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 #include "sd-path.h"
  40
  41 #include "alloc-util.h"
  42 #include "audit-fd.h"
  43 #include "boot-timestamps.h"
  44 #include "bus-common-errors.h"
  45 #include "bus-error.h"
  46 #include "bus-kernel.h"
  47 #include "bus-util.h"
  48 #include "clean-ipc.h"
  49 #include "dbus-job.h"
  50 #include "dbus-manager.h"
  51 #include "dbus-unit.h"
  52 #include "dbus.h"
  53 #include "dirent-util.h"
  54 #include "env-util.h"
  55 #include "escape.h"
  56 #include "execute.h"
  57 #include "exec-util.h"
  58 #include "exit-status.h"
  59 #include "fd-util.h"
  60 #include "fileio.h"
  61 #include "fs-util.h"
  62 #include "hashmap.h"
  63 #include "io-util.h"
  64 #include "locale-setup.h"
  65 #include "log.h"
  66 #include "macro.h"
  67 #include "manager.h"
  68 #include "missing.h"
  69 #include "mkdir.h"
  70 #include "parse-util.h"
  71 #include "path-lookup.h"
  72 #include "path-util.h"
  73 #include "process-util.h"
  74 #include "ratelimit.h"
  75 #include "rm-rf.h"
  76 #include "signal-util.h"
  77 #include "special.h"
  78 #include "stat-util.h"
  79 #include "string-table.h"
  80 #include "string-util.h"
  81 #include "strv.h"
  82 #include "terminal-util.h"
  83 #include "time-util.h"
  84 #include "transaction.h"
  85 #include "umask-util.h"
  86 #include "unit-name.h"
  87 #include "user-util.h"
  88 #include "util.h"
  89 #include "virt.h"
  90 #include "watchdog.h"
  91
  92 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
  93 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
  94
  95 /* Initial delay and the interval for printing status messages about running jobs */
  96 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
  97 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
  98 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
  99
 100 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 101 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 102 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 103 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 104 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 105 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 106 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
 107 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
 108 static int manager_run_environment_generators(Manager *m);
 109 static int manager_run_generators(Manager *m);
 110
 111 static void manager_watch_jobs_in_progress(Manager *m) {
 112         usec_t next;
 113         int r;
 114
 115         assert(m);
 116
 117         /* We do not want to show the cylon animation if the user
 118          * needs to confirm service executions otherwise confirmation
 119          * messages will be screwed by the cylon animation. */
 120         if (!manager_is_confirm_spawn_disabled(m))
 121                 return;
 122
 123         if (m->jobs_in_progress_event_source)
 124                 return;
 125
 126         next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
 127         r = sd_event_add_time(
 128                         m->event,
 129                         &m->jobs_in_progress_event_source,
 130                         CLOCK_MONOTONIC,
 131                         next, 0,
 132                         manager_dispatch_jobs_in_progress, m);
 133         if (r < 0)
 134                 return;
 135
 136         (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
 137 }
 138
 139 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
 140
 141 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
 142         char *p = buffer;
 143
 144         assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
 145         assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
 146
 147         if (pos > 1) {
 148                 if (pos > 2)
 149                         p = mempset(p, ' ', pos-2);
 150                 if (log_get_show_color())
 151                         p = stpcpy(p, ANSI_RED);
 152                 *p++ = '*';
 153         }
 154
 155         if (pos > 0 && pos <= width) {
 156                 if (log_get_show_color())
 157                         p = stpcpy(p, ANSI_HIGHLIGHT_RED);
 158                 *p++ = '*';
 159         }
 160
 161         if (log_get_show_color())
 162                 p = stpcpy(p, ANSI_NORMAL);
 163
 164         if (pos < width) {
 165                 if (log_get_show_color())
 166                         p = stpcpy(p, ANSI_RED);
 167                 *p++ = '*';
 168                 if (pos < width-1)
 169                         p = mempset(p, ' ', width-1-pos);
 170                 if (log_get_show_color())
 171                         strcpy(p, ANSI_NORMAL);
 172         }
 173 }
 174
 175 void manager_flip_auto_status(Manager *m, bool enable) {
 176         assert(m);
 177
 178         if (enable) {
 179                 if (m->show_status == SHOW_STATUS_AUTO)
 180                         manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
 181         } else {
 182                 if (m->show_status == SHOW_STATUS_TEMPORARY)
 183                         manager_set_show_status(m, SHOW_STATUS_AUTO);
 184         }
 185 }
 186
 187 static void manager_print_jobs_in_progress(Manager *m) {
 188         _cleanup_free_ char *job_of_n = NULL;
 189         Iterator i;
 190         Job *j;
 191         unsigned counter = 0, print_nr;
 192         char cylon[6 + CYLON_BUFFER_EXTRA + 1];
 193         unsigned cylon_pos;
 194         char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
 195         uint64_t x;
 196
 197         assert(m);
 198         assert(m->n_running_jobs > 0);
 199
 200         manager_flip_auto_status(m, true);
 201
 202         print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
 203
 204         HASHMAP_FOREACH(j, m->jobs, i)
 205                 if (j->state == JOB_RUNNING && counter++ == print_nr)
 206                         break;
 207
 208         /* m->n_running_jobs must be consistent with the contents of m->jobs,
 209          * so the above loop must have succeeded in finding j. */
 210         assert(counter == print_nr + 1);
 211         assert(j);
 212
 213         cylon_pos = m->jobs_in_progress_iteration % 14;
 214         if (cylon_pos >= 8)
 215                 cylon_pos = 14 - cylon_pos;
 216         draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
 217
 218         m->jobs_in_progress_iteration++;
 219
 220         if (m->n_running_jobs > 1) {
 221                 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
 222                         job_of_n = NULL;
 223         }
 224
 225         format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
 226         if (job_get_timeout(j, &x) > 0)
 227                 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
 228
 229         manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
 230                               "%sA %s job is running for %s (%s / %s)",
 231                               strempty(job_of_n),
 232                               job_type_to_string(j->type),
 233                               unit_description(j->unit),
 234                               time, limit);
 235 }
 236
 237 static int have_ask_password(void) {
 238         _cleanup_closedir_ DIR *dir;
 239         struct dirent *de;
 240
 241         dir = opendir("/run/systemd/ask-password");
 242         if (!dir) {
 243                 if (errno == ENOENT)
 244                         return false;
 245                 else
 246                         return -errno;
 247         }
 248
 249         FOREACH_DIRENT_ALL(de, dir, return -errno) {
 250                 if (startswith(de->d_name, "ask."))
 251                         return true;
 252         }
 253         return false;
 254 }
 255
 256 static int manager_dispatch_ask_password_fd(sd_event_source *source,
 257                                             int fd, uint32_t revents, void *userdata) {
 258         Manager *m = userdata;
 259
 260         assert(m);
 261
 262         flush_fd(fd);
 263
 264         m->have_ask_password = have_ask_password();
 265         if (m->have_ask_password < 0)
 266                 /* Log error but continue. Negative have_ask_password
 267                  * is treated as unknown status. */
 268                 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
 269
 270         return 0;
 271 }
 272
 273 static void manager_close_ask_password(Manager *m) {
 274         assert(m);
 275
 276         m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
 277         m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
 278         m->have_ask_password = -EINVAL;
 279 }
 280
 281 static int manager_check_ask_password(Manager *m) {
 282         int r;
 283
 284         assert(m);
 285
 286         if (!m->ask_password_event_source) {
 287                 assert(m->ask_password_inotify_fd < 0);
 288
 289                 mkdir_p_label("/run/systemd/ask-password", 0755);
 290
 291                 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
 292                 if (m->ask_password_inotify_fd < 0)
 293                         return log_error_errno(errno, "inotify_init1() failed: %m");
 294
 295                 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
 296                         log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
 297                         manager_close_ask_password(m);
 298                         return -errno;
 299                 }
 300
 301                 r = sd_event_add_io(m->event, &m->ask_password_event_source,
 302                                     m->ask_password_inotify_fd, EPOLLIN,
 303                                     manager_dispatch_ask_password_fd, m);
 304                 if (r < 0) {
 305                         log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
 306                         manager_close_ask_password(m);
 307                         return -errno;
 308                 }
 309
 310                 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
 311
 312                 /* Queries might have been added meanwhile... */
 313                 manager_dispatch_ask_password_fd(m->ask_password_event_source,
 314                                                  m->ask_password_inotify_fd, EPOLLIN, m);
 315         }
 316
 317         return m->have_ask_password;
 318 }
 319
 320 static int manager_watch_idle_pipe(Manager *m) {
 321         int r;
 322
 323         assert(m);
 324
 325         if (m->idle_pipe_event_source)
 326                 return 0;
 327
 328         if (m->idle_pipe[2] < 0)
 329                 return 0;
 330
 331         r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
 332         if (r < 0)
 333                 return log_error_errno(r, "Failed to watch idle pipe: %m");
 334
 335         (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
 336
 337         return 0;
 338 }
 339
 340 static void manager_close_idle_pipe(Manager *m) {
 341         assert(m);
 342
 343         m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
 344
 345         safe_close_pair(m->idle_pipe);
 346         safe_close_pair(m->idle_pipe + 2);
 347 }
 348
 349 static int manager_setup_time_change(Manager *m) {
 350         int r;
 351
 352         /* We only care for the cancellation event, hence we set the
 353          * timeout to the latest possible value. */
 354         struct itimerspec its = {
 355                 .it_value.tv_sec = TIME_T_MAX,
 356         };
 357
 358         assert(m);
 359         assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
 360
 361         if (m->test_run)
 362                 return 0;
 363
 364         /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
 365          * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
 366
 367         m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
 368         if (m->time_change_fd < 0)
 369                 return log_error_errno(errno, "Failed to create timerfd: %m");
 370
 371         if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
 372                 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
 373                 m->time_change_fd = safe_close(m->time_change_fd);
 374                 return 0;
 375         }
 376
 377         r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
 378         if (r < 0)
 379                 return log_error_errno(r, "Failed to create time change event source: %m");
 380
 381         (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
 382
 383         log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
 384
 385         return 0;
 386 }
 387
 388 static int enable_special_signals(Manager *m) {
 389         _cleanup_close_ int fd = -1;
 390
 391         assert(m);
 392
 393         if (m->test_run)
 394                 return 0;
 395
 396         /* Enable that we get SIGINT on control-alt-del. In containers
 397          * this will fail with EPERM (older) or EINVAL (newer), so
 398          * ignore that. */
 399         if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
 400                 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
 401
 402         fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
 403         if (fd < 0) {
 404                 /* Support systems without virtual console */
 405                 if (fd != -ENOENT)
 406                         log_warning_errno(errno, "Failed to open /dev/tty0: %m");
 407         } else {
 408                 /* Enable that we get SIGWINCH on kbrequest */
 409                 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
 410                         log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
 411         }
 412
 413         return 0;
 414 }
 415
 416 static int manager_setup_signals(Manager *m) {
 417         struct sigaction sa = {
 418                 .sa_handler = SIG_DFL,
 419                 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 420         };
 421         sigset_t mask;
 422         int r;
 423
 424         assert(m);
 425
 426         assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 427
 428         /* We make liberal use of realtime signals here. On
 429          * Linux/glibc we have 30 of them (with the exception of Linux
 430          * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
 431          * (aka SIGRTMAX). */
 432
 433         assert_se(sigemptyset(&mask) == 0);
 434         sigset_add_many(&mask,
 435                         SIGCHLD,     /* Child died */
 436                         SIGTERM,     /* Reexecute daemon */
 437                         SIGHUP,      /* Reload configuration */
 438                         SIGUSR1,     /* systemd/upstart: reconnect to D-Bus */
 439                         SIGUSR2,     /* systemd: dump status */
 440                         SIGINT,      /* Kernel sends us this on control-alt-del */
 441                         SIGWINCH,    /* Kernel sends us this on kbrequest (alt-arrowup) */
 442                         SIGPWR,      /* Some kernel drivers and upsd send us this on power failure */
 443
 444                         SIGRTMIN+0,  /* systemd: start default.target */
 445                         SIGRTMIN+1,  /* systemd: isolate rescue.target */
 446                         SIGRTMIN+2,  /* systemd: isolate emergency.target */
 447                         SIGRTMIN+3,  /* systemd: start halt.target */
 448                         SIGRTMIN+4,  /* systemd: start poweroff.target */
 449                         SIGRTMIN+5,  /* systemd: start reboot.target */
 450                         SIGRTMIN+6,  /* systemd: start kexec.target */
 451
 452                         /* ... space for more special targets ... */
 453
 454                         SIGRTMIN+13, /* systemd: Immediate halt */
 455                         SIGRTMIN+14, /* systemd: Immediate poweroff */
 456                         SIGRTMIN+15, /* systemd: Immediate reboot */
 457                         SIGRTMIN+16, /* systemd: Immediate kexec */
 458
 459                         /* ... space for more immediate system state changes ... */
 460
 461                         SIGRTMIN+20, /* systemd: enable status messages */
 462                         SIGRTMIN+21, /* systemd: disable status messages */
 463                         SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
 464                         SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
 465                         SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
 466
 467                         /* .. one free signal here ... */
 468
 469 #if !defined(__hppa64__) && !defined(__hppa__)
 470                         /* Apparently Linux on hppa has fewer RT
 471                          * signals (SIGRTMAX is SIGRTMIN+25 there),
 472                          * hence let's not try to make use of them
 473                          * here. Since these commands are accessible
 474                          * by different means and only really a safety
 475                          * net, the missing functionality on hppa
 476                          * shouldn't matter. */
 477
 478                         SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
 479                         SIGRTMIN+27, /* systemd: set log target to console */
 480                         SIGRTMIN+28, /* systemd: set log target to kmsg */
 481                         SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
 482
 483                         /* ... one free signal here SIGRTMIN+30 ... */
 484 #endif
 485                         -1);
 486         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 487
 488         m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 489         if (m->signal_fd < 0)
 490                 return -errno;
 491
 492         r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
 493         if (r < 0)
 494                 return r;
 495
 496         (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
 497
 498         /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
 499          * notify processing can still figure out to which process/service a message belongs, before we reap the
 500          * process. Also, process this before handling cgroup notifications, so that we always collect child exit
 501          * status information before detecting that there's no process in a cgroup. */
 502         r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
 503         if (r < 0)
 504                 return r;
 505
 506         if (MANAGER_IS_SYSTEM(m))
 507                 return enable_special_signals(m);
 508
 509         return 0;
 510 }
 511
 512 static void manager_clean_environment(Manager *m) {
 513         assert(m);
 514
 515         /* Let's remove some environment variables that we
 516          * need ourselves to communicate with our clients */
 517         strv_env_unset_many(
 518                         m->environment,
 519                         "NOTIFY_SOCKET",
 520                         "MAINPID",
 521                         "MANAGERPID",
 522                         "LISTEN_PID",
 523                         "LISTEN_FDS",
 524                         "LISTEN_FDNAMES",
 525                         "WATCHDOG_PID",
 526                         "WATCHDOG_USEC",
 527                         "INVOCATION_ID",
 528                         NULL);
 529 }
 530
 531 static int manager_default_environment(Manager *m) {
 532         assert(m);
 533
 534         if (MANAGER_IS_SYSTEM(m)) {
 535                 /* The system manager always starts with a clean
 536                  * environment for its children. It does not import
 537                  * the kernel's or the parents' exported variables.
