src/core/manager.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <assert.h>
  23 #include <errno.h>
  24 #include <string.h>
  25 #include <signal.h>
  26 #include <sys/wait.h>
  27 #include <unistd.h>
  28 #include <sys/poll.h>
  29 #include <sys/reboot.h>
  30 #include <sys/ioctl.h>
  31 #include <linux/kd.h>
  32 #include <termios.h>
  33 #include <fcntl.h>
  34 #include <sys/types.h>
  35 #include <sys/stat.h>
  36 #include <dirent.h>
  37 #include <sys/timerfd.h>
  38
  39 #ifdef HAVE_AUDIT
  40 #include <libaudit.h>
  41 #endif
  42
  43 #include "sd-daemon.h"
  44 #include "sd-id128.h"
  45 #include "sd-messages.h"
  46
  47 #include "manager.h"
  48 #include "transaction.h"
  49 #include "hashmap.h"
  50 #include "macro.h"
  51 #include "strv.h"
  52 #include "log.h"
  53 #include "util.h"
  54 #include "mkdir.h"
  55 #include "ratelimit.h"
  56 #include "locale-setup.h"
  57 #include "mount-setup.h"
  58 #include "unit-name.h"
  59 #include "missing.h"
  60 #include "path-lookup.h"
  61 #include "special.h"
  62 #include "exit-status.h"
  63 #include "virt.h"
  64 #include "watchdog.h"
  65 #include "cgroup-util.h"
  66 #include "path-util.h"
  67 #include "audit-fd.h"
  68 #include "boot-timestamps.h"
  69 #include "env-util.h"
  70 #include "bus-errors.h"
  71 #include "bus-error.h"
  72 #include "bus-util.h"
  73 #include "dbus.h"
  74 #include "dbus-unit.h"
  75 #include "dbus-job.h"
  76 #include "dbus-manager.h"
  77 #include "bus-kernel.h"
  78
  79 /* As soon as 5s passed since a unit was added to our GC queue, make sure to run a gc sweep */
  80 #define GC_QUEUE_USEC_MAX (10*USEC_PER_SEC)
  81
  82 /* Initial delay and the interval for printing status messages about running jobs */
  83 #define JOBS_IN_PROGRESS_WAIT_SEC 5
  84 #define JOBS_IN_PROGRESS_PERIOD_SEC 1
  85 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
  86
  87 /* Where clients shall send notification messages to */
  88 #define NOTIFY_SOCKET "@/org/freedesktop/systemd1/notify"
  89
  90 #define TIME_T_MAX (time_t)((1UL << ((sizeof(time_t) << 3) - 1)) - 1)
  91
  92 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  93 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  94 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  95 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  96 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
  97 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
  98
  99 static int manager_watch_jobs_in_progress(Manager *m) {
 100         assert(m);
 101
 102         if (m->jobs_in_progress_event_source)
 103                 return 0;
 104
 105         return sd_event_add_monotonic(m->event, JOBS_IN_PROGRESS_WAIT_SEC, 0, manager_dispatch_jobs_in_progress, m, &m->jobs_in_progress_event_source);
 106 }
 107
 108 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED_ON)-1) + sizeof(ANSI_HIGHLIGHT_RED_ON)-1 + 2*(sizeof(ANSI_HIGHLIGHT_OFF)-1))
 109
 110 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
 111         char *p = buffer;
 112
 113         assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
 114         assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
 115
 116         if (pos > 1) {
 117                 if (pos > 2)
 118                         p = mempset(p, ' ', pos-2);
 119                 p = stpcpy(p, ANSI_RED_ON);
 120                 *p++ = '*';
 121         }
 122
 123         if (pos > 0 && pos <= width) {
 124                 p = stpcpy(p, ANSI_HIGHLIGHT_RED_ON);
 125                 *p++ = '*';
 126         }
 127
 128         p = stpcpy(p, ANSI_HIGHLIGHT_OFF);
 129
 130         if (pos < width) {
 131                 p = stpcpy(p, ANSI_RED_ON);
 132                 *p++ = '*';
 133                 if (pos < width-1)
 134                         p = mempset(p, ' ', width-1-pos);
 135                 strcpy(p, ANSI_HIGHLIGHT_OFF);
 136         }
 137 }
 138
 139 static void manager_print_jobs_in_progress(Manager *m) {
 140         _cleanup_free_ char *job_of_n = NULL;
 141         Iterator i;
 142         Job *j;
 143         unsigned counter = 0, print_nr;
 144         char cylon[6 + CYLON_BUFFER_EXTRA + 1];
 145         unsigned cylon_pos;
 146
 147         assert(m);
 148
 149         print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
 150
 151         HASHMAP_FOREACH(j, m->jobs, i)
 152                 if (j->state == JOB_RUNNING && counter++ == print_nr)
 153                         break;
 154
 155         /* m->n_running_jobs must be consistent with the contents of m->jobs,
 156          * so the above loop must have succeeded in finding j. */
 157         assert(counter == print_nr + 1);
 158         assert(j);
 159
 160         cylon_pos = m->jobs_in_progress_iteration % 14;
 161         if (cylon_pos >= 8)
 162                 cylon_pos = 14 - cylon_pos;
 163         draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
 164
 165         if (m->n_running_jobs > 1)
 166                 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
 167                         job_of_n = NULL;
 168
 169         manager_status_printf(m, true, cylon, "%sA %s job is running for %s",
 170                               strempty(job_of_n), job_type_to_string(j->type), unit_description(j->unit));
 171
 172         m->jobs_in_progress_iteration++;
 173 }
 174
 175 static int manager_watch_idle_pipe(Manager *m) {
 176         int r;
 177
 178         assert(m);
 179
 180         if (m->idle_pipe_event_source)
 181                 return 0;
 182
 183         if (m->idle_pipe[2] < 0)
 184                 return 0;
 185
 186         r = sd_event_add_io(m->event, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m, &m->idle_pipe_event_source);
 187         if (r < 0) {
 188                 log_error("Failed to watch idle pipe: %s", strerror(-r));
 189                 return r;
 190         }
 191
 192         return 0;
 193 }
 194
 195 static void manager_close_idle_pipe(Manager *m) {
 196         assert(m);
 197
 198         close_pipe(m->idle_pipe);
 199         close_pipe(m->idle_pipe + 2);
 200 }
 201
 202 static int manager_setup_time_change(Manager *m) {
 203         int r;
 204
 205         /* We only care for the cancellation event, hence we set the
 206          * timeout to the latest possible value. */
 207         struct itimerspec its = {
 208                 .it_value.tv_sec = TIME_T_MAX,
 209         };
 210
 211         assert(m);
 212         assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
 213
 214         /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
 215          * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
 216
 217         m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
 218         if (m->time_change_fd < 0) {
 219                 log_error("Failed to create timerfd: %m");
 220                 return -errno;
 221         }
 222
 223         if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
 224                 log_debug("Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
 225                 close_nointr_nofail(m->time_change_fd);
 226                 m->time_change_fd = -1;
 227                 return 0;
 228         }
 229
 230         r = sd_event_add_io(m->event, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m, &m->time_change_event_source);
 231         if (r < 0) {
 232                 log_error("Failed to create time change event source: %s", strerror(-r));
 233                 return r;
 234         }
 235
 236         log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
 237
 238         return 0;
 239 }
 240
 241 static int enable_special_signals(Manager *m) {
 242         _cleanup_close_ int fd = -1;
 243
 244         assert(m);
 245
 246         /* Enable that we get SIGINT on control-alt-del. In containers
 247          * this will fail with EPERM (older) or EINVAL (newer), so
 248          * ignore that. */
 249         if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
 250                 log_warning("Failed to enable ctrl-alt-del handling: %m");
 251
 252         fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
 253         if (fd < 0) {
 254                 /* Support systems without virtual console */
 255                 if (fd != -ENOENT)
 256                         log_warning("Failed to open /dev/tty0: %m");
 257         } else {
 258                 /* Enable that we get SIGWINCH on kbrequest */
 259                 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
 260                         log_warning("Failed to enable kbrequest handling: %m");
 261         }
 262
 263         return 0;
 264 }
 265
 266 static int manager_setup_signals(Manager *m) {
 267         struct sigaction sa = {
 268                 .sa_handler = SIG_DFL,
 269                 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 270         };
 271         sigset_t mask;
 272         int r;
 273
 274         assert(m);
 275
 276         /* We are not interested in SIGSTOP and friends. */
 277         assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 278
 279         assert_se(sigemptyset(&mask) == 0);
 280
 281         sigset_add_many(&mask,
 282                         SIGCHLD,     /* Child died */
 283                         SIGTERM,     /* Reexecute daemon */
 284                         SIGHUP,      /* Reload configuration */
 285                         SIGUSR1,     /* systemd/upstart: reconnect to D-Bus */
 286                         SIGUSR2,     /* systemd: dump status */
 287                         SIGINT,      /* Kernel sends us this on control-alt-del */
 288                         SIGWINCH,    /* Kernel sends us this on kbrequest (alt-arrowup) */
 289                         SIGPWR,      /* Some kernel drivers and upsd send us this on power failure */
 290                         SIGRTMIN+0,  /* systemd: start default.target */
 291                         SIGRTMIN+1,  /* systemd: isolate rescue.target */
 292                         SIGRTMIN+2,  /* systemd: isolate emergency.target */
 293                         SIGRTMIN+3,  /* systemd: start halt.target */
 294                         SIGRTMIN+4,  /* systemd: start poweroff.target */
 295                         SIGRTMIN+5,  /* systemd: start reboot.target */
 296                         SIGRTMIN+6,  /* systemd: start kexec.target */
 297                         SIGRTMIN+13, /* systemd: Immediate halt */
 298                         SIGRTMIN+14, /* systemd: Immediate poweroff */
 299                         SIGRTMIN+15, /* systemd: Immediate reboot */
 300                         SIGRTMIN+16, /* systemd: Immediate kexec */
 301                         SIGRTMIN+20, /* systemd: enable status messages */
 302                         SIGRTMIN+21, /* systemd: disable status messages */
 303                         SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
 304                         SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
 305                         SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
 306                         SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
 307                         SIGRTMIN+27, /* systemd: set log target to console */
 308                         SIGRTMIN+28, /* systemd: set log target to kmsg */
 309                         SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg */
 310                         -1);
 311         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 312
 313         m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 314         if (m->signal_fd < 0)
 315                 return -errno;
 316
 317         r = sd_event_add_io(m->event, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m, &m->signal_event_source);
 318         if (r < 0)
 319                 return r;
 320
 321         /* Process signals a bit earlier than the rest of things, but
 322          * later that notify_fd processing, so that the notify
 323          * processing can still figure out to which process/service a
 324          * message belongs, before we reap the process. */
 325         r = sd_event_source_set_priority(m->signal_event_source, -5);
 326         if (r < 0)
 327                 return r;
 328
 329         if (m->running_as == SYSTEMD_SYSTEM)
 330                 return enable_special_signals(m);
 331
 332         return 0;
 333 }
 334
 335 static int manager_default_environment(Manager *m) {
 336         assert(m);
 337
 338         if (m->running_as == SYSTEMD_SYSTEM) {
 339                 /* The system manager always starts with a clean
 340                  * environment for its children. It does not import
 341                  * the kernel or the parents exported variables.
