src/machine/machine.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <unistd.h>
   5 #include <sys/stat.h>
   6
   7 #include "sd-messages.h"
   8
   9 #include "alloc-util.h"
  10 #include "bus-error.h"
  11 #include "bus-util.h"
  12 #include "env-file.h"
  13 #include "errno-util.h"
  14 #include "escape.h"
  15 #include "extract-word.h"
  16 #include "fd-util.h"
  17 #include "fileio.h"
  18 #include "format-util.h"
  19 #include "hashmap.h"
  20 #include "machine-dbus.h"
  21 #include "machine.h"
  22 #include "mkdir.h"
  23 #include "parse-util.h"
  24 #include "path-util.h"
  25 #include "process-util.h"
  26 #include "serialize.h"
  27 #include "special.h"
  28 #include "stdio-util.h"
  29 #include "string-table.h"
  30 #include "terminal-util.h"
  31 #include "tmpfile-util.h"
  32 #include "unit-name.h"
  33 #include "user-util.h"
  34 #include "util.h"
  35
  36 Machine* machine_new(Manager *manager, MachineClass class, const char *name) {
  37         Machine *m;
  38
  39         assert(manager);
  40         assert(class < _MACHINE_CLASS_MAX);
  41         assert(name);
  42
  43         /* Passing class == _MACHINE_CLASS_INVALID here is fine. It
  44          * means as much as "we don't know yet", and that we'll figure
  45          * it out later when loading the state file. */
  46
  47         m = new0(Machine, 1);
  48         if (!m)
  49                 return NULL;
  50
  51         m->name = strdup(name);
  52         if (!m->name)
  53                 goto fail;
  54
  55         if (class != MACHINE_HOST) {
  56                 m->state_file = path_join("/run/systemd/machines", m->name);
  57                 if (!m->state_file)
  58                         goto fail;
  59         }
  60
  61         m->class = class;
  62
  63         if (hashmap_put(manager->machines, m->name, m) < 0)
  64                 goto fail;
  65
  66         m->manager = manager;
  67
  68         return m;
  69
  70 fail:
  71         free(m->state_file);
  72         free(m->name);
  73         return mfree(m);
  74 }
  75
  76 Machine* machine_free(Machine *m) {
  77         if (!m)
  78                 return NULL;
  79
  80         while (m->operations)
  81                 operation_free(m->operations);
  82
  83         if (m->in_gc_queue)
  84                 LIST_REMOVE(gc_queue, m->manager->machine_gc_queue, m);
  85
  86         machine_release_unit(m);
  87
  88         free(m->scope_job);
  89
  90         (void) hashmap_remove(m->manager->machines, m->name);
  91
  92         if (m->manager->host_machine == m)
  93                 m->manager->host_machine = NULL;
  94
  95         if (m->leader > 0)
  96                 (void) hashmap_remove_value(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
  97
  98         sd_bus_message_unref(m->create_message);
  99
 100         free(m->name);
 101         free(m->state_file);
 102         free(m->service);
 103         free(m->root_directory);
 104         free(m->netif);
 105         return mfree(m);
 106 }
 107
 108 int machine_save(Machine *m) {
 109         _cleanup_free_ char *temp_path = NULL;
 110         _cleanup_fclose_ FILE *f = NULL;
 111         int r;
 112
 113         assert(m);
 114
 115         if (!m->state_file)
 116                 return 0;
 117
 118         if (!m->started)
 119                 return 0;
 120
 121         r = mkdir_safe_label("/run/systemd/machines", 0755, 0, 0, MKDIR_WARN_MODE);
 122         if (r < 0)
 123                 goto fail;
 124
 125         r = fopen_temporary(m->state_file, &f, &temp_path);
 126         if (r < 0)
 127                 goto fail;
 128
 129         (void) fchmod(fileno(f), 0644);
 130
 131         fprintf(f,
 132                 "# This is private data. Do not parse.\n"
 133                 "NAME=%s\n",
 134                 m->name);
 135
 136         if (m->unit) {
 137                 _cleanup_free_ char *escaped;
 138
 139                 escaped = cescape(m->unit);
 140                 if (!escaped) {
 141                         r = -ENOMEM;
 142                         goto fail;
 143                 }
 144
 145                 fprintf(f, "SCOPE=%s\n", escaped); /* We continue to call this "SCOPE=" because it is internal only, and we want to stay compatible with old files */
 146         }
 147
 148         if (m->scope_job)
 149                 fprintf(f, "SCOPE_JOB=%s\n", m->scope_job);
 150
 151         if (m->service) {
 152                 _cleanup_free_ char *escaped;
 153
 154                 escaped = cescape(m->service);
 155                 if (!escaped) {
 156                         r = -ENOMEM;
 157                         goto fail;
 158                 }
 159                 fprintf(f, "SERVICE=%s\n", escaped);
 160         }
 161
 162         if (m->root_directory) {
 163                 _cleanup_free_ char *escaped;
