src/core/manager.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2010 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <linux/kd.h>
  23 #include <signal.h>
  24 #include <string.h>
  25 #include <sys/epoll.h>
  26 #include <sys/inotify.h>
  27 #include <sys/ioctl.h>
  28 #include <sys/reboot.h>
  29 #include <sys/timerfd.h>
  30 #include <sys/wait.h>
  31 #include <unistd.h>
  32
  33 #ifdef HAVE_AUDIT
  34 #include <libaudit.h>
  35 #endif
  36
  37 #include "sd-daemon.h"
  38 #include "sd-messages.h"
  39
  40 #include "alloc-util.h"
  41 #include "audit-fd.h"
  42 #include "boot-timestamps.h"
  43 #include "bus-common-errors.h"
  44 #include "bus-error.h"
  45 #include "bus-kernel.h"
  46 #include "bus-util.h"
  47 #include "clean-ipc.h"
  48 #include "dbus-job.h"
  49 #include "dbus-manager.h"
  50 #include "dbus-unit.h"
  51 #include "dbus.h"
  52 #include "dirent-util.h"
  53 #include "env-util.h"
  54 #include "escape.h"
  55 #include "exit-status.h"
  56 #include "fd-util.h"
  57 #include "fileio.h"
  58 #include "fs-util.h"
  59 #include "hashmap.h"
  60 #include "io-util.h"
  61 #include "locale-setup.h"
  62 #include "log.h"
  63 #include "macro.h"
  64 #include "manager.h"
  65 #include "missing.h"
  66 #include "mkdir.h"
  67 #include "parse-util.h"
  68 #include "path-lookup.h"
  69 #include "path-util.h"
  70 #include "process-util.h"
  71 #include "ratelimit.h"
  72 #include "rm-rf.h"
  73 #include "signal-util.h"
  74 #include "special.h"
  75 #include "stat-util.h"
  76 #include "string-table.h"
  77 #include "string-util.h"
  78 #include "strv.h"
  79 #include "terminal-util.h"
  80 #include "time-util.h"
  81 #include "transaction.h"
  82 #include "umask-util.h"
  83 #include "unit-name.h"
  84 #include "user-util.h"
  85 #include "util.h"
  86 #include "virt.h"
  87 #include "watchdog.h"
  88
  89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
  90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
  91
  92 /* Initial delay and the interval for printing status messages about running jobs */
  93 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
  94 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
  95 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
  96
  97 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  98 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  99 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 100 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 101 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 102 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 103 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
 104 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
 105 static int manager_run_generators(Manager *m);
 106
 107 static void manager_watch_jobs_in_progress(Manager *m) {
 108         usec_t next;
 109         int r;
 110
 111         assert(m);
 112
 113         /* We do not want to show the cylon animation if the user
 114          * needs to confirm service executions otherwise confirmation
 115          * messages will be screwed by the cylon animation. */
 116         if (!manager_is_confirm_spawn_disabled(m))
 117                 return;
 118
 119         if (m->jobs_in_progress_event_source)
 120                 return;
 121
 122         next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
 123         r = sd_event_add_time(
 124                         m->event,
 125                         &m->jobs_in_progress_event_source,
 126                         CLOCK_MONOTONIC,
 127                         next, 0,
 128                         manager_dispatch_jobs_in_progress, m);
 129         if (r < 0)
 130                 return;
 131
 132         (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
 133 }
 134
 135 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
 136
 137 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
 138         char *p = buffer;
 139
 140         assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
 141         assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
 142
 143         if (pos > 1) {
 144                 if (pos > 2)
 145                         p = mempset(p, ' ', pos-2);
 146                 if (log_get_show_color())
 147                         p = stpcpy(p, ANSI_RED);
 148                 *p++ = '*';
 149         }
 150
 151         if (pos > 0 && pos <= width) {
 152                 if (log_get_show_color())
 153                         p = stpcpy(p, ANSI_HIGHLIGHT_RED);
 154                 *p++ = '*';
 155         }
 156
 157         if (log_get_show_color())
 158                 p = stpcpy(p, ANSI_NORMAL);
 159
 160         if (pos < width) {
 161                 if (log_get_show_color())
 162                         p = stpcpy(p, ANSI_RED);
 163                 *p++ = '*';
 164                 if (pos < width-1)
 165                         p = mempset(p, ' ', width-1-pos);
 166                 if (log_get_show_color())
 167                         strcpy(p, ANSI_NORMAL);
 168         }
 169 }
 170
 171 void manager_flip_auto_status(Manager *m, bool enable) {
 172         assert(m);
 173
 174         if (enable) {
 175                 if (m->show_status == SHOW_STATUS_AUTO)
 176                         manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
 177         } else {
 178                 if (m->show_status == SHOW_STATUS_TEMPORARY)
 179                         manager_set_show_status(m, SHOW_STATUS_AUTO);
 180         }
 181 }
 182
 183 static void manager_print_jobs_in_progress(Manager *m) {
 184         _cleanup_free_ char *job_of_n = NULL;
 185         Iterator i;
 186         Job *j;
 187         unsigned counter = 0, print_nr;
 188         char cylon[6 + CYLON_BUFFER_EXTRA + 1];
 189         unsigned cylon_pos;
 190         char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
 191         uint64_t x;
 192
 193         assert(m);
 194         assert(m->n_running_jobs > 0);
 195
 196         manager_flip_auto_status(m, true);
 197
 198         print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
 199
 200         HASHMAP_FOREACH(j, m->jobs, i)
 201                 if (j->state == JOB_RUNNING && counter++ == print_nr)
 202                         break;
 203
 204         /* m->n_running_jobs must be consistent with the contents of m->jobs,
 205          * so the above loop must have succeeded in finding j. */
 206         assert(counter == print_nr + 1);
 207         assert(j);
 208
 209         cylon_pos = m->jobs_in_progress_iteration % 14;
 210         if (cylon_pos >= 8)
 211                 cylon_pos = 14 - cylon_pos;
 212         draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
 213
 214         m->jobs_in_progress_iteration++;
 215
 216         if (m->n_running_jobs > 1) {
 217                 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
 218                         job_of_n = NULL;
 219         }
 220
 221         format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
 222         if (job_get_timeout(j, &x) > 0)
 223                 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
 224
 225         manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
 226                               "%sA %s job is running for %s (%s / %s)",
 227                               strempty(job_of_n),
 228                               job_type_to_string(j->type),
 229                               unit_description(j->unit),
 230                               time, limit);
 231 }
 232
 233 static int have_ask_password(void) {
 234         _cleanup_closedir_ DIR *dir;
 235         struct dirent *de;
 236
 237         dir = opendir("/run/systemd/ask-password");
 238         if (!dir) {
 239                 if (errno == ENOENT)
 240                         return false;
 241                 else
 242                         return -errno;
 243         }
 244
 245         FOREACH_DIRENT_ALL(de, dir, return -errno) {
 246                 if (startswith(de->d_name, "ask."))
 247                         return true;
 248         }
 249         return false;
 250 }
 251
 252 static int manager_dispatch_ask_password_fd(sd_event_source *source,
 253                                             int fd, uint32_t revents, void *userdata) {
 254         Manager *m = userdata;
 255
 256         assert(m);
 257
 258         flush_fd(fd);
 259
 260         m->have_ask_password = have_ask_password();
 261         if (m->have_ask_password < 0)
 262                 /* Log error but continue. Negative have_ask_password
 263                  * is treated as unknown status. */
 264                 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
 265
 266         return 0;
 267 }
 268
 269 static void manager_close_ask_password(Manager *m) {
 270         assert(m);
 271
 272         m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
 273         m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
 274         m->have_ask_password = -EINVAL;
 275 }
 276
 277 static int manager_check_ask_password(Manager *m) {
 278         int r;
 279
 280         assert(m);
 281
 282         if (!m->ask_password_event_source) {
 283                 assert(m->ask_password_inotify_fd < 0);
 284
 285                 mkdir_p_label("/run/systemd/ask-password", 0755);
 286
 287                 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
 288                 if (m->ask_password_inotify_fd < 0)
 289                         return log_error_errno(errno, "inotify_init1() failed: %m");
 290
 291                 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
 292                         log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
 293                         manager_close_ask_password(m);
 294                         return -errno;
 295                 }
 296
 297                 r = sd_event_add_io(m->event, &m->ask_password_event_source,
 298                                     m->ask_password_inotify_fd, EPOLLIN,
 299                                     manager_dispatch_ask_password_fd, m);
 300                 if (r < 0) {
 301                         log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
 302                         manager_close_ask_password(m);
 303                         return -errno;
 304                 }
 305
 306                 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
 307
 308                 /* Queries might have been added meanwhile... */
 309                 manager_dispatch_ask_password_fd(m->ask_password_event_source,
 310                                                  m->ask_password_inotify_fd, EPOLLIN, m);
 311         }
 312
 313         return m->have_ask_password;
 314 }
 315
 316 static int manager_watch_idle_pipe(Manager *m) {
 317         int r;
 318
 319         assert(m);
 320
 321         if (m->idle_pipe_event_source)
 322                 return 0;
 323
 324         if (m->idle_pipe[2] < 0)
 325                 return 0;
 326
 327         r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
 328         if (r < 0)
 329                 return log_error_errno(r, "Failed to watch idle pipe: %m");
 330
 331         (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
 332
 333         return 0;
 334 }
 335
 336 static void manager_close_idle_pipe(Manager *m) {
 337         assert(m);
 338
 339         m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
 340
 341         safe_close_pair(m->idle_pipe);
 342         safe_close_pair(m->idle_pipe + 2);
 343 }
 344
 345 static int manager_setup_time_change(Manager *m) {
 346         int r;
 347
 348         /* We only care for the cancellation event, hence we set the
 349          * timeout to the latest possible value. */
 350         struct itimerspec its = {
 351                 .it_value.tv_sec = TIME_T_MAX,
 352         };
 353
 354         assert(m);
 355         assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
 356
 357         if (m->test_run)
 358                 return 0;
 359
 360         /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
 361          * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
 362
 363         m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
 364         if (m->time_change_fd < 0)
 365                 return log_error_errno(errno, "Failed to create timerfd: %m");
 366
 367         if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
 368                 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
 369                 m->time_change_fd = safe_close(m->time_change_fd);
 370                 return 0;
 371         }
 372
 373         r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
 374         if (r < 0)
 375                 return log_error_errno(r, "Failed to create time change event source: %m");
 376
 377         (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
 378
 379         log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
 380
 381         return 0;
 382 }
 383
 384 static int enable_special_signals(Manager *m) {
 385         _cleanup_close_ int fd = -1;
 386
 387         assert(m);
 388
 389         if (m->test_run)
 390                 return 0;
 391
 392         /* Enable that we get SIGINT on control-alt-del. In containers
 393          * this will fail with EPERM (older) or EINVAL (newer), so
 394          * ignore that. */
 395         if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
 396                 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
 397
 398         fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
 399         if (fd < 0) {
 400                 /* Support systems without virtual console */
 401                 if (fd != -ENOENT)
 402                         log_warning_errno(errno, "Failed to open /dev/tty0: %m");
 403         } else {
 404                 /* Enable that we get SIGWINCH on kbrequest */
 405                 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
 406                         log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
 407         }
 408
 409         return 0;
 410 }
 411
 412 static int manager_setup_signals(Manager *m) {
 413         struct sigaction sa = {
 414                 .sa_handler = SIG_DFL,
 415                 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 416         };
 417         sigset_t mask;
 418         int r;
 419
 420         assert(m);
 421
 422         assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 423
 424         /* We make liberal use of realtime signals here. On
 425          * Linux/glibc we have 30 of them (with the exception of Linux
 426          * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
 427          * (aka SIGRTMAX). */
 428
 429         assert_se(sigemptyset(&mask) == 0);
 430         sigset_add_many(&mask,
 431                         SIGCHLD,     /* Child died */
 432                         SIGTERM,     /* Reexecute daemon */
 433                         SIGHUP,      /* Reload configuration */
 434                         SIGUSR1,     /* systemd/upstart: reconnect to D-Bus */
 435                         SIGUSR2,     /* systemd: dump status */
 436                         SIGINT,      /* Kernel sends us this on control-alt-del */
 437                         SIGWINCH,    /* Kernel sends us this on kbrequest (alt-arrowup) */
 438                         SIGPWR,      /* Some kernel drivers and upsd send us this on power failure */
 439
 440                         SIGRTMIN+0,  /* systemd: start default.target */
 441                         SIGRTMIN+1,  /* systemd: isolate rescue.target */
 442                         SIGRTMIN+2,  /* systemd: isolate emergency.target */
 443                         SIGRTMIN+3,  /* systemd: start halt.target */
 444                         SIGRTMIN+4,  /* systemd: start poweroff.target */
 445                         SIGRTMIN+5,  /* systemd: start reboot.target */
 446                         SIGRTMIN+6,  /* systemd: start kexec.target */
 447
 448                         /* ... space for more special targets ... */
 449
 450                         SIGRTMIN+13, /* systemd: Immediate halt */
 451                         SIGRTMIN+14, /* systemd: Immediate poweroff */
 452                         SIGRTMIN+15, /* systemd: Immediate reboot */
 453                         SIGRTMIN+16, /* systemd: Immediate kexec */
 454
 455                         /* ... space for more immediate system state changes ... */
 456
 457                         SIGRTMIN+20, /* systemd: enable status messages */
 458                         SIGRTMIN+21, /* systemd: disable status messages */
 459                         SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
 460                         SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
 461                         SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
 462
 463                         /* .. one free signal here ... */
 464
 465 #if !defined(__hppa64__) && !defined(__hppa__)
 466                         /* Apparently Linux on hppa has fewer RT
 467                          * signals (SIGRTMAX is SIGRTMIN+25 there),
 468                          * hence let's not try to make use of them
 469                          * here. Since these commands are accessible
 470                          * by different means and only really a safety
 471                          * net, the missing functionality on hppa
 472                          * shouldn't matter. */
 473
 474                         SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
 475                         SIGRTMIN+27, /* systemd: set log target to console */
 476                         SIGRTMIN+28, /* systemd: set log target to kmsg */
 477                         SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
 478
 479                         /* ... one free signal here SIGRTMIN+30 ... */
 480 #endif
 481                         -1);
 482         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 483
 484         m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 485         if (m->signal_fd < 0)
 486                 return -errno;
 487
 488         r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
 489         if (r < 0)
 490                 return r;
 491
 492         (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
 493
 494         /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
 495          * notify processing can still figure out to which process/service a message belongs, before we reap the
 496          * process. Also, process this before handling cgroup notifications, so that we always collect child exit
 497          * status information before detecting that there's no process in a cgroup. */
 498         r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
 499         if (r < 0)
 500                 return r;
 501
 502         if (MANAGER_IS_SYSTEM(m))
 503                 return enable_special_signals(m);
 504
 505         return 0;
 506 }
 507
 508 static void manager_clean_environment(Manager *m) {
 509         assert(m);
 510
 511         /* Let's remove some environment variables that we
 512          * need ourselves to communicate with our clients */
 513         strv_env_unset_many(
 514                         m->environment,
 515                         "NOTIFY_SOCKET",
 516                         "MAINPID",
 517                         "MANAGERPID",
 518                         "LISTEN_PID",
 519                         "LISTEN_FDS",
 520                         "LISTEN_FDNAMES",
 521                         "WATCHDOG_PID",
 522                         "WATCHDOG_USEC",
 523                         "INVOCATION_ID",
 524                         NULL);
 525 }
 526
 527 static int manager_default_environment(Manager *m) {
 528         assert(m);
 529
 530         if (MANAGER_IS_SYSTEM(m)) {
 531                 /* The system manager always starts with a clean
 532                  * environment for its children. It does not import
 533                  * the kernel or the parents exported variables.
