_errno(prio, r, "Failed to send child pidref: %m");
1633                                         _exit(EXIT_FAILURE);
1634                                 }
1635                         }
1636
1637                         _exit(EXIT_SUCCESS);
1638                 }
1639
1640                 /* We are in the parent process */
1641                 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1642
1643                 if (flags & FORK_WAIT) {
1644                         if (block_all) {
1645                                 /* undo everything except SIGCHLD */
1646                                 ss = saved_ss;
1647                                 assert_se(sigaddset(&ss, SIGCHLD) >= 0);
1648                                 (void) sigprocmask(SIG_SETMASK, &ss, NULL);
1649                         }
1650
1651                         r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1652                         if (r < 0)
1653                                 return r;
1654                         if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1655                                 return -EPROTO;
1656
1657                         /* If we are in the parent and successfully waited, then the process doesn't exist anymore. */
1658                         if (ret_pid)
1659                                 *ret_pid = PIDREF_NULL;
1660
1661                         return 1;
1662                 }
1663
1664                 if (ret_pid) {
1665                         if (FLAGS_SET(flags, FORK_PID_ONLY))
1666                                 *ret_pid = PIDREF_MAKE_FROM_PID(pid);
1667                         else {
1668                                 r = pidref_set_pid(ret_pid, pid);
1669                                 if (r < 0) /* Let's not fail for this, no matter what, the process exists after all, and that's key */
1670                                         *ret_pid = PIDREF_MAKE_FROM_PID(pid);
1671                         }
1672                 }
1673
1674                 return 1;
1675         }
1676
1677         /* We are in the child process */
1678
1679         pidref_transport_fds[1] = safe_close(pidref_transport_fds[1]);
1680
1681         /* Restore signal mask manually */
1682         saved_ssp = NULL;
1683
1684         if (flags & FORK_REOPEN_LOG) {
1685                 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1686                 log_close();
1687                 log_set_open_when_needed(true);
1688                 log_settle_target();
1689         }
1690
1691         if (name) {
1692                 r = rename_process(name);
1693                 if (r < 0)
1694                         log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1695                                        r, "Failed to rename process, ignoring: %m");
1696         }
1697
1698         if (flags & (FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGINT|FORK_DEATHSIG_SIGKILL))
1699                 if (prctl(PR_SET_PDEATHSIG, fork_flags_to_signal(flags)) < 0) {
1700                         log_full_errno(prio, errno, "Failed to set death signal: %m");
1701                         _exit(EXIT_FAILURE);
1702                 }
1703
1704         if (flags & FORK_RESET_SIGNALS) {
1705                 r = reset_all_signal_handlers();
1706                 if (r < 0) {
1707                         log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1708                         _exit(EXIT_FAILURE);
1709                 }
1710
1711                 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1712                 r = reset_signal_mask();
1713                 if (r < 0) {
1714                         log_full_errno(prio, r, "Failed to reset signal mask: %m");
1715                         _exit(EXIT_FAILURE);
1716                 }
1717         } else if (block_signals) { /* undo what we did above */
1718                 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1719                         log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1720                         _exit(EXIT_FAILURE);
1721                 }
1722         }
1723
1724         if (flags & (FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGKILL|FORK_DEATHSIG_SIGINT)) {
1725                 pid_t ppid;
1726                 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1727                  * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1728
1729                 ppid = getppid();
1730                 if (ppid == 0)
1731                         /* Parent is in a different PID namespace. */;
1732                 else if (ppid != original_pid) {
1733                         int sig = fork_flags_to_signal(flags);
1734                         log_debug("Parent died early, raising %s.", signal_to_string(sig));
1735                         (void) raise(sig);
1736                         _exit(EXIT_FAILURE);
1737                 }
1738         }
1739