 164
 165                 escaped = cescape(m->root_directory);
 166                 if (!escaped) {
 167                         r = -ENOMEM;
 168                         goto fail;
 169                 }
 170                 fprintf(f, "ROOT=%s\n", escaped);
 171         }
 172
 173         if (!sd_id128_is_null(m->id))
 174                 fprintf(f, "ID=" SD_ID128_FORMAT_STR "\n", SD_ID128_FORMAT_VAL(m->id));
 175
 176         if (m->leader != 0)
 177                 fprintf(f, "LEADER="PID_FMT"\n", m->leader);
 178
 179         if (m->class != _MACHINE_CLASS_INVALID)
 180                 fprintf(f, "CLASS=%s\n", machine_class_to_string(m->class));
 181
 182         if (dual_timestamp_is_set(&m->timestamp))
 183                 fprintf(f,
 184                         "REALTIME="USEC_FMT"\n"
 185                         "MONOTONIC="USEC_FMT"\n",
 186                         m->timestamp.realtime,
 187                         m->timestamp.monotonic);
 188
 189         if (m->n_netif > 0) {
 190                 size_t i;
 191
 192                 fputs("NETIF=", f);
 193
 194                 for (i = 0; i < m->n_netif; i++) {
 195                         if (i != 0)
 196                                 fputc(' ', f);
 197
 198                         fprintf(f, "%i", m->netif[i]);
 199                 }
 200
 201                 fputc('\n', f);
 202         }
 203
 204         r = fflush_and_check(f);
 205         if (r < 0)
 206                 goto fail;
 207
 208         if (rename(temp_path, m->state_file) < 0) {
 209                 r = -errno;
 210                 goto fail;
 211         }
 212
 213         if (m->unit) {
 214                 char *sl;
 215
 216                 /* Create a symlink from the unit name to the machine
 217                  * name, so that we can quickly find the machine for
 218                  * each given unit. Ignore error. */
 219                 sl = strjoina("/run/systemd/machines/unit:", m->unit);
 220                 (void) symlink(m->name, sl);
 221         }
 222
 223         return 0;
 224
 225 fail:
 226         (void) unlink(m->state_file);
 227
 228         if (temp_path)
 229                 (void) unlink(temp_path);
 230
 231         return log_error_errno(r, "Failed to save machine data %s: %m", m->state_file);
 232 }
 233
 234 static void machine_unlink(Machine *m) {
 235         assert(m);
 236
 237         if (m->unit) {
 238                 char *sl;
 239
 240                 sl = strjoina("/run/systemd/machines/unit:", m->unit);
 241                 (void) unlink(sl);
 242         }
 243
 244         if (m->state_file)
 245                 (void) unlink(m->state_file);
 246 }
 247
 248 int machine_load(Machine *m) {
 249         _cleanup_free_ char *realtime = NULL, *monotonic = NULL, *id = NULL, *leader = NULL, *class = NULL, *netif = NULL;
 250         int r;
 251
 252         assert(m);
 253
 254         if (!m->state_file)
 255                 return 0;
 256
 257         r = parse_env_file(NULL, m->state_file,
 258                            "SCOPE",     &m->unit,
 259                            "SCOPE_JOB", &m->scope_job,
 260                            "SERVICE",   &m->service,
 261                            "ROOT",      &m->root_directory,
 262                            "ID",        &id,
 263                            "LEADER",    &leader,
 264                            "CLASS",     &class,
 265                            "REALTIME",  &realtime,
 266                            "MONOTONIC", &monotonic,
 267                            "NETIF",     &netif);
 268         if (r < 0) {
 269                 if (r == -ENOENT)
 270                         return 0;
 271
 272                 return log_error_errno(r, "Failed to read %s: %m", m->state_file);
 273         }
 274
 275         if (id)
 276                 sd_id128_from_string(id, &m->id);
 277
 278         if (leader)
 279                 parse_pid(leader, &m->leader);
 280
 281         if (class) {
 282                 MachineClass c;
 283
 284                 c = machine_class_from_string(class);
 285                 if (c >= 0)
 286                         m->class = c;
 287         }
 288
 289         if (realtime)
 290                 (void) deserialize_usec(realtime, &m->timestamp.realtime);
 291         if (monotonic)
 292                 (void) deserialize_usec(monotonic, &m->timestamp.monotonic);
 293
 294         if (netif) {
 295                 size_t allocated = 0, nr = 0;
 296                 const char *p;
 297                 int *ni = NULL;
 298
 299                 p = netif;
 300                 for (;;) {
 301                         _cleanup_free_ char *word = NULL;
 302                         int ifi;
 303
 304                         r = extract_first_word(&p, &word, NULL, 0);
 305                         if (r == 0)