 534                  *
 535                  * The initial passed environ is untouched to keep
 536                  * /proc/self/environ valid; it is used for tagging
 537                  * the init process inside containers. */
 538                 m->environment = strv_new("PATH=" DEFAULT_PATH,
 539                                           NULL);
 540
 541                 /* Import locale variables LC_*= from configuration */
 542                 locale_setup(&m->environment);
 543         } else {
 544                 /* The user manager passes its own environment
 545                  * along to its children. */
 546                 m->environment = strv_copy(environ);
 547         }
 548
 549         if (!m->environment)
 550                 return -ENOMEM;
 551
 552         manager_clean_environment(m);
 553         strv_sort(m->environment);
 554
 555         return 0;
 556 }
 557
 558 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
 559         Manager *m;
 560         int r;
 561
 562         assert(_m);
 563         assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
 564
 565         m = new0(Manager, 1);
 566         if (!m)
 567                 return -ENOMEM;
 568
 569         m->unit_file_scope = scope;
 570         m->exit_code = _MANAGER_EXIT_CODE_INVALID;
 571         m->default_timer_accuracy_usec = USEC_PER_MINUTE;
 572         m->default_tasks_accounting = true;
 573         m->default_tasks_max = UINT64_MAX;
 574
 575 #ifdef ENABLE_EFI
 576         if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
 577                 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
 578 #endif
 579
 580         /* Prepare log fields we can use for structured logging */
 581         if (MANAGER_IS_SYSTEM(m)) {
 582                 m->unit_log_field = "UNIT=";
 583                 m->unit_log_format_string = "UNIT=%s";
 584
 585                 m->invocation_log_field = "INVOCATION_ID=";
 586                 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
 587         } else {
 588                 m->unit_log_field = "USER_UNIT=";
 589                 m->unit_log_format_string = "USER_UNIT=%s";
 590
 591                 m->invocation_log_field = "USER_INVOCATION_ID=";
 592                 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
 593         }
 594
 595         m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
 596
 597         m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
 598                 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
 599                 m->ask_password_inotify_fd = -1;
 600
 601         m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
 602
 603         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 604
 605         m->have_ask_password = -EINVAL; /* we don't know */
 606         m->first_boot = -1;
 607
 608         m->test_run = test_run;
 609
 610         /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
 611         RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
 612
 613         r = manager_default_environment(m);
 614         if (r < 0)
 615                 goto fail;
 616
 617         r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
 618         if (r < 0)
 619                 goto fail;
 620
 621         r = hashmap_ensure_allocated(&m->jobs, NULL);
 622         if (r < 0)
 623                 goto fail;
 624
 625         r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
 626         if (r < 0)
 627                 goto fail;
 628
 629         r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
 630         if (r < 0)
 631                 goto fail;
 632
 633         r = sd_event_default(&m->event);
 634         if (r < 0)
 635                 goto fail;
 636
 637         r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
 638         if (r < 0)
 639                 goto fail;
 640
 641         r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
 642         if (r < 0)
 643                 goto fail;
 644
 645         r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
 646         if (r < 0)
 647                 goto fail;
 648
 649         (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
 650
 651         r = manager_setup_signals(m);
 652         if (r < 0)
 653                 goto fail;
 654
 655         r = manager_setup_cgroup(m);
 656         if (r < 0)
 657                 goto fail;
 658
 659         r = manager_setup_time_change(m);
 660         if (r < 0)
 661                 goto fail;
 662
 663         m->udev = udev_new();
 664         if (!m->udev) {
 665                 r = -ENOMEM;
 666                 goto fail;
 667         }
 668
 669         /* Note that we do not set up the notify fd here. We do that after deserialization,
 670          * since they might have gotten serialized across the reexec. */
 671
 672         m->taint_usr = dir_is_empty("/usr") > 0;
 673
 674         *_m = m;
 675         return 0;
 676
 677 fail:
 678         manager_free(m);
 679         return r;
 680 }
 681
 682 static int manager_setup_notify(Manager *m) {
 683         int r;
 684
 685         if (m->test_run)
 686                 return 0;
 687
 688         if (m->notify_fd < 0) {
 689                 _cleanup_close_ int fd = -1;
 690                 union sockaddr_union sa = {
 691                         .sa.sa_family = AF_UNIX,
 692                 };
 693                 static const int one = 1;
 694                 const char *e;
 695
 696                 /* First free all secondary fields */
 697                 m->notify_socket = mfree(m->notify_socket);
 698                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
 699
 700                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 701                 if (fd < 0)
 702                         return log_error_errno(errno, "Failed to allocate notification socket: %m");
 703
 704                 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
 705
 706                 e = manager_get_runtime_prefix(m);
 707                 if (!e) {
 708                         log_error("Failed to determine runtime prefix.");
 709                         return -EINVAL;
 710                 }
 711
 712                 m->notify_socket = strappend(e, "/systemd/notify");
 713                 if (!m->notify_socket)
 714                         return log_oom();
 715
 716                 (void) mkdir_parents_label(m->notify_socket, 0755);
 717                 (void) unlink(m->notify_socket);
 718
 719                 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
 720                 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 721                 if (r < 0)
 722                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 723
 724                 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
 725                 if (r < 0)
 726                         return log_error_errno(errno, "SO_PASSCRED failed: %m");
 727
 728                 m->notify_fd = fd;
 729                 fd = -1;
 730
 731                 log_debug("Using notification socket %s", m->notify_socket);
 732         }
 733
 734         if (!m->notify_event_source) {
 735                 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
 736                 if (r < 0)
 737                         return log_error_errno(r, "Failed to allocate notify event source: %m");
 738
 739                 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
 740                  * service an exit message belongs. */
 741                 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
 742                 if (r < 0)
 743                         return log_error_errno(r, "Failed to set priority of notify event source: %m");
 744
 745                 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
 746         }
 747
 748         return 0;
 749 }
 750
 751 static int manager_setup_cgroups_agent(Manager *m) {
 752
 753         static const union sockaddr_union sa = {
 754                 .un.sun_family = AF_UNIX,
 755                 .un.sun_path = "/run/systemd/cgroups-agent",
 756         };
 757         int r;
 758
 759         /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
 760          * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
 761          * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
 762          * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
 763          * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
 764          * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
 765          * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
 766          * we thus won't lose messages.
 767          *
 768          * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
 769          * to it. The system instance hence listens on this special socket, but the user instances listen on the system
 770          * bus for these messages. */
 771
 772         if (m->test_run)
 773                 return 0;
 774
 775         if (!MANAGER_IS_SYSTEM(m))
 776                 return 0;
 777
 778         if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* We don't need this anymore on the unified hierarchy */
 779                 return 0;
 780
 781         if (m->cgroups_agent_fd < 0) {
 782                 _cleanup_close_ int fd = -1;
 783
 784                 /* First free all secondary fields */
 785                 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
 786
 787                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 788                 if (fd < 0)
 789                         return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
 790
 791                 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
 792
 793                 (void) unlink(sa.un.sun_path);
 794
 795                 /* Only allow root to connect to this socket */
 796                 RUN_WITH_UMASK(0077)
 797                         r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 798                 if (r < 0)
 799                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 800
 801                 m->cgroups_agent_fd = fd;
 802                 fd = -1;
 803         }
 804
 805         if (!m->cgroups_agent_event_source) {
 806                 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
 807                 if (r < 0)
 808                         return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
 809
 810                 /* Process cgroups notifications early, but after having processed service notification messages or
 811                  * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
 812                  * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
 813                  * cgroup inotify for the unified cgroup stuff. */
 814                 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
 815                 if (r < 0)
 816                         return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
 817
 818                 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
 819         }
 820
 821         return 0;
 822 }
 823
 824 static int manager_setup_user_lookup_fd(Manager *m) {
 825         int r;
 826
 827         assert(m);
 828
 829         /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
 830          * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
 831          * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
 832          * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
 833          * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
 834          * hence we establish this communication channel so that forked off processes can pass their UID/GID
 835          * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
 836          * datagram, along with their unit name, so that we can share one communication socket pair among all units for
 837          * this purpose.
