src/basic/process-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <ctype.h>
   4 #include <errno.h>
   5 #include <limits.h>
   6 #include <linux/oom.h>
   7 #include <pthread.h>
   8 #include <spawn.h>
   9 #include <stdbool.h>
  10 #include <stdio.h>
  11 #include <stdlib.h>
  12 #include <sys/mount.h>
  13 #include <sys/personality.h>
  14 #include <sys/prctl.h>
  15 #include <sys/types.h>
  16 #include <sys/wait.h>
  17 #include <syslog.h>
  18 #include <unistd.h>
  19 #if HAVE_VALGRIND_VALGRIND_H
  20 #include <valgrind/valgrind.h>
  21 #endif
  22
  23 #include "sd-messages.h"
  24
  25 #include "alloc-util.h"
  26 #include "architecture.h"
  27 #include "argv-util.h"
  28 #include "cgroup-util.h"
  29 #include "dirent-util.h"
  30 #include "env-file.h"
  31 #include "env-util.h"
  32 #include "errno-util.h"
  33 #include "escape.h"
  34 #include "fd-util.h"
  35 #include "fileio.h"
  36 #include "fs-util.h"
  37 #include "hostname-util.h"
  38 #include "locale-util.h"
  39 #include "log.h"
  40 #include "macro.h"
  41 #include "memory-util.h"
  42 #include "missing_sched.h"
  43 #include "missing_syscall.h"
  44 #include "missing_threads.h"
  45 #include "mountpoint-util.h"
  46 #include "namespace-util.h"
  47 #include "nulstr-util.h"
  48 #include "parse-util.h"
  49 #include "path-util.h"
  50 #include "process-util.h"
  51 #include "raw-clone.h"
  52 #include "rlimit-util.h"
  53 #include "signal-util.h"
  54 #include "stat-util.h"
  55 #include "stdio-util.h"
  56 #include "string-table.h"
  57 #include "string-util.h"
  58 #include "terminal-util.h"
  59 #include "user-util.h"
  60 #include "utf8.h"
  61
  62 /* The kernel limits userspace processes to TASK_COMM_LEN (16 bytes), but allows higher values for its own
  63  * workers, e.g. "kworker/u9:3-kcryptd/253:0". Let's pick a fixed smallish limit that will work for the kernel.
  64  */
  65 #define COMM_MAX_LEN 128
  66
  67 static int get_process_state(pid_t pid) {
  68         _cleanup_free_ char *line = NULL;
  69         const char *p;
  70         char state;
  71         int r;
  72
  73         assert(pid >= 0);
  74
  75         /* Shortcut: if we are enquired about our own state, we are obviously running */
  76         if (pid == 0 || pid == getpid_cached())
  77                 return (unsigned char) 'R';
  78
  79         p = procfs_file_alloca(pid, "stat");
  80
  81         r = read_one_line_file(p, &line);
  82         if (r == -ENOENT)
  83                 return -ESRCH;
  84         if (r < 0)
  85                 return r;
  86
  87         p = strrchr(line, ')');
  88         if (!p)
  89                 return -EIO;
  90
  91         p++;
  92
  93         if (sscanf(p, " %c", &state) != 1)
  94                 return -EIO;
  95
  96         return (unsigned char) state;
  97 }
  98
  99 int pid_get_comm(pid_t pid, char **ret) {
 100         _cleanup_free_ char *escaped = NULL, *comm = NULL;
 101         int r;
 102
 103         assert(ret);
 104         assert(pid >= 0);
 105
 106         if (pid == 0 || pid == getpid_cached()) {
 107                 comm = new0(char, TASK_COMM_LEN + 1); /* Must fit in 16 byte according to prctl(2) */
 108                 if (!comm)
 109                         return -ENOMEM;
 110
 111                 if (prctl(PR_GET_NAME, comm) < 0)
 112                         return -errno;
 113         } else {
 114                 const char *p;
 115
 116                 p = procfs_file_alloca(pid, "comm");
 117
 118                 /* Note that process names of kernel threads can be much longer than TASK_COMM_LEN */
 119                 r = read_one_line_file(p, &comm);
 120                 if (r == -ENOENT)
 121                         return -ESRCH;
 122                 if (r < 0)
 123                         return r;
 124         }
 125
 126         escaped = new(char, COMM_MAX_LEN);
 127         if (!escaped)
 128                 return -ENOMEM;
 129
 130         /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */
 131         cellescape(escaped, COMM_MAX_LEN, comm);
 132
 133         *ret = TAKE_PTR(escaped);
 134         return 0;
 135 }
 136
 137 int pidref_get_comm(const PidRef *pid, char **ret) {
 138         _cleanup_free_ char *comm = NULL;
 139         int r;
 140
 141         if (!pidref_is_set(pid))
 142                 return -ESRCH;
 143
 144         r = pid_get_comm(pid->pid, &comm);
 145         if (r < 0)
 146                 return r;
 147
 148         r = pidref_verify(pid);
 149         if (r < 0)
 150                 return r;
 151
 152         if (ret)
 153                 *ret = TAKE_PTR(comm);
 154         return 0;
 155 }
 156
 157 static int pid_get_cmdline_nulstr(
 158                 pid_t pid,
 159                 size_t max_size,
 160                 ProcessCmdlineFlags flags,
 161                 char **ret,
 162                 size_t *ret_size) {
 163
 164         _cleanup_free_ char *t = NULL;
 165         const char *p;
 166         size_t k;
 167         int r;
 168
 169         /* Retrieves a process' command line as a "sized nulstr", i.e. possibly without the last NUL, but
 170          * with a specified size.
 171          *
 172          * If PROCESS_CMDLINE_COMM_FALLBACK is specified in flags and the process has no command line set
 173          * (the case for kernel threads), or has a command line that resolves to the empty string, will
 174          * return the "comm" name of the process instead. This will use at most _SC_ARG_MAX bytes of input
 175          * data.
 176          *
 177          * Returns an error, 0 if output was read but is truncated, 1 otherwise.
 178          */
 179
 180         p = procfs_file_alloca(pid, "cmdline");
 181         r = read_virtual_file(p, max_size, &t, &k); /* Let's assume that each input byte results in >= 1
 182                                                      * columns of output. We ignore zero-width codepoints. */
 183         if (r == -ENOENT)
 184                 return -ESRCH;
 185         if (r < 0)
 186                 return r;
 187
 188         if (k == 0) {
 189                 if (!(flags & PROCESS_CMDLINE_COMM_FALLBACK))
 190                         return -ENOENT;
 191
 192                 /* Kernel threads have no argv[] */
 193                 _cleanup_free_ char *comm = NULL;
 194
 195                 r = pid_get_comm(pid, &comm);
 196                 if (r < 0)
 197                         return r;
 198
 199                 free(t);
 200                 t = strjoin("[", comm, "]");
 201                 if (!t)
 202                         return -ENOMEM;
 203
 204                 k = strlen(t);
 205                 r = k <= max_size;
 206                 if (r == 0) /* truncation */
 207                         t[max_size] = '\0';
 208         }
 209
 210         if (ret)
 211                 *ret = TAKE_PTR(t);
 212         if (ret_size)
 213                 *ret_size = k;
 214
 215         return r;
 216 }
 217
 218 int pid_get_cmdline(pid_t pid, size_t max_columns, ProcessCmdlineFlags flags, char **ret) {
 219         _cleanup_free_ char *t = NULL;
 220         size_t k;
 221         char *ans;
 222
 223         assert(pid >= 0);
 224         assert(ret);
 225
 226         /* Retrieve and format a command line. See above for discussion of retrieval options.
 227          *
 228          * There are two main formatting modes:
 229          *
 230          * - when PROCESS_CMDLINE_QUOTE is specified, output is quoted in C/Python style. If no shell special
 231          *   characters are present, this output can be copy-pasted into the terminal to execute. UTF-8
 232          *   output is assumed.
 233          *
 234          * - otherwise, a compact non-roundtrippable form is returned. Non-UTF8 bytes are replaced by �. The
 235          *   returned string is of the specified console width at most, abbreviated with an ellipsis.
 236          *
 237          * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
 238          * PROCESS_CMDLINE_COMM_FALLBACK is not specified). Returns 0 and sets *line otherwise. */
 239
 240         int full = pid_get_cmdline_nulstr(pid, max_columns, flags, &t, &k);
 241         if (full < 0)
 242                 return full;
 243
 244         if (flags & (PROCESS_CMDLINE_QUOTE | PROCESS_CMDLINE_QUOTE_POSIX)) {
 245                 ShellEscapeFlags shflags = SHELL_ESCAPE_EMPTY |
 246                         FLAGS_SET(flags, PROCESS_CMDLINE_QUOTE_POSIX) * SHELL_ESCAPE_POSIX;
 247
 248                 assert(!(flags & PROCESS_CMDLINE_USE_LOCALE));
 249
 250                 _cleanup_strv_free_ char **args = NULL;
 251
 252                 /* Drop trailing NULs, otherwise strv_parse_nulstr() adds additional empty strings at the end.