 306                                 break;
 307                         if (r == -ENOMEM)
 308                                 return log_oom();
 309                         if (r < 0) {
 310                                 log_warning_errno(r, "Failed to parse NETIF: %s", netif);
 311                                 break;
 312                         }
 313
 314                         if (parse_ifindex(word, &ifi) < 0)
 315                                 continue;
 316
 317                         if (!GREEDY_REALLOC(ni, allocated, nr+1)) {
 318                                 free(ni);
 319                                 return log_oom();
 320                         }
 321
 322                         ni[nr++] = ifi;
 323                 }
 324
 325                 free(m->netif);
 326                 m->netif = ni;
 327                 m->n_netif = nr;
 328         }
 329
 330         return r;
 331 }
 332
 333 static int machine_start_scope(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 334         assert(m);
 335         assert(m->class != MACHINE_HOST);
 336
 337         if (!m->unit) {
 338                 _cleanup_free_ char *escaped = NULL, *scope = NULL;
 339                 char *description, *job = NULL;
 340                 int r;
 341
 342                 escaped = unit_name_escape(m->name);
 343                 if (!escaped)
 344                         return log_oom();
 345
 346                 scope = strjoin("machine-", escaped, ".scope");
 347                 if (!scope)
 348                         return log_oom();
 349
 350                 description = strjoina(m->class == MACHINE_VM ? "Virtual Machine " : "Container ", m->name);
 351
 352                 r = manager_start_scope(m->manager, scope, m->leader, SPECIAL_MACHINE_SLICE, description, properties, error, &job);
 353                 if (r < 0)
 354                         return log_error_errno(r, "Failed to start machine scope: %s", bus_error_message(error, r));
 355
 356                 m->unit = TAKE_PTR(scope);
 357                 m->referenced = true;
 358                 free_and_replace(m->scope_job, job);
 359         }
 360
 361         if (m->unit)
 362                 hashmap_put(m->manager->machine_units, m->unit, m);
 363
 364         return 0;
 365 }
 366
 367 int machine_start(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 368         int r;
 369
 370         assert(m);
 371
 372         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 373                 return -EOPNOTSUPP;
 374
 375         if (m->started)
 376                 return 0;
 377
 378         r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
 379         if (r < 0)
 380                 return r;
 381
 382         /* Create cgroup */
 383         r = machine_start_scope(m, properties, error);
 384         if (r < 0)
 385                 return r;
 386
 387         log_struct(LOG_INFO,
 388                    "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
 389                    "NAME=%s", m->name,
 390                    "LEADER="PID_FMT, m->leader,
 391                    LOG_MESSAGE("New machine %s.", m->name));
 392
 393         if (!dual_timestamp_is_set(&m->timestamp))
 394                 dual_timestamp_get(&m->timestamp);
 395
 396         m->started = true;
 397
 398         /* Save new machine data */
 399         machine_save(m);
 400
 401         machine_send_signal(m, true);
 402         (void) manager_enqueue_nscd_cache_flush(m->manager);
 403
 404         return 0;
 405 }
 406
 407 static int machine_stop_scope(Machine *m) {
 408         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
 409         char *job = NULL;
 410         int r, q;
 411
 412         assert(m);
 413         assert(m->class != MACHINE_HOST);
 414
 415         if (!m->unit)
 416                 return 0;
 417
 418         r = manager_stop_unit(m->manager, m->unit, &error, &job);
 419         if (r < 0) {
 420                 log_error_errno(r, "Failed to stop machine scope: %s", bus_error_message(&error, r));
 421                 sd_bus_error_free(&error);
 422         } else
 423                 free_and_replace(m->scope_job, job);
 424
 425         if (m->referenced) {
 426                 q = manager_unref_unit(m->manager, m->unit, &error);
 427                 if (q < 0)
 428                         log_warning_errno(q, "Failed to drop reference to machine scope, ignoring: %s", bus_error_message(&error, r));
 429                 m->referenced = false;
 430         }
 431
 432         return r;
 433 }
 434
 435 int machine_stop(Machine *m) {
 436         int r;
 437         assert(m);
 438
 439         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 440                 return -EOPNOTSUPP;
 441
 442         r = machine_stop_scope(m);
 443