 538                  *
 539                  * The initial passed environment is untouched to keep
 540                  * /proc/self/environ valid; it is used for tagging
 541                  * the init process inside containers. */
 542                 m->environment = strv_new("PATH=" DEFAULT_PATH,
 543                                           NULL);
 544
 545                 /* Import locale variables LC_*= from configuration */
 546                 locale_setup(&m->environment);
 547         } else
 548                 /* The user manager passes its own environment
 549                  * along to its children. */
 550                 m->environment = strv_copy(environ);
 551
 552         if (!m->environment)
 553                 return -ENOMEM;
 554
 555         manager_clean_environment(m);
 556         strv_sort(m->environment);
 557
 558         return 0;
 559 }
 560
 561 static int manager_setup_prefix(Manager *m) {
 562         struct table_entry {
 563                 uint64_t type;
 564                 const char *suffix;
 565         };
 566
 567         static const struct table_entry paths_system[_EXEC_DIRECTORY_MAX] = {
 568                 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
 569                 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
 570                 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
 571                 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
 572                 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
 573         };
 574
 575         static const struct table_entry paths_user[_EXEC_DIRECTORY_MAX] = {
 576                 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
 577                 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
 578                 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
 579                 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_CONFIGURATION, "log" },
 580                 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
 581         };
 582
 583         const struct table_entry *p;
 584         ExecDirectoryType i;
 585         int r;
 586
 587         assert(m);
 588
 589         if (MANAGER_IS_SYSTEM(m))
 590                 p = paths_system;
 591         else
 592                 p = paths_user;
 593
 594         for (i = 0; i < _EXEC_DIRECTORY_MAX; i++) {
 595                 r = sd_path_home(p[i].type, p[i].suffix, &m->prefix[i]);
 596                 if (r < 0)
 597                         return r;
 598         }
 599
 600         return 0;
 601 }
 602
 603 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
 604         Manager *m;
 605         int r;
 606
 607         assert(_m);
 608         assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
 609
 610         m = new0(Manager, 1);
 611         if (!m)
 612                 return -ENOMEM;
 613
 614         m->unit_file_scope = scope;
 615         m->exit_code = _MANAGER_EXIT_CODE_INVALID;
 616         m->default_timer_accuracy_usec = USEC_PER_MINUTE;
 617         m->default_tasks_accounting = true;
 618         m->default_tasks_max = UINT64_MAX;
 619
 620 #ifdef ENABLE_EFI
 621         if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
 622                 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
 623 #endif
 624
 625         /* Prepare log fields we can use for structured logging */
 626         if (MANAGER_IS_SYSTEM(m)) {
 627                 m->unit_log_field = "UNIT=";
 628                 m->unit_log_format_string = "UNIT=%s";
 629
 630                 m->invocation_log_field = "INVOCATION_ID=";
 631                 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
 632         } else {
 633                 m->unit_log_field = "USER_UNIT=";
 634                 m->unit_log_format_string = "USER_UNIT=%s";
 635
 636                 m->invocation_log_field = "USER_INVOCATION_ID=";
 637                 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
 638         }
 639
 640         m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
 641
 642         m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
 643                 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
 644                 m->ask_password_inotify_fd = -1;
 645
 646         m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
 647
 648         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 649
 650         m->have_ask_password = -EINVAL; /* we don't know */
 651         m->first_boot = -1;
 652
 653         m->test_run = test_run;
 654
 655         /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
 656         RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
 657
 658         r = manager_default_environment(m);
 659         if (r < 0)
 660                 goto fail;
 661
 662         r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
 663         if (r < 0)
 664                 goto fail;
 665
 666         r = hashmap_ensure_allocated(&m->jobs, NULL);
 667         if (r < 0)
 668                 goto fail;
 669
 670         r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
 671         if (r < 0)
 672                 goto fail;
 673
 674         r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
 675         if (r < 0)
 676                 goto fail;
 677
 678         r = sd_event_default(&m->event);
 679         if (r < 0)
 680                 goto fail;
 681
 682         r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
 683         if (r < 0)
 684                 goto fail;
 685
 686         r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
 687         if (r < 0)
 688                 goto fail;
 689
 690         r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
 691         if (r < 0)
 692                 goto fail;
 693
 694         (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
 695
 696         r = manager_setup_signals(m);
 697         if (r < 0)
 698                 goto fail;
 699
 700         r = manager_setup_cgroup(m);
 701         if (r < 0)
 702                 goto fail;
 703
 704         r = manager_setup_time_change(m);
 705         if (r < 0)
 706                 goto fail;
 707
 708         m->udev = udev_new();
 709         if (!m->udev) {
 710                 r = -ENOMEM;
 711                 goto fail;
 712         }
 713
 714         /* Note that we do not set up the notify fd here. We do that after deserialization,
 715          * since they might have gotten serialized across the reexec. */
 716
 717         m->taint_usr = dir_is_empty("/usr") > 0;
 718
 719         r = manager_setup_prefix(m);
 720         if (r < 0)
 721                 goto fail;
 722
 723         *_m = m;
 724         return 0;
 725
 726 fail:
 727         manager_free(m);
 728         return r;
 729 }
 730
 731 static int manager_setup_notify(Manager *m) {
 732         int r;
 733
 734         if (m->test_run)
 735                 return 0;
 736
 737         if (m->notify_fd < 0) {
 738                 _cleanup_close_ int fd = -1;
 739                 union sockaddr_union sa = {
 740                         .sa.sa_family = AF_UNIX,
 741                 };
 742                 static const int one = 1;
 743
 744                 /* First free all secondary fields */
 745                 m->notify_socket = mfree(m->notify_socket);
 746                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
 747
 748                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 749                 if (fd < 0)
 750                         return log_error_errno(errno, "Failed to allocate notification socket: %m");
 751
 752                 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
 753
 754                 m->notify_socket = strappend(m->prefix[EXEC_DIRECTORY_RUNTIME], "/systemd/notify");
 755                 if (!m->notify_socket)
 756                         return log_oom();
 757
 758                 (void) mkdir_parents_label(m->notify_socket, 0755);
 759                 (void) unlink(m->notify_socket);
 760
 761                 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
 762                 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 763                 if (r < 0)
 764                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 765
 766                 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
 767                 if (r < 0)
 768                         return log_error_errno(errno, "SO_PASSCRED failed: %m");
 769
 770                 m->notify_fd = fd;
 771                 fd = -1;
 772
 773                 log_debug("Using notification socket %s", m->notify_socket);
 774         }
 775
 776         if (!m->notify_event_source) {
 777                 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
 778                 if (r < 0)
 779                         return log_error_errno(r, "Failed to allocate notify event source: %m");
 780
 781                 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
 782                  * service an exit message belongs. */
 783                 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
 784                 if (r < 0)
 785                         return log_error_errno(r, "Failed to set priority of notify event source: %m");
 786
 787                 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
 788         }
 789
 790         return 0;
 791 }
 792
 793 static int manager_setup_cgroups_agent(Manager *m) {
 794
 795         static const union sockaddr_union sa = {
 796                 .un.sun_family = AF_UNIX,
 797                 .un.sun_path = "/run/systemd/cgroups-agent",
 798         };
 799         int r;
 800
 801         /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
 802          * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
 803          * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
 804          * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
 805          * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
 806          * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
 807          * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
 808          * we thus won't lose messages.
 809          *
 810          * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
 811          * to it. The system instance hence listens on this special socket, but the user instances listen on the system
 812          * bus for these messages. */
 813
 814         if (m->test_run)
 815                 return 0;
 816
 817         if (!MANAGER_IS_SYSTEM(m))
 818                 return 0;
 819
 820         r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
 821         if (r < 0)
 822                 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
 823         if (r > 0) /* We don't need this anymore on the unified hierarchy */
 824                 return 0;
 825
 826         if (m->cgroups_agent_fd < 0) {
 827                 _cleanup_close_ int fd = -1;
 828
 829                 /* First free all secondary fields */
 830                 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
 831
 832                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 833                 if (fd < 0)
 834                         return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
 835
 836                 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
 837
 838                 (void) unlink(sa.un.sun_path);
 839
 840                 /* Only allow root to connect to this socket */
 841                 RUN_WITH_UMASK(0077)
 842                         r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 843                 if (r < 0)
 844                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 845
 846                 m->cgroups_agent_fd = fd;
 847                 fd = -1;
 848         }
 849
 850         if (!m->cgroups_agent_event_source) {
 851                 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
 852                 if (r < 0)
 853                         return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
 854
 855                 /* Process cgroups notifications early, but after having processed service notification messages or
 856                  * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
 857                  * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
 858                  * cgroup inotify for the unified cgroup stuff. */
 859                 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
 860                 if (r < 0)
 861                         return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
 862
 863                 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
 864         }
 865
 866         return 0;
 867 }
 868
 869 static int manager_setup_user_lookup_fd(Manager *m) {
 870         int r;
 871
 872         assert(m);
 873
 874         /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
 875          * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
 876          * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
 877          * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
 878          * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
 879          * hence we establish this communication channel so that forked off processes can pass their UID/GID
 880          * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
 881          * datagram, along with their unit name, so that we can share one communication socket pair among all units for
 882          * this purpose.