1740         if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) {
1741                 /* Optionally, make sure we never propagate mounts to the host. */
1742                 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
1743                         log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m");
1744                         _exit(EXIT_FAILURE);
1745                 }
1746         }
1747
1748         if (FLAGS_SET(flags, FORK_PRIVATE_TMP)) {
1749                 assert(FLAGS_SET(flags, FORK_NEW_MOUNTNS));
1750
1751                 /* Optionally, overmount new tmpfs instance on /tmp/. */
1752                 r = mount_nofollow("tmpfs", "/tmp", "tmpfs",
1753                                    MS_NOSUID|MS_NODEV,
1754                                    "mode=01777" TMPFS_LIMITS_RUN);
1755                 if (r < 0) {
1756                         log_full_errno(prio, r, "Failed to overmount /tmp/: %m");
1757                         _exit(EXIT_FAILURE);
1758                 }
1759         }
1760
1761         if (flags & FORK_REARRANGE_STDIO) {
1762                 if (stdio_fds) {
1763                         r = rearrange_stdio(stdio_fds[0], stdio_fds[1], stdio_fds[2]);
1764                         if (r < 0) {
1765                                 log_full_errno(prio, r, "Failed to rearrange stdio fds: %m");
1766                                 _exit(EXIT_FAILURE);
1767                         }
1768
1769                         /* Turn off O_NONBLOCK on the fdio fds, in case it was left on */
1770                         stdio_disable_nonblock();
1771                 } else {
1772                         r = make_null_stdio();
1773                         if (r < 0) {
1774                                 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1775                                 _exit(EXIT_FAILURE);
1776                         }
1777                 }
1778         } else if (flags & FORK_STDOUT_TO_STDERR) {
1779                 if (dup2(STDERR_FILENO, STDOUT_FILENO) < 0) {
1780                         log_full_errno(prio, errno, "Failed to connect stdout to stderr: %m");
1781                         _exit(EXIT_FAILURE);
1782                 }
1783         }
1784
1785         if (flags & FORK_CLOSE_ALL_FDS) {
1786                 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1787                 log_close();
1788
1789                 r = close_all_fds(except_fds, n_except_fds);
1790                 if (r < 0) {
1791                         log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1792                         _exit(EXIT_FAILURE);
1793                 }
1794         }
1795
1796         if (flags & FORK_PACK_FDS) {
1797                 /* FORK_CLOSE_ALL_FDS ensures that except_fds are the only FDs >= 3 that are
1798                  * open, this is including the log. This is required by pack_fds, which will
1799                  * get stuck in an infinite loop of any FDs other than except_fds are open. */
1800                 assert(FLAGS_SET(flags, FORK_CLOSE_ALL_FDS));
1801
1802                 r = pack_fds(except_fds, n_except_fds);
1803                 if (r < 0) {
1804                         log_full_errno(prio, r, "Failed to pack file descriptors: %m");
1805                         _exit(EXIT_FAILURE);
1806                 }
1807         }
1808
1809         if (flags & FORK_CLOEXEC_OFF) {
1810                 r = fd_cloexec_many(except_fds, n_except_fds, false);
1811                 if (r < 0) {
1812                         log_full_errno(prio, r, "Failed to turn off O_CLOEXEC on file descriptors: %m");
1813                         _exit(EXIT_FAILURE);
1814                 }
1815         }
1816
1817         /* When we were asked to reopen the logs, do so again now */
1818         if (flags & FORK_REOPEN_LOG) {
1819                 log_open();
1820                 log_set_open_when_needed(false);
1821         }
1822
1823         if (flags & FORK_RLIMIT_NOFILE_SAFE) {
1824                 r = rlimit_nofile_safe();
1825                 if (r < 0) {
1826                         log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m");
1827                         _exit(EXIT_FAILURE);
1828                 }
1829         }
1830
1831         if (!FLAGS_SET(flags, FORK_KEEP_NOTIFY_SOCKET)) {
1832                 r = RET_NERRNO(unsetenv("NOTIFY_SOCKET"));
1833                 if (r < 0) {
1834                         log_full_errno(prio, r, "Failed to unset $NOTIFY_SOCKET: %m");
1835                         _exit(EXIT_FAILURE);
1836                 }
1837         }
1838
1839         if (FLAGS_SET(flags, FORK_FREEZE))
1840                 freeze();
1841
1842         if (ret_pid) {
1843                 if (FLAGS_SET(flags, FORK_PID_ONLY))
1844                         *ret_pid = PIDREF_MAKE_FROM_PID(getpid_cached());
1845                 else {
1846                         r = pidref_set_self(ret_pid);
1847                         if (r < 0) {
