src/machine/machine.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <string.h>
   5 #include <unistd.h>
   6 #include <stdio_ext.h>
   7 #include <sys/stat.h>
   8
   9 #include "sd-messages.h"
  10
  11 #include "alloc-util.h"
  12 #include "bus-error.h"
  13 #include "bus-util.h"
  14 #include "env-file.h"
  15 #include "escape.h"
  16 #include "extract-word.h"
  17 #include "fd-util.h"
  18 #include "fileio.h"
  19 #include "format-util.h"
  20 #include "hashmap.h"
  21 #include "machine-dbus.h"
  22 #include "machine.h"
  23 #include "mkdir.h"
  24 #include "parse-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 = strappend("/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) __fsetlocking(f, FSETLOCKING_BYCALLER);
 130         (void) fchmod(fileno(f), 0644);
 131
 132         fprintf(f,
 133                 "# This is private data. Do not parse.\n"
 134                 "NAME=%s\n",
 135                 m->name);
 136
 137         if (m->unit) {
 138                 _cleanup_free_ char *escaped;
 139
 140                 escaped = cescape(m->unit);
 141                 if (!escaped) {
 142                         r = -ENOMEM;
 143                         goto fail;
 144                 }
 145
 146                 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 */
 147         }
 148
 149         if (m->scope_job)
 150                 fprintf(f, "SCOPE_JOB=%s\n", m->scope_job);
 151
 152         if (m->service) {
 153                 _cleanup_free_ char *escaped;
 154
 155                 escaped = cescape(m->service);
 156                 if (!escaped) {
 157                         r = -ENOMEM;
 158                         goto fail;
 159                 }
 160                 fprintf(f, "SERVICE=%s\n", escaped);
 161         }
 162
 163         if (m->root_directory) {
 164                 _cleanup_free_ char *escaped;
 165
 166                 escaped = cescape(m->root_directory);
 167                 if (!escaped) {
 168                         r = -ENOMEM;
 169                         goto fail;
 170                 }
 171                 fprintf(f, "ROOT=%s\n", escaped);
 172         }
 173
 174         if (!sd_id128_is_null(m->id))
 175                 fprintf(f, "ID=" SD_ID128_FORMAT_STR "\n", SD_ID128_FORMAT_VAL(m->id));
 176
 177         if (m->leader != 0)
 178                 fprintf(f, "LEADER="PID_FMT"\n", m->leader);
 179
 180         if (m->class != _MACHINE_CLASS_INVALID)
 181                 fprintf(f, "CLASS=%s\n", machine_class_to_string(m->class));
 182
 183         if (dual_timestamp_is_set(&m->timestamp))
 184                 fprintf(f,
 185                         "REALTIME="USEC_FMT"\n"
 186                         "MONOTONIC="USEC_FMT"\n",
 187                         m->timestamp.realtime,
 188                         m->timestamp.monotonic);
 189
 190         if (m->n_netif > 0) {
 191                 size_t i;
 192
 193                 fputs("NETIF=", f);
 194
 195                 for (i = 0; i < m->n_netif; i++) {
 196                         if (i != 0)
 197                                 fputc(' ', f);
 198
 199                         fprintf(f, "%i", m->netif[i]);
 200                 }
 201
 202                 fputc('\n', f);
 203         }
 204
 205         r = fflush_and_check(f);
 206         if (r < 0)
 207                 goto fail;
 208
 209         if (rename(temp_path, m->state_file) < 0) {
 210                 r = -errno;
 211                 goto fail;
 212         }
 213
 214         if (m->unit) {
 215                 char *sl;
 216
 217                 /* Create a symlink from the unit name to the machine
 218                  * name, so that we can quickly find the machine for
 219                  * each given unit. Ignore error. */
 220                 sl = strjoina("/run/systemd/machines/unit:", m->unit);
 221                 (void) symlink(m->name, sl);
 222         }
 223
 224         return 0;
 225
 226 fail:
 227         (void) unlink(m->state_file);
 228
 229         if (temp_path)
 230                 (void) unlink(temp_path);
 231
 232         return log_error_errno(r, "Failed to save machine data %s: %m", m->state_file);
 233 }
 234
 235 static void machine_unlink(Machine *m) {
 236         assert(m);
 237
 238         if (m->unit) {
 239                 char *sl;
 240
 241                 sl = strjoina("/run/systemd/machines/unit:", m->unit);
 242                 (void) unlink(sl);
 243         }
 244
 245         if (m->state_file)
 246                 (void) unlink(m->state_file);
 247 }
 248
 249 int machine_load(Machine *m) {
 250         _cleanup_free_ char *realtime = NULL, *monotonic = NULL, *id = NULL, *leader = NULL, *class = NULL, *netif = NULL;
 251         int r;