 883          *
 884          * You might wonder why we need a communication channel for this that is independent of the usual notification
 885          * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
 886          * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
 887          * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
 888          *
 889          * Note that this function is called under two circumstances: when we first initialize (in which case we
 890          * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
 891          * (in which case the socket pair already exists but we still need to allocate the event source for it). */
 892
 893         if (m->user_lookup_fds[0] < 0) {
 894
 895                 /* Free all secondary fields */
 896                 safe_close_pair(m->user_lookup_fds);
 897                 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
 898
 899                 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
 900                         return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
 901
 902                 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
 903         }
 904
 905         if (!m->user_lookup_event_source) {
 906                 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
 907                 if (r < 0)
 908                         return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
 909
 910                 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
 911                  * resolutions */
 912                 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
 913                 if (r < 0)
 914                         return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
 915
 916                 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
 917         }
 918
 919         return 0;
 920 }
 921
 922 static int manager_connect_bus(Manager *m, bool reexecuting) {
 923         bool try_bus_connect;
 924
 925         assert(m);
 926
 927         if (m->test_run)
 928                 return 0;
 929
 930         try_bus_connect =
 931                 reexecuting ||
 932                 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
 933
 934         /* Try to connect to the buses, if possible. */
 935         return bus_init(m, try_bus_connect);
 936 }
 937
 938 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
 939         Unit *u;
 940         unsigned n = 0;
 941
 942         assert(m);
 943
 944         while ((u = m->cleanup_queue)) {
 945                 assert(u->in_cleanup_queue);
 946
 947                 unit_free(u);
 948                 n++;
 949         }
 950
 951         return n;
 952 }
 953
 954 enum {
 955         GC_OFFSET_IN_PATH,  /* This one is on the path we were traveling */
 956         GC_OFFSET_UNSURE,   /* No clue */
 957         GC_OFFSET_GOOD,     /* We still need this unit */
 958         GC_OFFSET_BAD,      /* We don't need this unit anymore */
 959         _GC_OFFSET_MAX
 960 };
 961
 962 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
 963         Iterator i;
 964         Unit *other;
 965
 966         u->gc_marker = gc_marker + GC_OFFSET_GOOD;
 967
 968         /* Recursively mark referenced units as GOOD as well */
 969         SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
 970                 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
 971                         unit_gc_mark_good(other, gc_marker);
 972 }
 973
 974 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
 975         Iterator i;
 976         Unit *other;
 977         bool is_bad;
 978
 979         assert(u);
 980
 981         if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
 982             u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 983             u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
 984             u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
 985                 return;
 986
 987         if (u->in_cleanup_queue)
 988                 goto bad;
 989
 990         if (unit_check_gc(u))
 991                 goto good;
 992
 993         u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
 994
 995         is_bad = true;
 996
 997         SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
 998                 unit_gc_sweep(other, gc_marker);
 999
1000                 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1001                         goto good;
1002
1003                 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1004                         is_bad = false;
1005         }
1006
1007         if (is_bad)
1008                 goto bad;
1009
1010         /* We were unable to find anything out about this entry, so
1011          * let's investigate it later */
1012         u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1013         unit_add_to_gc_queue(u);
1014         return;
1015
1016 bad:
1017         /* We definitely know that this one is not useful anymore, so
1018          * let's mark it for deletion */
1019         u->gc_marker = gc_marker + GC_OFFSET_BAD;
1020         unit_add_to_cleanup_queue(u);
1021         return;
1022
1023 good:
1024         unit_gc_mark_good(u, gc_marker);
1025 }
1026
1027 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1028         unsigned n = 0, gc_marker;
1029         Unit *u;
1030
1031         assert(m);
1032
1033         /* log_debug("Running GC..."); */
1034
1035         m->gc_marker += _GC_OFFSET_MAX;
1036         if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1037                 m->gc_marker = 1;
1038
1039         gc_marker = m->gc_marker;
1040
1041         while ((u = m->gc_unit_queue)) {
1042                 assert(u->in_gc_queue);
1043
1044                 unit_gc_sweep(u, gc_marker);
1045
1046                 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1047                 u->in_gc_queue = false;
1048
1049                 n++;
1050
1051                 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1052                     u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1053                         if (u->id)
1054                                 log_unit_debug(u, "Collecting.");
1055                         u->gc_marker = gc_marker + GC_OFFSET_BAD;
1056                         unit_add_to_cleanup_queue(u);
1057                 }
1058         }
1059
1060         return n;
1061 }
1062
1063 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1064         unsigned n = 0;
1065         Job *j;
1066
1067         assert(m);
1068
1069         while ((j = m->gc_job_queue)) {
1070                 assert(j->in_gc_queue);
1071
1072                 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1073                 j->in_gc_queue = false;
1074
1075                 n++;
1076
1077                 if (job_check_gc(j))
1078                         continue;
1079
1080                 log_unit_debug(j->unit, "Collecting job.");
1081                 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1082         }
1083
1084         return n;
1085 }
1086
1087 static void manager_clear_jobs_and_units(Manager *m) {
1088         Unit *u;
1089
1090         assert(m);
1091
1092         while ((u = hashmap_first(m->units)))
1093                 unit_free(u);
1094
1095         manager_dispatch_cleanup_queue(m);
1096
1097         assert(!m->load_queue);
1098         assert(!m->run_queue);
1099         assert(!m->dbus_unit_queue);
1100         assert(!m->dbus_job_queue);
1101         assert(!m->cleanup_queue);
1102         assert(!m->gc_unit_queue);
1103         assert(!m->gc_job_queue);
1104
1105         assert(hashmap_isempty(m->jobs));
1106         assert(hashmap_isempty(m->units));
1107
1108         m->n_on_console = 0;
1109         m->n_running_jobs = 0;
1110 }
1111
1112 Manager* manager_free(Manager *m) {
1113         UnitType c;
1114         int i;
1115         ExecDirectoryType dt;
1116
1117         if (!m)
1118                 return NULL;
1119
1120         manager_clear_jobs_and_units(m);
1121
1122         for (c = 0; c < _UNIT_TYPE_MAX; c++)
1123                 if (unit_vtable[c]->shutdown)
1124                         unit_vtable[c]->shutdown(m);
1125
1126         /* If we reexecute ourselves, we keep the root cgroup
1127          * around */
1128         manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1129
1130         lookup_paths_flush_generator(&m->lookup_paths);
1131
1132         bus_done(m);
1133
1134         dynamic_user_vacuum(m, false);
1135         hashmap_free(m->dynamic_users);
1136
1137         hashmap_free(m->units);
1138         hashmap_free(m->units_by_invocation_id);
1139         hashmap_free(m->jobs);
1140         hashmap_free(m->watch_pids1);
1141         hashmap_free(m->watch_pids2);
1142         hashmap_free(m->watch_bus);
1143
1144         set_free(m->startup_units);
1145         set_free(m->failed_units);
1146
1147         sd_event_source_unref(m->signal_event_source);
1148         sd_event_source_unref(m->notify_event_source);
1149         sd_event_source_unref(m->cgroups_agent_event_source);
1150         sd_event_source_unref(m->time_change_event_source);
1151         sd_event_source_unref(m->jobs_in_progress_event_source);
1152         sd_event_source_unref(m->run_queue_event_source);
1153         sd_event_source_unref(m->user_lookup_event_source);
1154
1155         safe_close(m->signal_fd);
1156         safe_close(m->notify_fd);
1157         safe_close(m->cgroups_agent_fd);
1158         safe_close(m->time_change_fd);
1159         safe_close_pair(m->user_lookup_fds);
1160
1161         manager_close_ask_password(m);
1162
1163         manager_close_idle_pipe(m);
1164
1165         udev_unref(m->udev);
1166         sd_event_unref(m->event);
1167
1168         free(m->notify_socket);
1169
1170         lookup_paths_free(&m->lookup_paths);
1171         strv_free(m->environment);
1172
1173         hashmap_free(m->cgroup_unit);
1174         set_free_free(m->unit_path_cache);
1175
1176         free(m->switch_root);
1177         free(m->switch_root_init);
1178
1179         for (i = 0; i < _RLIMIT_MAX; i++)
1180                 m->rlimit[i] = mfree(m->rlimit[i]);
1181
1182         assert(hashmap_isempty(m->units_requiring_mounts_for));
1183         hashmap_free(m->units_requiring_mounts_for);
1184
1185         hashmap_free(m->uid_refs);
1186         hashmap_free(m->gid_refs);
1187
1188         for (dt = 0; dt < _EXEC_DIRECTORY_MAX; dt++)
1189                 m->prefix[dt] = mfree(m->prefix[dt]);
1190
1191         return mfree(m);
1192 }
1193
1194 void manager_enumerate(Manager *m) {
1195         UnitType c;
1196
1197         assert(m);
1198
1199         /* Let's ask every type to load all units from disk/kernel
1200          * that it might know */
1201         for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1202                 if (!unit_type_supported(c)) {
1203                         log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1204                         continue;
1205                 }
1206
1207                 if (!unit_vtable[c]->enumerate)
1208                         continue;
1209
1210                 unit_vtable[c]->enumerate(m);
1211         }
1212
1213         manager_dispatch_load_queue(m);
1214 }
1215
1216 static void manager_coldplug(Manager *m) {
1217         Iterator i;
1218         Unit *u;
1219         char *k;
1220         int r;
1221
1222         assert(m);
1223
1224         /* Then, let's set up their initial state. */
1225         HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1226
1227                 /* ignore aliases */
1228                 if (u->id != k)
1229                         continue;
1230
1231                 r = unit_coldplug(u);
1232                 if (r < 0)
1233                         log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1234         }
1235 }
1236
1237 static void manager_build_unit_path_cache(Manager *m) {
1238         char **i;
1239         int r;
1240
1241         assert(m);
1242
1243         set_free_free(m->unit_path_cache);
1244
1245         m->unit_path_cache = set_new(&string_hash_ops);
1246         if (!m->unit_path_cache) {
1247                 r = -ENOMEM;
1248                 goto fail;
1249         }
1250
1251         /* This simply builds a list of files we know exist, so that
1252          * we don't always have to go to disk */
1253
1254         STRV_FOREACH(i, m->lookup_paths.search_path) {
1255                 _cleanup_closedir_ DIR *d = NULL;
1256                 struct dirent *de;
1257
1258                 d = opendir(*i);
1259                 if (!d) {
1260                         if (errno != ENOENT)
1261                                 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1262                         continue;
1263                 }
1264
1265                 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1266                         char *p;
1267
1268                         p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1269                         if (!p) {
1270                                 r = -ENOMEM;
1271                                 goto fail;
1272                         }
1273
1274                         r = set_consume(m->unit_path_cache, p);
1275                         if (r < 0)
1276                                 goto fail;
1277                 }
1278         }
1279
1280         return;
1281
1282 fail:
1283         log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1284         m->unit_path_cache = set_free_free(m->unit_path_cache);
1285 }
1286
1287 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1288         Iterator i;
1289         Unit *u;
1290
1291         assert(m);
1292
1293         HASHMAP_FOREACH(u, m->units, i) {
1294
1295                 if (fdset_size(fds) <= 0)
1296                         break;
1297
1298                 if (!UNIT_VTABLE(u)->distribute_fds)
1299                         continue;
1300
1301                 UNIT_VTABLE(u)->distribute_fds(u, fds);
1302         }
1303 }
1304
1305 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1306         int r, q;
1307
1308         assert(m);
1309
1310         r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1311         if (r < 0)
1312                 return r;
1313
1314         r = manager_run_environment_generators(m);
1315         if (r < 0)
1316                 return r;
1317
1318         /* Make sure the transient directory always exists, so that it remains in the search path */
1319         if (!m->test_run) {
1320                 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1321                 if (r < 0)
1322                         return r;
1323         }
1324
1325         dual_timestamp_get(&m->generators_start_timestamp);
1326         r = manager_run_generators(m);
1327         dual_timestamp_get(&m->generators_finish_timestamp);
1328         if (r < 0)
1329                 return r;
1330
1331         lookup_paths_reduce(&m->lookup_paths);
1332         manager_build_unit_path_cache(m);
1333
1334         /* If we will deserialize make sure that during enumeration
1335          * this is already known, so we increase the counter here
1336          * already */
1337         if (serialization)
1338                 m->n_reloading++;
1339
1340         /* First, enumerate what we can from all config files */
1341         dual_timestamp_get(&m->units_load_start_timestamp);
1342         manager_enumerate(m);
1343         dual_timestamp_get(&m->units_load_finish_timestamp);
1344
1345         /* Second, deserialize if there is something to deserialize */
1346         if (serialization)
1347                 r = manager_deserialize(m, serialization, fds);
1348
1349         /* Any fds left? Find some unit which wants them. This is
1350          * useful to allow container managers to pass some file
1351          * descriptors to us pre-initialized. This enables
1352          * socket-based activation of entire containers. */
1353         manager_distribute_fds(m, fds);
1354
1355         /* We might have deserialized the notify fd, but if we didn't
1356          * then let's create the bus now */
1357         q = manager_setup_notify(m);
1358         if (q < 0 && r == 0)
1359                 r = q;
1360
1361         q = manager_setup_cgroups_agent(m);
1362         if (q < 0 && r == 0)
1363                 r = q;
1364
1365         q = manager_setup_user_lookup_fd(m);
1366         if (q < 0 && r == 0)
1367                 r = q;
1368
1369         /* Let's connect to the bus now. */
1370         (void) manager_connect_bus(m, !!serialization);
1371
1372         (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1373         m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1374
1375         /* Third, fire things up! */
1376         manager_coldplug(m);
1377
1378         /* Release any dynamic users no longer referenced */
1379         dynamic_user_vacuum(m, true);
1380
1381         /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1382         manager_vacuum_uid_refs(m);
1383         manager_vacuum_gid_refs(m);
1384
1385         if (serialization) {
1386                 assert(m->n_reloading > 0);
1387                 m->n_reloading--;
1388
1389                 /* Let's wait for the UnitNew/JobNew messages being
1390                  * sent, before we notify that the reload is
1391                  * finished */
1392                 m->send_reloading_done = true;
1393         }
1394
1395         return r;
1396 }
1397
1398 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1399         int r;
1400         Transaction *tr;
1401
1402         assert(m);
1403         assert(type < _JOB_TYPE_MAX);
1404         assert(unit);
1405         assert(mode < _JOB_MODE_MAX);
1406
1407         if (mode == JOB_ISOLATE && type != JOB_START)
1408                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1409
1410         if (mode == JOB_ISOLATE && !unit->allow_isolate)
1411                 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1412
1413         log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1414
1415         type = job_type_collapse(type, unit);
1416
1417         tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1418         if (!tr)