1848                                 log_full_errno(prio, r, "Failed to acquire PID reference on ourselves: %m");
1849                                 _exit(EXIT_FAILURE);
1850                         }
1851                 }
1852         }
1853
1854         return 0;
1855 }
1856
1857 int safe_fork_full(
1858                 const char *name,
1859                 const int stdio_fds[3],
1860                 int except_fds[],
1861                 size_t n_except_fds,
1862                 ForkFlags flags,
1863                 pid_t *ret_pid) {
1864
1865         _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
1866         int r;
1867
1868         /* Getting the detached child process pid without pidfd is racy, so don't allow it if not returning
1869          * a pidref to the caller. */
1870         assert(!FLAGS_SET(flags, FORK_DETACH) || !ret_pid);
1871
1872         r = pidref_safe_fork_full(name, stdio_fds, except_fds, n_except_fds, flags|FORK_PID_ONLY, ret_pid ? &pidref : NULL);
1873         if (r < 0 || !ret_pid)
1874                 return r;
1875
1876         *ret_pid = pidref.pid;
1877
1878         return r;
1879 }
1880
1881 int namespace_fork(
1882                 const char *outer_name,
1883                 const char *inner_name,
1884                 int except_fds[],
1885                 size_t n_except_fds,
1886                 ForkFlags flags,
1887                 int pidns_fd,
1888                 int mntns_fd,
1889                 int netns_fd,
1890                 int userns_fd,
1891                 int root_fd,
1892                 pid_t *ret_pid) {
1893
1894         int r;
1895
1896         /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle
1897          * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that
1898          * /proc/self/fd works correctly. */
1899
1900         r = safe_fork_full(outer_name,
1901                            NULL,
1902                            except_fds, n_except_fds,
1903                            (flags|FORK_DEATHSIG_SIGINT|FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGKILL) & ~(FORK_REOPEN_LOG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE), ret_pid);
1904         if (r < 0)
1905                 return r;
1906         if (r == 0) {
1907                 pid_t pid;
1908
1909                 /* Child */
1910
1911                 r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd);
1912                 if (r < 0) {
1913                         log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m");
1914                         _exit(EXIT_FAILURE);
1915                 }
1916
1917                 /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */
1918                 r = safe_fork_full(inner_name,
1919                                    NULL,
1920                                    except_fds, n_except_fds,
1921                                    flags & ~(FORK_WAIT|FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_REARRANGE_STDIO), &pid);
1922                 if (r < 0)
1923                         _exit(EXIT_FAILURE);
1924                 if (r == 0) {
1925                         /* Child */
1926                         if (ret_pid)
1927                                 *ret_pid = pid;
1928                         return 0;
1929                 }
1930
1931                 r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0);
1932                 if (r < 0)
1933                         _exit(EXIT_FAILURE);
1934
1935                 _exit(r);
1936         }
1937
1938         return 1;
1939 }
1940
1941 int set_oom_score_adjust(int value) {
1942         char t[DECIMAL_STR_MAX(int)];
1943
1944         if (!oom_score_adjust_is_valid(value))
1945                 return -EINVAL;
1946
1947         xsprintf(t, "%i", value);
1948
1949         return write_string_file("/proc/self/oom_score_adj", t,
1950                                  WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER);
1951 }
1952
1953 int get_oom_score_adjust(int *ret) {
1954         _cleanup_free_ char *t = NULL;
1955         int r, a;
1956
1957         r = read_virtual_file("/proc/self/oom_score_adj", SIZE_MAX, &t, NULL);
1958         if (r < 0)
1959                 return r;
1960
1961         delete_trailing_chars(t, WHITESPACE);
1962
1963         r = safe_atoi(t, &a);
1964         if (r < 0)
1965                 return r;
1966
1967         if (!oom_score_adjust_is_valid(a))
1968                 return -ENODATA;
1969
1970         if (ret)
1971                 *ret = a;
1972
1973         return 0;
1974 }
1975
1976 static int rlimit_to_nice(rlim_t limit) {
1977         if (limit <= 1)
1978                 return PRIO_MAX-1; /* i.e. 19 */
1979
1980         if (limit >= -PRIO_MIN + PRIO_MAX)
1981                 return PRIO_MIN; /* i.e. -20 */
1982
1983         return PRIO_MAX - (int) limit;
1984 }
1985
1986 int setpriority_closest(int priority) {
1987         struct rlimit highest;
1988         int r, current, limit;