 253                  * See also issue #21186. */
 254                 args = strv_parse_nulstr_full(t, k, /* drop_trailing_nuls = */ true);
 255                 if (!args)
 256                         return -ENOMEM;
 257
 258                 ans = quote_command_line(args, shflags);
 259                 if (!ans)
 260                         return -ENOMEM;
 261         } else {
 262                 /* Arguments are separated by NULs. Let's replace those with spaces. */
 263                 for (size_t i = 0; i < k - 1; i++)
 264                         if (t[i] == '\0')
 265                                 t[i] = ' ';
 266
 267                 delete_trailing_chars(t, WHITESPACE);
 268
 269                 bool eight_bit = (flags & PROCESS_CMDLINE_USE_LOCALE) && !is_locale_utf8();
 270
 271                 ans = escape_non_printable_full(t, max_columns,
 272                                                 eight_bit * XESCAPE_8_BIT | !full * XESCAPE_FORCE_ELLIPSIS);
 273                 if (!ans)
 274                         return -ENOMEM;
 275
 276                 ans = str_realloc(ans);
 277         }
 278
 279         *ret = ans;
 280         return 0;
 281 }
 282
 283 int pidref_get_cmdline(const PidRef *pid, size_t max_columns, ProcessCmdlineFlags flags, char **ret) {
 284         _cleanup_free_ char *s = NULL;
 285         int r;
 286
 287         if (!pidref_is_set(pid))
 288                 return -ESRCH;
 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         r = pid_get_cmdline_strv(pid->pid, flags, &args);
 333         if (r < 0)
 334                 return r;
 335
 336         r = pidref_verify(pid);
 337         if (r < 0)
 338                 return r;
 339
 340         if (ret)
 341                 *ret = TAKE_PTR(args);
 342
 343         return 0;
 344 }
 345
 346 int container_get_leader(const char *machine, pid_t *pid) {
 347         _cleanup_free_ char *s = NULL, *class = NULL;
 348         const char *p;
 349         pid_t leader;
 350         int r;
 351
 352         assert(machine);
 353         assert(pid);
 354
 355         if (streq(machine, ".host")) {
 356                 *pid = 1;
 357                 return 0;
 358         }
 359
 360         if (!hostname_is_valid(machine, 0))
 361                 return -EINVAL;
 362
 363         p = strjoina("/run/systemd/machines/", machine);
 364         r = parse_env_file(NULL, p,
 365                            "LEADER", &s,
 366                            "CLASS", &class);
 367         if (r == -ENOENT)
 368                 return -EHOSTDOWN;
 369         if (r < 0)
 370                 return r;
 371         if (!s)
 372                 return -EIO;
 373
 374         if (!streq_ptr(class, "container"))
 375                 return -EIO;
 376
 377         r = parse_pid(s, &leader);
 378         if (r < 0)
 379                 return r;
 380         if (leader <= 1)
 381                 return -EIO;
 382
 383         *pid = leader;
 384         return 0;
 385 }
 386
 387 int namespace_get_leader(pid_t pid, NamespaceType type, pid_t *ret) {
 388         int r;
 389
 390         assert(ret);
 391
 392         for (;;) {
 393                 pid_t ppid;
 394
 395                 r = get_process_ppid(pid, &ppid);
 396                 if (r < 0)
 397                         return r;
 398
 399                 r = in_same_namespace(pid, ppid, type);
 400                 if (r < 0)
 401                         return r;
 402                 if (r == 0) {
 403                         /* If the parent and the child are not in the same
 404                          * namespace, then the child is the leader we are
 405                          * looking for. */
 406                         *ret = pid;
 407                         return 0;
 408                 }
 409
 410                 pid = ppid;
 411         }
 412 }
 413
 414 int pid_is_kernel_thread(pid_t pid) {
 415         _cleanup_free_ char *line = NULL;
 416         unsigned long long flags;
 417         size_t l, i;
 418         const char *p;
 419         char *q;
 420         int r;
 421
 422         if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
 423                 return 0;
 424         if (!pid_is_valid(pid))
 425                 return -EINVAL;
 426
 427         p = procfs_file_alloca(pid, "stat");
 428         r = read_one_line_file(p, &line);
 429         if (r == -ENOENT)
 430                 return -ESRCH;
 431         if (r < 0)
 432                 return r;
 433
 434         /* Skip past the comm field */
 435         q = strrchr(line, ')');
 436         if (!q)
 437                 return -EINVAL;
 438         q++;
 439
 440         /* Skip 6 fields to reach the flags field */
 441         for (i = 0; i < 6; i++) {
 442                 l = strspn(q, WHITESPACE);
 443                 if (l < 1)
 444                         return -EINVAL;
 445                 q += l;
 446
 447                 l = strcspn(q, WHITESPACE);
 448                 if (l < 1)
 449                         return -EINVAL;
 450                 q += l;
 451         }
 452
 453         /* Skip preceding whitespace */
 454         l = strspn(q, WHITESPACE);
 455         if (l < 1)
 456                 return -EINVAL;
 457         q += l;
 458
 459         /* Truncate the rest */
 460         l = strcspn(q, WHITESPACE);
 461         if (l < 1)
 462                 return -EINVAL;
 463         q[l] = 0;
 464
 465         r = safe_atollu(q, &flags);
 466         if (r < 0)
 467                 return r;
 468
 469         return !!(flags & PF_KTHREAD);
 470 }
 471
 472 int pidref_is_kernel_thread(const PidRef *pid) {
 473         int result, r;
 474
 475         if (!pidref_is_set(pid))
 476                 return -ESRCH;
 477
 478         result = pid_is_kernel_thread(pid->pid);
 479         if (result < 0)
 480                 return result;
 481
 482         r = pidref_verify(pid); /* Verify that the PID wasn't reused since */
 483         if (r < 0)
 484                 return r;
 485
 486         return result;
 487 }
 488
 489 int get_process_capeff(pid_t pid, char **ret) {
 490         const char *p;
 491         int r;
 492
 493         assert(pid >= 0);
 494         assert(ret);
 495
 496         p = procfs_file_alloca(pid, "status");
 497
 498         r = get_proc_field(p, "CapEff", WHITESPACE, ret);
 499         if (r == -ENOENT)
 500                 return -ESRCH;
 501
 502         return r;
 503 }
 504
 505 static int get_process_link_contents(pid_t pid, const char *proc_file, char **ret) {
 506         const char *p;
 507         int r;
 508
 509         assert(proc_file);
 510
 511         p = procfs_file_alloca(pid, proc_file);
 512
 513         r = readlink_malloc(p, ret);
 514         return (r == -ENOENT && proc_mounted() > 0) ? -ESRCH : r;
 515 }
 516
 517 int get_process_exe(pid_t pid, char **ret) {
 518         char *d;
 519         int r;
 520
 521         assert(pid >= 0);
 522
 523         r = get_process_link_contents(pid, "exe", ret);
 524         if (r < 0)
 525                 return r;
 526
 527         if (ret) {
 528                 d = endswith(*ret, " (deleted)");
 529                 if (d)
 530                         *d = '\0';
 531         }
 532
 533         return 0;
 534 }
 535
 536 static int get_process_id(pid_t pid, const char *field, uid_t *ret) {
 537         _cleanup_fclose_ FILE *f = NULL;
 538         const char *p;
 539         int r;
 540
 541         assert(field);
 542         assert(ret);
 543
 544         if (pid < 0)
 545                 return -EINVAL;
 546
 547         p = procfs_file_alloca(pid, "status");
 548         r = fopen_unlocked(p, "re", &f);
 549         if (r == -ENOENT)
 550                 return -ESRCH;
 551         if (r < 0)
 552                 return r;
 553
 554         for (;;) {
 555                 _cleanup_free_ char *line = NULL;
 556                 char *l;
 557
 558                 r = read_stripped_line(f, LONG_LINE_MAX, &line);
 559                 if (r < 0)
 560                         return r;
 561                 if (r == 0)
 562                         break;
 563
 564                 l = startswith(line, field);
 565                 if (l) {
 566                         l += strspn(l, WHITESPACE);
 567
 568                         l[strcspn(l, WHITESPACE)] = 0;
 569
 570                         return parse_uid(l, ret);
 571                 }
 572         }
 573
 574         return -EIO;
 575 }
 576
 577 int pid_get_uid(pid_t pid, uid_t *ret) {
 578         assert(ret);
 579
 580         if (pid == 0 || pid == getpid_cached()) {
 581                 *ret = getuid();
 582                 return 0;
 583         }
 584
 585         return get_process_id(pid, "Uid:", ret);
 586 }
 587
 588 int pidref_get_uid(const PidRef *pid, uid_t *ret) {
 589         uid_t uid;
 590         int r;
 591
 592         if (!pidref_is_set(pid))
 593                 return -ESRCH;
 594
 595         r = pid_get_uid(pid->pid, &uid);
 596         if (r < 0)
 597                 return r;
 598
 599         r = pidref_verify(pid);
 600         if (r < 0)
 601                 return r;
 602
 603         if (ret)
 604                 *ret = uid;
 605         return 0;
 606 }
 607
 608 int get_process_gid(pid_t pid, gid_t *ret) {
 609
 610         if (pid == 0 || pid == getpid_cached()) {
 611                 *ret = getgid();
 612                 return 0;