 342                  *
 343                  * The initial passed environ is untouched to keep
 344                  * /proc/self/environ valid; it is used for tagging
 345                  * the init process inside containers. */
 346                 m->environment = strv_new("PATH=" DEFAULT_PATH,
 347                                           NULL);
 348
 349                 /* Import locale variables LC_*= from configuration */
 350                 locale_setup(&m->environment);
 351         } else
 352                 /* The user manager passes its own environment
 353                  * along to its children. */
 354                 m->environment = strv_copy(environ);
 355
 356         if (!m->environment)
 357                 return -ENOMEM;
 358
 359         strv_sort(m->environment);
 360
 361         return 0;
 362 }
 363
 364 int manager_new(SystemdRunningAs running_as, Manager **_m) {
 365         Manager *m;
 366         int r;
 367
 368         assert(_m);
 369         assert(running_as >= 0);
 370         assert(running_as < _SYSTEMD_RUNNING_AS_MAX);
 371
 372         m = new0(Manager, 1);
 373         if (!m)
 374                 return -ENOMEM;
 375
 376 #ifdef ENABLE_EFI
 377         if (detect_container(NULL) <= 0)
 378                 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
 379 #endif
 380
 381         m->running_as = running_as;
 382         m->exit_code = _MANAGER_EXIT_CODE_INVALID;
 383
 384         m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
 385
 386         m->pin_cgroupfs_fd = m->notify_fd = m->signal_fd = m->time_change_fd = m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = -1;
 387         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 388
 389         r = manager_default_environment(m);
 390         if (r < 0)
 391                 goto fail;
 392
 393         r = hashmap_ensure_allocated(&m->units, string_hash_func, string_compare_func);
 394         if (r < 0)
 395                 goto fail;
 396
 397         r = hashmap_ensure_allocated(&m->jobs, trivial_hash_func, trivial_compare_func);
 398         if (r < 0)
 399                 goto fail;
 400
 401         r = hashmap_ensure_allocated(&m->cgroup_unit, string_hash_func, string_compare_func);
 402         if (r < 0)
 403                 goto fail;
 404
 405         r = hashmap_ensure_allocated(&m->watch_pids, trivial_hash_func, trivial_compare_func);
 406         if (r < 0)
 407                 goto fail;
 408
 409         r = hashmap_ensure_allocated(&m->watch_bus, string_hash_func, string_compare_func);
 410         if (r < 0)
 411                 goto fail;
 412
 413         r = sd_event_default(&m->event);
 414         if (r < 0)
 415                 goto fail;
 416
 417         r = sd_event_add_defer(m->event, manager_dispatch_run_queue, m, &m->run_queue_event_source);
 418         if (r < 0)
 419                 goto fail;
 420
 421         r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
 422         if (r < 0)
 423                 goto fail;
 424
 425         r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
 426         if (r < 0)
 427                 goto fail;
 428
 429         r = manager_setup_signals(m);
 430         if (r < 0)
 431                 goto fail;
 432
 433         r = manager_setup_cgroup(m);
 434         if (r < 0)
 435                 goto fail;
 436
 437         r = manager_setup_time_change(m);
 438         if (r < 0)
 439                 goto fail;
 440
 441         m->udev = udev_new();
 442         if (!m->udev) {
 443                 r = -ENOMEM;
 444                 goto fail;
 445         }
 446
 447         /* Note that we set up neither kdbus, nor the notify fd
 448          * here. We do that after deserialization, since they might
 449          * have gotten serialized across the reexec. */
 450
 451         m->taint_usr = dir_is_empty("/usr") > 0;
 452
 453         *_m = m;
 454         return 0;
 455
 456 fail:
 457         manager_free(m);
 458         return r;
 459 }
 460
 461 static int manager_setup_notify(Manager *m) {
 462         union {
 463                 struct sockaddr sa;
 464                 struct sockaddr_un un;
 465         } sa = {
 466                 .sa.sa_family = AF_UNIX,
 467         };
 468         int one = 1, r;
 469
 470         if (m->notify_fd < 0) {
 471                 _cleanup_close_ int fd = -1;
 472
 473                 /* First free all secondary fields */
 474                 free(m->notify_socket);
 475                 m->notify_socket = NULL;
 476                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
 477
 478                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 479                 if (fd < 0) {
 480                         log_error("Failed to allocate notification socket: %m");
 481                         return -errno;
 482                 }
 483
 484                 if (getpid() != 1 || detect_container(NULL) > 0)
 485                         snprintf(sa.un.sun_path, sizeof(sa.un.sun_path), NOTIFY_SOCKET "/%" PRIx64, random_u64());
 486                 else
 487                         strncpy(sa.un.sun_path, NOTIFY_SOCKET, sizeof(sa.un.sun_path));
 488                 sa.un.sun_path[0] = 0;
 489
 490                 r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1));
 491                 if (r < 0) {
 492                         log_error("bind() failed: %m");
 493                         return -errno;
 494                 }
 495
 496                 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
 497                 if (r < 0) {
 498                         log_error("SO_PASSCRED failed: %m");
 499                         return -errno;
 500                 }
 501
 502                 sa.un.sun_path[0] = '@';
 503                 m->notify_socket = strdup(sa.un.sun_path);
 504                 if (!m->notify_socket)
 505                         return log_oom();
 506
 507                 m->notify_fd = fd;
 508                 fd = -1;
 509
 510                 log_debug("Using notification socket %s", m->notify_socket);
 511         }
 512
 513         if (!m->notify_event_source) {
 514                 r = sd_event_add_io(m->event, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m, &m->notify_event_source);
 515                 if (r < 0) {
 516                         log_error("Failed to allocate notify event source: %s", strerror(-r));
 517                         return -errno;
 518                 }
 519
 520                 /* Process signals a bit earlier than SIGCHLD, so that we can
 521                  * still identify to which service an exit message belongs */
 522                 r = sd_event_source_set_priority(m->notify_event_source, -7);
 523                 if (r < 0) {
 524                         log_error("Failed to set priority of notify event source: %s", strerror(-r));
 525                         return r;
 526                 }
 527         }
 528
 529         return 0;
 530 }
 531
 532 static int manager_setup_kdbus(Manager *m) {
 533         _cleanup_free_ char *p = NULL;
 534
 535         assert(m);
 536
 537 #ifdef ENABLE_KDBUS
 538         if (m->kdbus_fd >= 0)
 539                 return 0;
 540
 541         /* If there's already a bus address set, don't set up kdbus */
 542         if (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"))
 543                 return 0;
 544
 545         m->kdbus_fd = bus_kernel_create_bus(m->running_as == SYSTEMD_SYSTEM ? "system" : "user", &p);
 546         if (m->kdbus_fd < 0) {
 547                 log_debug("Failed to set up kdbus: %s", strerror(-m->kdbus_fd));
 548                 return m->kdbus_fd;
 549         }
 550
 551         log_debug("Successfully set up kdbus on %s", p);
 552
 553         /* Create the namespace directory here, so that the contents
 554          * of that directory is not visible to non-root users. This is
 555          * necessary to ensure that users cannot get access to busses
 556          * of virtualized users when no UID namespacing is used. */
 557         mkdir_p_label("/dev/kdbus/ns", 0700);
 558 #endif
 559
 560         return 0;
 561 }
 562
 563 static int manager_connect_bus(Manager *m, bool reexecuting) {
 564         bool try_bus_connect;
 565
 566         assert(m);
 567
 568         try_bus_connect =
 569                 m->kdbus_fd >= 0 ||
 570                 reexecuting ||
 571                 (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"));
 572
 573         /* Try to connect to the busses, if possible. */
 574         return bus_init(m, try_bus_connect);
 575 }
 576
 577 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
 578         Unit *u;
 579         unsigned n = 0;
 580
 581         assert(m);
 582
 583         while ((u = m->cleanup_queue)) {
 584                 assert(u->in_cleanup_queue);
 585
 586                 unit_free(u);
 587                 n++;
 588         }
 589
 590         return n;
 591 }
 592
 593 enum {
 594         GC_OFFSET_IN_PATH,  /* This one is on the path we were traveling */
 595         GC_OFFSET_UNSURE,   /* No clue */
 596         GC_OFFSET_GOOD,     /* We still need this unit */
 597         GC_OFFSET_BAD,      /* We don't need this unit anymore */
 598         _GC_OFFSET_MAX
 599 };
 600
 601 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
 602         Iterator i;
 603         Unit *other;
 604         bool is_bad;
 605
 606         assert(u);
 607
 608         if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
 609             u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 610             u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
 611                 return;
 612
 613         if (u->in_cleanup_queue)
 614                 goto bad;
 615
 616         if (unit_check_gc(u))
 617                 goto good;
 618
 619         u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
 620
 621         is_bad = true;
 622
 623         SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
 624                 unit_gc_sweep(other, gc_marker);
 625
 626                 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
 627                         goto good;
 628
 629                 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
 630                         is_bad = false;
 631         }
 632
 633         if (is_bad)
 634                 goto bad;
 635
 636         /* We were unable to find anything out about this entry, so
 637          * let's investigate it later */
 638         u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
 639         unit_add_to_gc_queue(u);
 640         return;
 641
 642 bad:
 643         /* We definitely know that this one is not useful anymore, so
 644          * let's mark it for deletion */
 645         u->gc_marker = gc_marker + GC_OFFSET_BAD;
 646         unit_add_to_cleanup_queue(u);
 647         return;
 648
 649 good:
 650         u->gc_marker = gc_marker + GC_OFFSET_GOOD;
 651 }
 652
 653 static unsigned manager_dispatch_gc_queue(Manager *m) {
 654         Unit *u;
 655         unsigned n = 0;
 656         unsigned gc_marker;
 657
 658         assert(m);
 659
 660         /* log_debug("Running GC..."); */
 661
 662         m->gc_marker += _GC_OFFSET_MAX;
 663         if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
 664                 m->gc_marker = 1;
 665
 666         gc_marker = m->gc_marker;
 667
 668         while ((u = m->gc_queue)) {
 669                 assert(u->in_gc_queue);
 670
 671                 unit_gc_sweep(u, gc_marker);
 672
 673                 LIST_REMOVE(gc_queue, m->gc_queue, u);
 674                 u->in_gc_queue = false;
 675
 676                 n++;
 677
 678                 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 679                     u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
 680                         log_debug_unit(u->id, "Collecting %s", u->id);
 681                         u->gc_marker = gc_marker + GC_OFFSET_BAD;
 682                         unit_add_to_cleanup_queue(u);
 683                 }
 684         }
 685
 686         m->n_in_gc_queue = 0;
 687
 688         return n;
 689 }
 690