 838          *
 839          * You might wonder why we need a communication channel for this that is independent of the usual notification
 840          * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
 841          * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
 842          * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
 843          *
 844          * Note that this function is called under two circumstances: when we first initialize (in which case we
 845          * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
 846          * (in which case the socket pair already exists but we still need to allocate the event source for it). */
 847
 848         if (m->user_lookup_fds[0] < 0) {
 849
 850                 /* Free all secondary fields */
 851                 safe_close_pair(m->user_lookup_fds);
 852                 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
 853
 854                 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
 855                         return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
 856
 857                 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
 858         }
 859
 860         if (!m->user_lookup_event_source) {
 861                 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
 862                 if (r < 0)
 863                         return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
 864
 865                 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
 866                  * resolutions */
 867                 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
 868                 if (r < 0)
 869                         return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
 870
 871                 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
 872         }
 873
 874         return 0;
 875 }
 876
 877 static int manager_connect_bus(Manager *m, bool reexecuting) {
 878         bool try_bus_connect;
 879
 880         assert(m);
 881
 882         if (m->test_run)
 883                 return 0;
 884
 885         try_bus_connect =
 886                 reexecuting ||
 887                 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
 888
 889         /* Try to connect to the buses, if possible. */
 890         return bus_init(m, try_bus_connect);
 891 }
 892
 893 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
 894         Unit *u;
 895         unsigned n = 0;
 896
 897         assert(m);
 898
 899         while ((u = m->cleanup_queue)) {
 900                 assert(u->in_cleanup_queue);
 901
 902                 unit_free(u);
 903                 n++;
 904         }
 905
 906         return n;
 907 }
 908
 909 enum {
 910         GC_OFFSET_IN_PATH,  /* This one is on the path we were traveling */
 911         GC_OFFSET_UNSURE,   /* No clue */
 912         GC_OFFSET_GOOD,     /* We still need this unit */
 913         GC_OFFSET_BAD,      /* We don't need this unit anymore */
 914         _GC_OFFSET_MAX
 915 };
 916
 917 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
 918         Iterator i;
 919         Unit *other;
 920
 921         u->gc_marker = gc_marker + GC_OFFSET_GOOD;
 922
 923         /* Recursively mark referenced units as GOOD as well */
 924         SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
 925                 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
 926                         unit_gc_mark_good(other, gc_marker);
 927 }
 928
 929 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
 930         Iterator i;
 931         Unit *other;
 932         bool is_bad;
 933
 934         assert(u);
 935
 936         if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
 937             u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 938             u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
 939             u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
 940                 return;
 941
 942         if (u->in_cleanup_queue)
 943                 goto bad;
 944
 945         if (unit_check_gc(u))
 946                 goto good;
 947
 948         u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
 949
 950         is_bad = true;
 951
 952         SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
 953                 unit_gc_sweep(other, gc_marker);
 954
 955                 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
 956                         goto good;
 957
 958                 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
 959                         is_bad = false;
 960         }
 961
 962         if (is_bad)
 963                 goto bad;
 964
 965         /* We were unable to find anything out about this entry, so
 966          * let's investigate it later */
 967         u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
 968         unit_add_to_gc_queue(u);
 969         return;
 970
 971 bad:
 972         /* We definitely know that this one is not useful anymore, so
 973          * let's mark it for deletion */
 974         u->gc_marker = gc_marker + GC_OFFSET_BAD;
 975         unit_add_to_cleanup_queue(u);
 976         return;
 977
 978 good:
 979         unit_gc_mark_good(u, gc_marker);
 980 }
 981
 982 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
 983         unsigned n = 0, gc_marker;
 984         Unit *u;
 985
 986         assert(m);
 987
 988         /* log_debug("Running GC..."); */
 989
 990         m->gc_marker += _GC_OFFSET_MAX;
 991         if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
 992                 m->gc_marker = 1;
 993
 994         gc_marker = m->gc_marker;
 995
 996         while ((u = m->gc_unit_queue)) {
 997                 assert(u->in_gc_queue);
 998
 999                 unit_gc_sweep(u, gc_marker);
1000
1001                 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1002                 u->in_gc_queue = false;
1003
1004                 n++;
1005
1006                 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1007                     u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1008                         if (u->id)
1009                                 log_unit_debug(u, "Collecting.");
1010                         u->gc_marker = gc_marker + GC_OFFSET_BAD;
1011                         unit_add_to_cleanup_queue(u);
1012                 }
1013         }
1014
1015         return n;
1016 }
1017
1018 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1019         unsigned n = 0;
1020         Job *j;
1021
1022         assert(m);
1023
1024         while ((j = m->gc_job_queue)) {
1025                 assert(j->in_gc_queue);
1026
1027                 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1028                 j->in_gc_queue = false;
1029
1030                 n++;
1031
1032                 if (job_check_gc(j))
1033                         continue;
1034
1035                 log_unit_debug(j->unit, "Collecting job.");
1036                 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1037         }
1038
1039         return n;
1040 }
1041
1042 static void manager_clear_jobs_and_units(Manager *m) {
1043         Unit *u;
1044
1045         assert(m);
1046
1047         while ((u = hashmap_first(m->units)))
1048                 unit_free(u);
1049
1050         manager_dispatch_cleanup_queue(m);
1051
1052         assert(!m->load_queue);
1053         assert(!m->run_queue);
1054         assert(!m->dbus_unit_queue);
1055         assert(!m->dbus_job_queue);
1056         assert(!m->cleanup_queue);
1057         assert(!m->gc_unit_queue);
1058         assert(!m->gc_job_queue);
1059
1060         assert(hashmap_isempty(m->jobs));
1061         assert(hashmap_isempty(m->units));
1062
1063         m->n_on_console = 0;
1064         m->n_running_jobs = 0;
1065 }
1066
1067 Manager* manager_free(Manager *m) {
1068         UnitType c;
1069         int i;
1070
1071         if (!m)
1072                 return NULL;
1073
1074         manager_clear_jobs_and_units(m);
1075
1076         for (c = 0; c < _UNIT_TYPE_MAX; c++)
1077                 if (unit_vtable[c]->shutdown)
1078                         unit_vtable[c]->shutdown(m);
1079
1080         /* If we reexecute ourselves, we keep the root cgroup
1081          * around */
1082         manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1083
1084         lookup_paths_flush_generator(&m->lookup_paths);
1085
1086         bus_done(m);
1087
1088         dynamic_user_vacuum(m, false);
1089         hashmap_free(m->dynamic_users);
1090
1091         hashmap_free(m->units);
1092         hashmap_free(m->units_by_invocation_id);
1093         hashmap_free(m->jobs);
1094         hashmap_free(m->watch_pids1);
1095         hashmap_free(m->watch_pids2);
1096         hashmap_free(m->watch_bus);
1097
1098         set_free(m->startup_units);
1099         set_free(m->failed_units);
1100
1101         sd_event_source_unref(m->signal_event_source);
1102         sd_event_source_unref(m->notify_event_source);
1103         sd_event_source_unref(m->cgroups_agent_event_source);
1104         sd_event_source_unref(m->time_change_event_source);
1105         sd_event_source_unref(m->jobs_in_progress_event_source);
1106         sd_event_source_unref(m->run_queue_event_source);
1107         sd_event_source_unref(m->user_lookup_event_source);
1108
1109         safe_close(m->signal_fd);
1110         safe_close(m->notify_fd);
1111         safe_close(m->cgroups_agent_fd);
1112         safe_close(m->time_change_fd);
1113         safe_close_pair(m->user_lookup_fds);
1114
1115         manager_close_ask_password(m);
1116
1117         manager_close_idle_pipe(m);
1118
1119         udev_unref(m->udev);
1120         sd_event_unref(m->event);
1121
1122         free(m->notify_socket);
1123
1124         lookup_paths_free(&m->lookup_paths);
1125         strv_free(m->environment);
1126
1127         hashmap_free(m->cgroup_unit);
1128         set_free_free(m->unit_path_cache);
1129
1130         free(m->switch_root);
1131         free(m->switch_root_init);
1132
1133         for (i = 0; i < _RLIMIT_MAX; i++)
1134                 m->rlimit[i] = mfree(m->rlimit[i]);
1135
1136         assert(hashmap_isempty(m->units_requiring_mounts_for));
1137         hashmap_free(m->units_requiring_mounts_for);
1138
1139         hashmap_free(m->uid_refs);
1140         hashmap_free(m->gid_refs);
1141
1142         return mfree(m);
1143 }
1144
1145 void manager_enumerate(Manager *m) {
1146         UnitType c;
1147
1148         assert(m);
1149
1150         /* Let's ask every type to load all units from disk/kernel
1151          * that it might know */
1152         for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1153                 if (!unit_type_supported(c)) {
1154                         log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1155                         continue;
1156                 }
1157
1158                 if (!unit_vtable[c]->enumerate)
1159                         continue;
1160
1161                 unit_vtable[c]->enumerate(m);
1162         }
1163
1164         manager_dispatch_load_queue(m);
1165 }
1166
1167 static void manager_coldplug(Manager *m) {
1168         Iterator i;
1169         Unit *u;
1170         char *k;
1171         int r;
1172
1173         assert(m);
1174
1175         /* Then, let's set up their initial state. */
1176         HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1177
1178                 /* ignore aliases */
1179                 if (u->id != k)
1180                         continue;
1181
1182                 r = unit_coldplug(u);
1183                 if (r < 0)
1184                         log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1185         }
1186 }
1187
1188 static void manager_build_unit_path_cache(Manager *m) {
1189         char **i;
1190         int r;
1191
1192         assert(m);
1193
1194         set_free_free(m->unit_path_cache);
1195
1196         m->unit_path_cache = set_new(&string_hash_ops);
1197         if (!m->unit_path_cache) {
1198                 r = -ENOMEM;
1199                 goto fail;
1200         }
1201
1202         /* This simply builds a list of files we know exist, so that
1203          * we don't always have to go to disk */
1204
1205         STRV_FOREACH(i, m->lookup_paths.search_path) {
1206                 _cleanup_closedir_ DIR *d = NULL;
1207                 struct dirent *de;
1208
1209                 d = opendir(*i);
1210                 if (!d) {
1211                         if (errno != ENOENT)
1212                                 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1213                         continue;
1214                 }
1215
1216                 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1217                         char *p;
1218
1219                         p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1220                         if (!p) {
1221                                 r = -ENOMEM;
1222                                 goto fail;
1223                         }
1224
1225                         r = set_consume(m->unit_path_cache, p);
1226                         if (r < 0)
1227                                 goto fail;
1228                 }
1229         }
1230
1231         return;
1232
1233 fail:
1234         log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1235         m->unit_path_cache = set_free_free(m->unit_path_cache);
1236 }
1237
1238 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1239         Iterator i;
1240         Unit *u;
1241
1242         assert(m);
1243
1244         HASHMAP_FOREACH(u, m->units, i) {
1245
1246                 if (fdset_size(fds) <= 0)
1247                         break;
1248
1249                 if (!UNIT_VTABLE(u)->distribute_fds)
1250                         continue;
1251
1252                 UNIT_VTABLE(u)->distribute_fds(u, fds);
1253         }
1254 }
1255
1256 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1257         int r, q;
1258
1259         assert(m);
1260
1261         r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1262         if (r < 0)
1263                 return r;
1264
1265         /* Make sure the transient directory always exists, so that it remains in the search path */
1266         if (!m->test_run) {
1267                 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1268                 if (r < 0)
1269                         return r;
1270         }
1271
1272         dual_timestamp_get(&m->generators_start_timestamp);
1273         r = manager_run_generators(m);
1274         dual_timestamp_get(&m->generators_finish_timestamp);
1275         if (r < 0)
1276                 return r;
1277
1278         lookup_paths_reduce(&m->lookup_paths);
1279         manager_build_unit_path_cache(m);
1280
1281         /* If we will deserialize make sure that during enumeration
1282          * this is already known, so we increase the counter here
1283          * already */
1284         if (serialization)
1285                 m->n_reloading++;
1286
1287         /* First, enumerate what we can from all config files */
1288         dual_timestamp_get(&m->units_load_start_timestamp);
1289         manager_enumerate(m);
1290         dual_timestamp_get(&m->units_load_finish_timestamp);
1291
1292         /* Second, deserialize if there is something to deserialize */
1293         if (serialization)
1294                 r = manager_deserialize(m, serialization, fds);
1295
1296         /* Any fds left? Find some unit which wants them. This is
1297          * useful to allow container managers to pass some file
1298          * descriptors to us pre-initialized. This enables
1299          * socket-based activation of entire containers. */
1300         manager_distribute_fds(m, fds);
1301
1302         /* We might have deserialized the notify fd, but if we didn't
1303          * then let's create the bus now */
1304         q = manager_setup_notify(m);
1305         if (q < 0 && r == 0)
1306                 r = q;
1307
1308         q = manager_setup_cgroups_agent(m);
1309         if (q < 0 && r == 0)
1310                 r = q;
1311
1312         q = manager_setup_user_lookup_fd(m);
1313         if (q < 0 && r == 0)
1314                 r = q;
1315
1316         /* Let's connect to the bus now. */
1317         (void) manager_connect_bus(m, !!serialization);
1318
1319         (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1320         m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1321
1322         /* Third, fire things up! */
1323         manager_coldplug(m);
1324
1325         /* Release any dynamic users no longer referenced */
1326         dynamic_user_vacuum(m, true);
1327
1328         /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1329         manager_vacuum_uid_refs(m);
1330         manager_vacuum_gid_refs(m);
1331
1332         if (serialization) {
1333                 assert(m->n_reloading > 0);
1334                 m->n_reloading--;
1335
1336                 /* Let's wait for the UnitNew/JobNew messages being
1337                  * sent, before we notify that the reload is
1338                  * finished */
1339                 m->send_reloading_done = true;
1340         }
1341
1342         return r;
1343 }
1344
1345 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1346         int r;
1347         Transaction *tr;
1348
1349         assert(m);
1350         assert(type < _JOB_TYPE_MAX);
1351         assert(unit);
1352         assert(mode < _JOB_MODE_MAX);
1353
1354         if (mode == JOB_ISOLATE && type != JOB_START)
1355                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1356
1357         if (mode == JOB_ISOLATE && !unit->allow_isolate)
1358                 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1359
1360         log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1361
1362         type = job_type_collapse(type, unit);
1363
1364         tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1365         if (!tr)
1366                 return -ENOMEM;
1367