1419                 return -ENOMEM;
1420
1421         r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1422                                                  mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1423                                                  mode == JOB_IGNORE_DEPENDENCIES, e);
1424         if (r < 0)
1425                 goto tr_abort;
1426
1427         if (mode == JOB_ISOLATE) {
1428                 r = transaction_add_isolate_jobs(tr, m);
1429                 if (r < 0)
1430                         goto tr_abort;
1431         }
1432
1433         r = transaction_activate(tr, m, mode, e);
1434         if (r < 0)
1435                 goto tr_abort;
1436
1437         log_unit_debug(unit,
1438                        "Enqueued job %s/%s as %u", unit->id,
1439                        job_type_to_string(type), (unsigned) tr->anchor_job->id);
1440
1441         if (_ret)
1442                 *_ret = tr->anchor_job;
1443
1444         transaction_free(tr);
1445         return 0;
1446
1447 tr_abort:
1448         transaction_abort(tr);
1449         transaction_free(tr);
1450         return r;
1451 }
1452
1453 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1454         Unit *unit = NULL;  /* just to appease gcc, initialization is not really necessary */
1455         int r;
1456
1457         assert(m);
1458         assert(type < _JOB_TYPE_MAX);
1459         assert(name);
1460         assert(mode < _JOB_MODE_MAX);
1461
1462         r = manager_load_unit(m, name, NULL, NULL, &unit);
1463         if (r < 0)
1464                 return r;
1465         assert(unit);
1466
1467         return manager_add_job(m, type, unit, mode, e, ret);
1468 }
1469
1470 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1471         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1472         int r;
1473
1474         assert(m);
1475         assert(type < _JOB_TYPE_MAX);
1476         assert(name);
1477         assert(mode < _JOB_MODE_MAX);
1478
1479         r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1480         if (r < 0)
1481                 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1482
1483         return r;
1484 }
1485
1486 Job *manager_get_job(Manager *m, uint32_t id) {
1487         assert(m);
1488
1489         return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1490 }
1491
1492 Unit *manager_get_unit(Manager *m, const char *name) {
1493         assert(m);
1494         assert(name);
1495
1496         return hashmap_get(m->units, name);
1497 }
1498
1499 unsigned manager_dispatch_load_queue(Manager *m) {
1500         Unit *u;
1501         unsigned n = 0;
1502
1503         assert(m);
1504
1505         /* Make sure we are not run recursively */
1506         if (m->dispatching_load_queue)
1507                 return 0;
1508
1509         m->dispatching_load_queue = true;
1510
1511         /* Dispatches the load queue. Takes a unit from the queue and
1512          * tries to load its data until the queue is empty */
1513
1514         while ((u = m->load_queue)) {
1515                 assert(u->in_load_queue);
1516
1517                 unit_load(u);
1518                 n++;
1519         }
1520
1521         m->dispatching_load_queue = false;
1522         return n;
1523 }
1524
1525 int manager_load_unit_prepare(
1526                 Manager *m,
1527                 const char *name,
1528                 const char *path,
1529                 sd_bus_error *e,
1530                 Unit **_ret) {
1531
1532         Unit *ret;
1533         UnitType t;
1534         int r;
1535
1536         assert(m);
1537         assert(name || path);
1538         assert(_ret);
1539
1540         /* This will prepare the unit for loading, but not actually
1541          * load anything from disk. */
1542
1543         if (path && !is_path(path))
1544                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1545
1546         if (!name)
1547                 name = basename(path);
1548
1549         t = unit_name_to_type(name);
1550
1551         if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1552                 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1553                         return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1554
1555                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1556         }
1557
1558         ret = manager_get_unit(m, name);
1559         if (ret) {
1560                 *_ret = ret;
1561                 return 1;
1562         }
1563
1564         ret = unit_new(m, unit_vtable[t]->object_size);
1565         if (!ret)
1566                 return -ENOMEM;
1567
1568         if (path) {
1569                 ret->fragment_path = strdup(path);
1570                 if (!ret->fragment_path) {
1571                         unit_free(ret);
1572                         return -ENOMEM;
1573                 }
1574         }
1575
1576         r = unit_add_name(ret, name);
1577         if (r < 0) {
1578                 unit_free(ret);
1579                 return r;
1580         }
1581
1582         unit_add_to_load_queue(ret);
1583         unit_add_to_dbus_queue(ret);
1584         unit_add_to_gc_queue(ret);
1585
1586         *_ret = ret;
1587
1588         return 0;
1589 }
1590
1591 int manager_load_unit(
1592                 Manager *m,
1593                 const char *name,
1594                 const char *path,
1595                 sd_bus_error *e,
1596                 Unit **_ret) {
1597
1598         int r;
1599
1600         assert(m);
1601         assert(_ret);
1602
1603         /* This will load the service information files, but not actually
1604          * start any services or anything. */
1605
1606         r = manager_load_unit_prepare(m, name, path, e, _ret);
1607         if (r != 0)
1608                 return r;
1609
1610         manager_dispatch_load_queue(m);
1611
1612         *_ret = unit_follow_merge(*_ret);
1613
1614         return 0;
1615 }
1616
1617 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1618         Iterator i;
1619         Job *j;
1620
1621         assert(s);
1622         assert(f);
1623
1624         HASHMAP_FOREACH(j, s->jobs, i)
1625                 job_dump(j, f, prefix);
1626 }
1627
1628 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1629         Iterator i;
1630         Unit *u;
1631         const char *t;
1632
1633         assert(s);
1634         assert(f);
1635
1636         HASHMAP_FOREACH_KEY(u, t, s->units, i)
1637                 if (u->id == t)
1638                         unit_dump(u, f, prefix);
1639 }
1640
1641 void manager_clear_jobs(Manager *m) {
1642         Job *j;
1643
1644         assert(m);
1645
1646         while ((j = hashmap_first(m->jobs)))
1647                 /* No need to recurse. We're cancelling all jobs. */
1648                 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1649 }
1650
1651 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1652         Manager *m = userdata;
1653         Job *j;
1654
1655         assert(source);
1656         assert(m);
1657
1658         while ((j = m->run_queue)) {
1659                 assert(j->installed);
1660                 assert(j->in_run_queue);
1661
1662                 job_run_and_invalidate(j);
1663         }
1664
1665         if (m->n_running_jobs > 0)
1666                 manager_watch_jobs_in_progress(m);
1667
1668         if (m->n_on_console > 0)
1669                 manager_watch_idle_pipe(m);
1670
1671         return 1;
1672 }
1673
1674 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1675         Job *j;
1676         Unit *u;
1677         unsigned n = 0;
1678
1679         assert(m);
1680
1681         if (m->dispatching_dbus_queue)
1682                 return 0;
1683
1684         m->dispatching_dbus_queue = true;
1685
1686         while ((u = m->dbus_unit_queue)) {
1687                 assert(u->in_dbus_queue);
1688
1689                 bus_unit_send_change_signal(u);
1690                 n++;
1691         }
1692
1693         while ((j = m->dbus_job_queue)) {
1694                 assert(j->in_dbus_queue);
1695
1696                 bus_job_send_change_signal(j);
1697                 n++;
1698         }
1699
1700         m->dispatching_dbus_queue = false;
1701
1702         if (m->send_reloading_done) {
1703                 m->send_reloading_done = false;
1704
1705                 bus_manager_send_reloading(m, false);
1706         }
1707
1708         if (m->queued_message)
1709                 bus_send_queued_message(m);
1710
1711         return n;
1712 }
1713
1714 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1715         Manager *m = userdata;
1716         char buf[PATH_MAX+1];
1717         ssize_t n;
1718
1719         n = recv(fd, buf, sizeof(buf), 0);
1720         if (n < 0)
1721                 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1722         if (n == 0) {
1723                 log_error("Got zero-length cgroups agent message, ignoring.");
1724                 return 0;
1725         }
1726         if ((size_t) n >= sizeof(buf)) {
1727                 log_error("Got overly long cgroups agent message, ignoring.");
1728                 return 0;
1729         }
1730
1731         if (memchr(buf, 0, n)) {
1732                 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1733                 return 0;
1734         }
1735         buf[n] = 0;
1736
1737         manager_notify_cgroup_empty(m, buf);
1738         bus_forward_agent_released(m, buf);
1739
1740         return 0;
1741 }
1742
1743 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1744         _cleanup_strv_free_ char **tags = NULL;
1745
1746         assert(m);
1747         assert(u);
1748         assert(buf);
1749
1750         tags = strv_split(buf, "\n\r");
1751         if (!tags) {
1752                 log_oom();
1753                 return;
1754         }
1755
1756         if (UNIT_VTABLE(u)->notify_message)
1757                 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1758         else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1759                 _cleanup_free_ char *x = NULL, *y = NULL;
1760
1761                 x = cescape(buf);
1762                 if (x)
1763                         y = ellipsize(x, 20, 90);
1764                 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1765         }
1766 }
1767
1768 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1769
1770         _cleanup_fdset_free_ FDSet *fds = NULL;
1771         Manager *m = userdata;
1772         char buf[NOTIFY_BUFFER_MAX+1];
1773         struct iovec iovec = {
1774                 .iov_base = buf,
1775                 .iov_len = sizeof(buf)-1,
1776         };
1777         union {
1778                 struct cmsghdr cmsghdr;
1779                 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1780                             CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1781         } control = {};
1782         struct msghdr msghdr = {
1783                 .msg_iov = &iovec,
1784                 .msg_iovlen = 1,
1785                 .msg_control = &control,
1786                 .msg_controllen = sizeof(control),
1787         };
1788
1789         struct cmsghdr *cmsg;
1790         struct ucred *ucred = NULL;
1791         Unit *u1, *u2, *u3;
1792         int r, *fd_array = NULL;
1793         unsigned n_fds = 0;
1794         ssize_t n;
1795
1796         assert(m);
1797         assert(m->notify_fd == fd);
1798
1799         if (revents != EPOLLIN) {
1800                 log_warning("Got unexpected poll event for notify fd.");
1801                 return 0;
1802         }
1803
1804         n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1805         if (n < 0) {
1806                 if (IN_SET(errno, EAGAIN, EINTR))
1807                         return 0; /* Spurious wakeup, try again */
1808
1809                 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1810                  * won't take notification messages anymore, but that's still better than busy looping around this:
1811                  * being woken up over and over again but being unable to actually read the message off the socket. */
1812                 return log_error_errno(errno, "Failed to receive notification message: %m");
1813         }
1814
1815         CMSG_FOREACH(cmsg, &msghdr) {
1816                 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1817
1818                         fd_array = (int*) CMSG_DATA(cmsg);
1819                         n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1820
1821                 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1822                            cmsg->cmsg_type == SCM_CREDENTIALS &&
1823                            cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1824
1825                         ucred = (struct ucred*) CMSG_DATA(cmsg);
1826                 }
1827         }
1828
1829         if (n_fds > 0) {
1830                 assert(fd_array);
1831
1832                 r = fdset_new_array(&fds, fd_array, n_fds);
1833                 if (r < 0) {
1834                         close_many(fd_array, n_fds);
1835                         log_oom();
1836                         return 0;
1837                 }
1838         }
1839
1840         if (!ucred || ucred->pid <= 0) {
1841                 log_warning("Received notify message without valid credentials. Ignoring.");
1842                 return 0;
1843         }
1844
1845         if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1846                 log_warning("Received notify message exceeded maximum size. Ignoring.");
1847                 return 0;
1848         }
1849
1850         /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1851          * trailing NUL byte in the message, but don't expect it. */
1852         if (n > 1 && memchr(buf, 0, n-1)) {
1853                 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1854                 return 0;
1855         }
1856
1857         /* Make sure it's NUL-terminated. */
1858         buf[n] = 0;
1859
1860         /* Notify every unit that might be interested, but try
1861          * to avoid notifying the same one multiple times. */
1862         u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1863         if (u1)
1864                 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1865
1866         u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1867         if (u2 && u2 != u1)
1868                 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1869
1870         u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1871         if (u3 && u3 != u2 && u3 != u1)
1872                 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1873
1874         if (!u1 && !u2 && !u3)
1875                 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1876
1877         if (fdset_size(fds) > 0)
1878                 log_warning("Got extra auxiliary fds with notification message, closing them.");
1879
1880         return 0;
1881 }
1882
1883 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1884         uint64_t iteration;
1885
1886         assert(m);
1887         assert(u);
1888         assert(si);
1889
1890         sd_event_get_iteration(m->event, &iteration);
1891
1892         log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1893
1894         unit_unwatch_pid(u, si->si_pid);
1895
1896         if (UNIT_VTABLE(u)->sigchld_event) {
1897                 if (set_size(u->pids) <= 1 ||
1898                     iteration != u->sigchldgen ||
1899                     unit_main_pid(u) == si->si_pid ||
1900                     unit_control_pid(u) == si->si_pid) {
1901                         UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1902                         u->sigchldgen = iteration;
1903                 } else
1904                         log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1905          }
1906 }
1907
1908 static int manager_dispatch_sigchld(Manager *m) {
1909         assert(m);
1910
1911         for (;;) {
1912                 siginfo_t si = {};
1913
1914                 /* First we call waitd() for a PID and do not reap the
1915                  * zombie. That way we can still access /proc/$PID for
1916                  * it while it is a zombie. */
1917                 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1918
1919                         if (errno == ECHILD)
1920                                 break;
1921
1922                         if (errno == EINTR)
1923                                 continue;
1924
1925                         return -errno;
1926                 }
1927
1928                 if (si.si_pid <= 0)
1929                         break;
1930
1931                 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1932                         _cleanup_free_ char *name = NULL;
1933                         Unit *u1, *u2, *u3;
1934
1935                         get_process_comm(si.si_pid, &name);
1936
1937                         log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1938                                   si.si_pid, strna(name),
1939                                   sigchld_code_to_string(si.si_code),
1940                                   si.si_status,
1941                                   strna(si.si_code == CLD_EXITED
1942                                         ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1943                                         : signal_to_string(si.si_status)));
1944
1945                         /* And now figure out the unit this belongs
1946                          * to, it might be multiple... */
1947                         u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1948                         if (u1)
1949                                 invoke_sigchld_event(m, u1, &si);
1950                         u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1951                         if (u2 && u2 != u1)
1952                                 invoke_sigchld_event(m, u2, &si);
1953                         u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1954                         if (u3 && u3 != u2 && u3 != u1)
1955                                 invoke_sigchld_event(m, u3, &si);
1956                 }
1957
1958                 /* And now, we actually reap the zombie. */