 691 static void manager_clear_jobs_and_units(Manager *m) {
 692         Unit *u;
 693
 694         assert(m);
 695
 696         while ((u = hashmap_first(m->units)))
 697                 unit_free(u);
 698
 699         manager_dispatch_cleanup_queue(m);
 700
 701         assert(!m->load_queue);
 702         assert(!m->run_queue);
 703         assert(!m->dbus_unit_queue);
 704         assert(!m->dbus_job_queue);
 705         assert(!m->cleanup_queue);
 706         assert(!m->gc_queue);
 707
 708         assert(hashmap_isempty(m->jobs));
 709         assert(hashmap_isempty(m->units));
 710
 711         m->n_on_console = 0;
 712         m->n_running_jobs = 0;
 713 }
 714
 715 void manager_free(Manager *m) {
 716         UnitType c;
 717         int i;
 718
 719         assert(m);
 720
 721         manager_clear_jobs_and_units(m);
 722
 723         for (c = 0; c < _UNIT_TYPE_MAX; c++)
 724                 if (unit_vtable[c]->shutdown)
 725                         unit_vtable[c]->shutdown(m);
 726
 727         /* If we reexecute ourselves, we keep the root cgroup
 728          * around */
 729         manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
 730
 731         manager_undo_generators(m);
 732
 733         bus_done(m);
 734
 735         hashmap_free(m->units);
 736         hashmap_free(m->jobs);
 737         hashmap_free(m->watch_pids);
 738         hashmap_free(m->watch_bus);
 739
 740         sd_event_source_unref(m->signal_event_source);
 741         sd_event_source_unref(m->notify_event_source);
 742         sd_event_source_unref(m->time_change_event_source);
 743         sd_event_source_unref(m->jobs_in_progress_event_source);
 744         sd_event_source_unref(m->idle_pipe_event_source);
 745         sd_event_source_unref(m->run_queue_event_source);
 746
 747         if (m->signal_fd >= 0)
 748                 close_nointr_nofail(m->signal_fd);
 749         if (m->notify_fd >= 0)
 750                 close_nointr_nofail(m->notify_fd);
 751         if (m->time_change_fd >= 0)
 752                 close_nointr_nofail(m->time_change_fd);
 753         if (m->kdbus_fd >= 0)
 754                 close_nointr_nofail(m->kdbus_fd);
 755
 756         manager_close_idle_pipe(m);
 757
 758         udev_unref(m->udev);
 759         sd_event_unref(m->event);
 760
 761         free(m->notify_socket);
 762
 763         lookup_paths_free(&m->lookup_paths);
 764         strv_free(m->environment);
 765
 766         hashmap_free(m->cgroup_unit);
 767         set_free_free(m->unit_path_cache);
 768
 769         free(m->switch_root);
 770         free(m->switch_root_init);
 771
 772         for (i = 0; i < RLIMIT_NLIMITS; i++)
 773                 free(m->rlimit[i]);
 774
 775         assert(hashmap_isempty(m->units_requiring_mounts_for));
 776         hashmap_free(m->units_requiring_mounts_for);
 777
 778         free(m);
 779 }
 780
 781 int manager_enumerate(Manager *m) {
 782         int r = 0, q;
 783         UnitType c;
 784
 785         assert(m);
 786
 787         /* Let's ask every type to load all units from disk/kernel
 788          * that it might know */
 789         for (c = 0; c < _UNIT_TYPE_MAX; c++)
 790                 if (unit_vtable[c]->enumerate) {
 791                         q = unit_vtable[c]->enumerate(m);
 792                         if (q < 0)
 793                                 r = q;
 794                 }
 795
 796         manager_dispatch_load_queue(m);
 797         return r;
 798 }
 799
 800 static int manager_coldplug(Manager *m) {
 801         int r = 0, q;
 802         Iterator i;
 803         Unit *u;
 804         char *k;
 805
 806         assert(m);
 807
 808         /* Then, let's set up their initial state. */
 809         HASHMAP_FOREACH_KEY(u, k, m->units, i) {
 810
 811                 /* ignore aliases */
 812                 if (u->id != k)
 813                         continue;
 814
 815                 if ((q = unit_coldplug(u)) < 0)
 816                         r = q;
 817         }
 818
 819         return r;
 820 }
 821
 822 static void manager_build_unit_path_cache(Manager *m) {
 823         char **i;
 824         _cleanup_free_ DIR *d = NULL;
 825         int r;
 826
 827         assert(m);
 828
 829         set_free_free(m->unit_path_cache);
 830
 831         m->unit_path_cache = set_new(string_hash_func, string_compare_func);
 832         if (!m->unit_path_cache) {
 833                 log_error("Failed to allocate unit path cache.");
 834                 return;
 835         }
 836
 837         /* This simply builds a list of files we know exist, so that
 838          * we don't always have to go to disk */
 839
 840         STRV_FOREACH(i, m->lookup_paths.unit_path) {
 841                 struct dirent *de;
 842
 843                 d = opendir(*i);
 844                 if (!d) {
 845                         if (errno != ENOENT)
 846                                 log_error("Failed to open directory %s: %m", *i);
 847                         continue;
 848                 }
 849
 850                 while ((de = readdir(d))) {
 851                         char *p;
 852
 853                         if (ignore_file(de->d_name))
 854                                 continue;
 855
 856                         p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
 857                         if (!p) {
 858                                 r = -ENOMEM;
 859                                 goto fail;
 860                         }
 861
 862                         r = set_consume(m->unit_path_cache, p);
 863                         if (r < 0)
 864                                 goto fail;
 865                 }
 866
 867                 closedir(d);
 868                 d = NULL;
 869         }
 870
 871         return;
 872
 873 fail:
 874         log_error("Failed to build unit path cache: %s", strerror(-r));
 875
 876         set_free_free(m->unit_path_cache);
 877         m->unit_path_cache = NULL;
 878 }
 879
 880
 881 static int manager_distribute_fds(Manager *m, FDSet *fds) {
 882         Unit *u;
 883         Iterator i;
 884         int r;
 885
 886         assert(m);
 887
 888         HASHMAP_FOREACH(u, m->units, i) {
 889
 890                 if (fdset_size(fds) <= 0)
 891                         break;
 892
 893                 if (UNIT_VTABLE(u)->distribute_fds) {
 894                         r = UNIT_VTABLE(u)->distribute_fds(u, fds);
 895                         if (r < 0)
 896                                 return r;
 897                 }
 898         }
 899
 900         return 0;
 901 }
 902
 903 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
 904         int r, q;
 905
 906         assert(m);
 907
 908         dual_timestamp_get(&m->generators_start_timestamp);
 909         manager_run_generators(m);
 910         dual_timestamp_get(&m->generators_finish_timestamp);
 911
 912         r = lookup_paths_init(
 913                         &m->lookup_paths, m->running_as, true,
 914                         m->generator_unit_path,
 915                         m->generator_unit_path_early,
 916                         m->generator_unit_path_late);
 917         if (r < 0)
 918                 return r;
 919
 920         manager_build_unit_path_cache(m);
 921
 922         /* If we will deserialize make sure that during enumeration
 923          * this is already known, so we increase the counter here
 924          * already */
 925         if (serialization)
 926                 m->n_reloading ++;
 927
 928         /* First, enumerate what we can from all config files */
 929         dual_timestamp_get(&m->units_load_start_timestamp);
 930         r = manager_enumerate(m);
 931         dual_timestamp_get(&m->units_load_finish_timestamp);
 932
 933         /* Second, deserialize if there is something to deserialize */
 934         if (serialization) {
 935                 q = manager_deserialize(m, serialization, fds);
 936                 if (q < 0)
 937                         r = q;
 938         }
 939
 940         /* Any fds left? Find some unit which wants them. This is
 941          * useful to allow container managers to pass some file
 942          * descriptors to us pre-initialized. This enables
 943          * socket-based activation of entire containers. */
 944         if (fdset_size(fds) > 0) {
 945                 q = manager_distribute_fds(m, fds);
 946                 if (q < 0)
 947                         r = q;
 948         }
 949
 950         /* We might have deserialized the notify fd, but if we didn't
 951          * then let's create the bus now */
 952         manager_setup_notify(m);
 953
 954         /* We might have deserialized the kdbus control fd, but if we
 955          * didn't, then let's create the bus now. */
 956         manager_setup_kdbus(m);
 957         manager_connect_bus(m, !!serialization);
 958
 959         /* Third, fire things up! */
 960         q = manager_coldplug(m);
 961         if (q < 0)
 962                 r = q;
 963
 964         if (serialization) {
 965                 assert(m->n_reloading > 0);
 966                 m->n_reloading --;
 967
 968                 /* Let's wait for the UnitNew/JobNew messages being
 969                  * sent, before we notify that the reload is
 970                  * finished */
 971                 m->send_reloading_done = true;
 972         }
 973
 974         return r;
 975 }
 976
 977 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
 978         int r;
 979         Transaction *tr;
 980
 981         assert(m);
 982         assert(type < _JOB_TYPE_MAX);
 983         assert(unit);
 984         assert(mode < _JOB_MODE_MAX);
 985
 986         if (mode == JOB_ISOLATE && type != JOB_START) {
 987                 sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
 988                 return -EINVAL;
 989         }
 990
 991         if (mode == JOB_ISOLATE && !unit->allow_isolate) {
 992                 sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
 993                 return -EPERM;
 994         }
 995
 996         log_debug_unit(unit->id,
 997                        "Trying to enqueue job %s/%s/%s", unit->id,
 998                        job_type_to_string(type), job_mode_to_string(mode));
 999
1000         job_type_collapse(&type, unit);
1001
1002         tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1003         if (!tr)
1004                 return -ENOMEM;
1005
1006         r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, override, false,
1007                                                  mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1008                                                  mode == JOB_IGNORE_DEPENDENCIES, e);
1009         if (r < 0)
1010                 goto tr_abort;
1011
1012         if (mode == JOB_ISOLATE) {
1013                 r = transaction_add_isolate_jobs(tr, m);
1014                 if (r < 0)
1015                         goto tr_abort;
1016         }
1017
1018         r = transaction_activate(tr, m, mode, e);
1019         if (r < 0)
1020                 goto tr_abort;
1021
1022         log_debug_unit(unit->id,
1023                        "Enqueued job %s/%s as %u", unit->id,
1024                        job_type_to_string(type), (unsigned) tr->anchor_job->id);
1025
1026         if (_ret)
1027                 *_ret = tr->anchor_job;
1028
1029         transaction_free(tr);
1030         return 0;
1031
1032 tr_abort:
1033         transaction_abort(tr);
1034         transaction_free(tr);
1035         return r;
1036 }
1037
1038 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1039         Unit *unit;
1040         int r;
1041
1042         assert(m);
1043         assert(type < _JOB_TYPE_MAX);
1044         assert(name);
1045         assert(mode < _JOB_MODE_MAX);
1046
1047         r = manager_load_unit(m, name, NULL, NULL, &unit);
1048         if (r < 0)
1049                 return r;
1050
1051         return manager_add_job(m, type, unit, mode, override, e, _ret);
1052 }
1053
1054 Job *manager_get_job(Manager *m, uint32_t id) {
1055         assert(m);
1056
1057         return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1058 }
1059
1060 Unit *manager_get_unit(Manager *m, const char *name) {