1368         r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1369                                                  mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1370                                                  mode == JOB_IGNORE_DEPENDENCIES, e);
1371         if (r < 0)
1372                 goto tr_abort;
1373
1374         if (mode == JOB_ISOLATE) {
1375                 r = transaction_add_isolate_jobs(tr, m);
1376                 if (r < 0)
1377                         goto tr_abort;
1378         }
1379
1380         r = transaction_activate(tr, m, mode, e);
1381         if (r < 0)
1382                 goto tr_abort;
1383
1384         log_unit_debug(unit,
1385                        "Enqueued job %s/%s as %u", unit->id,
1386                        job_type_to_string(type), (unsigned) tr->anchor_job->id);
1387
1388         if (_ret)
1389                 *_ret = tr->anchor_job;
1390
1391         transaction_free(tr);
1392         return 0;
1393
1394 tr_abort:
1395         transaction_abort(tr);
1396         transaction_free(tr);
1397         return r;
1398 }
1399
1400 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1401         Unit *unit = NULL;  /* just to appease gcc, initialization is not really necessary */
1402         int r;
1403
1404         assert(m);
1405         assert(type < _JOB_TYPE_MAX);
1406         assert(name);
1407         assert(mode < _JOB_MODE_MAX);
1408
1409         r = manager_load_unit(m, name, NULL, NULL, &unit);
1410         if (r < 0)
1411                 return r;
1412         assert(unit);
1413
1414         return manager_add_job(m, type, unit, mode, e, ret);
1415 }
1416
1417 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1418         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1419         int r;
1420
1421         assert(m);
1422         assert(type < _JOB_TYPE_MAX);
1423         assert(name);
1424         assert(mode < _JOB_MODE_MAX);
1425
1426         r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1427         if (r < 0)
1428                 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1429
1430         return r;
1431 }
1432
1433 Job *manager_get_job(Manager *m, uint32_t id) {
1434         assert(m);
1435
1436         return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1437 }
1438
1439 Unit *manager_get_unit(Manager *m, const char *name) {
1440         assert(m);
1441         assert(name);
1442
1443         return hashmap_get(m->units, name);
1444 }
1445
1446 unsigned manager_dispatch_load_queue(Manager *m) {
1447         Unit *u;
1448         unsigned n = 0;
1449
1450         assert(m);
1451
1452         /* Make sure we are not run recursively */
1453         if (m->dispatching_load_queue)
1454                 return 0;
1455
1456         m->dispatching_load_queue = true;
1457
1458         /* Dispatches the load queue. Takes a unit from the queue and
1459          * tries to load its data until the queue is empty */
1460
1461         while ((u = m->load_queue)) {
1462                 assert(u->in_load_queue);
1463
1464                 unit_load(u);
1465                 n++;
1466         }
1467
1468         m->dispatching_load_queue = false;
1469         return n;
1470 }
1471
1472 int manager_load_unit_prepare(
1473                 Manager *m,
1474                 const char *name,
1475                 const char *path,
1476                 sd_bus_error *e,
1477                 Unit **_ret) {
1478
1479         Unit *ret;
1480         UnitType t;
1481         int r;
1482
1483         assert(m);
1484         assert(name || path);
1485         assert(_ret);
1486
1487         /* This will prepare the unit for loading, but not actually
1488          * load anything from disk. */
1489
1490         if (path && !is_path(path))
1491                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1492
1493         if (!name)
1494                 name = basename(path);
1495
1496         t = unit_name_to_type(name);
1497
1498         if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1499                 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1500                         return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1501
1502                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1503         }
1504
1505         ret = manager_get_unit(m, name);
1506         if (ret) {
1507                 *_ret = ret;
1508                 return 1;
1509         }
1510
1511         ret = unit_new(m, unit_vtable[t]->object_size);
1512         if (!ret)
1513                 return -ENOMEM;
1514
1515         if (path) {
1516                 ret->fragment_path = strdup(path);
1517                 if (!ret->fragment_path) {
1518                         unit_free(ret);
1519                         return -ENOMEM;
1520                 }
1521         }
1522
1523         r = unit_add_name(ret, name);
1524         if (r < 0) {
1525                 unit_free(ret);
1526                 return r;
1527         }
1528
1529         unit_add_to_load_queue(ret);
1530         unit_add_to_dbus_queue(ret);
1531         unit_add_to_gc_queue(ret);
1532
1533         *_ret = ret;
1534
1535         return 0;
1536 }
1537
1538 int manager_load_unit(
1539                 Manager *m,
1540                 const char *name,
1541                 const char *path,
1542                 sd_bus_error *e,
1543                 Unit **_ret) {
1544
1545         int r;
1546
1547         assert(m);
1548         assert(_ret);
1549
1550         /* This will load the service information files, but not actually
1551          * start any services or anything. */
1552
1553         r = manager_load_unit_prepare(m, name, path, e, _ret);
1554         if (r != 0)
1555                 return r;
1556
1557         manager_dispatch_load_queue(m);
1558
1559         *_ret = unit_follow_merge(*_ret);
1560
1561         return 0;
1562 }
1563
1564 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1565         Iterator i;
1566         Job *j;
1567
1568         assert(s);
1569         assert(f);
1570
1571         HASHMAP_FOREACH(j, s->jobs, i)
1572                 job_dump(j, f, prefix);
1573 }
1574
1575 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1576         Iterator i;
1577         Unit *u;
1578         const char *t;
1579
1580         assert(s);
1581         assert(f);
1582
1583         HASHMAP_FOREACH_KEY(u, t, s->units, i)
1584                 if (u->id == t)
1585                         unit_dump(u, f, prefix);
1586 }
1587
1588 void manager_clear_jobs(Manager *m) {
1589         Job *j;
1590
1591         assert(m);
1592
1593         while ((j = hashmap_first(m->jobs)))
1594                 /* No need to recurse. We're cancelling all jobs. */
1595                 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1596 }
1597
1598 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1599         Manager *m = userdata;
1600         Job *j;
1601
1602         assert(source);
1603         assert(m);
1604
1605         while ((j = m->run_queue)) {
1606                 assert(j->installed);
1607                 assert(j->in_run_queue);
1608
1609                 job_run_and_invalidate(j);
1610         }
1611
1612         if (m->n_running_jobs > 0)
1613                 manager_watch_jobs_in_progress(m);
1614
1615         if (m->n_on_console > 0)
1616                 manager_watch_idle_pipe(m);
1617
1618         return 1;
1619 }
1620
1621 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1622         Job *j;
1623         Unit *u;
1624         unsigned n = 0;
1625
1626         assert(m);
1627
1628         if (m->dispatching_dbus_queue)
1629                 return 0;
1630
1631         m->dispatching_dbus_queue = true;
1632
1633         while ((u = m->dbus_unit_queue)) {
1634                 assert(u->in_dbus_queue);
1635
1636                 bus_unit_send_change_signal(u);
1637                 n++;
1638         }
1639
1640         while ((j = m->dbus_job_queue)) {
1641                 assert(j->in_dbus_queue);
1642
1643                 bus_job_send_change_signal(j);
1644                 n++;
1645         }
1646
1647         m->dispatching_dbus_queue = false;
1648
1649         if (m->send_reloading_done) {
1650                 m->send_reloading_done = false;
1651
1652                 bus_manager_send_reloading(m, false);
1653         }
1654
1655         if (m->queued_message)
1656                 bus_send_queued_message(m);
1657
1658         return n;
1659 }
1660
1661 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1662         Manager *m = userdata;
1663         char buf[PATH_MAX+1];
1664         ssize_t n;
1665
1666         n = recv(fd, buf, sizeof(buf), 0);
1667         if (n < 0)
1668                 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1669         if (n == 0) {
1670                 log_error("Got zero-length cgroups agent message, ignoring.");
1671                 return 0;
1672         }
1673         if ((size_t) n >= sizeof(buf)) {
1674                 log_error("Got overly long cgroups agent message, ignoring.");
1675                 return 0;
1676         }
1677
1678         if (memchr(buf, 0, n)) {
1679                 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1680                 return 0;
1681         }
1682         buf[n] = 0;
1683
1684         manager_notify_cgroup_empty(m, buf);
1685         bus_forward_agent_released(m, buf);
1686
1687         return 0;
1688 }
1689
1690 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1691         _cleanup_strv_free_ char **tags = NULL;
1692
1693         assert(m);
1694         assert(u);
1695         assert(buf);
1696
1697         tags = strv_split(buf, "\n\r");
1698         if (!tags) {
1699                 log_oom();
1700                 return;
1701         }
1702
1703         if (UNIT_VTABLE(u)->notify_message)
1704                 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1705         else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1706                 _cleanup_free_ char *x = NULL, *y = NULL;
1707
1708                 x = cescape(buf);
1709                 if (x)
1710                         y = ellipsize(x, 20, 90);
1711                 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1712         }
1713 }
1714
1715 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1716
1717         _cleanup_fdset_free_ FDSet *fds = NULL;
1718         Manager *m = userdata;
1719         char buf[NOTIFY_BUFFER_MAX+1];
1720         struct iovec iovec = {
1721                 .iov_base = buf,
1722                 .iov_len = sizeof(buf)-1,
1723         };
1724         union {
1725                 struct cmsghdr cmsghdr;
1726                 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1727                             CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1728         } control = {};
1729         struct msghdr msghdr = {
1730                 .msg_iov = &iovec,
1731                 .msg_iovlen = 1,
1732                 .msg_control = &control,
1733                 .msg_controllen = sizeof(control),
1734         };
1735
1736         struct cmsghdr *cmsg;
1737         struct ucred *ucred = NULL;
1738         Unit *u1, *u2, *u3;
1739         int r, *fd_array = NULL;
1740         unsigned n_fds = 0;
1741         ssize_t n;
1742
1743         assert(m);
1744         assert(m->notify_fd == fd);
1745
1746         if (revents != EPOLLIN) {
1747                 log_warning("Got unexpected poll event for notify fd.");
1748                 return 0;
1749         }
1750
1751         n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1752         if (n < 0) {
1753                 if (IN_SET(errno, EAGAIN, EINTR))
1754                         return 0; /* Spurious wakeup, try again */
1755
1756                 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1757                  * won't take notification messages anymore, but that's still better than busy looping around this:
1758                  * being woken up over and over again but being unable to actually read the message off the socket. */
1759                 return log_error_errno(errno, "Failed to receive notification message: %m");
1760         }
1761
1762         CMSG_FOREACH(cmsg, &msghdr) {
1763                 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1764
1765                         fd_array = (int*) CMSG_DATA(cmsg);
1766                         n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1767
1768                 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1769                            cmsg->cmsg_type == SCM_CREDENTIALS &&
1770                            cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1771
1772                         ucred = (struct ucred*) CMSG_DATA(cmsg);
1773                 }
1774         }
1775
1776         if (n_fds > 0) {
1777                 assert(fd_array);
1778
1779                 r = fdset_new_array(&fds, fd_array, n_fds);
1780                 if (r < 0) {
1781                         close_many(fd_array, n_fds);
1782                         log_oom();
1783                         return 0;
1784                 }
1785         }
1786
1787         if (!ucred || ucred->pid <= 0) {
1788                 log_warning("Received notify message without valid credentials. Ignoring.");
1789                 return 0;
1790         }
1791
1792         if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1793                 log_warning("Received notify message exceeded maximum size. Ignoring.");
1794                 return 0;
1795         }
1796
1797         /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1798          * trailing NUL byte in the message, but don't expect it. */
1799         if (n > 1 && memchr(buf, 0, n-1)) {
1800                 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1801                 return 0;
1802         }
1803
1804         /* Make sure it's NUL-terminated. */
1805         buf[n] = 0;
1806
1807         /* Notify every unit that might be interested, but try
1808          * to avoid notifying the same one multiple times. */
1809         u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1810         if (u1)
1811                 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1812
1813         u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1814         if (u2 && u2 != u1)
1815                 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1816
1817         u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1818         if (u3 && u3 != u2 && u3 != u1)
1819                 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1820
1821         if (!u1 && !u2 && !u3)
1822                 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1823
1824         if (fdset_size(fds) > 0)
1825                 log_warning("Got extra auxiliary fds with notification message, closing them.");
1826
1827         return 0;
1828 }
1829
1830 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1831         uint64_t iteration;
1832
1833         assert(m);
1834         assert(u);
1835         assert(si);
1836
1837         sd_event_get_iteration(m->event, &iteration);
1838
1839         log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1840
1841         unit_unwatch_pid(u, si->si_pid);
1842
1843         if (UNIT_VTABLE(u)->sigchld_event) {
1844                 if (set_size(u->pids) <= 1 ||
1845                     iteration != u->sigchldgen ||
1846                     unit_main_pid(u) == si->si_pid ||
1847                     unit_control_pid(u) == si->si_pid) {
1848                         UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1849                         u->sigchldgen = iteration;
1850                 } else
1851                         log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1852          }
1853 }
1854
1855 static int manager_dispatch_sigchld(Manager *m) {
1856         assert(m);
1857
1858         for (;;) {
1859                 siginfo_t si = {};
1860
1861                 /* First we call waitd() for a PID and do not reap the
1862                  * zombie. That way we can still access /proc/$PID for
1863                  * it while it is a zombie. */
1864                 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1865
1866                         if (errno == ECHILD)
1867                                 break;
1868
1869                         if (errno == EINTR)
1870                                 continue;
1871
1872                         return -errno;
1873                 }
1874
1875                 if (si.si_pid <= 0)
1876                         break;
1877
1878                 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1879                         _cleanup_free_ char *name = NULL;
1880                         Unit *u1, *u2, *u3;
1881
1882                         get_process_comm(si.si_pid, &name);
1883
1884                         log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1885                                   si.si_pid, strna(name),
1886                                   sigchld_code_to_string(si.si_code),
1887                                   si.si_status,
1888                                   strna(si.si_code == CLD_EXITED
1889                                         ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1890                                         : signal_to_string(si.si_status)));
1891
1892                         /* And now figure out the unit this belongs
1893                          * to, it might be multiple... */
1894                         u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1895                         if (u1)
1896                                 invoke_sigchld_event(m, u1, &si);
1897                         u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1898                         if (u2 && u2 != u1)
1899                                 invoke_sigchld_event(m, u2, &si);
1900                         u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1901                         if (u3 && u3 != u2 && u3 != u1)
1902                                 invoke_sigchld_event(m, u3, &si);
1903                 }
1904