 444         m->stopping = true;
 445
 446         machine_save(m);
 447         (void) manager_enqueue_nscd_cache_flush(m->manager);
 448
 449         return r;
 450 }
 451
 452 int machine_finalize(Machine *m) {
 453         assert(m);
 454
 455         if (m->started) {
 456                 log_struct(LOG_INFO,
 457                            "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
 458                            "NAME=%s", m->name,
 459                            "LEADER="PID_FMT, m->leader,
 460                            LOG_MESSAGE("Machine %s terminated.", m->name));
 461
 462                 m->stopping = true; /* The machine is supposed to be going away. Don't try to kill it. */
 463         }
 464
 465         machine_unlink(m);
 466         machine_add_to_gc_queue(m);
 467
 468         if (m->started) {
 469                 machine_send_signal(m, false);
 470                 m->started = false;
 471         }
 472
 473         return 0;
 474 }
 475
 476 bool machine_may_gc(Machine *m, bool drop_not_started) {
 477         assert(m);
 478
 479         if (m->class == MACHINE_HOST)
 480                 return false;
 481
 482         if (drop_not_started && !m->started)
 483                 return true;
 484
 485         if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
 486                 return false;
 487
 488         if (m->unit && manager_unit_is_active(m->manager, m->unit))
 489                 return false;
 490
 491         return true;
 492 }
 493
 494 void machine_add_to_gc_queue(Machine *m) {
 495         assert(m);
 496
 497         if (m->in_gc_queue)
 498                 return;
 499
 500         LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
 501         m->in_gc_queue = true;
 502 }
 503
 504 MachineState machine_get_state(Machine *s) {
 505         assert(s);
 506
 507         if (s->class == MACHINE_HOST)
 508                 return MACHINE_RUNNING;
 509
 510         if (s->stopping)
 511                 return MACHINE_CLOSING;
 512
 513         if (s->scope_job)
 514                 return MACHINE_OPENING;
 515
 516         return MACHINE_RUNNING;
 517 }
 518
 519 int machine_kill(Machine *m, KillWho who, int signo) {
 520         assert(m);
 521
 522         if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
 523                 return -EOPNOTSUPP;
 524
 525         if (!m->unit)
 526                 return -ESRCH;
 527
 528         if (who == KILL_LEADER) {
 529                 /* If we shall simply kill the leader, do so directly */
 530
 531                 if (kill(m->leader, signo) < 0)
 532                         return -errno;
 533
 534                 return 0;
 535         }
 536
 537         /* Otherwise, make PID 1 do it for us, for the entire cgroup */
 538         return manager_kill_unit(m->manager, m->unit, signo, NULL);
 539 }
 540
 541 int machine_openpt(Machine *m, int flags, char **ret_slave) {
 542         assert(m);
 543
 544         switch (m->class) {
 545
 546         case MACHINE_HOST:
 547
 548                 return openpt_allocate(flags, ret_slave);
 549
 550         case MACHINE_CONTAINER:
 551                 if (m->leader <= 0)
 552                         return -EINVAL;
 553
 554                 return openpt_allocate_in_namespace(m->leader, flags, ret_slave);
 555
 556         default:
 557                 return -EOPNOTSUPP;
 558         }
 559 }
 560
 561 int machine_open_terminal(Machine *m, const char *path, int mode) {
 562         assert(m);
 563
 564         switch (m->class) {
 565
 566         case MACHINE_HOST:
 567                 return open_terminal(path, mode);
 568
 569         case MACHINE_CONTAINER:
 570                 if (m->leader <= 0)
 571                         return -EINVAL;
 572
 573                 return open_terminal_in_namespace(m->leader, path, mode);
 574
 575         default:
 576                 return -EOPNOTSUPP;
 577         }
 578 }
 579
 580 void machine_release_unit(Machine *m) {
 581         assert(m);
 582
 583         if (!m->unit)
 584                 return;
 585
 586         (void) hashmap_remove(m->manager->machine_units, m->unit);
 587         m->unit = mfree(m->unit);
 588 }
 589
 590 int machine_get_uid_shift(Machine *m, uid_t *ret) {
 591         char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
 592         uid_t uid_base, uid_shift, uid_range;
 593         gid_t gid_base, gid_shift, gid_range;
 594         _cleanup_fclose_ FILE *f = NULL;
 595         int k, r;
 596
 597         assert(m);
 598         assert(ret);
 599
 600         /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
 601          * mappings. In most cases setups should be simple like this, and administrators should only care about the
 602          * basic offset a container has relative to the host. This is what this function exposes.