1061         assert(m);
1062         assert(name);
1063
1064         return hashmap_get(m->units, name);
1065 }
1066
1067 unsigned manager_dispatch_load_queue(Manager *m) {
1068         Unit *u;
1069         unsigned n = 0;
1070
1071         assert(m);
1072
1073         /* Make sure we are not run recursively */
1074         if (m->dispatching_load_queue)
1075                 return 0;
1076
1077         m->dispatching_load_queue = true;
1078
1079         /* Dispatches the load queue. Takes a unit from the queue and
1080          * tries to load its data until the queue is empty */
1081
1082         while ((u = m->load_queue)) {
1083                 assert(u->in_load_queue);
1084
1085                 unit_load(u);
1086                 n++;
1087         }
1088
1089         m->dispatching_load_queue = false;
1090         return n;
1091 }
1092
1093 int manager_load_unit_prepare(
1094                 Manager *m,
1095                 const char *name,
1096                 const char *path,
1097                 sd_bus_error *e,
1098                 Unit **_ret) {
1099
1100         Unit *ret;
1101         UnitType t;
1102         int r;
1103
1104         assert(m);
1105         assert(name || path);
1106
1107         /* This will prepare the unit for loading, but not actually
1108          * load anything from disk. */
1109
1110         if (path && !is_path(path))
1111                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1112
1113         if (!name)
1114                 name = basename(path);
1115
1116         t = unit_name_to_type(name);
1117
1118         if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, false))
1119                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1120
1121         ret = manager_get_unit(m, name);
1122         if (ret) {
1123                 *_ret = ret;
1124                 return 1;
1125         }
1126
1127         ret = unit_new(m, unit_vtable[t]->object_size);
1128         if (!ret)
1129                 return -ENOMEM;
1130
1131         if (path) {
1132                 ret->fragment_path = strdup(path);
1133                 if (!ret->fragment_path) {
1134                         unit_free(ret);
1135                         return -ENOMEM;
1136                 }
1137         }
1138
1139         r = unit_add_name(ret, name);
1140         if (r < 0) {
1141                 unit_free(ret);
1142                 return r;
1143         }
1144
1145         unit_add_to_load_queue(ret);
1146         unit_add_to_dbus_queue(ret);
1147         unit_add_to_gc_queue(ret);
1148
1149         if (_ret)
1150                 *_ret = ret;
1151
1152         return 0;
1153 }
1154
1155 int manager_load_unit(
1156                 Manager *m,
1157                 const char *name,
1158                 const char *path,
1159                 sd_bus_error *e,
1160                 Unit **_ret) {
1161
1162         int r;
1163
1164         assert(m);
1165
1166         /* This will load the service information files, but not actually
1167          * start any services or anything. */
1168
1169         r = manager_load_unit_prepare(m, name, path, e, _ret);
1170         if (r != 0)
1171                 return r;
1172
1173         manager_dispatch_load_queue(m);
1174
1175         if (_ret)
1176                 *_ret = unit_follow_merge(*_ret);
1177
1178         return 0;
1179 }
1180
1181 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1182         Iterator i;
1183         Job *j;
1184
1185         assert(s);
1186         assert(f);
1187
1188         HASHMAP_FOREACH(j, s->jobs, i)
1189                 job_dump(j, f, prefix);
1190 }
1191
1192 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1193         Iterator i;
1194         Unit *u;
1195         const char *t;
1196
1197         assert(s);
1198         assert(f);
1199
1200         HASHMAP_FOREACH_KEY(u, t, s->units, i)
1201                 if (u->id == t)
1202                         unit_dump(u, f, prefix);
1203 }
1204
1205 void manager_clear_jobs(Manager *m) {
1206         Job *j;
1207
1208         assert(m);
1209
1210         while ((j = hashmap_first(m->jobs)))
1211                 /* No need to recurse. We're cancelling all jobs. */
1212                 job_finish_and_invalidate(j, JOB_CANCELED, false);
1213 }
1214
1215 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1216         Manager *m = userdata;
1217         Job *j;
1218
1219         assert(source);
1220         assert(m);
1221
1222         while ((j = m->run_queue)) {
1223                 assert(j->installed);
1224                 assert(j->in_run_queue);
1225
1226                 job_run_and_invalidate(j);
1227         }
1228
1229         if (m->n_running_jobs > 0)
1230                 manager_watch_jobs_in_progress(m);
1231
1232         if (m->n_on_console > 0)
1233                 manager_watch_idle_pipe(m);
1234
1235         return 1;
1236 }
1237
1238 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1239         Job *j;
1240         Unit *u;
1241         unsigned n = 0;
1242
1243         assert(m);
1244
1245         if (m->dispatching_dbus_queue)
1246                 return 0;
1247
1248         m->dispatching_dbus_queue = true;
1249
1250         while ((u = m->dbus_unit_queue)) {
1251                 assert(u->in_dbus_queue);
1252
1253                 bus_unit_send_change_signal(u);
1254                 n++;
1255         }
1256
1257         while ((j = m->dbus_job_queue)) {
1258                 assert(j->in_dbus_queue);
1259
1260                 bus_job_send_change_signal(j);
1261                 n++;
1262         }
1263
1264         m->dispatching_dbus_queue = false;
1265
1266         if (m->send_reloading_done) {
1267                 m->send_reloading_done = false;
1268
1269                 bus_manager_send_reloading(m, false);
1270         }
1271
1272         if (m->queued_message)
1273                 bus_send_queued_message(m);
1274
1275         return n;
1276 }
1277
1278 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1279         Manager *m = userdata;
1280         ssize_t n;
1281
1282         assert(m);
1283         assert(m->notify_fd == fd);
1284
1285         if (revents != EPOLLIN) {
1286                 log_warning("Got unexpected poll event for notify fd.");
1287                 return 0;
1288         }
1289
1290         for (;;) {
1291                 char buf[4096];
1292                 struct iovec iovec = {
1293                         .iov_base = buf,
1294                         .iov_len = sizeof(buf)-1,
1295                 };
1296
1297                 union {
1298                         struct cmsghdr cmsghdr;
1299                         uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
1300                 } control = {};
1301
1302                 struct msghdr msghdr = {
1303                         .msg_iov = &iovec,
1304                         .msg_iovlen = 1,
1305                         .msg_control = &control,
1306                         .msg_controllen = sizeof(control),
1307                 };
1308                 struct ucred *ucred;
1309                 Unit *u;
1310                 _cleanup_strv_free_ char **tags = NULL;
1311
1312                 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT);
1313                 if (n <= 0) {
1314                         if (n == 0)
1315                                 return -EIO;
1316
1317                         if (errno == EAGAIN || errno == EINTR)
1318                                 break;
1319
1320                         return -errno;
1321                 }
1322
1323                 if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) ||
1324                     control.cmsghdr.cmsg_level != SOL_SOCKET ||
1325                     control.cmsghdr.cmsg_type != SCM_CREDENTIALS ||
1326                     control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) {
1327                         log_warning("Received notify message without credentials. Ignoring.");
1328                         continue;
1329                 }
1330
1331                 ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
1332
1333                 u = hashmap_get(m->watch_pids, LONG_TO_PTR(ucred->pid));
1334                 if (!u) {
1335                         u = manager_get_unit_by_pid(m, ucred->pid);
1336                         if (!u) {
1337                                 log_warning("Cannot find unit for notify message of PID %lu.", (unsigned long) ucred->pid);
1338                                 continue;
1339                         }
1340                 }
1341
1342                 assert((size_t) n < sizeof(buf));
1343                 buf[n] = 0;
1344                 tags = strv_split(buf, "\n\r");
1345                 if (!tags)
1346                         return log_oom();
1347
1348                 log_debug_unit(u->id, "Got notification message for unit %s", u->id);
1349
1350                 if (UNIT_VTABLE(u)->notify_message)
1351                         UNIT_VTABLE(u)->notify_message(u, ucred->pid, tags);
1352         }
1353
1354         return 0;
1355 }
1356
1357 static int manager_dispatch_sigchld(Manager *m) {
1358         assert(m);
1359
1360         for (;;) {
1361                 siginfo_t si = {};
1362                 Unit *u;
1363
1364                 /* First we call waitd() for a PID and do not reap the
1365                  * zombie. That way we can still access /proc/$PID for
1366                  * it while it is a zombie. */
1367                 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1368
1369                         if (errno == ECHILD)
1370                                 break;
1371
1372                         if (errno == EINTR)
1373                                 continue;
1374
1375                         return -errno;
1376                 }
1377
1378                 if (si.si_pid <= 0)
1379                         break;
1380
1381                 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1382                         _cleanup_free_ char *name = NULL;
1383
1384                         get_process_comm(si.si_pid, &name);
1385                         log_debug("Got SIGCHLD for process %lu (%s)", (unsigned long) si.si_pid, strna(name));
1386                 }
1387
1388                 /* And now figure out the unit this belongs to */
1389                 u = hashmap_get(m->watch_pids, LONG_TO_PTR(si.si_pid));
1390                 if (!u)
1391                         u = manager_get_unit_by_pid(m, si.si_pid);
1392
1393                 /* And now, we actually reap the zombie. */
1394                 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1395                         if (errno == EINTR)
1396                                 continue;
1397
1398                         return -errno;
1399                 }
1400
1401                 if (si.si_code != CLD_EXITED && si.si_code != CLD_KILLED && si.si_code != CLD_DUMPED)
1402                         continue;
1403
1404                 log_debug("Child %lu died (code=%s, status=%i/%s)",
1405                           (long unsigned) si.si_pid,
1406                           sigchld_code_to_string(si.si_code),
1407                           si.si_status,
1408                           strna(si.si_code == CLD_EXITED
1409                                 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1410                                 : signal_to_string(si.si_status)));
1411
1412                 if (!u)
1413                         continue;
1414
1415                 log_debug_unit(u->id,
1416                                "Child %lu belongs to %s", (long unsigned) si.si_pid, u->id);
1417
1418                 hashmap_remove(m->watch_pids, LONG_TO_PTR(si.si_pid));
1419                 UNIT_VTABLE(u)->sigchld_event(u, si.si_pid, si.si_code, si.si_status);
1420         }
1421
1422         return 0;
1423 }
1424
1425 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1426         _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1427         int r;
1428
1429         log_debug_unit(name, "Activating special unit %s", name);
1430
1431         r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL);
1432         if (r < 0)