1989
1990         /* Try to set requested nice level */
1991         r = RET_NERRNO(setpriority(PRIO_PROCESS, 0, priority));
1992         if (r >= 0)
1993                 return 1;
1994         if (!ERRNO_IS_NEG_PRIVILEGE(r))
1995                 return r;
1996
1997         errno = 0;
1998         current = getpriority(PRIO_PROCESS, 0);
1999         if (errno != 0)
2000                 return -errno;
2001
2002         if (priority == current)
2003                 return 1;
2004
2005        /* Hmm, we'd expect that raising the nice level from our status quo would always work. If it doesn't,
2006         * then the whole setpriority() system call is blocked to us, hence let's propagate the error
2007         * right-away */
2008         if (priority > current)
2009                 return r;
2010
2011         if (getrlimit(RLIMIT_NICE, &highest) < 0)
2012                 return -errno;
2013
2014         limit = rlimit_to_nice(highest.rlim_cur);
2015
2016         /* Push to the allowed limit if we're higher than that. Note that we could also be less nice than
2017          * limit allows us, but still higher than what's requested. In that case our current value is
2018          * the best choice. */
2019         if (current > limit)
2020                 if (setpriority(PRIO_PROCESS, 0, limit) < 0)
2021                         return -errno;
2022
2023         log_debug("Cannot set requested nice level (%i), using next best (%i).", priority, MIN(current, limit));
2024         return 0;
2025 }
2026
2027 _noreturn_ void freeze(void) {
2028         log_close();
2029
2030         /* Make sure nobody waits for us (i.e. on one of our sockets) anymore. Note that we use
2031          * close_all_fds_without_malloc() instead of plain close_all_fds() here, since we want this function
2032          * to be compatible with being called from signal handlers. */
2033         (void) close_all_fds_without_malloc(NULL, 0);
2034
2035         /* Let's not freeze right away, but keep reaping zombies. */
2036         for (;;) {
2037                 siginfo_t si = {};
2038
2039                 if (waitid(P_ALL, 0, &si, WEXITED) < 0 && errno != EINTR)
2040                         break;
2041         }
2042
2043         /* waitid() failed with an ECHLD error (because there are no left-over child processes) or any other
2044          * (unexpected) error. Freeze for good now! */
2045         for (;;)
2046                 pause();
2047 }
2048
2049 int get_process_threads(pid_t pid) {
2050         _cleanup_free_ char *t = NULL;
2051         int n, r;
2052
2053         if (pid < 0)
2054                 return -EINVAL;
2055
2056         r = procfs_file_get_field(pid, "status", "Threads", &t);
2057         if (r == -ENOENT)
2058                 return -ESRCH;
2059         if (r < 0)
2060                 return r;
2061
2062         r = safe_atoi(t, &n);
2063         if (r < 0)
2064                 return r;
2065         if (n < 0)
2066                 return -EINVAL;
2067
2068         return n;
2069 }
2070
2071 int is_reaper_process(void) {
2072         int b = 0;
2073
2074         /* Checks if we are running in a reaper process, i.e. if we are expected to deal with processes
2075          * reparented to us. This simply checks if we are PID 1 or if PR_SET_CHILD_SUBREAPER was called. */
2076
2077         if (getpid_cached() == 1)
2078                 return true;
2079
2080         if (prctl(PR_GET_CHILD_SUBREAPER, (unsigned long) &b, 0UL, 0UL, 0UL) < 0)
2081                 return -errno;
2082
2083         return b != 0;
2084 }
2085
2086 int make_reaper_process(bool b) {
2087
2088         if (getpid_cached() == 1) {
2089
2090                 if (!b)
2091                         return -EINVAL;
2092
2093                 return 0;
2094         }
2095
2096         /* Some prctl()s insist that all 5 arguments are specified, others do not. Let's always specify all,
2097          * to avoid any ambiguities */
2098         if (prctl(PR_SET_CHILD_SUBREAPER, (unsigned long) b, 0UL, 0UL, 0UL) < 0)
2099                 return -errno;
2100
2101         return 0;
2102 }
2103
2104 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(posix_spawnattr_t*, posix_spawnattr_destroy, NULL);
2105
2106 int posix_spawn_wrapper(
2107                 const char *path,
2108                 char * const *argv,
2109                 char * const *envp,
2110                 const char *cgroup,
2111                 PidRef *ret_pidref) {
2112
2113         short flags = POSIX_SPAWN_SETSIGMASK;
2114         posix_spawnattr_t attr;
2115         sigset_t mask;
2116         int r;
2117
2118         /* Forks and invokes 'path' with 'argv' and 'envp' using CLONE_VM and CLONE_VFORK, which means the
2119          * caller will be blocked until the child either exits or exec's. The memory of the child will be
2120          * fully shared with the memory of the parent, so that there are no copy-on-write or memory.max
2121          * issues.