 252
 253         assert(m);
 254
 255         if (!m->state_file)
 256                 return 0;
 257
 258         r = parse_env_file(NULL, m->state_file,
 259                            "SCOPE",     &m->unit,
 260                            "SCOPE_JOB", &m->scope_job,
 261                            "SERVICE",   &m->service,
 262                            "ROOT",      &m->root_directory,
 263                            "ID",        &id,
 264                            "LEADER",    &leader,
 265                            "CLASS",     &class,
 266                            "REALTIME",  &realtime,
 267                            "MONOTONIC", &monotonic,
 268                            "NETIF",     &netif);
 269         if (r < 0) {
 270                 if (r == -ENOENT)
 271                         return 0;
 272
 273                 return log_error_errno(r, "Failed to read %s: %m", m->state_file);
 274         }
 275
 276         if (id)
 277                 sd_id128_from_string(id, &m->id);
 278
 279         if (leader)
 280                 parse_pid(leader, &m->leader);
 281
 282         if (class) {
 283                 MachineClass c;
 284
 285                 c = machine_class_from_string(class);
 286                 if (c >= 0)
 287                         m->class = c;
 288         }
 289
 290         if (realtime)
 291                 (void) deserialize_usec(realtime, &m->timestamp.realtime);
 292         if (monotonic)
 293                 (void) deserialize_usec(monotonic, &m->timestamp.monotonic);
 294
 295         if (netif) {
 296                 size_t allocated = 0, nr = 0;
 297                 const char *p;
 298                 int *ni = NULL;
 299
 300                 p = netif;
 301                 for (;;) {
 302                         _cleanup_free_ char *word = NULL;
 303                         int ifi;
 304
 305                         r = extract_first_word(&p, &word, NULL, 0);
 306                         if (r == 0)
 307                                 break;
 308                         if (r == -ENOMEM)
 309                                 return log_oom();
 310                         if (r < 0) {
 311                                 log_warning_errno(r, "Failed to parse NETIF: %s", netif);
 312                                 break;
 313                         }
 314
 315                         if (parse_ifindex(word, &ifi) < 0)
 316                                 continue;
 317
 318                         if (!GREEDY_REALLOC(ni, allocated, nr+1)) {
 319                                 free(ni);
 320                                 return log_oom();
 321                         }
 322
 323                         ni[nr++] = ifi;
 324                 }
 325
 326                 free(m->netif);
 327                 m->netif = ni;
 328                 m->n_netif = nr;
 329         }
 330
 331         return r;
 332 }
 333
 334 static int machine_start_scope(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 335         assert(m);
 336         assert(m->class != MACHINE_HOST);
 337
 338         if (!m->unit) {
 339                 _cleanup_free_ char *escaped = NULL, *scope = NULL;
 340                 char *description, *job = NULL;
 341                 int r;
 342
 343                 escaped = unit_name_escape(m->name);
 344                 if (!escaped)
 345                         return log_oom();
 346
 347                 scope = strjoin("machine-", escaped, ".scope");
 348                 if (!scope)
 349                         return log_oom();
 350
 351                 description = strjoina(m->class == MACHINE_VM ? "Virtual Machine " : "Container ", m->name);
 352
 353                 r = manager_start_scope(m->manager, scope, m->leader, SPECIAL_MACHINE_SLICE, description, properties, error, &job);
 354                 if (r < 0)
 355                         return log_error_errno(r, "Failed to start machine scope: %s", bus_error_message(error, r));
 356
 357                 m->unit = TAKE_PTR(scope);
 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         q = manager_unref_unit(m->manager, m->unit, &error);
 426         if (q < 0)
 427                 log_warning_errno(q, "Failed to drop reference to machine scope, ignoring: %s", bus_error_message(&error, r));
 428
 429         return r;
 430 }
 431
 432 int machine_stop(Machine *m) {
 433         int r;
 434         assert(m);
 435
 436         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 437                 return -EOPNOTSUPP;
 438
 439         r = machine_stop_scope(m);
 440
 441         m->stopping = true;
 442
 443         machine_save(m);
 444         (void) manager_enqueue_nscd_cache_flush(m->manager);
 445
 446         return r;
 447 }