1433                 log_error_unit(name, "Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1434
1435         return r;
1436 }
1437
1438 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1439         Manager *m = userdata;
1440         ssize_t n;
1441         struct signalfd_siginfo sfsi;
1442         bool sigchld = false;
1443
1444         assert(m);
1445         assert(m->signal_fd == fd);
1446
1447         if (revents != EPOLLIN) {
1448                 log_warning("Got unexpected events from signal file descriptor.");
1449                 return 0;
1450         }
1451
1452         for (;;) {
1453                 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1454                 if (n != sizeof(sfsi)) {
1455
1456                         if (n >= 0)
1457                                 return -EIO;
1458
1459                         if (errno == EINTR || errno == EAGAIN)
1460                                 break;
1461
1462                         return -errno;
1463                 }
1464
1465                 if (sfsi.ssi_pid > 0) {
1466                         char *p = NULL;
1467
1468                         get_process_comm(sfsi.ssi_pid, &p);
1469
1470                         log_debug("Received SIG%s from PID %lu (%s).",
1471                                   signal_to_string(sfsi.ssi_signo),
1472                                   (unsigned long) sfsi.ssi_pid, strna(p));
1473                         free(p);
1474                 } else
1475                         log_debug("Received SIG%s.", signal_to_string(sfsi.ssi_signo));
1476
1477                 switch (sfsi.ssi_signo) {
1478
1479                 case SIGCHLD:
1480                         sigchld = true;
1481                         break;
1482
1483                 case SIGTERM:
1484                         if (m->running_as == SYSTEMD_SYSTEM) {
1485                                 /* This is for compatibility with the
1486                                  * original sysvinit */
1487                                 m->exit_code = MANAGER_REEXECUTE;
1488                                 break;
1489                         }
1490
1491                         /* Fall through */
1492
1493                 case SIGINT:
1494                         if (m->running_as == SYSTEMD_SYSTEM) {
1495                                 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1496                                 break;
1497                         }
1498
1499                         /* Run the exit target if there is one, if not, just exit. */
1500                         if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1501                                 m->exit_code = MANAGER_EXIT;
1502                                 return 0;
1503                         }
1504
1505                         break;
1506
1507                 case SIGWINCH:
1508                         if (m->running_as == SYSTEMD_SYSTEM)
1509                                 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
1510
1511                         /* This is a nop on non-init */
1512                         break;
1513
1514                 case SIGPWR:
1515                         if (m->running_as == SYSTEMD_SYSTEM)
1516                                 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
1517
1518                         /* This is a nop on non-init */
1519                         break;
1520
1521                 case SIGUSR1: {
1522                         Unit *u;
1523
1524                         u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1525
1526                         if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
1527                                 log_info("Trying to reconnect to bus...");
1528                                 bus_init(m, true);
1529                         }
1530
1531                         if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
1532                                 log_info("Loading D-Bus service...");
1533                                 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
1534                         }
1535
1536                         break;
1537                 }
1538
1539                 case SIGUSR2: {
1540                         _cleanup_free_ char *dump = NULL;
1541                         _cleanup_fclose_ FILE *f = NULL;
1542                         size_t size;
1543
1544                         f = open_memstream(&dump, &size);
1545                         if (!f) {
1546                                 log_warning("Failed to allocate memory stream.");
1547                                 break;
1548                         }
1549
1550                         manager_dump_units(m, f, "\t");
1551                         manager_dump_jobs(m, f, "\t");
1552
1553                         if (ferror(f)) {
1554                                 log_warning("Failed to write status stream");
1555                                 break;
1556                         }
1557
1558                         log_dump(LOG_INFO, dump);
1559                         break;
1560                 }
1561
1562                 case SIGHUP:
1563                         m->exit_code = MANAGER_RELOAD;
1564                         break;
1565
1566                 default: {
1567
1568                         /* Starting SIGRTMIN+0 */
1569                         static const char * const target_table[] = {
1570                                 [0] = SPECIAL_DEFAULT_TARGET,
1571                                 [1] = SPECIAL_RESCUE_TARGET,
1572                                 [2] = SPECIAL_EMERGENCY_TARGET,
1573                                 [3] = SPECIAL_HALT_TARGET,
1574                                 [4] = SPECIAL_POWEROFF_TARGET,
1575                                 [5] = SPECIAL_REBOOT_TARGET,
1576                                 [6] = SPECIAL_KEXEC_TARGET
1577                         };
1578
1579                         /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1580                         static const ManagerExitCode code_table[] = {
1581                                 [0] = MANAGER_HALT,
1582                                 [1] = MANAGER_POWEROFF,
1583                                 [2] = MANAGER_REBOOT,
1584                                 [3] = MANAGER_KEXEC
1585                         };
1586
1587                         if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
1588                             (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
1589                                 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
1590                                 manager_start_target(m, target_table[idx],
1591                                                      (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
1592                                 break;
1593                         }
1594
1595                         if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
1596                             (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
1597                                 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
1598                                 break;
1599                         }
1600
1601                         switch (sfsi.ssi_signo - SIGRTMIN) {
1602
1603                         case 20:
1604                                 log_debug("Enabling showing of status.");
1605                                 manager_set_show_status(m, true);
1606                                 break;
1607
1608                         case 21:
1609                                 log_debug("Disabling showing of status.");
1610                                 manager_set_show_status(m, false);
1611                                 break;
1612
1613                         case 22:
1614                                 log_set_max_level(LOG_DEBUG);
1615                                 log_notice("Setting log level to debug.");
1616                                 break;
1617
1618                         case 23:
1619                                 log_set_max_level(LOG_INFO);
1620                                 log_notice("Setting log level to info.");
1621                                 break;
1622
1623                         case 24:
1624                                 if (m->running_as == SYSTEMD_USER) {
1625                                         m->exit_code = MANAGER_EXIT;
1626                                         return 0;
1627                                 }
1628
1629                                 /* This is a nop on init */
1630                                 break;
1631
1632                         case 26:
1633                                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1634                                 log_notice("Setting log target to journal-or-kmsg.");
1635                                 break;
1636
1637                         case 27:
1638                                 log_set_target(LOG_TARGET_CONSOLE);
1639                                 log_notice("Setting log target to console.");
1640                                 break;
1641
1642                         case 28:
1643                                 log_set_target(LOG_TARGET_KMSG);
1644                                 log_notice("Setting log target to kmsg.");
1645                                 break;
1646
1647                         case 29:
1648                                 log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
1649                                 log_notice("Setting log target to syslog-or-kmsg.");
1650                                 break;
1651
1652                         default:
1653                                 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
1654                         }
1655                 }
1656                 }
1657         }
1658
1659         if (sigchld)
1660                 manager_dispatch_sigchld(m);
1661
1662         return 0;
1663 }
1664
1665 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1666         Manager *m = userdata;
1667         Iterator i;
1668         Unit *u;
1669
1670         assert(m);
1671         assert(m->time_change_fd == fd);
1672
1673         log_struct(LOG_INFO,
1674                    MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
1675                    "MESSAGE=Time has been changed",
1676                    NULL);
1677
1678         /* Restart the watch */
1679         m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
1680
1681         close_nointr_nofail(m->time_change_fd);
1682         m->time_change_fd = -1;
1683
1684         manager_setup_time_change(m);
1685
1686         HASHMAP_FOREACH(u, m->units, i)
1687                 if (UNIT_VTABLE(u)->time_change)
1688                         UNIT_VTABLE(u)->time_change(u);
1689
1690         return 0;
1691 }
1692
1693 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1694         Manager *m = userdata;
1695
1696         assert(m);
1697         assert(m->idle_pipe[2] == fd);
1698
1699         m->no_console_output = m->n_on_console > 0;
1700
1701         m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
1702         manager_close_idle_pipe(m);
1703
1704         return 0;
1705 }
1706
1707 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
1708         Manager *m = userdata;
1709
1710         assert(m);
1711
1712         manager_print_jobs_in_progress(m);
1713         return 0;
1714 }
1715
1716 int manager_loop(Manager *m) {
1717         int r;
1718
1719         RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
1720
1721         assert(m);
1722         m->exit_code = MANAGER_RUNNING;
1723
1724         /* Release the path cache */
1725         set_free_free(m->unit_path_cache);
1726         m->unit_path_cache = NULL;
1727
1728         manager_check_finished(m);
1729
1730         /* There might still be some zombies hanging around from
1731          * before we were exec()'ed. Let's reap them. */
1732         r = manager_dispatch_sigchld(m);
1733         if (r < 0)
1734                 return r;
1735
1736         while (m->exit_code == MANAGER_RUNNING) {
1737                 usec_t wait_usec;
1738