 613         }
 614
 615         assert_cc(sizeof(uid_t) == sizeof(gid_t));
 616         return get_process_id(pid, "Gid:", ret);
 617 }
 618
 619 int get_process_cwd(pid_t pid, char **ret) {
 620         assert(pid >= 0);
 621
 622         if (pid == 0 || pid == getpid_cached())
 623                 return safe_getcwd(ret);
 624
 625         return get_process_link_contents(pid, "cwd", ret);
 626 }
 627
 628 int get_process_root(pid_t pid, char **ret) {
 629         assert(pid >= 0);
 630         return get_process_link_contents(pid, "root", ret);
 631 }
 632
 633 #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U)
 634
 635 int get_process_environ(pid_t pid, char **ret) {
 636         _cleanup_fclose_ FILE *f = NULL;
 637         _cleanup_free_ char *outcome = NULL;
 638         size_t sz = 0;
 639         const char *p;
 640         int r;
 641
 642         assert(pid >= 0);
 643         assert(ret);
 644
 645         p = procfs_file_alloca(pid, "environ");
 646
 647         r = fopen_unlocked(p, "re", &f);
 648         if (r == -ENOENT)
 649                 return -ESRCH;
 650         if (r < 0)
 651                 return r;
 652
 653         for (;;) {
 654                 char c;
 655
 656                 if (sz >= ENVIRONMENT_BLOCK_MAX)
 657                         return -ENOBUFS;
 658
 659                 if (!GREEDY_REALLOC(outcome, sz + 5))
 660                         return -ENOMEM;
 661
 662                 r = safe_fgetc(f, &c);
 663                 if (r < 0)
 664                         return r;
 665                 if (r == 0)
 666                         break;
 667
 668                 if (c == '\0')
 669                         outcome[sz++] = '\n';
 670                 else
 671                         sz += cescape_char(c, outcome + sz);
 672         }
 673
 674         outcome[sz] = '\0';
 675         *ret = TAKE_PTR(outcome);
 676
 677         return 0;
 678 }
 679
 680 int get_process_ppid(pid_t pid, pid_t *ret) {
 681         _cleanup_free_ char *line = NULL;
 682         unsigned long ppid;
 683         const char *p;
 684         int r;
 685
 686         assert(pid >= 0);
 687
 688         if (pid == 0 || pid == getpid_cached()) {
 689                 if (ret)
 690                         *ret = getppid();
 691                 return 0;
 692         }
 693
 694         if (pid == 1) /* PID 1 has no parent, shortcut this case */
 695                 return -EADDRNOTAVAIL;
 696
 697         p = procfs_file_alloca(pid, "stat");
 698         r = read_one_line_file(p, &line);
 699         if (r == -ENOENT)
 700                 return -ESRCH;
 701         if (r < 0)
 702                 return r;
 703
 704         /* Let's skip the pid and comm fields. The latter is enclosed in () but does not escape any () in its
 705          * value, so let's skip over it manually */
 706
 707         p = strrchr(line, ')');
 708         if (!p)
 709                 return -EIO;
 710
 711         p++;
 712
 713         if (sscanf(p, " "
 714                    "%*c "  /* state */
 715                    "%lu ", /* ppid */
 716                    &ppid) != 1)
 717                 return -EIO;
 718
 719         /* If ppid is zero the process has no parent. Which might be the case for PID 1 but also for
 720          * processes originating in other namespaces that are inserted into a pidns. Return a recognizable
 721          * error in this case. */
 722         if (ppid == 0)
 723                 return -EADDRNOTAVAIL;
 724
 725         if ((pid_t) ppid < 0 || (unsigned long) (pid_t) ppid != ppid)
 726                 return -ERANGE;
 727
 728         if (ret)
 729                 *ret = (pid_t) ppid;
 730
 731         return 0;
 732 }
 733
 734 int pid_get_start_time(pid_t pid, uint64_t *ret) {
 735         _cleanup_free_ char *line = NULL;
 736         const char *p;
 737         int r;
 738
 739         assert(pid >= 0);
 740
 741         p = procfs_file_alloca(pid, "stat");
 742         r = read_one_line_file(p, &line);
 743         if (r == -ENOENT)
 744                 return -ESRCH;
 745         if (r < 0)
 746                 return r;
 747
 748         /* Let's skip the pid and comm fields. The latter is enclosed in () but does not escape any () in its
 749          * value, so let's skip over it manually */
 750
 751         p = strrchr(line, ')');
 752         if (!p)
 753                 return -EIO;
 754
 755         p++;
 756
 757         unsigned long llu;
 758
 759         if (sscanf(p, " "
 760                    "%*c "  /* state */
 761                    "%*u " /* ppid */
 762                    "%*u " /* pgrp */
 763                    "%*u " /* session */
 764                    "%*u " /* tty_nr */
 765                    "%*u " /* tpgid */
 766                    "%*u " /* flags */
 767                    "%*u " /* minflt */
 768                    "%*u " /* cminflt */
 769                    "%*u " /* majflt */
 770                    "%*u " /* cmajflt */
 771                    "%*u " /* utime */
 772                    "%*u " /* stime */
 773                    "%*u " /* cutime */
 774                    "%*u " /* cstime */
 775                    "%*i " /* priority */
 776                    "%*i " /* nice */
 777                    "%*u " /* num_threads */
 778                    "%*u " /* itrealvalue */
 779                    "%lu ", /* starttime */
 780                    &llu) != 1)
 781                 return -EIO;
 782
 783         if (ret)
 784                 *ret = llu;
 785
 786         return 0;
 787 }
 788
 789 int pidref_get_start_time(const PidRef *pid, uint64_t *ret) {
 790         uint64_t t;
 791         int r;
 792
 793         if (!pidref_is_set(pid))
 794                 return -ESRCH;
 795
 796         r = pid_get_start_time(pid->pid, ret ? &t : NULL);
 797         if (r < 0)
 798                 return r;
 799
 800         r = pidref_verify(pid);
 801         if (r < 0)
 802                 return r;
 803
 804         if (ret)
 805                 *ret = t;
 806
 807         return 0;
 808 }
 809
 810 int get_process_umask(pid_t pid, mode_t *ret) {
 811         _cleanup_free_ char *m = NULL;
 812         const char *p;
 813         int r;
 814
 815         assert(pid >= 0);
 816         assert(ret);
 817
 818         p = procfs_file_alloca(pid, "status");
 819
 820         r = get_proc_field(p, "Umask", WHITESPACE, &m);
 821         if (r == -ENOENT)
 822                 return -ESRCH;
 823         if (r < 0)
 824                 return r;
 825
 826         return parse_mode(m, ret);
 827 }
 828
 829 int wait_for_terminate(pid_t pid, siginfo_t *status) {
 830         siginfo_t dummy;
 831
 832         assert(pid >= 1);
 833
 834         if (!status)
 835                 status = &dummy;
 836
 837         for (;;) {
 838                 zero(*status);
 839
 840                 if (waitid(P_PID, pid, status, WEXITED) < 0) {
 841
 842                         if (errno == EINTR)
 843                                 continue;
 844
 845                         return negative_errno();
 846                 }
 847
 848                 return 0;
 849         }
 850 }
 851
 852 /*
 853  * Return values:
 854  * < 0 : wait_for_terminate() failed to get the state of the
 855  *       process, the process was terminated by a signal, or
 856  *       failed for an unknown reason.
 857  * >=0 : The process terminated normally, and its exit code is
 858  *       returned.
 859  *
 860  * That is, success is indicated by a return value of zero, and an
 861  * error is indicated by a non-zero value.
 862  *
 863  * A warning is emitted if the process terminates abnormally,
 864  * and also if it returns non-zero unless check_exit_code is true.
 865  */
 866 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
 867         _cleanup_free_ char *buffer = NULL;
 868         siginfo_t status;
 869         int r, prio;
 870
 871         assert(pid > 1);
 872
 873         if (!name) {
 874                 r = pid_get_comm(pid, &buffer);
 875                 if (r < 0)
 876                         log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
 877                 else
 878                         name = buffer;
 879         }
 880
 881         prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
 882
 883         r = wait_for_terminate(pid, &status);
 884         if (r < 0)
 885                 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
 886
 887         if (status.si_code == CLD_EXITED) {
 888                 if (status.si_status != EXIT_SUCCESS)
 889                         log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
 890                                  "%s failed with exit status %i.", strna(name), status.si_status);
 891                 else
 892                         log_debug("%s succeeded.", name);
 893
 894                 return status.si_status;
 895
 896         } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
 897
 898                 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
 899                 return -EPROTO;
 900         }
 901
 902         log_full(prio, "%s failed due to unknown reason.", strna(name));
 903         return -EPROTO;
 904 }
 905
 906 /*
 907  * Return values:
 908  *
 909  * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process
 910  *       was terminated by a signal, or failed for an unknown reason.