1959                 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1960                         if (errno == EINTR)
1961                                 continue;
1962
1963                         return -errno;
1964                 }
1965         }
1966
1967         return 0;
1968 }
1969
1970 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1971         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1972         int r;
1973
1974         log_debug("Activating special unit %s", name);
1975
1976         r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1977         if (r < 0)
1978                 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1979
1980         return r;
1981 }
1982
1983 static void manager_handle_ctrl_alt_del(Manager *m) {
1984         /* If the user presses C-A-D more than
1985          * 7 times within 2s, we reboot/shutdown immediately,
1986          * unless it was disabled in system.conf */
1987
1988         if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
1989                 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1990         else
1991                 emergency_action(m, m->cad_burst_action, NULL,
1992                                 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1993 }
1994
1995 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1996         Manager *m = userdata;
1997         ssize_t n;
1998         struct signalfd_siginfo sfsi;
1999         bool sigchld = false;
2000         int r;
2001
2002         assert(m);
2003         assert(m->signal_fd == fd);
2004
2005         if (revents != EPOLLIN) {
2006                 log_warning("Got unexpected events from signal file descriptor.");
2007                 return 0;
2008         }
2009
2010         for (;;) {
2011                 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2012                 if (n != sizeof(sfsi)) {
2013                         if (n >= 0) {
2014                                 log_warning("Truncated read from signal fd (%zu bytes)!", n);
2015                                 return 0;
2016                         }
2017
2018                         if (IN_SET(errno, EINTR, EAGAIN))
2019                                 break;
2020
2021                         /* We return an error here, which will kill this handler,
2022                          * to avoid a busy loop on read error. */
2023                         return log_error_errno(errno, "Reading from signal fd failed: %m");
2024                 }
2025
2026                 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2027                                     (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2028                                     ? LOG_DEBUG : LOG_INFO,
2029                                     &sfsi);
2030
2031                 switch (sfsi.ssi_signo) {
2032
2033                 case SIGCHLD:
2034                         sigchld = true;
2035                         break;
2036
2037                 case SIGTERM:
2038                         if (MANAGER_IS_SYSTEM(m)) {
2039                                 /* This is for compatibility with the
2040                                  * original sysvinit */
2041                                 r = verify_run_space_and_log("Refusing to reexecute");
2042                                 if (r >= 0)
2043                                         m->exit_code = MANAGER_REEXECUTE;
2044                                 break;
2045                         }
2046
2047                         /* Fall through */
2048
2049                 case SIGINT:
2050                         if (MANAGER_IS_SYSTEM(m)) {
2051                                 manager_handle_ctrl_alt_del(m);
2052                                 break;
2053                         }
2054
2055                         /* Run the exit target if there is one, if not, just exit. */
2056                         if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2057                                 m->exit_code = MANAGER_EXIT;
2058                                 return 0;
2059                         }
2060
2061                         break;
2062
2063                 case SIGWINCH:
2064                         if (MANAGER_IS_SYSTEM(m))
2065                                 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2066
2067                         /* This is a nop on non-init */
2068                         break;
2069
2070                 case SIGPWR:
2071                         if (MANAGER_IS_SYSTEM(m))
2072                                 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2073
2074                         /* This is a nop on non-init */
2075                         break;
2076
2077                 case SIGUSR1: {
2078                         Unit *u;
2079
2080                         u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2081
2082                         if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2083                                 log_info("Trying to reconnect to bus...");
2084                                 bus_init(m, true);
2085                         }
2086
2087                         if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2088                                 log_info("Loading D-Bus service...");
2089                                 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2090                         }
2091
2092                         break;
2093                 }
2094
2095                 case SIGUSR2: {
2096                         _cleanup_free_ char *dump = NULL;
2097                         _cleanup_fclose_ FILE *f = NULL;
2098                         size_t size;
2099
2100                         f = open_memstream(&dump, &size);
2101                         if (!f) {
2102                                 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2103                                 break;
2104                         }
2105
2106                         manager_dump_units(m, f, "\t");
2107                         manager_dump_jobs(m, f, "\t");
2108
2109                         r = fflush_and_check(f);
2110                         if (r < 0) {
2111                                 log_warning_errno(r, "Failed to write status stream: %m");
2112                                 break;
2113                         }
2114
2115                         log_dump(LOG_INFO, dump);
2116                         break;
2117                 }
2118
2119                 case SIGHUP:
2120                         r = verify_run_space_and_log("Refusing to reload");
2121                         if (r >= 0)
2122                                 m->exit_code = MANAGER_RELOAD;
2123                         break;
2124
2125                 default: {
2126
2127                         /* Starting SIGRTMIN+0 */
2128                         static const char * const target_table[] = {
2129                                 [0] = SPECIAL_DEFAULT_TARGET,
2130                                 [1] = SPECIAL_RESCUE_TARGET,
2131                                 [2] = SPECIAL_EMERGENCY_TARGET,
2132                                 [3] = SPECIAL_HALT_TARGET,
2133                                 [4] = SPECIAL_POWEROFF_TARGET,
2134                                 [5] = SPECIAL_REBOOT_TARGET,
2135                                 [6] = SPECIAL_KEXEC_TARGET
2136                         };
2137
2138                         /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2139                         static const ManagerExitCode code_table[] = {
2140                                 [0] = MANAGER_HALT,
2141                                 [1] = MANAGER_POWEROFF,
2142                                 [2] = MANAGER_REBOOT,
2143                                 [3] = MANAGER_KEXEC
2144                         };
2145
2146                         if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2147                             (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2148                                 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2149                                 manager_start_target(m, target_table[idx],
2150                                                      (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2151                                 break;
2152                         }
2153
2154                         if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2155                             (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2156                                 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2157                                 break;
2158                         }
2159
2160                         switch (sfsi.ssi_signo - SIGRTMIN) {
2161
2162                         case 20:
2163                                 manager_set_show_status(m, SHOW_STATUS_YES);
2164                                 break;
2165
2166                         case 21:
2167                                 manager_set_show_status(m, SHOW_STATUS_NO);
2168                                 break;
2169
2170                         case 22:
2171                                 log_set_max_level(LOG_DEBUG);
2172                                 log_info("Setting log level to debug.");
2173                                 break;
2174
2175                         case 23:
2176                                 log_set_max_level(LOG_INFO);
2177                                 log_info("Setting log level to info.");
2178                                 break;
2179
2180                         case 24:
2181                                 if (MANAGER_IS_USER(m)) {
2182                                         m->exit_code = MANAGER_EXIT;
2183                                         return 0;
2184                                 }
2185
2186                                 /* This is a nop on init */
2187                                 break;
2188
2189                         case 26:
2190                         case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2191                                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2192                                 log_notice("Setting log target to journal-or-kmsg.");
2193                                 break;
2194
2195                         case 27:
2196                                 log_set_target(LOG_TARGET_CONSOLE);
2197                                 log_notice("Setting log target to console.");
2198                                 break;
2199
2200                         case 28:
2201                                 log_set_target(LOG_TARGET_KMSG);
2202                                 log_notice("Setting log target to kmsg.");
2203                                 break;
2204
2205                         default:
2206                                 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2207                         }
2208                 }
2209                 }
2210         }
2211
2212         if (sigchld)
2213                 manager_dispatch_sigchld(m);
2214
2215         return 0;
2216 }
2217
2218 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2219         Manager *m = userdata;
2220         Iterator i;
2221         Unit *u;
2222
2223         assert(m);
2224         assert(m->time_change_fd == fd);
2225
2226         log_struct(LOG_DEBUG,
2227                    "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2228                    LOG_MESSAGE("Time has been changed"),
2229                    NULL);
2230
2231         /* Restart the watch */
2232         m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2233         m->time_change_fd = safe_close(m->time_change_fd);
2234
2235         manager_setup_time_change(m);
2236
2237         HASHMAP_FOREACH(u, m->units, i)
2238                 if (UNIT_VTABLE(u)->time_change)
2239                         UNIT_VTABLE(u)->time_change(u);
2240
2241         return 0;
2242 }
2243
2244 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2245         Manager *m = userdata;
2246
2247         assert(m);
2248         assert(m->idle_pipe[2] == fd);
2249
2250         m->no_console_output = m->n_on_console > 0;
2251
2252         manager_close_idle_pipe(m);
2253
2254         return 0;
2255 }
2256
2257 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2258         Manager *m = userdata;
2259         int r;
2260         uint64_t next;
2261
2262         assert(m);
2263         assert(source);
2264
2265         manager_print_jobs_in_progress(m);
2266
2267         next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2268         r = sd_event_source_set_time(source, next);
2269         if (r < 0)
2270                 return r;
2271
2272         return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2273 }
2274
2275 int manager_loop(Manager *m) {
2276         int r;
2277
2278         RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2279
2280         assert(m);
2281         m->exit_code = MANAGER_OK;
2282
2283         /* Release the path cache */
2284         m->unit_path_cache = set_free_free(m->unit_path_cache);
2285
2286         manager_check_finished(m);
2287
2288         /* There might still be some zombies hanging around from
2289          * before we were exec()'ed. Let's reap them. */
2290         r = manager_dispatch_sigchld(m);
2291         if (r < 0)
2292                 return r;
2293
2294         while (m->exit_code == MANAGER_OK) {
2295                 usec_t wait_usec;
2296
2297                 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2298                         watchdog_ping();
2299
2300                 if (!ratelimit_test(&rl)) {
2301                         /* Yay, something is going seriously wrong, pause a little */
2302                         log_warning("Looping too fast. Throttling execution a little.");
2303                         sleep(1);
2304                 }
2305
2306                 if (manager_dispatch_load_queue(m) > 0)
2307                         continue;
2308
2309                 if (manager_dispatch_gc_job_queue(m) > 0)
2310                         continue;
2311
2312                 if (manager_dispatch_gc_unit_queue(m) > 0)
2313                         continue;
2314
2315                 if (manager_dispatch_cleanup_queue(m) > 0)
2316                         continue;
2317
2318                 if (manager_dispatch_cgroup_queue(m) > 0)
2319                         continue;
2320
2321                 if (manager_dispatch_dbus_queue(m) > 0)
2322                         continue;
2323
2324                 /* Sleep for half the watchdog time */
2325                 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2326                         wait_usec = m->runtime_watchdog / 2;
2327                         if (wait_usec <= 0)
2328                                 wait_usec = 1;
2329                 } else
2330                         wait_usec = USEC_INFINITY;
2331
2332                 r = sd_event_run(m->event, wait_usec);
2333                 if (r < 0)
2334                         return log_error_errno(r, "Failed to run event loop: %m");
2335         }
2336
2337         return m->exit_code;
2338 }
2339
2340 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2341         _cleanup_free_ char *n = NULL;
2342         sd_id128_t invocation_id;
2343         Unit *u;
2344         int r;
2345
2346         assert(m);
2347         assert(s);
2348         assert(_u);
2349
2350         r = unit_name_from_dbus_path(s, &n);
2351         if (r < 0)
2352                 return r;
2353
2354         /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2355          * as invocation ID. */
2356         r = sd_id128_from_string(n, &invocation_id);
2357         if (r >= 0) {
2358                 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2359                 if (u) {
2360                         *_u = u;
2361                         return 0;
2362                 }
2363
2364                 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));
2365         }
2366
2367         /* If this didn't work, we check if this is a unit name */
2368         if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
2369                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is neither a valid invocation ID nor unit name.", n);
2370
2371         r = manager_load_unit(m, n, NULL, e, &u);
2372         if (r < 0)
2373                 return r;
2374
2375         *_u = u;
2376         return 0;
2377 }
2378
2379 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2380         const char *p;
2381         unsigned id;
2382         Job *j;
2383         int r;
2384
2385         assert(m);
2386         assert(s);
2387         assert(_j);
2388
2389         p = startswith(s, "/org/freedesktop/systemd1/job/");
2390         if (!p)
2391                 return -EINVAL;
2392
2393         r = safe_atou(p, &id);
2394         if (r < 0)
2395                 return r;
2396
2397         j = manager_get_job(m, id);
2398         if (!j)
2399                 return -ENOENT;
2400
2401         *_j = j;
2402
2403         return 0;
2404 }
2405
2406 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2407
2408 #ifdef HAVE_AUDIT
2409         _cleanup_free_ char *p = NULL;
2410         const char *msg;
2411         int audit_fd, r;
2412
2413         if (!MANAGER_IS_SYSTEM(m))
2414                 return;
2415
2416         audit_fd = get_audit_fd();
2417         if (audit_fd < 0)
2418                 return;
2419
2420         /* Don't generate audit events if the service was already
2421          * started and we're just deserializing */
2422         if (MANAGER_IS_RELOADING(m))
2423                 return;
2424
2425         if (u->type != UNIT_SERVICE)
2426                 return;
2427
2428         r = unit_name_to_prefix_and_instance(u->id, &p);
2429         if (r < 0) {
2430                 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2431                 return;
2432         }
2433
2434         msg = strjoina("unit=", p);
2435         if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2436                 if (errno == EPERM)
2437                         /* We aren't allowed to send audit messages?