1739                 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM)
1740                         watchdog_ping();
1741
1742                 if (!ratelimit_test(&rl)) {
1743                         /* Yay, something is going seriously wrong, pause a little */
1744                         log_warning("Looping too fast. Throttling execution a little.");
1745                         sleep(1);
1746                         continue;
1747                 }
1748
1749                 if (manager_dispatch_load_queue(m) > 0)
1750                         continue;
1751
1752                 if (manager_dispatch_gc_queue(m) > 0)
1753                         continue;
1754
1755                 if (manager_dispatch_cleanup_queue(m) > 0)
1756                         continue;
1757
1758                 if (manager_dispatch_cgroup_queue(m) > 0)
1759                         continue;
1760
1761                 if (manager_dispatch_dbus_queue(m) > 0)
1762                         continue;
1763
1764                 /* Sleep for half the watchdog time */
1765                 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) {
1766                         wait_usec = m->runtime_watchdog / 2;
1767                         if (wait_usec <= 0)
1768                                 wait_usec = 1;
1769                 } else
1770                         wait_usec = (usec_t) -1;
1771
1772                 r = sd_event_run(m->event, wait_usec);
1773                 if (r < 0) {
1774                         log_error("Failed to run event loop: %s", strerror(-r));
1775                         return r;
1776                 }
1777         }
1778
1779         return m->exit_code;
1780 }
1781
1782 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
1783         _cleanup_free_ char *n = NULL;
1784         Unit *u;
1785         int r;
1786
1787         assert(m);
1788         assert(s);
1789         assert(_u);
1790
1791         r = unit_name_from_dbus_path(s, &n);
1792         if (r < 0)
1793                 return r;
1794
1795         r = manager_load_unit(m, n, NULL, e, &u);
1796         if (r < 0)
1797                 return r;
1798
1799         *_u = u;
1800
1801         return 0;
1802 }
1803
1804 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
1805         const char *p;
1806         unsigned id;
1807         Job *j;
1808         int r;
1809
1810         assert(m);
1811         assert(s);
1812         assert(_j);
1813
1814         p = startswith(s, "/org/freedesktop/systemd1/job/");
1815         if (!p)
1816                 return -EINVAL;
1817
1818         r = safe_atou(p, &id);
1819         if (r < 0)
1820                 return r;
1821
1822         j = manager_get_job(m, id);
1823         if (!j)
1824                 return -ENOENT;
1825
1826         *_j = j;
1827
1828         return 0;
1829 }
1830
1831 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
1832
1833 #ifdef HAVE_AUDIT
1834         char *p;
1835         int audit_fd;
1836
1837         audit_fd = get_audit_fd();
1838         if (audit_fd < 0)
1839                 return;
1840
1841         /* Don't generate audit events if the service was already
1842          * started and we're just deserializing */
1843         if (m->n_reloading > 0)
1844                 return;
1845
1846         if (m->running_as != SYSTEMD_SYSTEM)
1847                 return;
1848
1849         if (u->type != UNIT_SERVICE)
1850                 return;
1851
1852         p = unit_name_to_prefix_and_instance(u->id);
1853         if (!p) {
1854                 log_error_unit(u->id,
1855                                "Failed to allocate unit name for audit message: %s", strerror(ENOMEM));
1856                 return;
1857         }
1858
1859         if (audit_log_user_comm_message(audit_fd, type, "", p, NULL, NULL, NULL, success) < 0) {
1860                 if (errno == EPERM) {
1861                         /* We aren't allowed to send audit messages?
1862                          * Then let's not retry again. */
1863                         close_audit_fd();
1864                 } else
1865                         log_warning("Failed to send audit message: %m");
1866         }
1867
1868         free(p);
1869 #endif
1870
1871 }
1872
1873 void manager_send_unit_plymouth(Manager *m, Unit *u) {
1874         int fd = -1;
1875         union sockaddr_union sa;
1876         int n = 0;
1877         char *message = NULL;
1878
1879         /* Don't generate plymouth events if the service was already
1880          * started and we're just deserializing */
1881         if (m->n_reloading > 0)
1882                 return;
1883
1884         if (m->running_as != SYSTEMD_SYSTEM)
1885                 return;
1886
1887         if (detect_container(NULL) > 0)
1888                 return;
1889
1890         if (u->type != UNIT_SERVICE &&
1891             u->type != UNIT_MOUNT &&
1892             u->type != UNIT_SWAP)
1893                 return;
1894
1895         /* We set SOCK_NONBLOCK here so that we rather drop the
1896          * message then wait for plymouth */
1897         fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1898         if (fd < 0) {
1899                 log_error("socket() failed: %m");
1900                 return;
1901         }
1902
1903         zero(sa);
1904         sa.sa.sa_family = AF_UNIX;
1905         strncpy(sa.un.sun_path+1, "/org/freedesktop/plymouthd", sizeof(sa.un.sun_path)-1);
1906         if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) {
1907
1908                 if (errno != EPIPE &&
1909                     errno != EAGAIN &&
1910                     errno != ENOENT &&
1911                     errno != ECONNREFUSED &&
1912                     errno != ECONNRESET &&
1913                     errno != ECONNABORTED)
1914                         log_error("connect() failed: %m");
1915
1916                 goto finish;
1917         }
1918
1919         if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
1920                 log_oom();
1921                 goto finish;
1922         }
1923
1924         errno = 0;
1925         if (write(fd, message, n + 1) != n + 1) {
1926
1927                 if (errno != EPIPE &&
1928                     errno != EAGAIN &&
1929                     errno != ENOENT &&
1930                     errno != ECONNREFUSED &&
1931                     errno != ECONNRESET &&
1932                     errno != ECONNABORTED)
1933                         log_error("Failed to write Plymouth message: %m");
1934
1935                 goto finish;
1936         }
1937
1938 finish:
1939         if (fd >= 0)
1940                 close_nointr_nofail(fd);
1941
1942         free(message);
1943 }
1944
1945 void manager_dispatch_bus_name_owner_changed(
1946                 Manager *m,
1947                 const char *name,
1948                 const char* old_owner,
1949                 const char *new_owner) {
1950
1951         Unit *u;
1952
1953         assert(m);
1954         assert(name);
1955
1956         u = hashmap_get(m->watch_bus, name);
1957         if (!u)
1958                 return;
1959
1960         UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner);
1961 }
1962
1963 int manager_open_serialization(Manager *m, FILE **_f) {
1964         _cleanup_free_ char *path = NULL;
1965         int fd = -1;
1966         FILE *f;
1967
1968         assert(_f);
1969
1970         if (m->running_as == SYSTEMD_SYSTEM)
1971                 asprintf(&path, "/run/systemd/dump-%lu-XXXXXX", (unsigned long) getpid());
1972         else
1973                 asprintf(&path, "/tmp/systemd-dump-%lu-XXXXXX", (unsigned long) getpid());
1974
1975         if (!path)
1976                 return -ENOMEM;
1977
1978         RUN_WITH_UMASK(0077) {
1979                 fd = mkostemp(path, O_RDWR|O_CLOEXEC);
1980         }
1981
1982         if (fd < 0)
1983                 return -errno;
1984
1985         unlink(path);
1986         log_debug("Serializing state to %s", path);
1987
1988         f = fdopen(fd, "w+");
1989         if (!f) {
1990                 close_nointr_nofail(fd);
1991                 return -errno;
1992         }
1993
1994         *_f = f;
1995
1996         return 0;
1997 }
1998
1999 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2000         Iterator i;
2001         Unit *u;
2002         const char *t;
2003         char **e;
2004         int r;
2005
2006         assert(m);
2007         assert(f);
2008         assert(fds);
2009
2010         m->n_reloading ++;
2011
2012         fprintf(f, "current-job-id=%i\n", m->current_job_id);
2013         fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2014         fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2015         fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2016
2017         dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2018         dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2019         dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2020         dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2021
2022         if (!in_initrd()) {
2023                 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2024                 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2025                 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2026                 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2027                 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2028                 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2029                 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2030                 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2031         }
2032
2033         if (!switching_root) {
2034                 STRV_FOREACH(e, m->environment) {
2035                         _cleanup_free_ char *ce;
2036
2037                         ce = cescape(*e);
2038                         if (!ce)
2039                                 return -ENOMEM;
2040
2041                         fprintf(f, "env=%s\n", *e);
2042                 }
2043         }
2044
2045         if (m->notify_fd >= 0) {
2046                 int copy;
2047
2048                 copy = fdset_put_dup(fds, m->notify_fd);
2049                 if (copy < 0)
2050                         return copy;
2051
2052                 fprintf(f, "notify-fd=%i\n", copy);
2053                 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2054         }
2055
2056         if (m->kdbus_fd >= 0) {
2057                 int copy;
2058
2059                 copy = fdset_put_dup(fds, m->kdbus_fd);
2060                 if (copy < 0)
2061                         return copy;
2062
2063                 fprintf(f, "kdbus-fd=%i\n", copy);
2064         }
2065
2066         bus_serialize(m, f);
2067
2068         fputc('\n', f);
2069
2070         HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2071                 if (u->id != t)
2072                         continue;
2073
2074                 if (!unit_can_serialize(u))
2075                         continue;
2076
2077                 /* Start marker */
2078                 fputs(u->id, f);
2079                 fputc('\n', f);
2080
2081                 r = unit_serialize(u, f, fds, !switching_root);
2082                 if (r < 0) {
2083                         m->n_reloading --;
2084                         return r;
2085                 }
2086         }
2087
2088         assert(m->n_reloading > 0);
2089         m->n_reloading --;
2090
2091         if (ferror(f))
2092                 return -EIO;
2093
2094         r = bus_fdset_add_all(m, fds);
2095         if (r < 0)
2096                 return r;
2097
2098         return 0;
2099 }
2100
2101 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2102         int r = 0;
2103
2104         assert(m);
2105         assert(f);
2106
2107         log_debug("Deserializing state...");
2108
2109         m->n_reloading ++;
2110
2111         for (;;) {
2112                 char line[LINE_MAX], *l;
2113
2114                 if (!fgets(line, sizeof(line), f)) {
2115                         if (feof(f))
2116                                 r = 0;
2117                         else
2118                                 r = -errno;
2119
2120                         goto finish;