 911  *
 912  * >=0 : The process terminated normally with no failures.
 913  *
 914  * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure
 915  * states are indicated by error is indicated by a non-zero value.
 916  *
 917  * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off
 918  * to remain entirely race-free.
 919  */
 920 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
 921         sigset_t mask;
 922         int r;
 923         usec_t until;
 924
 925         assert_se(sigemptyset(&mask) == 0);
 926         assert_se(sigaddset(&mask, SIGCHLD) == 0);
 927
 928         /* Drop into a sigtimewait-based timeout. Waiting for the
 929          * pid to exit. */
 930         until = usec_add(now(CLOCK_MONOTONIC), timeout);
 931         for (;;) {
 932                 usec_t n;
 933                 siginfo_t status = {};
 934
 935                 n = now(CLOCK_MONOTONIC);
 936                 if (n >= until)
 937                         break;
 938
 939                 r = RET_NERRNO(sigtimedwait(&mask, NULL, TIMESPEC_STORE(until - n)));
 940                 /* Assuming we woke due to the child exiting. */
 941                 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
 942                         if (status.si_pid == pid) {
 943                                 /* This is the correct child. */
 944                                 if (status.si_code == CLD_EXITED)
 945                                         return status.si_status == 0 ? 0 : -EPROTO;
 946                                 else
 947                                         return -EPROTO;
 948                         }
 949                 }
 950                 /* Not the child, check for errors and proceed appropriately */
 951                 if (r < 0) {
 952                         switch (r) {
 953                         case -EAGAIN:
 954                                 /* Timed out, child is likely hung. */
 955                                 return -ETIMEDOUT;
 956                         case -EINTR:
 957                                 /* Received a different signal and should retry */
 958                                 continue;
 959                         default:
 960                                 /* Return any unexpected errors */
 961                                 return r;
 962                         }
 963                 }
 964         }
 965
 966         return -EPROTO;
 967 }
 968
 969 void sigkill_wait(pid_t pid) {
 970         assert(pid > 1);
 971
 972         (void) kill(pid, SIGKILL);
 973         (void) wait_for_terminate(pid, NULL);
 974 }
 975
 976 void sigkill_waitp(pid_t *pid) {
 977         PROTECT_ERRNO;
 978
 979         if (!pid)
 980                 return;
 981         if (*pid <= 1)
 982                 return;
 983
 984         sigkill_wait(*pid);
 985 }
 986
 987 void sigterm_wait(pid_t pid) {
 988         assert(pid > 1);
 989
 990         (void) kill_and_sigcont(pid, SIGTERM);
 991         (void) wait_for_terminate(pid, NULL);
 992 }
 993
 994 void sigkill_nowait(pid_t pid) {
 995         assert(pid > 1);
 996
 997         (void) kill(pid, SIGKILL);
 998 }
 999
1000 void sigkill_nowaitp(pid_t *pid) {
1001         PROTECT_ERRNO;
1002
1003         if (!pid)
1004                 return;
1005         if (*pid <= 1)
1006                 return;
1007
1008         sigkill_nowait(*pid);
1009 }
1010
1011 int kill_and_sigcont(pid_t pid, int sig) {
1012         int r;
1013
1014         r = RET_NERRNO(kill(pid, sig));
1015
1016         /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
1017          * affected by a process being suspended anyway. */
1018         if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
1019                 (void) kill(pid, SIGCONT);
1020
1021         return r;
1022 }
1023
1024 int getenv_for_pid(pid_t pid, const char *field, char **ret) {
1025         _cleanup_fclose_ FILE *f = NULL;
1026         const char *path;
1027         size_t sum = 0;
1028         int r;
1029
1030         assert(pid >= 0);
1031         assert(field);
1032         assert(ret);
1033
1034         if (pid == 0 || pid == getpid_cached())
1035                 return strdup_to_full(ret, getenv(field));
1036
1037         if (!pid_is_valid(pid))
1038                 return -EINVAL;
1039
1040         path = procfs_file_alloca(pid, "environ");
1041
1042         r = fopen_unlocked(path, "re", &f);
1043         if (r == -ENOENT)
1044                 return -ESRCH;
1045         if (r < 0)
1046                 return r;
1047
1048         for (;;) {
1049                 _cleanup_free_ char *line = NULL;
1050                 const char *match;
1051
1052                 if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */
1053                         return -ENOBUFS;
1054
1055                 r = read_nul_string(f, LONG_LINE_MAX, &line);
1056                 if (r < 0)
1057                         return r;
1058                 if (r == 0)  /* EOF */
1059                         break;
1060
1061                 sum += r;
1062
1063                 match = startswith(line, field);
1064                 if (match && *match == '=')
1065                         return strdup_to_full(ret, match + 1);
1066         }
1067
1068         *ret = NULL;
1069         return 0;
1070 }
1071
1072 int pid_is_my_child(pid_t pid) {
1073         pid_t ppid;
1074         int r;
1075
1076         if (pid < 0)
1077                 return -ESRCH;
1078
1079         if (pid <= 1)
1080                 return false;
1081
1082         r = get_process_ppid(pid, &ppid);
1083         if (r < 0)
1084                 return r;
1085
1086         return ppid == getpid_cached();
1087 }
1088
1089 int pidref_is_my_child(const PidRef *pid) {
1090         int r, result;
1091
1092         if (!pidref_is_set(pid))
1093                 return -ESRCH;
1094
1095         result = pid_is_my_child(pid->pid);
1096         if (result < 0)
1097                 return result;
1098
1099         r = pidref_verify(pid);
1100         if (r < 0)
1101                 return r;
1102
1103         return result;
1104 }
1105
1106 int pid_is_unwaited(pid_t pid) {
1107         /* Checks whether a PID is still valid at all, including a zombie */
1108
1109         if (pid < 0)
1110                 return -ESRCH;
1111
1112         if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
1113                 return true;
1114
1115         if (pid == getpid_cached())
1116                 return true;
1117
1118         if (kill(pid, 0) >= 0)
1119                 return true;
1120
1121         return errno != ESRCH;
1122 }
1123
1124 int pidref_is_unwaited(const PidRef *pid) {
1125         int r;
1126
1127         if (!pidref_is_set(pid))
1128                 return -ESRCH;
1129
1130         if (pid->pid == 1 || pidref_is_self(pid))
1131                 return true;
1132
1133         r = pidref_kill(pid, 0);
1134         if (r == -ESRCH)
1135                 return false;
1136         if (r < 0)
1137                 return r;
1138
1139         return true;
1140 }
1141
1142 int pid_is_alive(pid_t pid) {
1143         int r;
1144
1145         /* Checks whether a PID is still valid and not a zombie */
1146
1147         if (pid < 0)
1148                 return -ESRCH;
1149
1150         if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
1151                 return true;
1152
1153         if (pid == getpid_cached())
1154                 return true;
1155
1156         r = get_process_state(pid);
1157         if (r == -ESRCH)
1158                 return false;
1159         if (r < 0)
1160                 return r;
1161
1162         return r != 'Z';
1163 }
1164
1165 int pidref_is_alive(const PidRef *pidref) {
1166         int r, result;
1167
1168         if (!pidref_is_set(pidref))
1169                 return -ESRCH;
1170
1171         result = pid_is_alive(pidref->pid);
1172         if (result < 0) {
1173                 assert(result != -ESRCH);
1174                 return result;
1175         }
1176
1177         r = pidref_verify(pidref);
1178         if (r == -ESRCH)
1179                 return false;
1180         if (r < 0)
1181                 return r;
1182
1183         return result;
1184 }
1185
1186 int pid_from_same_root_fs(pid_t pid) {
1187         const char *root;
1188
1189         if (pid < 0)
1190                 return false;
1191
1192         if (pid == 0 || pid == getpid_cached())
1193                 return true;
1194
1195         root = procfs_file_alloca(pid, "root");
1196
1197         return inode_same(root, "/proc/1/root", 0);
1198 }
1199
1200 bool is_main_thread(void) {
1201         static thread_local int cached = 0;
1202
1203         if (_unlikely_(cached == 0))
1204                 cached = getpid_cached() == gettid() ? 1 : -1;
1205
1206         return cached > 0;
1207 }
1208
1209 bool oom_score_adjust_is_valid(int oa) {
1210         return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
1211 }
1212
1213 unsigned long personality_from_string(const char *p) {
1214         Architecture architecture;
1215
1216         if (!p)
1217                 return PERSONALITY_INVALID;
1218
1219         /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
1220          * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
1221          * the same register size. */
1222
1223         architecture = architecture_from_string(p);
1224         if (architecture < 0)
1225                 return PERSONALITY_INVALID;
1226
1227         if (architecture == native_architecture())
1228                 return PER_LINUX;
1229 #ifdef ARCHITECTURE_SECONDARY
1230         if (architecture == ARCHITECTURE_SECONDARY)
1231                 return PER_LINUX32;
1232 #endif
1233
1234         return PERSONALITY_INVALID;
1235 }
1236
1237 const char* personality_to_string(unsigned long p) {
1238         Architecture architecture = _ARCHITECTURE_INVALID;
1239
1240         if (p == PER_LINUX)
1241                 architecture = native_architecture();
1242 #ifdef ARCHITECTURE_SECONDARY
1243         else if (p == PER_LINUX32)
1244                 architecture = ARCHITECTURE_SECONDARY;
1245 #endif
1246
1247         if (architecture < 0)
1248                 return NULL;
1249
1250         return architecture_to_string(architecture);
1251 }
1252
1253 int safe_personality(unsigned long p) {
1254         int ret;
1255
1256         /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1257          * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1258          * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1259          * the return value indicating the same issue, so that we are definitely on the safe side.