2438                          * Then let's not retry again. */
2439                         close_audit_fd();
2440                 else
2441                         log_warning_errno(errno, "Failed to send audit message: %m");
2442         }
2443 #endif
2444
2445 }
2446
2447 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2448         static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2449         _cleanup_free_ char *message = NULL;
2450         _cleanup_close_ int fd = -1;
2451         int n = 0;
2452
2453         /* Don't generate plymouth events if the service was already
2454          * started and we're just deserializing */
2455         if (MANAGER_IS_RELOADING(m))
2456                 return;
2457
2458         if (!MANAGER_IS_SYSTEM(m))
2459                 return;
2460
2461         if (detect_container() > 0)
2462                 return;
2463
2464         if (u->type != UNIT_SERVICE &&
2465             u->type != UNIT_MOUNT &&
2466             u->type != UNIT_SWAP)
2467                 return;
2468
2469         /* We set SOCK_NONBLOCK here so that we rather drop the
2470          * message then wait for plymouth */
2471         fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2472         if (fd < 0) {
2473                 log_error_errno(errno, "socket() failed: %m");
2474                 return;
2475         }
2476
2477         if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2478
2479                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2480                         log_error_errno(errno, "connect() failed: %m");
2481                 return;
2482         }
2483
2484         if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2485                 log_oom();
2486                 return;
2487         }
2488
2489         errno = 0;
2490         if (write(fd, message, n + 1) != n + 1)
2491                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2492                         log_error_errno(errno, "Failed to write Plymouth message: %m");
2493 }
2494
2495 int manager_open_serialization(Manager *m, FILE **_f) {
2496         int fd;
2497         FILE *f;
2498
2499         assert(_f);
2500
2501         fd = open_serialization_fd("systemd-state");
2502         if (fd < 0)
2503                 return fd;
2504
2505         f = fdopen(fd, "w+");
2506         if (!f) {
2507                 safe_close(fd);
2508                 return -errno;
2509         }
2510
2511         *_f = f;
2512         return 0;
2513 }
2514
2515 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2516         Iterator i;
2517         Unit *u;
2518         const char *t;
2519         int r;
2520
2521         assert(m);
2522         assert(f);
2523         assert(fds);
2524
2525         m->n_reloading++;
2526
2527         fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2528         fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2529         fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2530         fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2531
2532         dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2533         dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2534         dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2535         dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2536
2537         if (!in_initrd()) {
2538                 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2539                 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2540                 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2541                 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2542                 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2543                 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2544                 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2545                 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2546         }
2547
2548         if (!switching_root)
2549                 (void) serialize_environment(f, m->environment);
2550
2551         if (m->notify_fd >= 0) {
2552                 int copy;
2553
2554                 copy = fdset_put_dup(fds, m->notify_fd);
2555                 if (copy < 0)
2556                         return copy;
2557
2558                 fprintf(f, "notify-fd=%i\n", copy);
2559                 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2560         }
2561
2562         if (m->cgroups_agent_fd >= 0) {
2563                 int copy;
2564
2565                 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2566                 if (copy < 0)
2567                         return copy;
2568
2569                 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2570         }
2571
2572         if (m->user_lookup_fds[0] >= 0) {
2573                 int copy0, copy1;
2574
2575                 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2576                 if (copy0 < 0)
2577                         return copy0;
2578
2579                 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2580                 if (copy1 < 0)
2581                         return copy1;
2582
2583                 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2584         }
2585
2586         bus_track_serialize(m->subscribed, f, "subscribed");
2587
2588         r = dynamic_user_serialize(m, f, fds);
2589         if (r < 0)
2590                 return r;
2591
2592         manager_serialize_uid_refs(m, f);
2593         manager_serialize_gid_refs(m, f);
2594
2595         fputc('\n', f);
2596
2597         HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2598                 if (u->id != t)
2599                         continue;
2600
2601                 /* Start marker */
2602                 fputs(u->id, f);
2603                 fputc('\n', f);
2604
2605                 r = unit_serialize(u, f, fds, !switching_root);
2606                 if (r < 0) {
2607                         m->n_reloading--;
2608                         return r;
2609                 }
2610         }
2611
2612         assert(m->n_reloading > 0);
2613         m->n_reloading--;
2614
2615         if (ferror(f))
2616                 return -EIO;
2617
2618         r = bus_fdset_add_all(m, fds);
2619         if (r < 0)
2620                 return r;
2621
2622         return 0;
2623 }
2624
2625 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2626         int r = 0;
2627
2628         assert(m);
2629         assert(f);
2630
2631         log_debug("Deserializing state...");
2632
2633         m->n_reloading++;
2634
2635         for (;;) {
2636                 char line[LINE_MAX];
2637                 const char *val, *l;
2638
2639                 if (!fgets(line, sizeof(line), f)) {
2640                         if (feof(f))
2641                                 r = 0;
2642                         else
2643                                 r = -errno;
2644
2645                         goto finish;
2646                 }
2647
2648                 char_array_0(line);
2649                 l = strstrip(line);
2650
2651                 if (l[0] == 0)
2652                         break;
2653
2654                 if ((val = startswith(l, "current-job-id="))) {
2655                         uint32_t id;
2656
2657                         if (safe_atou32(val, &id) < 0)
2658                                 log_notice("Failed to parse current job id value %s", val);
2659                         else
2660                                 m->current_job_id = MAX(m->current_job_id, id);
2661
2662                 } else if ((val = startswith(l, "n-installed-jobs="))) {
2663                         uint32_t n;
2664
2665                         if (safe_atou32(val, &n) < 0)
2666                                 log_notice("Failed to parse installed jobs counter %s", val);
2667                         else
2668                                 m->n_installed_jobs += n;
2669
2670                 } else if ((val = startswith(l, "n-failed-jobs="))) {
2671                         uint32_t n;
2672
2673                         if (safe_atou32(val, &n) < 0)
2674                                 log_notice("Failed to parse failed jobs counter %s", val);
2675                         else
2676                                 m->n_failed_jobs += n;
2677
2678                 } else if ((val = startswith(l, "taint-usr="))) {
2679                         int b;
2680
2681                         b = parse_boolean(val);
2682                         if (b < 0)
2683                                 log_notice("Failed to parse taint /usr flag %s", val);
2684                         else
2685                                 m->taint_usr = m->taint_usr || b;
2686
2687                 } else if ((val = startswith(l, "firmware-timestamp=")))
2688                         dual_timestamp_deserialize(val, &m->firmware_timestamp);
2689                 else if ((val = startswith(l, "loader-timestamp=")))
2690                         dual_timestamp_deserialize(val, &m->loader_timestamp);
2691                 else if ((val = startswith(l, "kernel-timestamp=")))
2692                         dual_timestamp_deserialize(val, &m->kernel_timestamp);
2693                 else if ((val = startswith(l, "initrd-timestamp=")))
2694                         dual_timestamp_deserialize(val, &m->initrd_timestamp);
2695                 else if ((val = startswith(l, "userspace-timestamp=")))
2696                         dual_timestamp_deserialize(val, &m->userspace_timestamp);
2697                 else if ((val = startswith(l, "finish-timestamp=")))
2698                         dual_timestamp_deserialize(val, &m->finish_timestamp);
2699                 else if ((val = startswith(l, "security-start-timestamp=")))
2700                         dual_timestamp_deserialize(val, &m->security_start_timestamp);
2701                 else if ((val = startswith(l, "security-finish-timestamp=")))
2702                         dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2703                 else if ((val = startswith(l, "generators-start-timestamp=")))
2704                         dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2705                 else if ((val = startswith(l, "generators-finish-timestamp=")))
2706                         dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2707                 else if ((val = startswith(l, "units-load-start-timestamp=")))
2708                         dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2709                 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2710                         dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2711                 else if (startswith(l, "env=")) {
2712                         r = deserialize_environment(&m->environment, l);
2713                         if (r == -ENOMEM)
2714                                 goto finish;
2715                         if (r < 0)
2716                                 log_notice_errno(r, "Failed to parse environment entry: \"%s\": %m", l);
2717
2718                 } else if ((val = startswith(l, "notify-fd="))) {
2719                         int fd;
2720
2721                         if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2722                                 log_notice("Failed to parse notify fd: \"%s\"", val);
2723                         else {
2724                                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2725                                 safe_close(m->notify_fd);
2726                                 m->notify_fd = fdset_remove(fds, fd);
2727                         }
2728
2729                 } else if ((val = startswith(l, "notify-socket="))) {
2730                         char *n;
2731
2732                         n = strdup(val);
2733                         if (!n) {
2734                                 r = -ENOMEM;
2735                                 goto finish;
2736                         }
2737
2738                         free(m->notify_socket);
2739                         m->notify_socket = n;
2740
2741                 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2742                         int fd;
2743
2744                         if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2745                                 log_notice("Failed to parse cgroups agent fd: %s", val);
2746                         else {
2747                                 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2748                                 safe_close(m->cgroups_agent_fd);
2749                                 m->cgroups_agent_fd = fdset_remove(fds, fd);
2750                         }
2751
2752                 } else if ((val = startswith(l, "user-lookup="))) {
2753                         int fd0, fd1;
2754
2755                         if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2756                                 log_notice("Failed to parse user lookup fd: %s", val);
2757                         else {
2758                                 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2759                                 safe_close_pair(m->user_lookup_fds);
2760                                 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2761                                 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2762                         }
2763
2764                 } else if ((val = startswith(l, "dynamic-user=")))
2765                         dynamic_user_deserialize_one(m, val, fds);
2766                 else if ((val = startswith(l, "destroy-ipc-uid=")))
2767                         manager_deserialize_uid_refs_one(m, val);
2768                 else if ((val = startswith(l, "destroy-ipc-gid=")))
2769                         manager_deserialize_gid_refs_one(m, val);
2770                 else if ((val = startswith(l, "subscribed="))) {
2771
2772                         if (strv_extend(&m->deserialized_subscribed, val) < 0)
2773                                 log_oom();
2774
2775                 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2776                         log_notice("Unknown serialization item '%s'", l);
2777         }
2778
2779         for (;;) {
2780                 Unit *u;
2781                 char name[UNIT_NAME_MAX+2];
2782
2783                 /* Start marker */
2784                 if (!fgets(name, sizeof(name), f)) {
2785                         if (feof(f))
2786                                 r = 0;
2787                         else
2788                                 r = -errno;
2789
2790                         goto finish;
2791                 }
2792
2793                 char_array_0(name);
2794
2795                 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2796                 if (r < 0)
2797                         goto finish;
2798
2799                 r = unit_deserialize(u, f, fds);
2800                 if (r < 0)
2801                         goto finish;
2802         }
2803
2804 finish:
2805         if (ferror(f))
2806                 r = -EIO;
2807
2808         assert(m->n_reloading > 0);
2809         m->n_reloading--;
2810
2811         return r;
2812 }
2813
2814 int manager_reload(Manager *m) {
2815         int r, q;
2816         _cleanup_fclose_ FILE *f = NULL;
2817         _cleanup_fdset_free_ FDSet *fds = NULL;
2818
2819         assert(m);
2820
2821         r = manager_open_serialization(m, &f);
2822         if (r < 0)
2823                 return r;
2824
2825         m->n_reloading++;
2826         bus_manager_send_reloading(m, true);
2827
2828         fds = fdset_new();
2829         if (!fds) {
2830                 m->n_reloading--;
2831                 return -ENOMEM;
2832         }
2833
2834         r = manager_serialize(m, f, fds, false);
2835         if (r < 0) {