2121                 }
2122
2123                 char_array_0(line);
2124                 l = strstrip(line);
2125
2126                 if (l[0] == 0)
2127                         break;
2128
2129                 if (startswith(l, "current-job-id=")) {
2130                         uint32_t id;
2131
2132                         if (safe_atou32(l+15, &id) < 0)
2133                                 log_debug("Failed to parse current job id value %s", l+15);
2134                         else
2135                                 m->current_job_id = MAX(m->current_job_id, id);
2136
2137                 } else if (startswith(l, "n-installed-jobs=")) {
2138                         uint32_t n;
2139
2140                         if (safe_atou32(l+17, &n) < 0)
2141                                 log_debug("Failed to parse installed jobs counter %s", l+17);
2142                         else
2143                                 m->n_installed_jobs += n;
2144
2145                 } else if (startswith(l, "n-failed-jobs=")) {
2146                         uint32_t n;
2147
2148                         if (safe_atou32(l+14, &n) < 0)
2149                                 log_debug("Failed to parse failed jobs counter %s", l+14);
2150                         else
2151                                 m->n_failed_jobs += n;
2152
2153                 } else if (startswith(l, "taint-usr=")) {
2154                         int b;
2155
2156                         b = parse_boolean(l+10);
2157                         if (b < 0)
2158                                 log_debug("Failed to parse taint /usr flag %s", l+10);
2159                         else
2160                                 m->taint_usr = m->taint_usr || b;
2161
2162                 } else if (startswith(l, "firmware-timestamp="))
2163                         dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2164                 else if (startswith(l, "loader-timestamp="))
2165                         dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2166                 else if (startswith(l, "kernel-timestamp="))
2167                         dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2168                 else if (startswith(l, "initrd-timestamp="))
2169                         dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2170                 else if (startswith(l, "userspace-timestamp="))
2171                         dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2172                 else if (startswith(l, "finish-timestamp="))
2173                         dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2174                 else if (startswith(l, "security-start-timestamp="))
2175                         dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2176                 else if (startswith(l, "security-finish-timestamp="))
2177                         dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2178                 else if (startswith(l, "generators-start-timestamp="))
2179                         dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2180                 else if (startswith(l, "generators-finish-timestamp="))
2181                         dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2182                 else if (startswith(l, "units-load-start-timestamp="))
2183                         dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2184                 else if (startswith(l, "units-load-finish-timestamp="))
2185                         dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2186                 else if (startswith(l, "env=")) {
2187                         _cleanup_free_ char *uce = NULL;
2188                         char **e;
2189
2190                         uce = cunescape(l+4);
2191                         if (!uce) {
2192                                 r = -ENOMEM;
2193                                 goto finish;
2194                         }
2195
2196                         e = strv_env_set(m->environment, uce);
2197                         if (!e) {
2198                                 r = -ENOMEM;
2199                                 goto finish;
2200                         }
2201
2202                         strv_free(m->environment);
2203                         m->environment = e;
2204
2205                 } else if (startswith(l, "notify-fd=")) {
2206                         int fd;
2207
2208                         if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2209                                 log_debug("Failed to parse notify fd: %s", l + 10);
2210                         else {
2211                                 if (m->notify_fd >= 0)
2212                                         close_nointr_nofail(m->notify_fd);
2213
2214                                 m->notify_fd = fdset_remove(fds, fd);
2215                         }
2216
2217                 } else if (startswith(l, "notify-socket=")) {
2218                         char *n;
2219
2220                         n = strdup(l+14);
2221                         if (!n) {
2222                                 r = -ENOMEM;
2223                                 goto finish;
2224                         }
2225
2226                         free(m->notify_socket);
2227                         m->notify_socket = n;
2228
2229                 } else if (startswith(l, "kdbus-fd=")) {
2230                         int fd;
2231
2232                         if (safe_atoi(l + 9, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2233                                 log_debug("Failed to parse kdbus fd: %s", l + 9);
2234                         else {
2235                                 if (m->kdbus_fd >= 0)
2236                                         close_nointr_nofail(m->kdbus_fd);
2237
2238                                 m->kdbus_fd = fdset_remove(fds, fd);
2239                         }
2240
2241                 } else if (bus_deserialize_item(m, l) == 0)
2242                         log_debug("Unknown serialization item '%s'", l);
2243         }
2244
2245         for (;;) {
2246                 Unit *u;
2247                 char name[UNIT_NAME_MAX+2];
2248
2249                 /* Start marker */
2250                 if (!fgets(name, sizeof(name), f)) {
2251                         if (feof(f))
2252                                 r = 0;
2253                         else
2254                                 r = -errno;
2255
2256                         goto finish;
2257                 }
2258
2259                 char_array_0(name);
2260
2261                 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2262                 if (r < 0)
2263                         goto finish;
2264
2265                 r = unit_deserialize(u, f, fds);
2266                 if (r < 0)
2267                         goto finish;
2268         }
2269
2270 finish:
2271         if (ferror(f))
2272                 r = -EIO;
2273
2274         assert(m->n_reloading > 0);
2275         m->n_reloading --;
2276
2277         return r;
2278 }
2279
2280 int manager_reload(Manager *m) {
2281         int r, q;
2282         _cleanup_fclose_ FILE *f = NULL;
2283         _cleanup_fdset_free_ FDSet *fds = NULL;
2284
2285         assert(m);
2286
2287         r = manager_open_serialization(m, &f);
2288         if (r < 0)
2289                 return r;
2290
2291         m->n_reloading ++;
2292         bus_manager_send_reloading(m, true);
2293
2294         fds = fdset_new();
2295         if (!fds) {
2296                 m->n_reloading --;
2297                 return -ENOMEM;
2298         }
2299
2300         r = manager_serialize(m, f, fds, false);
2301         if (r < 0) {
2302                 m->n_reloading --;
2303                 return r;
2304         }
2305
2306         if (fseeko(f, 0, SEEK_SET) < 0) {
2307                 m->n_reloading --;
2308                 return -errno;
2309         }
2310
2311         /* From here on there is no way back. */
2312         manager_clear_jobs_and_units(m);
2313         manager_undo_generators(m);
2314         lookup_paths_free(&m->lookup_paths);
2315
2316         /* Find new unit paths */
2317         manager_run_generators(m);
2318
2319         q = lookup_paths_init(
2320                         &m->lookup_paths, m->running_as, true,
2321                         m->generator_unit_path,
2322                         m->generator_unit_path_early,
2323                         m->generator_unit_path_late);
2324         if (q < 0)
2325                 r = q;
2326
2327         manager_build_unit_path_cache(m);
2328
2329         /* First, enumerate what we can from all config files */
2330         q = manager_enumerate(m);
2331         if (q < 0)
2332                 r = q;
2333
2334         /* Second, deserialize our stored data */
2335         q = manager_deserialize(m, f, fds);
2336         if (q < 0)
2337                 r = q;
2338
2339         fclose(f);
2340         f = NULL;
2341
2342         /* Third, fire things up! */
2343         q = manager_coldplug(m);
2344         if (q < 0)
2345                 r = q;
2346
2347         assert(m->n_reloading > 0);
2348         m->n_reloading--;
2349
2350         m->send_reloading_done = true;
2351
2352         return r;
2353 }
2354
2355 static bool manager_is_booting_or_shutting_down(Manager *m) {
2356         Unit *u;
2357
2358         assert(m);
2359
2360         /* Is the initial job still around? */
2361         if (manager_get_job(m, m->default_unit_job_id))
2362                 return true;
2363
2364         /* Is there a job for the shutdown target? */
2365         u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
2366         if (u)
2367                 return !!u->job;
2368
2369         return false;
2370 }
2371
2372 bool manager_is_reloading_or_reexecuting(Manager *m) {
2373         assert(m);
2374
2375         return m->n_reloading != 0;
2376 }
2377
2378 void manager_reset_failed(Manager *m) {
2379         Unit *u;
2380         Iterator i;
2381
2382         assert(m);
2383
2384         HASHMAP_FOREACH(u, m->units, i)
2385                 unit_reset_failed(u);
2386 }
2387
2388 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2389         Unit *u;
2390
2391         assert(m);
2392         assert(name);
2393
2394         /* Returns true if the unit is inactive or going down */
2395         u = manager_get_unit(m, name);
2396         if (!u)
2397                 return true;
2398
2399         return unit_inactive_or_pending(u);
2400 }
2401
2402 void manager_check_finished(Manager *m) {
2403         char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2404         usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2405
2406         assert(m);
2407
2408         if (m->n_running_jobs == 0)
2409                 m->jobs_in_progress_event_source = sd_event_source_unref(m->jobs_in_progress_event_source);
2410
2411         if (hashmap_size(m->jobs) > 0) {
2412                 if (m->jobs_in_progress_event_source)
2413                         sd_event_source_set_time(m->jobs_in_progress_event_source, JOBS_IN_PROGRESS_PERIOD_SEC);
2414                 return;
2415         }
2416
2417         /* Notify Type=idle units that we are done now */
2418         m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
2419         manager_close_idle_pipe(m);
2420
2421         /* Turn off confirm spawn now */
2422         m->confirm_spawn = false;
2423
2424         if (dual_timestamp_is_set(&m->finish_timestamp))
2425                 return;
2426
2427         dual_timestamp_get(&m->finish_timestamp);
2428
2429         if (m->running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0) {
2430
2431                 /* Note that m->kernel_usec.monotonic is always at 0,
2432                  * and m->firmware_usec.monotonic and
2433                  * m->loader_usec.monotonic should be considered
2434                  * negative values. */
2435
2436                 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2437                 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2438                 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2439                 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2440