1905                 /* And now, we actually reap the zombie. */
1906                 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1907                         if (errno == EINTR)
1908                                 continue;
1909
1910                         return -errno;
1911                 }
1912         }
1913
1914         return 0;
1915 }
1916
1917 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1918         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1919         int r;
1920
1921         log_debug("Activating special unit %s", name);
1922
1923         r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1924         if (r < 0)
1925                 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1926
1927         return r;
1928 }
1929
1930 static void manager_handle_ctrl_alt_del(Manager *m) {
1931         /* If the user presses C-A-D more than
1932          * 7 times within 2s, we reboot/shutdown immediately,
1933          * unless it was disabled in system.conf */
1934
1935         if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
1936                 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1937         else
1938                 emergency_action(m, m->cad_burst_action, NULL,
1939                                 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1940 }
1941
1942 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1943         Manager *m = userdata;
1944         ssize_t n;
1945         struct signalfd_siginfo sfsi;
1946         bool sigchld = false;
1947         int r;
1948
1949         assert(m);
1950         assert(m->signal_fd == fd);
1951
1952         if (revents != EPOLLIN) {
1953                 log_warning("Got unexpected events from signal file descriptor.");
1954                 return 0;
1955         }
1956
1957         for (;;) {
1958                 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1959                 if (n != sizeof(sfsi)) {
1960                         if (n >= 0) {
1961                                 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1962                                 return 0;
1963                         }
1964
1965                         if (IN_SET(errno, EINTR, EAGAIN))
1966                                 break;
1967
1968                         /* We return an error here, which will kill this handler,
1969                          * to avoid a busy loop on read error. */
1970                         return log_error_errno(errno, "Reading from signal fd failed: %m");
1971                 }
1972
1973                 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1974                                     (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1975                                     ? LOG_DEBUG : LOG_INFO,
1976                                     &sfsi);
1977
1978                 switch (sfsi.ssi_signo) {
1979
1980                 case SIGCHLD:
1981                         sigchld = true;
1982                         break;
1983
1984                 case SIGTERM:
1985                         if (MANAGER_IS_SYSTEM(m)) {
1986                                 /* This is for compatibility with the
1987                                  * original sysvinit */
1988                                 r = verify_run_space_and_log("Refusing to reexecute");
1989                                 if (r >= 0)
1990                                         m->exit_code = MANAGER_REEXECUTE;
1991                                 break;
1992                         }
1993
1994                         /* Fall through */
1995
1996                 case SIGINT:
1997                         if (MANAGER_IS_SYSTEM(m)) {
1998                                 manager_handle_ctrl_alt_del(m);
1999                                 break;
2000                         }
2001
2002                         /* Run the exit target if there is one, if not, just exit. */
2003                         if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2004                                 m->exit_code = MANAGER_EXIT;
2005                                 return 0;
2006                         }
2007
2008                         break;
2009
2010                 case SIGWINCH:
2011                         if (MANAGER_IS_SYSTEM(m))
2012                                 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2013
2014                         /* This is a nop on non-init */
2015                         break;
2016
2017                 case SIGPWR:
2018                         if (MANAGER_IS_SYSTEM(m))
2019                                 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2020
2021                         /* This is a nop on non-init */
2022                         break;
2023
2024                 case SIGUSR1: {
2025                         Unit *u;
2026
2027                         u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2028
2029                         if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2030                                 log_info("Trying to reconnect to bus...");
2031                                 bus_init(m, true);
2032                         }
2033
2034                         if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2035                                 log_info("Loading D-Bus service...");
2036                                 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2037                         }
2038
2039                         break;
2040                 }
2041
2042                 case SIGUSR2: {
2043                         _cleanup_free_ char *dump = NULL;
2044                         _cleanup_fclose_ FILE *f = NULL;
2045                         size_t size;
2046
2047                         f = open_memstream(&dump, &size);
2048                         if (!f) {
2049                                 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2050                                 break;
2051                         }
2052
2053                         manager_dump_units(m, f, "\t");
2054                         manager_dump_jobs(m, f, "\t");
2055
2056                         r = fflush_and_check(f);
2057                         if (r < 0) {
2058                                 log_warning_errno(r, "Failed to write status stream: %m");
2059                                 break;
2060                         }
2061
2062                         log_dump(LOG_INFO, dump);
2063                         break;
2064                 }
2065
2066                 case SIGHUP:
2067                         r = verify_run_space_and_log("Refusing to reload");
2068                         if (r >= 0)
2069                                 m->exit_code = MANAGER_RELOAD;
2070                         break;
2071
2072                 default: {
2073
2074                         /* Starting SIGRTMIN+0 */
2075                         static const char * const target_table[] = {
2076                                 [0] = SPECIAL_DEFAULT_TARGET,
2077                                 [1] = SPECIAL_RESCUE_TARGET,
2078                                 [2] = SPECIAL_EMERGENCY_TARGET,
2079                                 [3] = SPECIAL_HALT_TARGET,
2080                                 [4] = SPECIAL_POWEROFF_TARGET,
2081                                 [5] = SPECIAL_REBOOT_TARGET,
2082                                 [6] = SPECIAL_KEXEC_TARGET
2083                         };
2084
2085                         /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2086                         static const ManagerExitCode code_table[] = {
2087                                 [0] = MANAGER_HALT,
2088                                 [1] = MANAGER_POWEROFF,
2089                                 [2] = MANAGER_REBOOT,
2090                                 [3] = MANAGER_KEXEC
2091                         };
2092
2093                         if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2094                             (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2095                                 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2096                                 manager_start_target(m, target_table[idx],
2097                                                      (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2098                                 break;
2099                         }
2100
2101                         if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2102                             (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2103                                 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2104                                 break;
2105                         }
2106
2107                         switch (sfsi.ssi_signo - SIGRTMIN) {
2108
2109                         case 20:
2110                                 manager_set_show_status(m, SHOW_STATUS_YES);
2111                                 break;
2112
2113                         case 21:
2114                                 manager_set_show_status(m, SHOW_STATUS_NO);
2115                                 break;
2116
2117                         case 22:
2118                                 log_set_max_level(LOG_DEBUG);
2119                                 log_info("Setting log level to debug.");
2120                                 break;
2121
2122                         case 23:
2123                                 log_set_max_level(LOG_INFO);
2124                                 log_info("Setting log level to info.");
2125                                 break;
2126
2127                         case 24:
2128                                 if (MANAGER_IS_USER(m)) {
2129                                         m->exit_code = MANAGER_EXIT;
2130                                         return 0;
2131                                 }
2132
2133                                 /* This is a nop on init */
2134                                 break;
2135
2136                         case 26:
2137                         case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2138                                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2139                                 log_notice("Setting log target to journal-or-kmsg.");
2140                                 break;
2141
2142                         case 27:
2143                                 log_set_target(LOG_TARGET_CONSOLE);
2144                                 log_notice("Setting log target to console.");
2145                                 break;
2146
2147                         case 28:
2148                                 log_set_target(LOG_TARGET_KMSG);
2149                                 log_notice("Setting log target to kmsg.");
2150                                 break;
2151
2152                         default:
2153                                 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2154                         }
2155                 }
2156                 }
2157         }
2158
2159         if (sigchld)
2160                 manager_dispatch_sigchld(m);
2161
2162         return 0;
2163 }
2164
2165 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2166         Manager *m = userdata;
2167         Iterator i;
2168         Unit *u;
2169
2170         assert(m);
2171         assert(m->time_change_fd == fd);
2172
2173         log_struct(LOG_DEBUG,
2174                    "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2175                    LOG_MESSAGE("Time has been changed"),
2176                    NULL);
2177
2178         /* Restart the watch */
2179         m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2180         m->time_change_fd = safe_close(m->time_change_fd);
2181
2182         manager_setup_time_change(m);
2183
2184         HASHMAP_FOREACH(u, m->units, i)
2185                 if (UNIT_VTABLE(u)->time_change)
2186                         UNIT_VTABLE(u)->time_change(u);
2187
2188         return 0;
2189 }
2190
2191 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2192         Manager *m = userdata;
2193
2194         assert(m);
2195         assert(m->idle_pipe[2] == fd);
2196
2197         m->no_console_output = m->n_on_console > 0;
2198
2199         manager_close_idle_pipe(m);
2200
2201         return 0;
2202 }
2203
2204 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2205         Manager *m = userdata;
2206         int r;
2207         uint64_t next;
2208
2209         assert(m);
2210         assert(source);
2211
2212         manager_print_jobs_in_progress(m);
2213
2214         next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2215         r = sd_event_source_set_time(source, next);
2216         if (r < 0)
2217                 return r;
2218
2219         return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2220 }
2221
2222 int manager_loop(Manager *m) {
2223         int r;
2224
2225         RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2226
2227         assert(m);
2228         m->exit_code = MANAGER_OK;
2229
2230         /* Release the path cache */
2231         m->unit_path_cache = set_free_free(m->unit_path_cache);
2232
2233         manager_check_finished(m);
2234
2235         /* There might still be some zombies hanging around from
2236          * before we were exec()'ed. Let's reap them. */
2237         r = manager_dispatch_sigchld(m);
2238         if (r < 0)
2239                 return r;
2240
2241         while (m->exit_code == MANAGER_OK) {
2242                 usec_t wait_usec;
2243
2244                 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2245                         watchdog_ping();
2246
2247                 if (!ratelimit_test(&rl)) {
2248                         /* Yay, something is going seriously wrong, pause a little */
2249                         log_warning("Looping too fast. Throttling execution a little.");
2250                         sleep(1);
2251                 }
2252
2253                 if (manager_dispatch_load_queue(m) > 0)
2254                         continue;
2255
2256                 if (manager_dispatch_gc_job_queue(m) > 0)
2257                         continue;
2258
2259                 if (manager_dispatch_gc_unit_queue(m) > 0)
2260                         continue;
2261
2262                 if (manager_dispatch_cleanup_queue(m) > 0)
2263                         continue;
2264
2265                 if (manager_dispatch_cgroup_queue(m) > 0)
2266                         continue;
2267
2268                 if (manager_dispatch_dbus_queue(m) > 0)
2269                         continue;
2270
2271                 /* Sleep for half the watchdog time */
2272                 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2273                         wait_usec = m->runtime_watchdog / 2;
2274                         if (wait_usec <= 0)
2275                                 wait_usec = 1;
2276                 } else
2277                         wait_usec = USEC_INFINITY;
2278
2279                 r = sd_event_run(m->event, wait_usec);
2280                 if (r < 0)
2281                         return log_error_errno(r, "Failed to run event loop: %m");
2282         }
2283
2284         return m->exit_code;
2285 }
2286
2287 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2288         _cleanup_free_ char *n = NULL;
2289         sd_id128_t invocation_id;
2290         Unit *u;
2291         int r;
2292
2293         assert(m);
2294         assert(s);
2295         assert(_u);
2296
2297         r = unit_name_from_dbus_path(s, &n);
2298         if (r < 0)
2299                 return r;
2300
2301         /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2302          * as invocation ID. */
2303         r = sd_id128_from_string(n, &invocation_id);
2304         if (r >= 0) {
2305                 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2306                 if (u) {
2307                         *_u = u;
2308                         return 0;
2309                 }
2310
2311                 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.", SD_ID128_FORMAT_VAL(invocation_id));
2312         }
2313
2314         /* If this didn't work, we use the suffix as unit name. */
2315         r = manager_load_unit(m, n, NULL, e, &u);
2316         if (r < 0)
2317                 return r;
2318
2319         *_u = u;
2320         return 0;
2321 }
2322
2323 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2324         const char *p;
2325         unsigned id;
2326         Job *j;
2327         int r;
2328
2329         assert(m);
2330         assert(s);
2331         assert(_j);
2332
2333         p = startswith(s, "/org/freedesktop/systemd1/job/");
2334         if (!p)
2335                 return -EINVAL;
2336
2337         r = safe_atou(p, &id);
2338         if (r < 0)
2339                 return r;
2340
2341         j = manager_get_job(m, id);
2342         if (!j)
2343                 return -ENOENT;
2344
2345         *_j = j;
2346
2347         return 0;
2348 }
2349
2350 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2351
2352 #ifdef HAVE_AUDIT
2353         _cleanup_free_ char *p = NULL;
2354         const char *msg;
2355         int audit_fd, r;
2356
2357         if (!MANAGER_IS_SYSTEM(m))
2358                 return;
2359
2360         audit_fd = get_audit_fd();
2361         if (audit_fd < 0)
2362                 return;
2363
2364         /* Don't generate audit events if the service was already
2365          * started and we're just deserializing */
2366         if (MANAGER_IS_RELOADING(m))
2367                 return;
2368
2369         if (u->type != UNIT_SERVICE)
2370                 return;
2371
2372         r = unit_name_to_prefix_and_instance(u->id, &p);
2373         if (r < 0) {
2374                 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2375                 return;
2376         }
2377
2378         msg = strjoina("unit=", p);
2379         if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2380                 if (errno == EPERM)
2381                         /* We aren't allowed to send audit messages?