2836                 m->n_reloading--;
2837                 return r;
2838         }
2839
2840         if (fseeko(f, 0, SEEK_SET) < 0) {
2841                 m->n_reloading--;
2842                 return -errno;
2843         }
2844
2845         /* From here on there is no way back. */
2846         manager_clear_jobs_and_units(m);
2847         lookup_paths_flush_generator(&m->lookup_paths);
2848         lookup_paths_free(&m->lookup_paths);
2849         dynamic_user_vacuum(m, false);
2850         m->uid_refs = hashmap_free(m->uid_refs);
2851         m->gid_refs = hashmap_free(m->gid_refs);
2852
2853         q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2854         if (q < 0 && r >= 0)
2855                 r = q;
2856
2857         q = manager_run_environment_generators(m);
2858         if (q < 0 && r >= 0)
2859                 r = q;
2860
2861         /* Find new unit paths */
2862         q = manager_run_generators(m);
2863         if (q < 0 && r >= 0)
2864                 r = q;
2865
2866         lookup_paths_reduce(&m->lookup_paths);
2867         manager_build_unit_path_cache(m);
2868
2869         /* First, enumerate what we can from all config files */
2870         manager_enumerate(m);
2871
2872         /* Second, deserialize our stored data */
2873         q = manager_deserialize(m, f, fds);
2874         if (q < 0 && r >= 0)
2875                 r = q;
2876
2877         fclose(f);
2878         f = NULL;
2879
2880         /* Re-register notify_fd as event source */
2881         q = manager_setup_notify(m);
2882         if (q < 0 && r >= 0)
2883                 r = q;
2884
2885         q = manager_setup_cgroups_agent(m);
2886         if (q < 0 && r >= 0)
2887                 r = q;
2888
2889         q = manager_setup_user_lookup_fd(m);
2890         if (q < 0 && r >= 0)
2891                 r = q;
2892
2893         /* Third, fire things up! */
2894         manager_coldplug(m);
2895
2896         /* Release any dynamic users no longer referenced */
2897         dynamic_user_vacuum(m, true);
2898
2899         /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2900         manager_vacuum_uid_refs(m);
2901         manager_vacuum_gid_refs(m);
2902
2903         /* Sync current state of bus names with our set of listening units */
2904         if (m->api_bus)
2905                 manager_sync_bus_names(m, m->api_bus);
2906
2907         assert(m->n_reloading > 0);
2908         m->n_reloading--;
2909
2910         m->send_reloading_done = true;
2911
2912         return r;
2913 }
2914
2915 void manager_reset_failed(Manager *m) {
2916         Unit *u;
2917         Iterator i;
2918
2919         assert(m);
2920
2921         HASHMAP_FOREACH(u, m->units, i)
2922                 unit_reset_failed(u);
2923 }
2924
2925 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2926         Unit *u;
2927
2928         assert(m);
2929         assert(name);
2930
2931         /* Returns true if the unit is inactive or going down */
2932         u = manager_get_unit(m, name);
2933         if (!u)
2934                 return true;
2935
2936         return unit_inactive_or_pending(u);
2937 }
2938
2939 static void manager_notify_finished(Manager *m) {
2940         char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2941         usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2942
2943         if (m->test_run)
2944                 return;
2945
2946         if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2947
2948                 /* Note that m->kernel_usec.monotonic is always at 0,
2949                  * and m->firmware_usec.monotonic and
2950                  * m->loader_usec.monotonic should be considered
2951                  * negative values. */
2952
2953                 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2954                 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2955                 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2956                 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2957
2958                 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2959
2960                         kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2961                         initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2962
2963                         log_struct(LOG_INFO,
2964                                    "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2965                                    "KERNEL_USEC="USEC_FMT, kernel_usec,
2966                                    "INITRD_USEC="USEC_FMT, initrd_usec,
2967                                    "USERSPACE_USEC="USEC_FMT, userspace_usec,
2968                                    LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2969                                                format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2970                                                format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2971                                                format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2972                                                format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2973                                    NULL);
2974                 } else {
2975                         kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2976                         initrd_usec = 0;
2977
2978                         log_struct(LOG_INFO,
2979                                    "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2980                                    "KERNEL_USEC="USEC_FMT, kernel_usec,
2981                                    "USERSPACE_USEC="USEC_FMT, userspace_usec,
2982                                    LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2983                                                format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2984                                                format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2985                                                format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2986                                    NULL);
2987                 }
2988         } else {
2989                 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2990                 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2991
2992                 log_struct(LOG_INFO,
2993                            "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
2994                            "USERSPACE_USEC="USEC_FMT, userspace_usec,
2995                            LOG_MESSAGE("Startup finished in %s.",
2996                                        format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2997                            NULL);
2998         }
2999
3000         bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3001
3002         sd_notifyf(false,
3003                    "READY=1\n"
3004                    "STATUS=Startup finished in %s.",
3005                    format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
3006 }
3007
3008 void manager_check_finished(Manager *m) {
3009         assert(m);
3010
3011         if (MANAGER_IS_RELOADING(m))
3012                 return;
3013
3014         /* Verify that we are actually running currently. Initially
3015          * the exit code is set to invalid, and during operation it is
3016          * then set to MANAGER_OK */
3017         if (m->exit_code != MANAGER_OK)
3018                 return;
3019
3020         if (hashmap_size(m->jobs) > 0) {
3021                 if (m->jobs_in_progress_event_source)
3022                         /* Ignore any failure, this is only for feedback */
3023                         (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
3024
3025                 return;
3026         }
3027
3028         manager_flip_auto_status(m, false);
3029
3030         /* Notify Type=idle units that we are done now */
3031         manager_close_idle_pipe(m);
3032
3033         /* Turn off confirm spawn now */
3034         m->confirm_spawn = NULL;
3035
3036         /* No need to update ask password status when we're going non-interactive */
3037         manager_close_ask_password(m);
3038
3039         /* This is no longer the first boot */
3040         manager_set_first_boot(m, false);
3041
3042         if (dual_timestamp_is_set(&m->finish_timestamp))
3043                 return;
3044
3045         dual_timestamp_get(&m->finish_timestamp);
3046
3047         manager_notify_finished(m);
3048
3049         manager_invalidate_startup_units(m);
3050 }
3051
3052 static bool generator_path_any(const char* const* paths) {
3053         char **path;
3054         bool found = false;
3055
3056         /* Optimize by skipping the whole process by not creating output directories
3057          * if no generators are found. */
3058         STRV_FOREACH(path, (char**) paths)
3059                 if (access(*path, F_OK) == 0)
3060                         found = true;
3061                 else if (errno != ENOENT)
3062                         log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3063
3064         return found;
3065 }
3066
3067 static const char* system_env_generator_binary_paths[] = {
3068         "/run/systemd/system-environment-generators",
3069         "/etc/systemd/system-environment-generators",
3070         "/usr/local/lib/systemd/system-environment-generators",
3071         SYSTEM_ENV_GENERATOR_PATH,
3072         NULL
3073 };
3074
3075 static const char* user_env_generator_binary_paths[] = {
3076         "/run/systemd/user-environment-generators",
3077         "/etc/systemd/user-environment-generators",
3078         "/usr/local/lib/systemd/user-environment-generators",
3079         USER_ENV_GENERATOR_PATH,
3080         NULL
3081 };
3082
3083 static int manager_run_environment_generators(Manager *m) {
3084         char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3085         const char **paths;
3086         void* args[] = {&tmp, &tmp, &m->environment};
3087
3088         if (m->test_run)
3089                 return 0;
3090
3091         paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3092
3093         if (!generator_path_any(paths))
3094                 return 0;
3095
3096         return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3097 }
3098
3099 static int manager_run_generators(Manager *m) {
3100         _cleanup_strv_free_ char **paths = NULL;
3101         const char *argv[5];
3102         int r;
3103
3104         assert(m);
3105
3106         if (m->test_run)
3107                 return 0;
3108
3109         paths = generator_binary_paths(m->unit_file_scope);
3110         if (!paths)
3111                 return log_oom();
3112
3113         if (!generator_path_any((const char* const*) paths))
3114                 return 0;
3115
3116         r = lookup_paths_mkdir_generator(&m->lookup_paths);
3117         if (r < 0)
3118                 goto finish;
3119
3120         argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3121         argv[1] = m->lookup_paths.generator;
3122         argv[2] = m->lookup_paths.generator_early;
3123         argv[3] = m->lookup_paths.generator_late;
3124         argv[4] = NULL;
3125
3126         RUN_WITH_UMASK(0022)
3127                 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3128                                     NULL, NULL, (char**) argv);
3129
3130 finish:
3131         lookup_paths_trim_generator(&m->lookup_paths);
3132         return r;
3133 }
3134
3135 int manager_environment_add(Manager *m, char **minus, char **plus) {
3136         char **a = NULL, **b = NULL, **l;
3137         assert(m);
3138
3139         l = m->environment;
3140
3141         if (!strv_isempty(minus)) {
3142                 a = strv_env_delete(l, 1, minus);
3143                 if (!a)
3144                         return -ENOMEM;
3145
3146                 l = a;
3147         }
3148
3149         if (!strv_isempty(plus)) {
3150                 b = strv_env_merge(2, l, plus);
3151                 if (!b) {
3152                         strv_free(a);
3153                         return -ENOMEM;
3154                 }
3155
3156                 l = b;
3157         }
3158
3159         if (m->environment != l)
3160                 strv_free(m->environment);
3161         if (a != l)
3162                 strv_free(a);
3163         if (b != l)
3164                 strv_free(b);
3165
3166         m->environment = l;
3167         manager_clean_environment(m);
3168         strv_sort(m->environment);
3169
3170         return 0;
3171 }
3172
3173 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3174         int i;
3175
3176         assert(m);
3177
3178         for (i = 0; i < _RLIMIT_MAX; i++) {
3179                 m->rlimit[i] = mfree(m->rlimit[i]);
3180
3181                 if (!default_rlimit[i])
3182                         continue;
3183
3184                 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3185                 if (!m->rlimit[i])
3186                         return log_oom();
3187         }
3188
3189         return 0;
3190 }
3191
3192 void manager_recheck_journal(Manager *m) {
3193         Unit *u;
3194
3195         assert(m);
3196
3197         if (!MANAGER_IS_SYSTEM(m))
3198                 return;
3199
3200         u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3201         if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3202                 log_close_journal();
3203                 return;
3204         }
3205
3206         u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3207         if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3208                 log_close_journal();
3209                 return;
3210         }
3211
3212         /* Hmm, OK, so the socket is fully up and the service is up
3213          * too, then let's make use of the thing. */
3214         log_open();
3215 }
3216
3217 void manager_set_show_status(Manager *m, ShowStatus mode) {
3218         assert(m);
3219         assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3220
3221         if (!MANAGER_IS_SYSTEM(m))
3222                 return;
3223
3224         if (m->show_status != mode)
3225                 log_debug("%s showing of status.",
3226                           mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3227         m->show_status = mode;
3228
3229         if (mode > 0)
3230                 (void) touch("/run/systemd/show-status");
3231         else
3232                 (void) unlink("/run/systemd/show-status");
3233 }
3234
3235 static bool manager_get_show_status(Manager *m, StatusType type) {
3236         assert(m);
3237
3238         if (!MANAGER_IS_SYSTEM(m))
3239                 return false;
3240
3241         if (m->no_console_output)
3242                 return false;
3243
3244         if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3245                 return false;
3246
3247         /* If we cannot find out the status properly, just proceed. */
3248         if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3249                 return false;
3250
3251         if (m->show_status > 0)
3252                 return true;
3253
3254         return false;
3255 }
3256
3257 const char *manager_get_confirm_spawn(Manager *m) {
3258         static int last_errno = 0;
3259         const char *vc = m->confirm_spawn;
3260         struct stat st;
3261         int r;
3262
3263         /* Here's the deal: we want to test the validity of the console but don't want
3264          * PID1 to go through the whole console process which might block. But we also
3265          * want to warn the user only once if something is wrong with the console so we
3266          * cannot do the sanity checks after spawning our children. So here we simply do
3267          * really basic tests to hopefully trap common errors.