 448
 449 int machine_finalize(Machine *m) {
 450         assert(m);
 451
 452         if (m->started)
 453                 log_struct(LOG_INFO,
 454                            "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
 455                            "NAME=%s", m->name,
 456                            "LEADER="PID_FMT, m->leader,
 457                            LOG_MESSAGE("Machine %s terminated.", m->name));
 458
 459         machine_unlink(m);
 460         machine_add_to_gc_queue(m);
 461
 462         if (m->started) {
 463                 machine_send_signal(m, false);
 464                 m->started = false;
 465         }
 466
 467         return 0;
 468 }
 469
 470 bool machine_may_gc(Machine *m, bool drop_not_started) {
 471         assert(m);
 472
 473         if (m->class == MACHINE_HOST)
 474                 return false;
 475
 476         if (drop_not_started && !m->started)
 477                 return true;
 478
 479         if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
 480                 return false;
 481
 482         if (m->unit && manager_unit_is_active(m->manager, m->unit))
 483                 return false;
 484
 485         return true;
 486 }
 487
 488 void machine_add_to_gc_queue(Machine *m) {
 489         assert(m);
 490
 491         if (m->in_gc_queue)
 492                 return;
 493
 494         LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
 495         m->in_gc_queue = true;
 496 }
 497
 498 MachineState machine_get_state(Machine *s) {
 499         assert(s);
 500
 501         if (s->class == MACHINE_HOST)
 502                 return MACHINE_RUNNING;
 503
 504         if (s->stopping)
 505                 return MACHINE_CLOSING;
 506
 507         if (s->scope_job)
 508                 return MACHINE_OPENING;
 509
 510         return MACHINE_RUNNING;
 511 }
 512
 513 int machine_kill(Machine *m, KillWho who, int signo) {
 514         assert(m);
 515
 516         if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
 517                 return -EOPNOTSUPP;
 518
 519         if (!m->unit)
 520                 return -ESRCH;
 521
 522         if (who == KILL_LEADER) {
 523                 /* If we shall simply kill the leader, do so directly */
 524
 525                 if (kill(m->leader, signo) < 0)
 526                         return -errno;
 527
 528                 return 0;
 529         }
 530
 531         /* Otherwise, make PID 1 do it for us, for the entire cgroup */
 532         return manager_kill_unit(m->manager, m->unit, signo, NULL);
 533 }
 534
 535 int machine_openpt(Machine *m, int flags) {
 536         assert(m);
 537
 538         switch (m->class) {
 539
 540         case MACHINE_HOST: {
 541                 int fd;
 542
 543                 fd = posix_openpt(flags);
 544                 if (fd < 0)
 545                         return -errno;
 546
 547                 if (unlockpt(fd) < 0)
 548                         return -errno;
 549
 550                 return fd;
 551         }
 552
 553         case MACHINE_CONTAINER:
 554                 if (m->leader <= 0)
 555                         return -EINVAL;
 556
 557                 return openpt_in_namespace(m->leader, flags);
 558
 559         default:
 560                 return -EOPNOTSUPP;
 561         }
 562 }
 563
 564 int machine_open_terminal(Machine *m, const char *path, int mode) {
 565         assert(m);
 566
 567         switch (m->class) {
 568
 569         case MACHINE_HOST:
 570                 return open_terminal(path, mode);
 571
 572         case MACHINE_CONTAINER:
 573                 if (m->leader <= 0)
 574                         return -EINVAL;
 575
 576                 return open_terminal_in_namespace(m->leader, path, mode);
 577
 578         default:
 579                 return -EOPNOTSUPP;
 580         }
 581 }
 582
 583 void machine_release_unit(Machine *m) {
 584         assert(m);
 585
 586         if (!m->unit)
 587                 return;
 588
 589         (void) hashmap_remove(m->manager->machine_units, m->unit);
 590         m->unit = mfree(m->unit);
 591 }
 592
 593 int machine_get_uid_shift(Machine *m, uid_t *ret) {
 594         char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
 595         uid_t uid_base, uid_shift, uid_range;
 596         gid_t gid_base, gid_shift, gid_range;
 597         _cleanup_fclose_ FILE *f = NULL;
 598         int k, r;
 599
 600         assert(m);
 601         assert(ret);
 602
 603         /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
 604          * mappings. In most cases setups should be simple like this, and administrators should only care about the
 605          * basic offset a container has relative to the host. This is what this function exposes.