2382                          * Then let's not retry again. */
2383                         close_audit_fd();
2384                 else
2385                         log_warning_errno(errno, "Failed to send audit message: %m");
2386         }
2387 #endif
2388
2389 }
2390
2391 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2392         static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2393         _cleanup_free_ char *message = NULL;
2394         _cleanup_close_ int fd = -1;
2395         int n = 0;
2396
2397         /* Don't generate plymouth events if the service was already
2398          * started and we're just deserializing */
2399         if (MANAGER_IS_RELOADING(m))
2400                 return;
2401
2402         if (!MANAGER_IS_SYSTEM(m))
2403                 return;
2404
2405         if (detect_container() > 0)
2406                 return;
2407
2408         if (u->type != UNIT_SERVICE &&
2409             u->type != UNIT_MOUNT &&
2410             u->type != UNIT_SWAP)
2411                 return;
2412
2413         /* We set SOCK_NONBLOCK here so that we rather drop the
2414          * message then wait for plymouth */
2415         fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2416         if (fd < 0) {
2417                 log_error_errno(errno, "socket() failed: %m");
2418                 return;
2419         }
2420
2421         if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2422
2423                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2424                         log_error_errno(errno, "connect() failed: %m");
2425                 return;
2426         }
2427
2428         if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2429                 log_oom();
2430                 return;
2431         }
2432
2433         errno = 0;
2434         if (write(fd, message, n + 1) != n + 1)
2435                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2436                         log_error_errno(errno, "Failed to write Plymouth message: %m");
2437 }
2438
2439 int manager_open_serialization(Manager *m, FILE **_f) {
2440         int fd = -1;
2441         FILE *f;
2442
2443         assert(_f);
2444
2445         fd = memfd_create("systemd-serialization", MFD_CLOEXEC);
2446         if (fd < 0) {
2447                 const char *path;
2448
2449                 path = MANAGER_IS_SYSTEM(m) ? "/run/systemd" : "/tmp";
2450                 fd = open_tmpfile_unlinkable(path, O_RDWR|O_CLOEXEC);
2451                 if (fd < 0)
2452                         return -errno;
2453                 log_debug("Serializing state to %s.", path);
2454         } else
2455                 log_debug("Serializing state to memfd.");
2456
2457         f = fdopen(fd, "w+");
2458         if (!f) {
2459                 safe_close(fd);
2460                 return -errno;
2461         }
2462
2463         *_f = f;
2464
2465         return 0;
2466 }
2467
2468 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2469         Iterator i;
2470         Unit *u;
2471         const char *t;
2472         char **e;
2473         int r;
2474
2475         assert(m);
2476         assert(f);
2477         assert(fds);
2478
2479         m->n_reloading++;
2480
2481         fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2482         fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2483         fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2484         fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2485
2486         dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2487         dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2488         dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2489         dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2490
2491         if (!in_initrd()) {
2492                 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2493                 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2494                 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2495                 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2496                 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2497                 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2498                 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2499                 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2500         }
2501
2502         if (!switching_root) {
2503                 STRV_FOREACH(e, m->environment) {
2504                         _cleanup_free_ char *ce;
2505
2506                         ce = cescape(*e);
2507                         if (!ce)
2508                                 return -ENOMEM;
2509
2510                         fprintf(f, "env=%s\n", *e);
2511                 }
2512         }
2513
2514         if (m->notify_fd >= 0) {
2515                 int copy;
2516
2517                 copy = fdset_put_dup(fds, m->notify_fd);
2518                 if (copy < 0)
2519                         return copy;
2520
2521                 fprintf(f, "notify-fd=%i\n", copy);
2522                 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2523         }
2524
2525         if (m->cgroups_agent_fd >= 0) {
2526                 int copy;
2527
2528                 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2529                 if (copy < 0)
2530                         return copy;
2531
2532                 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2533         }
2534
2535         if (m->user_lookup_fds[0] >= 0) {
2536                 int copy0, copy1;
2537
2538                 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2539                 if (copy0 < 0)
2540                         return copy0;
2541
2542                 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2543                 if (copy1 < 0)
2544                         return copy1;
2545
2546                 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2547         }
2548
2549         bus_track_serialize(m->subscribed, f, "subscribed");
2550
2551         r = dynamic_user_serialize(m, f, fds);
2552         if (r < 0)
2553                 return r;
2554
2555         manager_serialize_uid_refs(m, f);
2556         manager_serialize_gid_refs(m, f);
2557
2558         fputc('\n', f);
2559
2560         HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2561                 if (u->id != t)
2562                         continue;
2563
2564                 /* Start marker */
2565                 fputs(u->id, f);
2566                 fputc('\n', f);
2567
2568                 r = unit_serialize(u, f, fds, !switching_root);
2569                 if (r < 0) {
2570                         m->n_reloading--;
2571                         return r;
2572                 }
2573         }
2574
2575         assert(m->n_reloading > 0);
2576         m->n_reloading--;
2577
2578         if (ferror(f))
2579                 return -EIO;
2580
2581         r = bus_fdset_add_all(m, fds);
2582         if (r < 0)
2583                 return r;
2584
2585         return 0;
2586 }
2587
2588 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2589         int r = 0;
2590
2591         assert(m);
2592         assert(f);
2593
2594         log_debug("Deserializing state...");
2595
2596         m->n_reloading++;
2597
2598         for (;;) {
2599                 char line[LINE_MAX], *l;
2600                 const char *val;
2601
2602                 if (!fgets(line, sizeof(line), f)) {
2603                         if (feof(f))
2604                                 r = 0;
2605                         else
2606                                 r = -errno;
2607
2608                         goto finish;
2609                 }
2610
2611                 char_array_0(line);
2612                 l = strstrip(line);
2613
2614                 if (l[0] == 0)
2615                         break;
2616
2617                 if ((val = startswith(l, "current-job-id="))) {
2618                         uint32_t id;
2619
2620                         if (safe_atou32(val, &id) < 0)
2621                                 log_debug("Failed to parse current job id value %s", val);
2622                         else
2623                                 m->current_job_id = MAX(m->current_job_id, id);
2624
2625                 } else if ((val = startswith(l, "n-installed-jobs="))) {
2626                         uint32_t n;
2627
2628                         if (safe_atou32(val, &n) < 0)
2629                                 log_debug("Failed to parse installed jobs counter %s", val);
2630                         else
2631                                 m->n_installed_jobs += n;
2632
2633                 } else if ((val = startswith(l, "n-failed-jobs="))) {
2634                         uint32_t n;
2635
2636                         if (safe_atou32(val, &n) < 0)
2637                                 log_debug("Failed to parse failed jobs counter %s", val);
2638                         else
2639                                 m->n_failed_jobs += n;
2640
2641                 } else if ((val = startswith(l, "taint-usr="))) {
2642                         int b;
2643
2644                         b = parse_boolean(val);
2645                         if (b < 0)
2646                                 log_debug("Failed to parse taint /usr flag %s", val);
2647                         else
2648                                 m->taint_usr = m->taint_usr || b;
2649
2650                 } else if ((val = startswith(l, "firmware-timestamp=")))
2651                         dual_timestamp_deserialize(val, &m->firmware_timestamp);
2652                 else if ((val = startswith(l, "loader-timestamp=")))
2653                         dual_timestamp_deserialize(val, &m->loader_timestamp);
2654                 else if ((val = startswith(l, "kernel-timestamp=")))
2655                         dual_timestamp_deserialize(val, &m->kernel_timestamp);
2656                 else if ((val = startswith(l, "initrd-timestamp=")))
2657                         dual_timestamp_deserialize(val, &m->initrd_timestamp);
2658                 else if ((val = startswith(l, "userspace-timestamp=")))
2659                         dual_timestamp_deserialize(val, &m->userspace_timestamp);
2660                 else if ((val = startswith(l, "finish-timestamp=")))
2661                         dual_timestamp_deserialize(val, &m->finish_timestamp);
2662                 else if ((val = startswith(l, "security-start-timestamp=")))
2663                         dual_timestamp_deserialize(val, &m->security_start_timestamp);
2664                 else if ((val = startswith(l, "security-finish-timestamp=")))
2665                         dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2666                 else if ((val = startswith(l, "generators-start-timestamp=")))
2667                         dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2668                 else if ((val = startswith(l, "generators-finish-timestamp=")))
2669                         dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2670                 else if ((val = startswith(l, "units-load-start-timestamp=")))
2671                         dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2672                 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2673                         dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2674                 else if (startswith(l, "env=")) {
2675                         _cleanup_free_ char *uce = NULL;
2676                         char **e;
2677
2678                         r = cunescape(l + 4, UNESCAPE_RELAX, &uce);
2679                         if (r < 0)
2680                                 goto finish;
2681
2682                         e = strv_env_set(m->environment, uce);
2683                         if (!e) {
2684                                 r = -ENOMEM;
2685                                 goto finish;
2686                         }
2687
2688                         strv_free(m->environment);
2689                         m->environment = e;
2690
2691                 } else if ((val = startswith(l, "notify-fd="))) {
2692                         int fd;
2693
2694                         if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2695                                 log_debug("Failed to parse notify fd: %s", val);
2696                         else {
2697                                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2698                                 safe_close(m->notify_fd);
2699                                 m->notify_fd = fdset_remove(fds, fd);
2700                         }
2701
2702                 } else if ((val = startswith(l, "notify-socket="))) {
2703                         char *n;
2704
2705                         n = strdup(val);
2706                         if (!n) {
2707                                 r = -ENOMEM;
2708                                 goto finish;
2709                         }
2710
2711                         free(m->notify_socket);
2712                         m->notify_socket = n;
2713
2714                 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2715                         int fd;
2716
2717                         if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2718                                 log_debug("Failed to parse cgroups agent fd: %s", val);
2719                         else {
2720                                 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2721                                 safe_close(m->cgroups_agent_fd);
2722                                 m->cgroups_agent_fd = fdset_remove(fds, fd);
2723                         }
2724
2725                 } else if ((val = startswith(l, "user-lookup="))) {
2726                         int fd0, fd1;
2727
2728                         if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2729                                 log_debug("Failed to parse user lookup fd: %s", val);
2730                         else {
2731                                 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2732                                 safe_close_pair(m->user_lookup_fds);
2733                                 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2734                                 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2735                         }
2736
2737                 } else if ((val = startswith(l, "dynamic-user=")))
2738                         dynamic_user_deserialize_one(m, val, fds);
2739                 else if ((val = startswith(l, "destroy-ipc-uid=")))
2740                         manager_deserialize_uid_refs_one(m, val);
2741                 else if ((val = startswith(l, "destroy-ipc-gid=")))
2742                         manager_deserialize_gid_refs_one(m, val);
2743                 else if ((val = startswith(l, "subscribed="))) {
2744
2745                         if (strv_extend(&m->deserialized_subscribed, val) < 0)
2746                                 log_oom();
2747
2748                 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2749                         log_debug("Unknown serialization item '%s'", l);
2750         }
2751
2752         for (;;) {
2753                 Unit *u;
2754                 char name[UNIT_NAME_MAX+2];
2755
2756                 /* Start marker */
2757                 if (!fgets(name, sizeof(name), f)) {
2758                         if (feof(f))
2759                                 r = 0;
2760                         else