1260          *
1261          * See https://github.com/systemd/systemd/issues/6737 */
1262
1263         errno = 0;
1264         ret = personality(p);
1265         if (ret < 0) {
1266                 if (errno != 0)
1267                         return -errno;
1268
1269                 errno = -ret;
1270         }
1271
1272         return ret;
1273 }
1274
1275 int opinionated_personality(unsigned long *ret) {
1276         int current;
1277
1278         /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1279          * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1280          * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1281
1282         current = safe_personality(PERSONALITY_INVALID);
1283         if (current < 0)
1284                 return current;
1285
1286         if (((unsigned long) current & OPINIONATED_PERSONALITY_MASK) == PER_LINUX32)
1287                 *ret = PER_LINUX32;
1288         else
1289                 *ret = PER_LINUX;
1290
1291         return 0;
1292 }
1293
1294 void valgrind_summary_hack(void) {
1295 #if HAVE_VALGRIND_VALGRIND_H
1296         if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1297                 pid_t pid;
1298                 pid = raw_clone(SIGCHLD);
1299                 if (pid < 0)
1300                         log_struct_errno(
1301                                 LOG_EMERG, errno,
1302                                 "MESSAGE_ID=" SD_MESSAGE_VALGRIND_HELPER_FORK_STR,
1303                                 LOG_MESSAGE( "Failed to fork off valgrind helper: %m"));
1304                 else if (pid == 0)
1305                         exit(EXIT_SUCCESS);
1306                 else {
1307                         log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1308                         (void) wait_for_terminate(pid, NULL);
1309                 }
1310         }
1311 #endif
1312 }
1313
1314 int pid_compare_func(const pid_t *a, const pid_t *b) {
1315         /* Suitable for usage in qsort() */
1316         return CMP(*a, *b);
1317 }
1318
1319 /* The cached PID, possible values:
1320  *
1321  *     == UNSET [0]  → cache not initialized yet
1322  *     == BUSY [-1]  → some thread is initializing it at the moment
1323  *     any other     → the cached PID
1324  */
1325
1326 #define CACHED_PID_UNSET ((pid_t) 0)
1327 #define CACHED_PID_BUSY ((pid_t) -1)
1328
1329 static pid_t cached_pid = CACHED_PID_UNSET;
1330
1331 void reset_cached_pid(void) {
1332         /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1333         cached_pid = CACHED_PID_UNSET;
1334 }
1335
1336 pid_t getpid_cached(void) {
1337         static bool installed = false;
1338         pid_t current_value = CACHED_PID_UNSET;
1339
1340         /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1341          * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1342          * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1343          * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1344          *
1345          * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1346          * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1347          */
1348
1349         (void) __atomic_compare_exchange_n(
1350                         &cached_pid,
1351                         &current_value,
1352                         CACHED_PID_BUSY,
1353                         false,
1354                         __ATOMIC_SEQ_CST,
1355                         __ATOMIC_SEQ_CST);
1356
1357         switch (current_value) {
1358
1359         case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1360                 pid_t new_pid;
1361
1362                 new_pid = raw_getpid();
1363
1364                 if (!installed) {
1365                         /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
1366                          * only half-documented (glibc doesn't document it but LSB does — though only superficially)
1367                          * we'll check for errors only in the most generic fashion possible. */
1368
1369                         if (pthread_atfork(NULL, NULL, reset_cached_pid) != 0) {
1370                                 /* OOM? Let's try again later */
1371                                 cached_pid = CACHED_PID_UNSET;
1372                                 return new_pid;
1373                         }
1374
1375                         installed = true;
1376                 }
1377
1378                 cached_pid = new_pid;
1379                 return new_pid;
1380         }
1381
1382         case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1383                 return raw_getpid();
1384
1385         default: /* Properly initialized */
1386                 return current_value;
1387         }
1388 }
1389
1390 int must_be_root(void) {
1391
1392         if (geteuid() == 0)
1393                 return 0;
1394
1395         return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root.");
1396 }
1397
1398 static void restore_sigsetp(sigset_t **ssp) {
1399         if (*ssp)
1400                 (void) sigprocmask(SIG_SETMASK, *ssp, NULL);
1401 }
1402
1403 pid_t clone_with_nested_stack(int (*fn)(void *), int flags, void *userdata) {
1404         size_t ps;
1405         pid_t pid;
1406         void *mystack;
1407
1408         /* A wrapper around glibc's clone() call that automatically sets up a "nested" stack. Only supports
1409          * invocations without CLONE_VM, so that we can continue to use the parent's stack mapping.
1410          *
1411          * Note: glibc's clone() wrapper does not synchronize malloc() locks. This means that if the parent
1412          * is threaded these locks will be in an undefined state in the child, and hence memory allocations
1413          * are likely going to run into deadlocks. Hence: if you use this function make sure your parent is
1414          * strictly single-threaded or your child never calls malloc(). */
1415
1416         assert((flags & (CLONE_VM|CLONE_PARENT_SETTID|CLONE_CHILD_SETTID|
1417                          CLONE_CHILD_CLEARTID|CLONE_SETTLS)) == 0);
1418
1419         /* We allocate some space on the stack to use as the stack for the child (hence "nested"). Note that
1420          * the net effect is that the child will have the start of its stack inside the stack of the parent,
1421          * but since they are a CoW copy of each other that's fine. We allocate one page-aligned page. But
1422          * since we don't want to deal with differences between systems where the stack grows backwards or
1423          * forwards we'll allocate one more and place the stack address in the middle. Except that we also
1424          * want it page aligned, hence we'll allocate one page more. Makes 3. */
1425
1426         ps = page_size();
1427         mystack = alloca(ps*3);
1428         mystack = (uint8_t*) mystack + ps; /* move pointer one page ahead since stacks usually grow backwards */
1429         mystack = (void*) ALIGN_TO((uintptr_t) mystack, ps); /* align to page size (moving things further ahead) */
1430
1431 #if HAVE_CLONE
1432         pid = clone(fn, mystack, flags, userdata);
1433 #else
1434         pid = __clone2(fn, mystack, ps, flags, userdata);
1435 #endif
1436         if (pid < 0)
1437                 return -errno;
1438
1439         return pid;
1440 }
1441
1442 static int fork_flags_to_signal(ForkFlags flags) {
1443         return (flags & FORK_DEATHSIG_SIGTERM) ? SIGTERM :
1444                 (flags & FORK_DEATHSIG_SIGINT) ? SIGINT :
1445                                                  SIGKILL;
1446 }
1447
1448 int safe_fork_full(
1449                 const char *name,
1450                 const int stdio_fds[3],
1451                 int except_fds[],
1452                 size_t n_except_fds,
1453                 ForkFlags flags,
1454                 pid_t *ret_pid) {
1455
1456         pid_t original_pid, pid;
1457         sigset_t saved_ss, ss;
1458         _unused_ _cleanup_(restore_sigsetp) sigset_t *saved_ssp = NULL;
1459         bool block_signals = false, block_all = false, intermediary = false;
1460         int prio, r;
1461
1462         assert(!FLAGS_SET(flags, FORK_DETACH) || !ret_pid);
1463         assert(!FLAGS_SET(flags, FORK_DETACH|FORK_WAIT));
1464
1465         /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1466          * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1467
1468         prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1469
1470         original_pid = getpid_cached();
1471
1472         if (flags & FORK_FLUSH_STDIO) {
1473                 fflush(stdout);
1474                 fflush(stderr); /* This one shouldn't be necessary, stderr should be unbuffered anyway, but let's better be safe than sorry */
1475         }
1476
1477         if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGINT)) {
1478                 /* We temporarily block all signals, so that the new child has them blocked initially. This
1479                  * way, we can be sure that SIGTERMs are not lost we might send to the child. (Note that for
1480                  * FORK_DEATHSIG_SIGKILL we don't bother, since it cannot be blocked anyway.) */
1481
1482                 assert_se(sigfillset(&ss) >= 0);
1483                 block_signals = block_all = true;
1484
1485         } else if (flags & FORK_WAIT) {
1486                 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1487
1488                 assert_se(sigemptyset(&ss) >= 0);
1489                 assert_se(sigaddset(&ss, SIGCHLD) >= 0);
1490                 block_signals = true;
1491         }
1492
1493         if (block_signals) {
1494                 if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
1495                         return log_full_errno(prio, errno, "Failed to set signal mask: %m");
1496                 saved_ssp = &saved_ss;
1497         }
1498
1499         if (FLAGS_SET(flags, FORK_DETACH)) {
1500                 assert(!FLAGS_SET(flags, FORK_WAIT));
1501                 assert(!ret_pid);
1502
1503                 /* Fork off intermediary child if needed */
1504
1505                 r = is_reaper_process();
1506                 if (r < 0)
1507                         return log_full_errno(prio, r, "Failed to determine if we are a reaper process: %m");
1508
1509                 if (!r) {
1510                         /* Not a reaper process, hence do a double fork() so we are reparented to one */
1511
1512                         pid = fork();
1513                         if (pid < 0)
1514                                 return log_full_errno(prio, errno, "Failed to fork off '%s': %m", strna(name));