3268          *
3269          * If the console suddenly disappear at the time our children will really it
3270          * then they will simply fail to acquire it and a positive answer will be
3271          * assumed. New children will fallback to /dev/console though.
3272          *
3273          * Note: TTYs are devices that can come and go any time, and frequently aren't
3274          * available yet during early boot (consider a USB rs232 dongle...). If for any
3275          * reason the configured console is not ready, we fallback to the default
3276          * console. */
3277
3278         if (!vc || path_equal(vc, "/dev/console"))
3279                 return vc;
3280
3281         r = stat(vc, &st);
3282         if (r < 0)
3283                 goto fail;
3284
3285         if (!S_ISCHR(st.st_mode)) {
3286                 errno = ENOTTY;
3287                 goto fail;
3288         }
3289
3290         last_errno = 0;
3291         return vc;
3292 fail:
3293         if (last_errno != errno) {
3294                 last_errno = errno;
3295                 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3296         }
3297         return "/dev/console";
3298 }
3299
3300 void manager_set_first_boot(Manager *m, bool b) {
3301         assert(m);
3302
3303         if (!MANAGER_IS_SYSTEM(m))
3304                 return;
3305
3306         if (m->first_boot != (int) b) {
3307                 if (b)
3308                         (void) touch("/run/systemd/first-boot");
3309                 else
3310                         (void) unlink("/run/systemd/first-boot");
3311         }
3312
3313         m->first_boot = b;
3314 }
3315
3316 void manager_disable_confirm_spawn(void) {
3317         (void) touch("/run/systemd/confirm_spawn_disabled");
3318 }
3319
3320 bool manager_is_confirm_spawn_disabled(Manager *m) {
3321         if (!m->confirm_spawn)
3322                 return true;
3323
3324         return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3325 }
3326
3327 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3328         va_list ap;
3329
3330         /* If m is NULL, assume we're after shutdown and let the messages through. */
3331
3332         if (m && !manager_get_show_status(m, type))
3333                 return;
3334
3335         /* XXX We should totally drop the check for ephemeral here
3336          * and thus effectively make 'Type=idle' pointless. */
3337         if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3338                 return;
3339
3340         va_start(ap, format);
3341         status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3342         va_end(ap);
3343 }
3344
3345 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3346         char p[strlen(path)+1];
3347
3348         assert(m);
3349         assert(path);
3350
3351         strcpy(p, path);
3352         path_kill_slashes(p);
3353
3354         return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3355 }
3356
3357 int manager_set_exec_params(Manager *m, ExecParameters *p) {
3358         assert(m);
3359         assert(p);
3360
3361         p->environment = m->environment;
3362         p->confirm_spawn = manager_get_confirm_spawn(m);
3363         p->cgroup_supported = m->cgroup_supported;
3364         p->prefix = m->prefix;
3365
3366         return 0;
3367 }
3368
3369 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3370         unsigned size;
3371         int r;
3372
3373         assert(m);
3374         assert(u->manager == m);
3375
3376         size = set_size(m->failed_units);
3377
3378         if (failed) {
3379                 r = set_ensure_allocated(&m->failed_units, NULL);
3380                 if (r < 0)
3381                         return log_oom();
3382
3383                 if (set_put(m->failed_units, u) < 0)
3384                         return log_oom();
3385         } else
3386                 (void) set_remove(m->failed_units, u);
3387
3388         if (set_size(m->failed_units) != size)
3389                 bus_manager_send_change_signal(m);
3390
3391         return 0;
3392 }
3393
3394 ManagerState manager_state(Manager *m) {
3395         Unit *u;
3396
3397         assert(m);
3398
3399         /* Did we ever finish booting? If not then we are still starting up */
3400         if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3401
3402                 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3403                 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3404                         return MANAGER_INITIALIZING;
3405
3406                 return MANAGER_STARTING;
3407         }
3408
3409         /* Is the special shutdown target queued? If so, we are in shutdown state */
3410         u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3411         if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3412                 return MANAGER_STOPPING;
3413
3414         /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3415         u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3416         if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3417                   (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3418                 return MANAGER_MAINTENANCE;
3419
3420         u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3421         if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3422                   (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3423                 return MANAGER_MAINTENANCE;
3424
3425         /* Are there any failed units? If so, we are in degraded mode */
3426         if (set_size(m->failed_units) > 0)
3427                 return MANAGER_DEGRADED;
3428
3429         return MANAGER_RUNNING;
3430 }
3431
3432 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3433
3434 static void manager_unref_uid_internal(
3435                 Manager *m,
3436                 Hashmap **uid_refs,
3437                 uid_t uid,
3438                 bool destroy_now,
3439                 int (*_clean_ipc)(uid_t uid)) {
3440
3441         uint32_t c, n;
3442
3443         assert(m);
3444         assert(uid_refs);
3445         assert(uid_is_valid(uid));
3446         assert(_clean_ipc);
3447
3448         /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3449          * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3450          *
3451          * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3452          * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3453          * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3454          * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3455
3456         assert_cc(sizeof(uid_t) == sizeof(gid_t));
3457         assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3458
3459         if (uid == 0) /* We don't keep track of root, and will never destroy it */
3460                 return;
3461
3462         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3463
3464         n = c & ~DESTROY_IPC_FLAG;
3465         assert(n > 0);
3466         n--;
3467
3468         if (destroy_now && n == 0) {
3469                 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3470
3471                 if (c & DESTROY_IPC_FLAG) {
3472                         log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3473                                   _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3474                                   uid);
3475                         (void) _clean_ipc(uid);
3476                 }
3477         } else {
3478                 c = n | (c & DESTROY_IPC_FLAG);
3479                 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3480         }
3481 }
3482
3483 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3484         manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3485 }
3486
3487 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3488         manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3489 }
3490
3491 static int manager_ref_uid_internal(
3492                 Manager *m,
3493                 Hashmap **uid_refs,
3494                 uid_t uid,
3495                 bool clean_ipc) {
3496
3497         uint32_t c, n;
3498         int r;
3499
3500         assert(m);
3501         assert(uid_refs);
3502         assert(uid_is_valid(uid));
3503
3504         /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3505          * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3506
3507         assert_cc(sizeof(uid_t) == sizeof(gid_t));
3508         assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3509
3510         if (uid == 0) /* We don't keep track of root, and will never destroy it */
3511                 return 0;
3512
3513         r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3514         if (r < 0)
3515                 return r;
3516
3517         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3518
3519         n = c & ~DESTROY_IPC_FLAG;
3520         n++;
3521
3522         if (n & DESTROY_IPC_FLAG) /* check for overflow */
3523                 return -EOVERFLOW;
3524
3525         c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3526
3527         return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3528 }
3529
3530 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3531         return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3532 }
3533
3534 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3535         return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3536 }
3537
3538 static void manager_vacuum_uid_refs_internal(
3539                 Manager *m,
3540                 Hashmap **uid_refs,
3541                 int (*_clean_ipc)(uid_t uid)) {
3542
3543         Iterator i;
3544         void *p, *k;
3545
3546         assert(m);
3547         assert(uid_refs);
3548         assert(_clean_ipc);
3549
3550         HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3551                 uint32_t c, n;
3552                 uid_t uid;
3553
3554                 uid = PTR_TO_UID(k);
3555                 c = PTR_TO_UINT32(p);
3556
3557                 n = c & ~DESTROY_IPC_FLAG;
3558                 if (n > 0)
3559                         continue;
3560
3561                 if (c & DESTROY_IPC_FLAG) {
3562                         log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3563                                   _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3564                                   uid);
3565                         (void) _clean_ipc(uid);
3566                 }
3567
3568                 assert_se(hashmap_remove(*uid_refs, k) == p);
3569         }
3570 }
3571
3572 void manager_vacuum_uid_refs(Manager *m) {
3573         manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3574 }
3575
3576 void manager_vacuum_gid_refs(Manager *m) {
3577         manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3578 }
3579
3580 static void manager_serialize_uid_refs_internal(
3581                 Manager *m,
3582                 FILE *f,
3583                 Hashmap **uid_refs,
3584                 const char *field_name) {
3585
3586         Iterator i;
3587         void *p, *k;
3588
3589         assert(m);
3590         assert(f);
3591         assert(uid_refs);
3592         assert(field_name);
3593
3594         /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3595          * of it is better rebuild after a reload/reexec. */
3596
3597         HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3598                 uint32_t c;
3599                 uid_t uid;
3600
3601                 uid = PTR_TO_UID(k);
3602                 c = PTR_TO_UINT32(p);
3603
3604                 if (!(c & DESTROY_IPC_FLAG))
3605                         continue;
3606
3607                 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3608         }
3609 }
3610
3611 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3612         manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3613 }
3614
3615 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3616         manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3617 }
3618
3619 static void manager_deserialize_uid_refs_one_internal(
3620                 Manager *m,
3621                 Hashmap** uid_refs,
3622                 const char *value) {
3623
3624         uid_t uid;
3625         uint32_t c;
3626         int r;
3627
3628         assert(m);
3629         assert(uid_refs);
3630         assert(value);
3631
3632         r = parse_uid(value, &uid);
3633         if (r < 0 || uid == 0) {
3634                 log_debug("Unable to parse UID reference serialization");
3635                 return;
3636         }
3637
3638         r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3639         if (r < 0) {
3640                 log_oom();
3641                 return;
3642         }
3643
3644         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3645         if (c & DESTROY_IPC_FLAG)
3646                 return;
3647
3648         c |= DESTROY_IPC_FLAG;
3649
3650         r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3651         if (r < 0) {
3652                 log_debug("Failed to add UID reference entry");
3653                 return;
3654         }
3655 }
3656
3657 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3658         manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3659 }
3660
3661 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3662         manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3663 }
3664
3665 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3666         struct buffer {
3667                 uid_t uid;
3668                 gid_t gid;
3669                 char unit_name[UNIT_NAME_MAX+1];
3670         } _packed_ buffer;
3671
3672         Manager *m = userdata;
3673         ssize_t l;
3674         size_t n;
3675         Unit *u;
3676
3677         assert_se(source);
3678         assert_se(m);
3679
3680         /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3681          * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3682          * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3683
3684         l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3685         if (l < 0) {
3686                 if (errno == EINTR || errno == EAGAIN)
3687                         return 0;
3688
3689                 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3690         }
3691
3692         if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3693                 log_warning("Received too short user lookup message, ignoring.");
3694                 return 0;
3695         }
3696
3697         if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3698                 log_warning("Received too long user lookup message, ignoring.");
3699                 return 0;
3700         }
3701
3702         if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3703                 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3704                 return 0;
3705         }
3706
3707         n = (size_t) l - offsetof(struct buffer, unit_name);
3708         if (memchr(buffer.unit_name, 0, n)) {
3709                 log_warning("Received lookup message with embedded NUL character, ignoring.");
3710                 return 0;
3711         }
3712
3713         buffer.unit_name[n] = 0;
3714         u = manager_get_unit(m, buffer.unit_name);
3715         if (!u) {
3716                 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3717                 return 0;
3718         }
3719
3720         log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3721
3722         unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3723         return 0;
3724 }
3725
3726 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3727         [MANAGER_INITIALIZING] = "initializing",
3728         [MANAGER_STARTING] = "starting",
3729         [MANAGER_RUNNING] = "running",
3730         [MANAGER_DEGRADED] = "degraded",
3731         [MANAGER_MAINTENANCE] = "maintenance",
3732         [MANAGER_STOPPING] = "stopping",
3733 };
3734
3735 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);