2761                                 r = -errno;
2762
2763                         goto finish;
2764                 }
2765
2766                 char_array_0(name);
2767
2768                 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2769                 if (r < 0)
2770                         goto finish;
2771
2772                 r = unit_deserialize(u, f, fds);
2773                 if (r < 0)
2774                         goto finish;
2775         }
2776
2777 finish:
2778         if (ferror(f))
2779                 r = -EIO;
2780
2781         assert(m->n_reloading > 0);
2782         m->n_reloading--;
2783
2784         return r;
2785 }
2786
2787 int manager_reload(Manager *m) {
2788         int r, q;
2789         _cleanup_fclose_ FILE *f = NULL;
2790         _cleanup_fdset_free_ FDSet *fds = NULL;
2791
2792         assert(m);
2793
2794         r = manager_open_serialization(m, &f);
2795         if (r < 0)
2796                 return r;
2797
2798         m->n_reloading++;
2799         bus_manager_send_reloading(m, true);
2800
2801         fds = fdset_new();
2802         if (!fds) {
2803                 m->n_reloading--;
2804                 return -ENOMEM;
2805         }
2806
2807         r = manager_serialize(m, f, fds, false);
2808         if (r < 0) {
2809                 m->n_reloading--;
2810                 return r;
2811         }
2812
2813         if (fseeko(f, 0, SEEK_SET) < 0) {
2814                 m->n_reloading--;
2815                 return -errno;
2816         }
2817
2818         /* From here on there is no way back. */
2819         manager_clear_jobs_and_units(m);
2820         lookup_paths_flush_generator(&m->lookup_paths);
2821         lookup_paths_free(&m->lookup_paths);
2822         dynamic_user_vacuum(m, false);
2823         m->uid_refs = hashmap_free(m->uid_refs);
2824         m->gid_refs = hashmap_free(m->gid_refs);
2825
2826         q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2827         if (q < 0 && r >= 0)
2828                 r = q;
2829
2830         /* Find new unit paths */
2831         q = manager_run_generators(m);
2832         if (q < 0 && r >= 0)
2833                 r = q;
2834
2835         lookup_paths_reduce(&m->lookup_paths);
2836         manager_build_unit_path_cache(m);
2837
2838         /* First, enumerate what we can from all config files */
2839         manager_enumerate(m);
2840
2841         /* Second, deserialize our stored data */
2842         q = manager_deserialize(m, f, fds);
2843         if (q < 0 && r >= 0)
2844                 r = q;
2845
2846         fclose(f);
2847         f = NULL;
2848
2849         /* Re-register notify_fd as event source */
2850         q = manager_setup_notify(m);
2851         if (q < 0 && r >= 0)
2852                 r = q;
2853
2854         q = manager_setup_cgroups_agent(m);
2855         if (q < 0 && r >= 0)
2856                 r = q;
2857
2858         q = manager_setup_user_lookup_fd(m);
2859         if (q < 0 && r >= 0)
2860                 r = q;
2861
2862         /* Third, fire things up! */
2863         manager_coldplug(m);
2864
2865         /* Release any dynamic users no longer referenced */
2866         dynamic_user_vacuum(m, true);
2867
2868         /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2869         manager_vacuum_uid_refs(m);
2870         manager_vacuum_gid_refs(m);
2871
2872         /* Sync current state of bus names with our set of listening units */
2873         if (m->api_bus)
2874                 manager_sync_bus_names(m, m->api_bus);
2875
2876         assert(m->n_reloading > 0);
2877         m->n_reloading--;
2878
2879         m->send_reloading_done = true;
2880
2881         return r;
2882 }
2883
2884 void manager_reset_failed(Manager *m) {
2885         Unit *u;
2886         Iterator i;
2887
2888         assert(m);
2889
2890         HASHMAP_FOREACH(u, m->units, i)
2891                 unit_reset_failed(u);
2892 }
2893
2894 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2895         Unit *u;
2896
2897         assert(m);
2898         assert(name);
2899
2900         /* Returns true if the unit is inactive or going down */
2901         u = manager_get_unit(m, name);
2902         if (!u)
2903                 return true;
2904
2905         return unit_inactive_or_pending(u);
2906 }
2907
2908 static void manager_notify_finished(Manager *m) {
2909         char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2910         usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2911
2912         if (m->test_run)
2913                 return;
2914
2915         if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2916
2917                 /* Note that m->kernel_usec.monotonic is always at 0,
2918                  * and m->firmware_usec.monotonic and
2919                  * m->loader_usec.monotonic should be considered
2920                  * negative values. */
2921
2922                 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2923                 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2924                 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2925                 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2926
2927                 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2928
2929                         kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2930                         initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2931
2932                         log_struct(LOG_INFO,
2933                                    "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2934                                    "KERNEL_USEC="USEC_FMT, kernel_usec,
2935                                    "INITRD_USEC="USEC_FMT, initrd_usec,
2936                                    "USERSPACE_USEC="USEC_FMT, userspace_usec,
2937                                    LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2938                                                format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2939                                                format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2940                                                format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2941                                                format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2942                                    NULL);
2943                 } else {
2944                         kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2945                         initrd_usec = 0;
2946
2947                         log_struct(LOG_INFO,
2948                                    "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2949                                    "KERNEL_USEC="USEC_FMT, kernel_usec,
2950                                    "USERSPACE_USEC="USEC_FMT, userspace_usec,
2951                                    LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2952                                                format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2953                                                format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2954                                                format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2955                                    NULL);
2956                 }
2957         } else {
2958                 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2959                 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2960
2961                 log_struct(LOG_INFO,
2962                            "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
2963                            "USERSPACE_USEC="USEC_FMT, userspace_usec,
2964                            LOG_MESSAGE("Startup finished in %s.",
2965                                        format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2966                            NULL);
2967         }
2968
2969         bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2970
2971         sd_notifyf(false,
2972                    "READY=1\n"
2973                    "STATUS=Startup finished in %s.",
2974                    format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2975 }
2976
2977 void manager_check_finished(Manager *m) {
2978         assert(m);
2979
2980         if (MANAGER_IS_RELOADING(m))
2981                 return;
2982
2983         /* Verify that we are actually running currently. Initially
2984          * the exit code is set to invalid, and during operation it is
2985          * then set to MANAGER_OK */
2986         if (m->exit_code != MANAGER_OK)
2987                 return;
2988
2989         if (hashmap_size(m->jobs) > 0) {
2990                 if (m->jobs_in_progress_event_source)
2991                         /* Ignore any failure, this is only for feedback */
2992                         (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2993
2994                 return;
2995         }
2996
2997         manager_flip_auto_status(m, false);
2998
2999         /* Notify Type=idle units that we are done now */
3000         manager_close_idle_pipe(m);
3001
3002         /* Turn off confirm spawn now */
3003         m->confirm_spawn = NULL;
3004
3005         /* No need to update ask password status when we're going non-interactive */
3006         manager_close_ask_password(m);
3007
3008         /* This is no longer the first boot */
3009         manager_set_first_boot(m, false);
3010
3011         if (dual_timestamp_is_set(&m->finish_timestamp))
3012                 return;
3013
3014         dual_timestamp_get(&m->finish_timestamp);
3015
3016         manager_notify_finished(m);
3017
3018         manager_invalidate_startup_units(m);
3019 }
3020
3021 static int manager_run_generators(Manager *m) {
3022         _cleanup_strv_free_ char **paths = NULL;
3023         const char *argv[5];
3024         char **path;
3025         int r;
3026
3027         assert(m);
3028
3029         if (m->test_run)
3030                 return 0;
3031
3032         paths = generator_binary_paths(m->unit_file_scope);
3033         if (!paths)
3034                 return log_oom();
3035
3036         /* Optimize by skipping the whole process by not creating output directories
3037          * if no generators are found. */
3038         STRV_FOREACH(path, paths) {
3039                 if (access(*path, F_OK) >= 0)
3040                         goto found;
3041                 if (errno != ENOENT)
3042                         log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3043         }
3044
3045         return 0;
3046
3047  found:
3048         r = lookup_paths_mkdir_generator(&m->lookup_paths);
3049         if (r < 0)
3050                 goto finish;
3051
3052         argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3053         argv[1] = m->lookup_paths.generator;
3054         argv[2] = m->lookup_paths.generator_early;
3055         argv[3] = m->lookup_paths.generator_late;
3056         argv[4] = NULL;
3057
3058         RUN_WITH_UMASK(0022)
3059                 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, (char**) argv);
3060
3061 finish:
3062         lookup_paths_trim_generator(&m->lookup_paths);
3063         return r;
3064 }
3065
3066 int manager_environment_add(Manager *m, char **minus, char **plus) {
3067         char **a = NULL, **b = NULL, **l;
3068         assert(m);
3069
3070         l = m->environment;
3071
3072         if (!strv_isempty(minus)) {
3073                 a = strv_env_delete(l, 1, minus);
3074                 if (!a)
3075                         return -ENOMEM;
3076
3077                 l = a;
3078         }
3079
3080         if (!strv_isempty(plus)) {
3081                 b = strv_env_merge(2, l, plus);
3082                 if (!b) {
3083                         strv_free(a);
3084                         return -ENOMEM;
3085                 }
3086
3087                 l = b;
3088         }
3089
3090         if (m->environment != l)
3091                 strv_free(m->environment);
3092         if (a != l)
3093                 strv_free(a);
3094         if (b != l)
3095                 strv_free(b);
3096
3097         m->environment = l;
3098         manager_clean_environment(m);
3099         strv_sort(m->environment);
3100
3101         return 0;
3102 }
3103
3104 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3105         int i;
3106
3107         assert(m);
3108
3109         for (i = 0; i < _RLIMIT_MAX; i++) {
3110                 m->rlimit[i] = mfree(m->rlimit[i]);
3111
3112                 if (!default_rlimit[i])
3113                         continue;
3114
3115                 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3116                 if (!m->rlimit[i])
3117                         return log_oom();
3118         }
3119
3120         return 0;
3121 }
3122
3123 void manager_recheck_journal(Manager *m) {
3124         Unit *u;
3125
3126         assert(m);
3127
3128         if (!MANAGER_IS_SYSTEM(m))
3129                 return;
3130
3131         u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3132         if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3133                 log_close_journal();
3134                 return;
3135         }
3136
3137         u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3138         if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3139                 log_close_journal();
3140                 return;
3141         }
3142
3143         /* Hmm, OK, so the socket is fully up and the service is up
3144          * too, then let's make use of the thing. */
3145         log_open();
3146 }
3147
3148 void manager_set_show_status(Manager *m, ShowStatus mode) {
3149         assert(m);
3150         assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3151
3152         if (!MANAGER_IS_SYSTEM(m))
3153                 return;
3154
3155         if (m->show_status != mode)
3156                 log_debug("%s showing of status.",
3157                           mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3158         m->show_status = mode;
3159
3160         if (mode > 0)
3161                 (void) touch("/run/systemd/show-status");
3162         else
3163                 (void) unlink("/run/systemd/show-status");
3164 }
3165
3166 static bool manager_get_show_status(Manager *m, StatusType type) {
3167         assert(m);
3168
3169         if (!MANAGER_IS_SYSTEM(m))
3170                 return false;
3171
3172         if (m->no_console_output)
3173                 return false;
3174
3175         if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3176                 return false;
3177
3178         /* If we cannot find out the status properly, just proceed. */
3179         if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3180                 return false;
3181
3182         if (m->show_status > 0)
3183                 return true;
3184
3185         return false;
3186 }
3187
3188 const char *manager_get_confirm_spawn(Manager *m) {
3189         static int last_errno = 0;
3190         const char *vc = m->confirm_spawn;
3191         struct stat st;
3192         int r;
3193
3194         /* Here's the deal: we want to test the validity of the console but don't want
3195          * PID1 to go through the whole console process which might block. But we also
3196          * want to warn the user only once if something is wrong with the console so we
3197          * cannot do the sanity checks after spawning our children. So here we simply do
3198          * really basic tests to hopefully trap common errors.