1515                         if (pid > 0) {
1516                                 log_debug("Successfully forked off intermediary '%s' as PID " PID_FMT ".", strna(name), pid);
1517                                 return 1; /* return in the parent */
1518                         }
1519
1520                         intermediary = true;
1521                 }
1522         }
1523
1524         if ((flags & (FORK_NEW_MOUNTNS|FORK_NEW_USERNS|FORK_NEW_NETNS)) != 0)
1525                 pid = raw_clone(SIGCHLD|
1526                                 (FLAGS_SET(flags, FORK_NEW_MOUNTNS) ? CLONE_NEWNS : 0) |
1527                                 (FLAGS_SET(flags, FORK_NEW_USERNS) ? CLONE_NEWUSER : 0) |
1528                                 (FLAGS_SET(flags, FORK_NEW_NETNS) ? CLONE_NEWNET : 0));
1529         else
1530                 pid = fork();
1531         if (pid < 0)
1532                 return log_full_errno(prio, errno, "Failed to fork off '%s': %m", strna(name));
1533         if (pid > 0) {
1534
1535                 /* If we are in the intermediary process, exit now */
1536                 if (intermediary)
1537                         _exit(EXIT_SUCCESS);
1538
1539                 /* We are in the parent process */
1540                 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1541
1542                 if (flags & FORK_WAIT) {
1543                         if (block_all) {
1544                                 /* undo everything except SIGCHLD */
1545                                 ss = saved_ss;
1546                                 assert_se(sigaddset(&ss, SIGCHLD) >= 0);
1547                                 (void) sigprocmask(SIG_SETMASK, &ss, NULL);
1548                         }
1549
1550                         r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1551                         if (r < 0)
1552                                 return r;
1553                         if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1554                                 return -EPROTO;
1555                 }
1556
1557                 if (ret_pid)
1558                         *ret_pid = pid;
1559
1560                 return 1;
1561         }
1562
1563         /* We are in the child process */
1564
1565         /* Restore signal mask manually */
1566         saved_ssp = NULL;
1567
1568         if (flags & FORK_REOPEN_LOG) {
1569                 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1570                 log_close();
1571                 log_set_open_when_needed(true);
1572                 log_settle_target();
1573         }
1574
1575         if (name) {
1576                 r = rename_process(name);
1577                 if (r < 0)
1578                         log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1579                                        r, "Failed to rename process, ignoring: %m");
1580         }
1581
1582         if (flags & (FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGINT|FORK_DEATHSIG_SIGKILL))
1583                 if (prctl(PR_SET_PDEATHSIG, fork_flags_to_signal(flags)) < 0) {
1584                         log_full_errno(prio, errno, "Failed to set death signal: %m");
1585                         _exit(EXIT_FAILURE);
1586                 }
1587
1588         if (flags & FORK_RESET_SIGNALS) {
1589                 r = reset_all_signal_handlers();
1590                 if (r < 0) {
1591                         log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1592                         _exit(EXIT_FAILURE);
1593                 }
1594
1595                 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1596                 r = reset_signal_mask();
1597                 if (r < 0) {
1598                         log_full_errno(prio, r, "Failed to reset signal mask: %m");
1599                         _exit(EXIT_FAILURE);
1600                 }
1601         } else if (block_signals) { /* undo what we did above */
1602                 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1603                         log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1604                         _exit(EXIT_FAILURE);
1605                 }
1606         }
1607
1608         if (flags & (FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGKILL|FORK_DEATHSIG_SIGINT)) {
1609                 pid_t ppid;
1610                 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1611                  * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1612
1613                 ppid = getppid();
1614                 if (ppid == 0)
1615                         /* Parent is in a different PID namespace. */;
1616                 else if (ppid != original_pid) {
1617                         int sig = fork_flags_to_signal(flags);
1618                         log_debug("Parent died early, raising %s.", signal_to_string(sig));
1619                         (void) raise(sig);
1620                         _exit(EXIT_FAILURE);
1621                 }
1622         }
1623
1624         if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) {
1625                 /* Optionally, make sure we never propagate mounts to the host. */
1626                 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
1627                         log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m");
1628                         _exit(EXIT_FAILURE);
1629                 }
1630         }
1631
1632         if (FLAGS_SET(flags, FORK_PRIVATE_TMP)) {
1633                 assert(FLAGS_SET(flags, FORK_NEW_MOUNTNS));
1634
1635                 /* Optionally, overmount new tmpfs instance on /tmp/. */
1636                 r = mount_nofollow("tmpfs", "/tmp", "tmpfs",
1637                                    MS_NOSUID|MS_NODEV,
1638                                    "mode=01777" TMPFS_LIMITS_RUN);
1639                 if (r < 0) {
1640                         log_full_errno(prio, r, "Failed to overmount /tmp/: %m");
1641                         _exit(EXIT_FAILURE);
1642                 }
1643         }
1644
1645         if (flags & FORK_REARRANGE_STDIO) {
1646                 if (stdio_fds) {
1647                         r = rearrange_stdio(stdio_fds[0], stdio_fds[1], stdio_fds[2]);
1648                         if (r < 0) {
1649                                 log_full_errno(prio, r, "Failed to rearrange stdio fds: %m");
1650                                 _exit(EXIT_FAILURE);
1651                         }
1652
1653                         /* Turn off O_NONBLOCK on the fdio fds, in case it was left on */
1654                         stdio_disable_nonblock();
1655                 } else {
1656                         r = make_null_stdio();
1657                         if (r < 0) {
1658                                 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1659                                 _exit(EXIT_FAILURE);
1660                         }
1661                 }
1662         } else if (flags & FORK_STDOUT_TO_STDERR) {
1663                 if (dup2(STDERR_FILENO, STDOUT_FILENO) < 0) {
1664                         log_full_errno(prio, errno, "Failed to connect stdout to stderr: %m");
1665                         _exit(EXIT_FAILURE);
1666                 }
1667         }
1668
1669         if (flags & FORK_CLOSE_ALL_FDS) {
1670                 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1671                 log_close();
1672
1673                 r = close_all_fds(except_fds, n_except_fds);
1674                 if (r < 0) {
1675                         log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1676                         _exit(EXIT_FAILURE);
1677                 }
1678         }
1679
1680         if (flags & FORK_PACK_FDS) {
1681                 /* FORK_CLOSE_ALL_FDS ensures that except_fds are the only FDs >= 3 that are
1682                  * open, this is including the log. This is required by pack_fds, which will
1683                  * get stuck in an infinite loop of any FDs other than except_fds are open. */
1684                 assert(FLAGS_SET(flags, FORK_CLOSE_ALL_FDS));
1685
1686                 r = pack_fds(except_fds, n_except_fds);
1687                 if (r < 0) {
1688                         log_full_errno(prio, r, "Failed to pack file descriptors: %m");
1689                         _exit(EXIT_FAILURE);
1690                 }
1691         }
1692
1693         if (flags & FORK_CLOEXEC_OFF) {
1694                 r = fd_cloexec_many(except_fds, n_except_fds, false);
1695                 if (r < 0) {
1696                         log_full_errno(prio, r, "Failed to turn off O_CLOEXEC on file descriptors: %m");
1697                         _exit(EXIT_FAILURE);
1698                 }
1699         }
1700
1701         /* When we were asked to reopen the logs, do so again now */
1702         if (flags & FORK_REOPEN_LOG) {
1703                 log_open();
1704                 log_set_open_when_needed(false);
1705         }
1706
1707         if (flags & FORK_RLIMIT_NOFILE_SAFE) {
1708                 r = rlimit_nofile_safe();
1709                 if (r < 0) {
1710                         log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m");
1711                         _exit(EXIT_FAILURE);
1712                 }
1713         }
1714
1715         if (!FLAGS_SET(flags, FORK_KEEP_NOTIFY_SOCKET)) {
1716                 r = RET_NERRNO(unsetenv("NOTIFY_SOCKET"));
1717                 if (r < 0) {
1718                         log_full_errno(prio, r, "Failed to unset $NOTIFY_SOCKET: %m");
1719                         _exit(EXIT_FAILURE);
1720                 }
1721         }
1722
1723         if (ret_pid)
1724                 *ret_pid = getpid_cached();
1725
1726         return 0;
1727 }
1728
1729 int pidref_safe_fork_full(
1730                 const char *name,
1731                 const int stdio_fds[3],
1732                 int except_fds[],
1733                 size_t n_except_fds,
1734                 ForkFlags flags,
1735                 PidRef *ret_pid) {
1736
1737         pid_t pid;
1738         int r, q;
1739
1740         assert(!FLAGS_SET(flags, FORK_WAIT));
1741
1742         r = safe_fork_full(name, stdio_fds, except_fds, n_except_fds, flags, &pid);
1743         if (r < 0)
1744                 return r;
1745
1746         q = pidref_set_pid(ret_pid, pid);
1747         if (q < 0) /* Let's not fail for this, no matter what, the process exists after all, and that's key */
1748                 *ret_pid = PIDREF_MAKE_FROM_PID(pid);
1749
1750         return r;
1751 }
1752
1753 int namespace_fork(