2122          *
2123          * Also, move the newly-created process into 'cgroup' through POSIX_SPAWN_SETCGROUP (clone3())
2124          * if available.
2125          * returns 1: We're already in the right cgroup
2126          *         0: 'cgroup' not specified or POSIX_SPAWN_SETCGROUP is not supported. The caller
2127          *            needs to call 'cg_attach' on their own */
2128
2129         assert(path);
2130         assert(argv);
2131         assert(ret_pidref);
2132
2133         assert_se(sigfillset(&mask) >= 0);
2134
2135         r = posix_spawnattr_init(&attr);
2136         if (r != 0)
2137                 return -r; /* These functions return a positive errno on failure */
2138
2139         /* Initialization needs to succeed before we can set up a destructor. */
2140         _unused_ _cleanup_(posix_spawnattr_destroyp) posix_spawnattr_t *attr_destructor = &attr;
2141
2142 #if HAVE_PIDFD_SPAWN
2143         static bool have_clone_into_cgroup = true; /* kernel 5.7+ */
2144         _cleanup_close_ int cgroup_fd = -EBADF;
2145
2146         if (cgroup && have_clone_into_cgroup) {
2147                 _cleanup_free_ char *resolved_cgroup = NULL;
2148
2149                 r = cg_get_path_and_check(
2150                                 SYSTEMD_CGROUP_CONTROLLER,
2151                                 cgroup,
2152                                 /* suffix= */ NULL,
2153                                 &resolved_cgroup);
2154                 if (r < 0)
2155                         return r;
2156
2157                 cgroup_fd = open(resolved_cgroup, O_PATH|O_DIRECTORY|O_CLOEXEC);
2158                 if (cgroup_fd < 0)
2159                         return -errno;
2160
2161                 r = posix_spawnattr_setcgroup_np(&attr, cgroup_fd);
2162                 if (r != 0)
2163                         return -r;
2164
2165                 flags |= POSIX_SPAWN_SETCGROUP;
2166         }
2167 #endif
2168
2169         r = posix_spawnattr_setflags(&attr, flags);
2170         if (r != 0)
2171                 return -r;
2172         r = posix_spawnattr_setsigmask(&attr, &mask);
2173         if (r != 0)
2174                 return -r;
2175
2176 #if HAVE_PIDFD_SPAWN
2177         _cleanup_close_ int pidfd = -EBADF;
2178
2179         r = pidfd_spawn(&pidfd, path, NULL, &attr, argv, envp);
2180         if (ERRNO_IS_NOT_SUPPORTED(r) && FLAGS_SET(flags, POSIX_SPAWN_SETCGROUP) && cg_is_threaded(cgroup) > 0)
2181                 return -EUCLEAN; /* clone3() could also return EOPNOTSUPP if the target cgroup is in threaded mode,
2182                                     turn that into something recognizable */
2183         if ((ERRNO_IS_NOT_SUPPORTED(r) || ERRNO_IS_PRIVILEGE(r) || r == E2BIG) &&
2184             FLAGS_SET(flags, POSIX_SPAWN_SETCGROUP)) {
2185                 /* Compiled on a newer host, or seccomp&friends blocking clone3()? Fallback, but
2186                  * need to disable POSIX_SPAWN_SETCGROUP, which is what redirects to clone3().