 606          *
 607          * If we encounter any more complex mappings we politely refuse this with ENXIO. */
 608
 609         if (m->class == MACHINE_HOST) {
 610                 *ret = 0;
 611                 return 0;
 612         }
 613
 614         if (m->class != MACHINE_CONTAINER)
 615                 return -EOPNOTSUPP;
 616
 617         xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
 618         f = fopen(p, "re");
 619         if (!f) {
 620                 if (errno == ENOENT) {
 621                         /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
 622                         *ret = 0;
 623                         return 0;
 624                 }
 625
 626                 return -errno;
 627         }
 628
 629         /* Read the first line. There's at least one. */
 630         errno = 0;
 631         k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
 632         if (k != 3) {
 633                 if (ferror(f))
 634                         return -errno;
 635
 636                 return -EBADMSG;
 637         }
 638
 639         /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
 640         if (uid_base != 0)
 641                 return -ENXIO;
 642         /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
 643         if (uid_range < UID_NOBODY)
 644                 return -ENXIO;
 645
 646         /* If there's more than one line, then we don't support this mapping. */
 647         r = safe_fgetc(f, NULL);
 648         if (r < 0)
 649                 return r;
 650         if (r != 0) /* Insist on EOF */
 651                 return -ENXIO;
 652
 653         fclose(f);
 654
 655         xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
 656         f = fopen(p, "re");
 657         if (!f)
 658                 return -errno;
 659
 660         /* Read the first line. There's at least one. */
 661         errno = 0;
 662         k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
 663         if (k != 3) {
 664                 if (ferror(f))
 665                         return -errno;
 666
 667                 return -EBADMSG;
 668         }
 669
 670         /* If there's more than one line, then we don't support this file. */
 671         r = safe_fgetc(f, NULL);
 672         if (r < 0)
 673                 return r;
 674         if (r != 0) /* Insist on EOF */
 675                 return -ENXIO;
 676
 677         /* If the UID and GID mapping doesn't match, we don't support this mapping. */
 678         if (uid_base != (uid_t) gid_base)
 679                 return -ENXIO;
 680         if (uid_shift != (uid_t) gid_shift)
 681                 return -ENXIO;
 682         if (uid_range != (uid_t) gid_range)
 683                 return -ENXIO;
 684
 685         *ret = uid_shift;
 686         return 0;
 687 }
 688
 689 static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
 690         [MACHINE_CONTAINER] = "container",
 691         [MACHINE_VM] = "vm",
 692         [MACHINE_HOST] = "host",
 693 };
 694
 695 DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);
 696
 697 static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
 698         [MACHINE_OPENING] = "opening",
 699         [MACHINE_RUNNING] = "running",
 700         [MACHINE_CLOSING] = "closing"
 701 };
 702
 703 DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);
 704
 705 static const char* const kill_who_table[_KILL_WHO_MAX] = {
 706         [KILL_LEADER] = "leader",
 707         [KILL_ALL] = "all"
 708 };
 709
 710 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);