1754                 const char *outer_name,
1755                 const char *inner_name,
1756                 int except_fds[],
1757                 size_t n_except_fds,
1758                 ForkFlags flags,
1759                 int pidns_fd,
1760                 int mntns_fd,
1761                 int netns_fd,
1762                 int userns_fd,
1763                 int root_fd,
1764                 pid_t *ret_pid) {
1765
1766         int r;
1767
1768         /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle
1769          * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that
1770          * /proc/self/fd works correctly. */
1771
1772         r = safe_fork_full(outer_name,
1773                            NULL,
1774                            except_fds, n_except_fds,
1775                            (flags|FORK_DEATHSIG_SIGINT|FORK_DEATHSIG_SIGTERM|FORK_DEATHSIG_SIGKILL) & ~(FORK_REOPEN_LOG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE), ret_pid);
1776         if (r < 0)
1777                 return r;
1778         if (r == 0) {
1779                 pid_t pid;
1780
1781                 /* Child */
1782
1783                 r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd);
1784                 if (r < 0) {
1785                         log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m");
1786                         _exit(EXIT_FAILURE);
1787                 }
1788
1789                 /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */
1790                 r = safe_fork_full(inner_name,
1791                                    NULL,
1792                                    except_fds, n_except_fds,
1793                                    flags & ~(FORK_WAIT|FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_REARRANGE_STDIO), &pid);
1794                 if (r < 0)
1795                         _exit(EXIT_FAILURE);
1796                 if (r == 0) {
1797                         /* Child */
1798                         if (ret_pid)
1799                                 *ret_pid = pid;
1800                         return 0;
1801                 }
1802
1803                 r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0);
1804                 if (r < 0)
1805                         _exit(EXIT_FAILURE);
1806
1807                 _exit(r);
1808         }
1809
1810         return 1;
1811 }
1812
1813 int set_oom_score_adjust(int value) {
1814         char t[DECIMAL_STR_MAX(int)];
1815
1816         xsprintf(t, "%i", value);
1817
1818         return write_string_file("/proc/self/oom_score_adj", t,
1819                                  WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER);
1820 }
1821
1822 int get_oom_score_adjust(int *ret) {
1823         _cleanup_free_ char *t = NULL;
1824         int r, a;
1825
1826         r = read_virtual_file("/proc/self/oom_score_adj", SIZE_MAX, &t, NULL);
1827         if (r < 0)
1828                 return r;
1829
1830         delete_trailing_chars(t, WHITESPACE);
1831
1832         assert_se(safe_atoi(t, &a) >= 0);
1833         assert_se(oom_score_adjust_is_valid(a));
1834
1835         if (ret)
1836                 *ret = a;
1837         return 0;
1838 }
1839
1840 int pidfd_get_pid(int fd, pid_t *ret) {
1841         char path[STRLEN("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)];
1842         _cleanup_free_ char *fdinfo = NULL;
1843         char *p;
1844         int r;
1845
1846         /* Converts a pidfd into a pid. Well known errors:
1847          *
1848          *    -EBADF   → fd invalid
1849          *    -ENOSYS  → /proc/ not mounted
1850          *    -ENOTTY  → fd valid, but not a pidfd
1851          *    -EREMOTE → fd valid, but pid is in another namespace we cannot translate to the local one
1852          *    -ESRCH   → fd valid, but process is already reaped
1853          */
1854
1855         if (fd < 0)
1856                 return -EBADF;
1857
1858         xsprintf(path, "/proc/self/fdinfo/%i", fd);
1859
1860         r = read_full_virtual_file(path, &fdinfo, NULL);
1861         if (r == -ENOENT) /* if fdinfo doesn't exist we assume the process does not exist */
1862                 return proc_mounted() > 0 ? -EBADF : -ENOSYS;
1863         if (r < 0)
1864                 return r;
1865
1866         p = find_line_startswith(fdinfo, "Pid:");
1867         if (!p)
1868                 return -ENOTTY; /* not a pidfd? */
1869
1870         p += strspn(p, WHITESPACE);
1871         p[strcspn(p, WHITESPACE)] = 0;
1872
1873         if (streq(p, "0"))
1874                 return -EREMOTE; /* PID is in foreign PID namespace? */
1875         if (streq(p, "-1"))
1876                 return -ESRCH;   /* refers to reaped process? */
1877
1878         return parse_pid(p, ret);
1879 }
1880
1881 int pidfd_verify_pid(int pidfd, pid_t pid) {
1882         pid_t current_pid;
1883         int r;
1884
1885         assert(pidfd >= 0);
1886         assert(pid > 0);
1887
1888         r = pidfd_get_pid(pidfd, &current_pid);
1889         if (r < 0)
1890                 return r;
1891
1892         return current_pid != pid ? -ESRCH : 0;
1893 }
1894
1895 static int rlimit_to_nice(rlim_t limit) {
1896         if (limit <= 1)
1897                 return PRIO_MAX-1; /* i.e. 19 */
1898
1899         if (limit >= -PRIO_MIN + PRIO_MAX)
1900                 return PRIO_MIN; /* i.e. -20 */
1901
1902         return PRIO_MAX - (int) limit;
1903 }
1904
1905 int setpriority_closest(int priority) {
1906         int current, limit, saved_errno;
1907         struct rlimit highest;
1908
1909         /* Try to set requested nice level */
1910         if (setpriority(PRIO_PROCESS, 0, priority) >= 0)
1911                 return 1;
1912
1913         /* Permission failed */
1914         saved_errno = -errno;
1915         if (!ERRNO_IS_PRIVILEGE(saved_errno))
1916                 return saved_errno;
1917
1918         errno = 0;
1919         current = getpriority(PRIO_PROCESS, 0);
1920         if (errno != 0)
1921                 return -errno;
1922
1923         if (priority == current)
1924                 return 1;
1925
1926        /* Hmm, we'd expect that raising the nice level from our status quo would always work. If it doesn't,
1927         * then the whole setpriority() system call is blocked to us, hence let's propagate the error
1928         * right-away */
1929         if (priority > current)
1930                 return saved_errno;
1931
1932         if (getrlimit(RLIMIT_NICE, &highest) < 0)
1933                 return -errno;
1934
1935         limit = rlimit_to_nice(highest.rlim_cur);
1936
1937         /* We are already less nice than limit allows us */
1938         if (current < limit) {
1939                 log_debug("Cannot raise nice level, permissions and the resource limit do not allow it.");
1940                 return 0;
1941         }
1942
1943         /* Push to the allowed limit */
1944         if (setpriority(PRIO_PROCESS, 0, limit) < 0)
1945                 return -errno;
1946
1947         log_debug("Cannot set requested nice level (%i), used next best (%i).", priority, limit);
1948         return 0;
1949 }
1950
1951 _noreturn_ void freeze(void) {
1952         log_close();
1953
1954         /* Make sure nobody waits for us (i.e. on one of our sockets) anymore. Note that we use
1955          * close_all_fds_without_malloc() instead of plain close_all_fds() here, since we want this function
1956          * to be compatible with being called from signal handlers. */
1957         (void) close_all_fds_without_malloc(NULL, 0);
1958
1959         /* Let's not freeze right away, but keep reaping zombies. */
1960         for (;;) {
1961                 siginfo_t si = {};
1962
1963                 if (waitid(P_ALL, 0, &si, WEXITED) < 0 && errno != EINTR)
1964                         break;
1965         }
1966
1967         /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
1968         for (;;)
1969                 pause();
1970 }
1971
1972 int get_process_threads(pid_t pid) {
1973         _cleanup_free_ char *t = NULL;
1974         const char *p;
1975         int n, r;
1976
1977         if (pid < 0)
1978                 return -EINVAL;
1979
1980         p = procfs_file_alloca(pid, "status");
1981
1982         r = get_proc_field(p, "Threads", WHITESPACE, &t);
1983         if (r == -ENOENT)
1984                 return proc_mounted() == 0 ? -ENOSYS : -ESRCH;
1985         if (r < 0)
1986                 return r;
1987
1988         r = safe_atoi(t, &n);
1989         if (r < 0)
1990                 return r;
1991         if (n < 0)
1992                 return -EINVAL;
1993
1994         return n;
1995 }
1996
1997 int is_reaper_process(void) {
1998         int b = 0;
1999
2000         /* Checks if we are running in a reaper process, i.e. if we are expected to deal with processes
2001          * reparented to us. This simply checks if we are PID 1 or if PR_SET_CHILD_SUBREAPER was called. */
2002
2003         if (getpid_cached() == 1)
2004                 return true;
2005
2006         if (prctl(PR_GET_CHILD_SUBREAPER, (unsigned long) &b, 0UL, 0UL, 0UL) < 0)
2007                 return -errno;
2008
2009         return b != 0;
2010 }
2011
2012 int make_reaper_process(bool b) {
2013
2014         if (getpid_cached() == 1) {
2015
2016                 if (!b)
2017                         return -EINVAL;
2018
2019                 return 0;
2020         }
2021
2022         /* Some prctl()s insist that all 5 arguments are specified, others do not. Let's always specify all,
2023          * to avoid any ambiguities */
2024         if (prctl(PR_SET_CHILD_SUBREAPER, (unsigned long) b, 0UL, 0UL, 0UL) < 0)
2025                 return -errno;
2026
2027         return 0;
2028 }
2029
2030 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(posix_spawnattr_t*, posix_spawnattr_destroy, NULL);
2031
2032 int posix_spawn_wrapper(
2033                 const char *path,
2034                 char * const *argv,
2035                 char * const *envp,
2036                 const char *cgroup,
2037                 PidRef *ret_pidref) {
2038
2039         short flags = POSIX_SPAWN_SETSIGMASK|POSIX_SPAWN_SETSIGDEF;
2040         posix_spawnattr_t attr;
2041         sigset_t mask;
2042         int r;
2043
2044         /* Forks and invokes 'path' with 'argv' and 'envp' using CLONE_VM and CLONE_VFORK, which means the
2045          * caller will be blocked until the child either exits or exec's. The memory of the child will be
2046          * fully shared with the memory of the parent, so that there are no copy-on-write or memory.max
2047          * issues.