2441                 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2442
2443                         kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2444                         initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2445
2446                         if (!log_on_console())
2447                                 log_struct(LOG_INFO,
2448                                            MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2449                                            "KERNEL_USEC=%llu", (unsigned long long) kernel_usec,
2450                                            "INITRD_USEC=%llu", (unsigned long long) initrd_usec,
2451                                            "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec,
2452                                            "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2453                                            format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2454                                            format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2455                                            format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2456                                            format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2457                                            NULL);
2458                 } else {
2459                         kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2460                         initrd_usec = 0;
2461
2462                         if (!log_on_console())
2463                                 log_struct(LOG_INFO,
2464                                            MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2465                                            "KERNEL_USEC=%llu", (unsigned long long) kernel_usec,
2466                                            "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec,
2467                                            "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.",
2468                                            format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2469                                            format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2470                                            format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2471                                            NULL);
2472                 }
2473         } else {
2474                 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2475                 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2476
2477                 if (!log_on_console())
2478                         log_struct(LOG_INFO,
2479                                    MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2480                                    "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec,
2481                                    "MESSAGE=Startup finished in %s.",
2482                                    format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2483                                    NULL);
2484         }
2485
2486         bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2487
2488         sd_notifyf(false,
2489                    "READY=1\nSTATUS=Startup finished in %s.",
2490                    format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2491 }
2492
2493 static int create_generator_dir(Manager *m, char **generator, const char *name) {
2494         char *p;
2495         int r;
2496
2497         assert(m);
2498         assert(generator);
2499         assert(name);
2500
2501         if (*generator)
2502                 return 0;
2503
2504         if (m->running_as == SYSTEMD_SYSTEM && getpid() == 1) {
2505
2506                 p = strappend("/run/systemd/", name);
2507                 if (!p)
2508                         return log_oom();
2509
2510                 r = mkdir_p_label(p, 0755);
2511                 if (r < 0) {
2512                         log_error("Failed to create generator directory %s: %s",
2513                                   p, strerror(-r));
2514                         free(p);
2515                         return r;
2516                 }
2517         } else {
2518                 p = strjoin("/tmp/systemd-", name, ".XXXXXX", NULL);
2519                 if (!p)
2520                         return log_oom();
2521
2522                 if (!mkdtemp(p)) {
2523                         log_error("Failed to create generator directory %s: %m",
2524                                   p);
2525                         free(p);
2526                         return -errno;
2527                 }
2528         }
2529
2530         *generator = p;
2531         return 0;
2532 }
2533
2534 static void trim_generator_dir(Manager *m, char **generator) {
2535         assert(m);
2536         assert(generator);
2537
2538         if (!*generator)
2539                 return;
2540
2541         if (rmdir(*generator) >= 0) {
2542                 free(*generator);
2543                 *generator = NULL;
2544         }
2545
2546         return;
2547 }
2548
2549 void manager_run_generators(Manager *m) {
2550         _cleanup_closedir_ DIR *d = NULL;
2551         const char *generator_path;
2552         const char *argv[5];
2553         int r;
2554
2555         assert(m);
2556
2557         generator_path = m->running_as == SYSTEMD_SYSTEM ? SYSTEM_GENERATOR_PATH : USER_GENERATOR_PATH;
2558         d = opendir(generator_path);
2559         if (!d) {
2560                 if (errno == ENOENT)
2561                         return;
2562
2563                 log_error("Failed to enumerate generator directory %s: %m",
2564                           generator_path);
2565                 return;
2566         }
2567
2568         r = create_generator_dir(m, &m->generator_unit_path, "generator");
2569         if (r < 0)
2570                 goto finish;
2571
2572         r = create_generator_dir(m, &m->generator_unit_path_early, "generator.early");
2573         if (r < 0)
2574                 goto finish;
2575
2576         r = create_generator_dir(m, &m->generator_unit_path_late, "generator.late");
2577         if (r < 0)
2578                 goto finish;
2579
2580         argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
2581         argv[1] = m->generator_unit_path;
2582         argv[2] = m->generator_unit_path_early;
2583         argv[3] = m->generator_unit_path_late;
2584         argv[4] = NULL;
2585
2586         RUN_WITH_UMASK(0022)
2587                 execute_directory(generator_path, d, (char**) argv);
2588
2589 finish:
2590         trim_generator_dir(m, &m->generator_unit_path);
2591         trim_generator_dir(m, &m->generator_unit_path_early);
2592         trim_generator_dir(m, &m->generator_unit_path_late);
2593 }
2594
2595 static void remove_generator_dir(Manager *m, char **generator) {
2596         assert(m);
2597         assert(generator);
2598
2599         if (!*generator)
2600                 return;
2601
2602         strv_remove(m->lookup_paths.unit_path, *generator);
2603         rm_rf(*generator, false, true, false);
2604
2605         free(*generator);
2606         *generator = NULL;
2607 }
2608
2609 void manager_undo_generators(Manager *m) {
2610         assert(m);
2611
2612         remove_generator_dir(m, &m->generator_unit_path);
2613         remove_generator_dir(m, &m->generator_unit_path_early);
2614         remove_generator_dir(m, &m->generator_unit_path_late);
2615 }
2616
2617 int manager_environment_add(Manager *m, char **minus, char **plus) {
2618         char **a = NULL, **b = NULL, **l;
2619         assert(m);
2620
2621         l = m->environment;
2622
2623         if (!strv_isempty(minus)) {
2624                 a = strv_env_delete(l, 1, minus);
2625                 if (!a)
2626                         return -ENOMEM;
2627
2628                 l = a;
2629         }
2630
2631         if (!strv_isempty(plus)) {
2632                 b = strv_env_merge(2, l, plus);
2633                 if (!b)
2634                         return -ENOMEM;
2635
2636                 l = b;
2637         }
2638
2639         if (m->environment != l)
2640                 strv_free(m->environment);
2641         if (a != l)
2642                 strv_free(a);
2643         if (b != l)
2644                 strv_free(b);
2645
2646         m->environment = strv_sort(l);
2647         return 0;
2648 }
2649
2650 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
2651         int i;
2652
2653         assert(m);
2654
2655         for (i = 0; i < RLIMIT_NLIMITS; i++) {
2656                 if (!default_rlimit[i])
2657                         continue;
2658
2659                 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
2660                 if (!m->rlimit[i])
2661                         return -ENOMEM;
2662         }
2663
2664         return 0;
2665 }
2666
2667 void manager_recheck_journal(Manager *m) {
2668         Unit *u;
2669
2670         assert(m);
2671
2672         if (m->running_as != SYSTEMD_SYSTEM)
2673                 return;
2674
2675         u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
2676         if (u && SOCKET(u)->state != SOCKET_RUNNING) {
2677                 log_close_journal();
2678                 return;
2679         }
2680
2681         u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
2682         if (u && SERVICE(u)->state != SERVICE_RUNNING) {
2683                 log_close_journal();
2684                 return;
2685         }
2686
2687         /* Hmm, OK, so the socket is fully up and the service is up
2688          * too, then let's make use of the thing. */
2689         log_open();
2690 }
2691
2692 void manager_set_show_status(Manager *m, bool b) {
2693         assert(m);
2694
2695         if (m->running_as != SYSTEMD_SYSTEM)
2696                 return;
2697
2698         m->show_status = b;
2699
2700         if (b)
2701                 touch("/run/systemd/show-status");
2702         else
2703                 unlink("/run/systemd/show-status");
2704 }
2705
2706 static bool manager_get_show_status(Manager *m) {
2707         assert(m);
2708
2709         if (m->running_as != SYSTEMD_SYSTEM)
2710                 return false;
2711
2712         if (m->no_console_output)
2713                 return false;
2714
2715         if (m->show_status)
2716                 return true;
2717
2718         /* If Plymouth is running make sure we show the status, so
2719          * that there's something nice to see when people press Esc */
2720
2721         return plymouth_running();
2722 }
2723
2724 void manager_status_printf(Manager *m, bool ephemeral, const char *status, const char *format, ...) {
2725         va_list ap;
2726
2727         if (!manager_get_show_status(m))
2728                 return;
2729
2730         /* XXX We should totally drop the check for ephemeral here
2731          * and thus effectively make 'Type=idle' pointless. */
2732         if (ephemeral && m->n_on_console > 0)
2733                 return;
2734
2735         if (!manager_is_booting_or_shutting_down(m))
2736                 return;
2737
2738         va_start(ap, format);
2739         status_vprintf(status, true, ephemeral, format, ap);
2740         va_end(ap);
2741 }
2742
2743 int manager_get_unit_by_path(Manager *m, const char *path, const char *suffix, Unit **_found) {
2744         _cleanup_free_ char *p = NULL;
2745         Unit *found;
2746
2747         assert(m);
2748         assert(path);
2749         assert(suffix);
2750         assert(_found);
2751
2752         p = unit_name_from_path(path, suffix);
2753         if (!p)
2754                 return -ENOMEM;
2755
2756         found = manager_get_unit(m, p);
2757         if (!found) {
2758                 *_found = NULL;
2759                 return 0;
2760         }
2761
2762         *_found = found;
2763         return 1;
2764 }
2765
2766 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
2767         char p[strlen(path)+1];
2768
2769         assert(m);
2770         assert(path);
2771
2772         strcpy(p, path);
2773         path_kill_slashes(p);
2774
2775         return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
2776 }