3199          *
3200          * If the console suddenly disappear at the time our children will really it
3201          * then they will simply fail to acquire it and a positive answer will be
3202          * assumed. New children will fallback to /dev/console though.
3203          *
3204          * Note: TTYs are devices that can come and go any time, and frequently aren't
3205          * available yet during early boot (consider a USB rs232 dongle...). If for any
3206          * reason the configured console is not ready, we fallback to the default
3207          * console. */
3208
3209         if (!vc || path_equal(vc, "/dev/console"))
3210                 return vc;
3211
3212         r = stat(vc, &st);
3213         if (r < 0)
3214                 goto fail;
3215
3216         if (!S_ISCHR(st.st_mode)) {
3217                 errno = ENOTTY;
3218                 goto fail;
3219         }
3220
3221         last_errno = 0;
3222         return vc;
3223 fail:
3224         if (last_errno != errno) {
3225                 last_errno = errno;
3226                 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3227         }
3228         return "/dev/console";
3229 }
3230
3231 void manager_set_first_boot(Manager *m, bool b) {
3232         assert(m);
3233
3234         if (!MANAGER_IS_SYSTEM(m))
3235                 return;
3236
3237         if (m->first_boot != (int) b) {
3238                 if (b)
3239                         (void) touch("/run/systemd/first-boot");
3240                 else
3241                         (void) unlink("/run/systemd/first-boot");
3242         }
3243
3244         m->first_boot = b;
3245 }
3246
3247 void manager_disable_confirm_spawn(void) {
3248         (void) touch("/run/systemd/confirm_spawn_disabled");
3249 }
3250
3251 bool manager_is_confirm_spawn_disabled(Manager *m) {
3252         if (!m->confirm_spawn)
3253                 return true;
3254
3255         return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3256 }
3257
3258 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3259         va_list ap;
3260
3261         /* If m is NULL, assume we're after shutdown and let the messages through. */
3262
3263         if (m && !manager_get_show_status(m, type))
3264                 return;
3265
3266         /* XXX We should totally drop the check for ephemeral here
3267          * and thus effectively make 'Type=idle' pointless. */
3268         if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3269                 return;
3270
3271         va_start(ap, format);
3272         status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3273         va_end(ap);
3274 }
3275
3276 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3277         char p[strlen(path)+1];
3278
3279         assert(m);
3280         assert(path);
3281
3282         strcpy(p, path);
3283         path_kill_slashes(p);
3284
3285         return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3286 }
3287
3288 const char *manager_get_runtime_prefix(Manager *m) {
3289         assert(m);
3290
3291         return MANAGER_IS_SYSTEM(m) ?
3292                "/run" :
3293                getenv("XDG_RUNTIME_DIR");
3294 }
3295
3296 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3297         unsigned size;
3298         int r;
3299
3300         assert(m);
3301         assert(u->manager == m);
3302
3303         size = set_size(m->failed_units);
3304
3305         if (failed) {
3306                 r = set_ensure_allocated(&m->failed_units, NULL);
3307                 if (r < 0)
3308                         return log_oom();
3309
3310                 if (set_put(m->failed_units, u) < 0)
3311                         return log_oom();
3312         } else
3313                 (void) set_remove(m->failed_units, u);
3314
3315         if (set_size(m->failed_units) != size)
3316                 bus_manager_send_change_signal(m);
3317
3318         return 0;
3319 }
3320
3321 ManagerState manager_state(Manager *m) {
3322         Unit *u;
3323
3324         assert(m);
3325
3326         /* Did we ever finish booting? If not then we are still starting up */
3327         if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3328
3329                 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3330                 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3331                         return MANAGER_INITIALIZING;
3332
3333                 return MANAGER_STARTING;
3334         }
3335
3336         /* Is the special shutdown target queued? If so, we are in shutdown state */
3337         u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3338         if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3339                 return MANAGER_STOPPING;
3340
3341         /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3342         u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3343         if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3344                   (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3345                 return MANAGER_MAINTENANCE;
3346
3347         u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3348         if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3349                   (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3350                 return MANAGER_MAINTENANCE;
3351
3352         /* Are there any failed units? If so, we are in degraded mode */
3353         if (set_size(m->failed_units) > 0)
3354                 return MANAGER_DEGRADED;
3355
3356         return MANAGER_RUNNING;
3357 }
3358
3359 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3360
3361 static void manager_unref_uid_internal(
3362                 Manager *m,
3363                 Hashmap **uid_refs,
3364                 uid_t uid,
3365                 bool destroy_now,
3366                 int (*_clean_ipc)(uid_t uid)) {
3367
3368         uint32_t c, n;
3369
3370         assert(m);
3371         assert(uid_refs);
3372         assert(uid_is_valid(uid));
3373         assert(_clean_ipc);
3374
3375         /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3376          * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3377          *
3378          * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3379          * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3380          * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3381          * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3382
3383         assert_cc(sizeof(uid_t) == sizeof(gid_t));
3384         assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3385
3386         if (uid == 0) /* We don't keep track of root, and will never destroy it */
3387                 return;
3388
3389         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3390
3391         n = c & ~DESTROY_IPC_FLAG;
3392         assert(n > 0);
3393         n--;
3394
3395         if (destroy_now && n == 0) {
3396                 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3397
3398                 if (c & DESTROY_IPC_FLAG) {
3399                         log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3400                                   _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3401                                   uid);
3402                         (void) _clean_ipc(uid);
3403                 }
3404         } else {
3405                 c = n | (c & DESTROY_IPC_FLAG);
3406                 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3407         }
3408 }
3409
3410 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3411         manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3412 }
3413
3414 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3415         manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3416 }
3417
3418 static int manager_ref_uid_internal(
3419                 Manager *m,
3420                 Hashmap **uid_refs,
3421                 uid_t uid,
3422                 bool clean_ipc) {
3423
3424         uint32_t c, n;
3425         int r;
3426
3427         assert(m);
3428         assert(uid_refs);
3429         assert(uid_is_valid(uid));
3430
3431         /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3432          * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3433
3434         assert_cc(sizeof(uid_t) == sizeof(gid_t));
3435         assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3436
3437         if (uid == 0) /* We don't keep track of root, and will never destroy it */
3438                 return 0;
3439
3440         r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3441         if (r < 0)
3442                 return r;
3443
3444         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3445
3446         n = c & ~DESTROY_IPC_FLAG;
3447         n++;
3448
3449         if (n & DESTROY_IPC_FLAG) /* check for overflow */
3450                 return -EOVERFLOW;
3451
3452         c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3453
3454         return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3455 }
3456
3457 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3458         return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3459 }
3460
3461 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3462         return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3463 }
3464
3465 static void manager_vacuum_uid_refs_internal(
3466                 Manager *m,
3467                 Hashmap **uid_refs,
3468                 int (*_clean_ipc)(uid_t uid)) {
3469
3470         Iterator i;
3471         void *p, *k;
3472
3473         assert(m);
3474         assert(uid_refs);
3475         assert(_clean_ipc);
3476
3477         HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3478                 uint32_t c, n;
3479                 uid_t uid;
3480
3481                 uid = PTR_TO_UID(k);
3482                 c = PTR_TO_UINT32(p);
3483
3484                 n = c & ~DESTROY_IPC_FLAG;
3485                 if (n > 0)
3486                         continue;
3487
3488                 if (c & DESTROY_IPC_FLAG) {
3489                         log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3490                                   _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3491                                   uid);
3492                         (void) _clean_ipc(uid);
3493                 }
3494
3495                 assert_se(hashmap_remove(*uid_refs, k) == p);
3496         }
3497 }
3498
3499 void manager_vacuum_uid_refs(Manager *m) {
3500         manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3501 }
3502
3503 void manager_vacuum_gid_refs(Manager *m) {
3504         manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3505 }
3506
3507 static void manager_serialize_uid_refs_internal(
3508                 Manager *m,
3509                 FILE *f,
3510                 Hashmap **uid_refs,
3511                 const char *field_name) {
3512
3513         Iterator i;
3514         void *p, *k;
3515
3516         assert(m);
3517         assert(f);
3518         assert(uid_refs);
3519         assert(field_name);
3520
3521         /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3522          * of it is better rebuild after a reload/reexec. */
3523
3524         HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3525                 uint32_t c;
3526                 uid_t uid;
3527
3528                 uid = PTR_TO_UID(k);
3529                 c = PTR_TO_UINT32(p);
3530
3531                 if (!(c & DESTROY_IPC_FLAG))
3532                         continue;
3533
3534                 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3535         }
3536 }
3537
3538 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3539         manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3540 }
3541
3542 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3543         manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3544 }
3545
3546 static void manager_deserialize_uid_refs_one_internal(
3547                 Manager *m,
3548                 Hashmap** uid_refs,
3549                 const char *value) {
3550
3551         uid_t uid;
3552         uint32_t c;
3553         int r;
3554
3555         assert(m);
3556         assert(uid_refs);
3557         assert(value);
3558
3559         r = parse_uid(value, &uid);
3560         if (r < 0 || uid == 0) {
3561                 log_debug("Unable to parse UID reference serialization");
3562                 return;
3563         }
3564
3565         r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3566         if (r < 0) {
3567                 log_oom();
3568                 return;
3569         }
3570
3571         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3572         if (c & DESTROY_IPC_FLAG)
3573                 return;
3574
3575         c |= DESTROY_IPC_FLAG;
3576
3577         r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3578         if (r < 0) {
3579                 log_debug("Failed to add UID reference entry");
3580                 return;
3581         }
3582 }
3583
3584 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3585         manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3586 }
3587
3588 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3589         manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3590 }
3591
3592 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3593         struct buffer {
3594                 uid_t uid;
3595                 gid_t gid;
3596                 char unit_name[UNIT_NAME_MAX+1];
3597         } _packed_ buffer;
3598
3599         Manager *m = userdata;
3600         ssize_t l;
3601         size_t n;
3602         Unit *u;
3603
3604         assert_se(source);
3605         assert_se(m);
3606
3607         /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3608          * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3609          * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3610
3611         l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3612         if (l < 0) {
3613                 if (errno == EINTR || errno == EAGAIN)
3614                         return 0;
3615
3616                 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3617         }
3618
3619         if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3620                 log_warning("Received too short user lookup message, ignoring.");
3621                 return 0;
3622         }
3623
3624         if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3625                 log_warning("Received too long user lookup message, ignoring.");
3626                 return 0;
3627         }
3628
3629         if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3630                 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3631                 return 0;
3632         }
3633
3634         n = (size_t) l - offsetof(struct buffer, unit_name);
3635         if (memchr(buffer.unit_name, 0, n)) {
3636                 log_warning("Received lookup message with embedded NUL character, ignoring.");
3637                 return 0;
3638         }
3639
3640         buffer.unit_name[n] = 0;
3641         u = manager_get_unit(m, buffer.unit_name);
3642         if (!u) {
3643                 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3644                 return 0;
3645         }
3646
3647         log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3648
3649         unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3650         return 0;
3651 }
3652
3653 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3654         [MANAGER_INITIALIZING] = "initializing",
3655         [MANAGER_STARTING] = "starting",
3656         [MANAGER_RUNNING] = "running",
3657         [MANAGER_DEGRADED] = "degraded",
3658         [MANAGER_MAINTENANCE] = "maintenance",
3659         [MANAGER_STOPPING] = "stopping",
3660 };
3661
3662 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);