 603          *
 604          * If we encounter any more complex mappings we politely refuse this with ENXIO. */
 605
 606         if (m->class == MACHINE_HOST) {
 607                 *ret = 0;
 608                 return 0;
 609         }
 610
 611         if (m->class != MACHINE_CONTAINER)
 612                 return -EOPNOTSUPP;
 613
 614         xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
 615         f = fopen(p, "re");
 616         if (!f) {
 617                 if (errno == ENOENT) {
 618                         /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
 619                         *ret = 0;
 620                         return 0;
 621                 }
 622
 623                 return -errno;
 624         }
 625
 626         /* Read the first line. There's at least one. */
 627         errno = 0;
 628         k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
 629         if (k != 3) {
 630                 if (ferror(f))
 631                         return errno_or_else(EIO);
 632
 633                 return -EBADMSG;
 634         }
 635
 636         /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
 637         if (uid_base != 0)
 638                 return -ENXIO;
 639         /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
 640         if (uid_range < UID_NOBODY)
 641                 return -ENXIO;
 642
 643         /* If there's more than one line, then we don't support this mapping. */
 644         r = safe_fgetc(f, NULL);
 645         if (r < 0)
 646                 return r;
 647         if (r != 0) /* Insist on EOF */
 648                 return -ENXIO;
 649
 650         fclose(f);
 651
 652         xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
 653         f = fopen(p, "re");
 654         if (!f)
 655                 return -errno;
 656
 657         /* Read the first line. There's at least one. */
 658         errno = 0;
 659         k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
 660         if (k != 3) {
 661                 if (ferror(f))
 662                         return errno_or_else(EIO);
 663
 664                 return -EBADMSG;
 665         }
 666
 667         /* If there's more than one line, then we don't support this file. */
 668         r = safe_fgetc(f, NULL);
 669         if (r < 0)
 670                 return r;
 671         if (r != 0) /* Insist on EOF */
 672                 return -ENXIO;
 673
 674         /* If the UID and GID mapping doesn't match, we don't support this mapping. */
 675         if (uid_base != (uid_t) gid_base)
 676                 return -ENXIO;
 677         if (uid_shift != (uid_t) gid_shift)
 678                 return -ENXIO;
 679         if (uid_range != (uid_t) gid_range)
 680                 return -ENXIO;
 681
 682         *ret = uid_shift;
 683         return 0;
 684 }
 685
 686 static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
 687         [MACHINE_CONTAINER] = "container",
 688         [MACHINE_VM] = "vm",
 689         [MACHINE_HOST] = "host",
 690 };
 691
 692 DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);
 693
 694 static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
 695         [MACHINE_OPENING] = "opening",
 696         [MACHINE_RUNNING] = "running",
 697         [MACHINE_CLOSING] = "closing"
 698 };
 699
 700 DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);
 701
 702 static const char* const kill_who_table[_KILL_WHO_MAX] = {
 703         [KILL_LEADER] = "leader",
 704         [KILL_ALL] = "all"
 705 };
 706
 707 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);