2048          *
2049          * Also, move the newly-created process into 'cgroup' through POSIX_SPAWN_SETCGROUP (clone3())
2050          * if available. Note that CLONE_INTO_CGROUP is only supported on cgroup v2.
2051          * returns 1: We're already in the right cgroup
2052          *         0: 'cgroup' not specified or POSIX_SPAWN_SETCGROUP is not supported. The caller
2053          *            needs to call 'cg_attach' on their own */
2054
2055         assert(path);
2056         assert(argv);
2057         assert(ret_pidref);
2058
2059         assert_se(sigfillset(&mask) >= 0);
2060
2061         r = posix_spawnattr_init(&attr);
2062         if (r != 0)
2063                 return -r; /* These functions return a positive errno on failure */
2064
2065         /* Initialization needs to succeed before we can set up a destructor. */
2066         _unused_ _cleanup_(posix_spawnattr_destroyp) posix_spawnattr_t *attr_destructor = &attr;
2067
2068 #if HAVE_PIDFD_SPAWN
2069         _cleanup_close_ int cgroup_fd = -EBADF;
2070
2071         if (cgroup) {
2072                 _cleanup_free_ char *resolved_cgroup = NULL;
2073
2074                 r = cg_get_path_and_check(
2075                                 SYSTEMD_CGROUP_CONTROLLER,
2076                                 cgroup,
2077                                 /* suffix= */ NULL,
2078                                 &resolved_cgroup);
2079                 if (r < 0)
2080                         return r;
2081
2082                 cgroup_fd = open(resolved_cgroup, O_PATH|O_DIRECTORY|O_CLOEXEC);
2083                 if (cgroup_fd < 0)
2084                         return -errno;
2085
2086                 r = posix_spawnattr_setcgroup_np(&attr, cgroup_fd);
2087                 if (r != 0)
2088                         return -r;
2089
2090                 flags |= POSIX_SPAWN_SETCGROUP;
2091         }
2092 #endif
2093
2094         r = posix_spawnattr_setflags(&attr, flags);
2095         if (r != 0)
2096                 return -r;
2097         r = posix_spawnattr_setsigmask(&attr, &mask);
2098         if (r != 0)
2099                 return -r;
2100
2101 #if HAVE_PIDFD_SPAWN
2102         _cleanup_close_ int pidfd = -EBADF;
2103
2104         r = pidfd_spawn(&pidfd, path, NULL, &attr, argv, envp);
2105         if (r == 0) {
2106                 r = pidref_set_pidfd_consume(ret_pidref, TAKE_FD(pidfd));
2107                 if (r < 0)
2108                         return r;
2109
2110                 return FLAGS_SET(flags, POSIX_SPAWN_SETCGROUP);
2111         }
2112         if (ERRNO_IS_NOT_SUPPORTED(r)) {
2113                 /* clone3() could also return EOPNOTSUPP if the target cgroup is in threaded mode. */
2114                 if (cgroup && cg_is_threaded(cgroup) > 0)
2115                         return -EUCLEAN;
2116
2117                 /* clone3() not available? */
2118         } else if (!ERRNO_IS_PRIVILEGE(r))
2119                 return -r;
2120
2121         /* Compiled on a newer host, or seccomp&friends blocking clone3()? Fallback, but need to change the
2122          * flags to remove the cgroup one, which is what redirects to clone3() */
2123         flags &= ~POSIX_SPAWN_SETCGROUP;
2124         r = posix_spawnattr_setflags(&attr, flags);
2125         if (r != 0)
2126                 return -r;
2127 #endif
2128
2129         pid_t pid;
2130         r = posix_spawn(&pid, path, NULL, &attr, argv, envp);
2131         if (r != 0)
2132                 return -r;
2133
2134         r = pidref_set_pid(ret_pidref, pid);
2135         if (r < 0)
2136                 return r;
2137
2138         return 0; /* We did not use CLONE_INTO_CGROUP so return 0, the caller will have to move the child */
2139 }
2140
2141 int proc_dir_open(DIR **ret) {
2142         DIR *d;
2143
2144         assert(ret);
2145
2146         d = opendir("/proc");
2147         if (!d)
2148                 return -errno;
2149
2150         *ret = d;
2151         return 0;
2152 }
2153
2154 int proc_dir_read(DIR *d, pid_t *ret) {
2155         assert(d);
2156
2157         for (;;) {
2158                 struct dirent *de;
2159
2160                 errno = 0;
2161                 de = readdir_no_dot(d);
2162                 if (!de) {
2163                         if (errno != 0)
2164                                 return -errno;
2165
2166                         break;
2167                 }
2168
2169                 if (!IN_SET(de->d_type, DT_DIR, DT_UNKNOWN))
2170                         continue;
2171
2172                 if (parse_pid(de->d_name, ret) >= 0)
2173                         return 1;
2174         }
2175
2176         if (ret)
2177                 *ret = 0;
2178         return 0;
2179 }
2180
2181 int proc_dir_read_pidref(DIR *d, PidRef *ret) {
2182         int r;
2183
2184         assert(d);
2185
2186         for (;;) {
2187                 pid_t pid;
2188
2189                 r = proc_dir_read(d, &pid);
2190                 if (r < 0)
2191                         return r;
2192                 if (r == 0)
2193                         break;
2194
2195                 r = pidref_set_pid(ret, pid);
2196                 if (r == -ESRCH) /* gone by now? skip it */
2197                         continue;
2198                 if (r < 0)
2199                         return r;
2200
2201                 return 1;
2202         }
2203
2204         if (ret)
2205                 *ret = PIDREF_NULL;
2206         return 0;
2207 }
2208
2209 static const char *const sigchld_code_table[] = {
2210         [CLD_EXITED] = "exited",
2211         [CLD_KILLED] = "killed",
2212         [CLD_DUMPED] = "dumped",
2213         [CLD_TRAPPED] = "trapped",
2214         [CLD_STOPPED] = "stopped",
2215         [CLD_CONTINUED] = "continued",
2216 };
2217
2218 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
2219
2220 static const char* const sched_policy_table[] = {
2221         [SCHED_OTHER] = "other",
2222         [SCHED_BATCH] = "batch",
2223         [SCHED_IDLE] = "idle",
2224         [SCHED_FIFO] = "fifo",
2225         [SCHED_RR] = "rr",
2226 };
2227
2228 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);