2187                  * Note that we might get E2BIG here since some kernels (e.g. 5.4) support clone3()
2188                  * but not CLONE_INTO_CGROUP. */
2189
2190                 /* CLONE_INTO_CGROUP definitely won't work, hence remember the fact so that we don't
2191                  * retry every time. */
2192                 have_clone_into_cgroup = false;
2193
2194                 flags &= ~POSIX_SPAWN_SETCGROUP;
2195                 r = posix_spawnattr_setflags(&attr, flags);
2196                 if (r != 0)
2197                         return -r;
2198
2199                 r = pidfd_spawn(&pidfd, path, NULL, &attr, argv, envp);
2200         }
2201         if (r != 0)
2202                 return -r;
2203
2204         r = pidref_set_pidfd_consume(ret_pidref, TAKE_FD(pidfd));
2205         if (r < 0)
2206                 return r;
2207
2208         return FLAGS_SET(flags, POSIX_SPAWN_SETCGROUP);
2209 #else
2210         pid_t pid;
2211
2212         r = posix_spawn(&pid, path, NULL, &attr, argv, envp);
2213         if (r != 0)
2214                 return -r;
2215
2216         r = pidref_set_pid(ret_pidref, pid);
2217         if (r < 0)
2218                 return r;
2219
2220         return 0; /* We did not use CLONE_INTO_CGROUP so return 0, the caller will have to move the child */
2221 #endif
2222 }
2223
2224 int proc_dir_open(DIR **ret) {
2225         DIR *d;
2226
2227         assert(ret);
2228
2229         d = opendir("/proc");
2230         if (!d)
2231                 return -errno;
2232
2233         *ret = d;
2234         return 0;
2235 }
2236
2237 int proc_dir_read(DIR *d, pid_t *ret) {
2238         assert(d);
2239
2240         for (;;) {
2241                 struct dirent *de;
2242
2243                 errno = 0;
2244                 de = readdir_no_dot(d);
2245                 if (!de) {
2246                         if (errno != 0)
2247                                 return -errno;
2248
2249                         break;
2250                 }
2251
2252                 if (!IN_SET(de->d_type, DT_DIR, DT_UNKNOWN))
2253                         continue;
2254
2255                 if (parse_pid(de->d_name, ret) >= 0)
2256                         return 1;
2257         }
2258
2259         if (ret)
2260                 *ret = 0;
2261         return 0;
2262 }
2263
2264 int proc_dir_read_pidref(DIR *d, PidRef *ret) {
2265         int r;
2266
2267         assert(d);
2268
2269         for (;;) {
2270                 pid_t pid;
2271
2272                 r = proc_dir_read(d, &pid);
2273                 if (r < 0)
2274                         return r;
2275                 if (r == 0)
2276                         break;
2277
2278                 r = pidref_set_pid(ret, pid);
2279                 if (r == -ESRCH) /* gone by now? skip it */
2280                         continue;
2281                 if (r < 0)
2282                         return r;
2283
2284                 return 1;
2285         }
2286
2287         if (ret)
2288                 *ret = PIDREF_NULL;
2289         return 0;
2290 }
2291
2292 static const char *const sigchld_code_table[] = {
2293         [CLD_EXITED] = "exited",
2294         [CLD_KILLED] = "killed",
2295         [CLD_DUMPED] = "dumped",
2296         [CLD_TRAPPED] = "trapped",
2297         [CLD_STOPPED] = "stopped",
2298         [CLD_CONTINUED] = "continued",
2299 };
2300
2301 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
2302
2303 static const char* const sched_policy_table[] = {
2304         [SCHED_OTHER] = "other",
2305         [SCHED_BATCH] = "batch",
2306         [SCHED_IDLE] = "idle",
2307         [SCHED_FIFO] = "fifo",
2308         [SCHED_RR] = "rr",
2309 };
2310
2311 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);
2312
2313 _noreturn_ void report_errno_and_exit(int errno_fd, int error) {
2314         int r;
2315
2316         if (error >= 0)
2317                 _exit(EXIT_SUCCESS);
2318
2319         assert(errno_fd >= 0);
2320
2321         r = loop_write(errno_fd, &error, sizeof(error));
2322         if (r < 0)
2323                 log_debug_errno(r, "Failed to write errno to errno_fd=%d: %m", errno_fd);
2324
2325         _exit(EXIT_FAILURE);
2326 }
2327
2328 int read_errno(int errno_fd) {
2329         int r;
2330
2331         assert(errno_fd >= 0);
2332
2333         /* The issue here is that it's impossible to distinguish between an error code returned by child and
2334          * IO error arose when reading it. So, the function logs errors and return EIO for the later case. */
2335
2336         ssize_t n = loop_read(errno_fd, &r, sizeof(r), /* do_poll = */ false);
2337         if (n < 0) {
2338                 log_debug_errno(n, "Failed to read errno: %m");
2339                 return -EIO;
2340         }
2341         if (n == sizeof(r)) {
2342                 if (r == 0)
2343                         return 0;
2344                 if (r < 0) /* child process reported an error, return it */
2345                         return log_debug_errno(r, "Child process failed with errno: %m");
2346                 return log_debug_errno(SYNTHETIC_ERRNO(EIO), "Received an errno, but it's a positive value.");
2347         }
2348         if (n != 0)
2349                 return log_debug_errno(SYNTHETIC_ERRNO(EIO), "Received unexpected amount of bytes while reading errno.");
2350
2351         /* the process exited without reporting an error, assuming success */
2352         return 0;
2353 }