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 <sys/stat.h>
   7
   8 #include "sd-messages.h"
   9
  10 #include "alloc-util.h"
  11 #include "bus-error.h"
  12 #include "bus-util.h"
  13 #include "env-file.h"
  14 #include "errno-util.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) 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                 free_and_replace(m->scope_job, job);
 358         }
 359
 360         if (m->unit)
 361                 hashmap_put(m->manager->machine_units, m->unit, m);
 362
 363         return 0;
 364 }
 365
 366 int machine_start(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 367         int r;
 368
 369         assert(m);
 370
 371         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 372                 return -EOPNOTSUPP;
 373
 374         if (m->started)
 375                 return 0;
 376
 377         r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
 378         if (r < 0)
 379                 return r;
 380
 381         /* Create cgroup */
 382         r = machine_start_scope(m, properties, error);
 383         if (r < 0)
 384                 return r;
 385
 386         log_struct(LOG_INFO,
 387                    "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
 388                    "NAME=%s", m->name,
 389                    "LEADER="PID_FMT, m->leader,
 390                    LOG_MESSAGE("New machine %s.", m->name));
 391
 392         if (!dual_timestamp_is_set(&m->timestamp))
 393                 dual_timestamp_get(&m->timestamp);
 394
 395         m->started = true;
 396
 397         /* Save new machine data */
 398         machine_save(m);
 399
 400         machine_send_signal(m, true);
 401         (void) manager_enqueue_nscd_cache_flush(m->manager);
 402
 403         return 0;
 404 }
 405
 406 static int machine_stop_scope(Machine *m) {
 407         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
 408         char *job = NULL;
 409         int r, q;
 410
 411         assert(m);
 412         assert(m->class != MACHINE_HOST);
 413
 414         if (!m->unit)
 415                 return 0;
 416
 417         r = manager_stop_unit(m->manager, m->unit, &error, &job);
 418         if (r < 0) {
 419                 log_error_errno(r, "Failed to stop machine scope: %s", bus_error_message(&error, r));
 420                 sd_bus_error_free(&error);
 421         } else
 422                 free_and_replace(m->scope_job, job);
 423
 424         q = manager_unref_unit(m->manager, m->unit, &error);
 425         if (q < 0)
 426                 log_warning_errno(q, "Failed to drop reference to machine scope, ignoring: %s", bus_error_message(&error, r));
 427
 428         return r;
 429 }
 430
 431 int machine_stop(Machine *m) {
 432         int r;
 433         assert(m);
 434
 435         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 436                 return -EOPNOTSUPP;
 437
 438         r = machine_stop_scope(m);
 439
 440         m->stopping = true;
 441
 442         machine_save(m);
 443         (void) manager_enqueue_nscd_cache_flush(m->manager);
 444
 445         return r;
 446 }
 447
 448 int machine_finalize(Machine *m) {
 449         assert(m);
 450
 451         if (m->started)
 452                 log_struct(LOG_INFO,
 453                            "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
 454                            "NAME=%s", m->name,
 455                            "LEADER="PID_FMT, m->leader,
 456                            LOG_MESSAGE("Machine %s terminated.", m->name));
 457
 458         machine_unlink(m);
 459         machine_add_to_gc_queue(m);
 460
 461         if (m->started) {
 462                 machine_send_signal(m, false);
 463                 m->started = false;
 464         }
 465
 466         return 0;
 467 }
 468
 469 bool machine_may_gc(Machine *m, bool drop_not_started) {
 470         assert(m);
 471
 472         if (m->class == MACHINE_HOST)
 473                 return false;
 474
 475         if (drop_not_started && !m->started)
 476                 return true;
 477
 478         if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
 479                 return false;
 480
 481         if (m->unit && manager_unit_is_active(m->manager, m->unit))
 482                 return false;
 483
 484         return true;
 485 }
 486
 487 void machine_add_to_gc_queue(Machine *m) {
 488         assert(m);
 489
 490         if (m->in_gc_queue)
 491                 return;
 492
 493         LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
 494         m->in_gc_queue = true;
 495 }
 496
 497 MachineState machine_get_state(Machine *s) {
 498         assert(s);
 499
 500         if (s->class == MACHINE_HOST)
 501                 return MACHINE_RUNNING;
 502
 503         if (s->stopping)
 504                 return MACHINE_CLOSING;
 505
 506         if (s->scope_job)
 507                 return MACHINE_OPENING;
 508
 509         return MACHINE_RUNNING;
 510 }
 511
 512 int machine_kill(Machine *m, KillWho who, int signo) {
 513         assert(m);
 514
 515         if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
 516                 return -EOPNOTSUPP;
 517
 518         if (!m->unit)
 519                 return -ESRCH;
 520
 521         if (who == KILL_LEADER) {
 522                 /* If we shall simply kill the leader, do so directly */
 523
 524                 if (kill(m->leader, signo) < 0)
 525                         return -errno;
 526
 527                 return 0;
 528         }
 529
 530         /* Otherwise, make PID 1 do it for us, for the entire cgroup */
 531         return manager_kill_unit(m->manager, m->unit, signo, NULL);
 532 }
 533
 534 int machine_openpt(Machine *m, int flags, char **ret_slave) {
 535         assert(m);
 536
 537         switch (m->class) {
 538
 539         case MACHINE_HOST:
 540
 541                 return openpt_allocate(flags, ret_slave);
 542
 543         case MACHINE_CONTAINER:
 544                 if (m->leader <= 0)
 545                         return -EINVAL;
 546
 547                 return openpt_allocate_in_namespace(m->leader, flags, ret_slave);
 548
 549         default:
 550                 return -EOPNOTSUPP;
 551         }
 552 }
 553
 554 int machine_open_terminal(Machine *m, const char *path, int mode) {
 555         assert(m);
 556
 557         switch (m->class) {
 558
 559         case MACHINE_HOST:
 560                 return open_terminal(path, mode);
 561
 562         case MACHINE_CONTAINER:
 563                 if (m->leader <= 0)
 564                         return -EINVAL;
 565
 566                 return open_terminal_in_namespace(m->leader, path, mode);
 567
 568         default:
 569                 return -EOPNOTSUPP;
 570         }
 571 }
 572
 573 void machine_release_unit(Machine *m) {
 574         assert(m);
 575
 576         if (!m->unit)
 577                 return;
 578
 579         (void) hashmap_remove(m->manager->machine_units, m->unit);
 580         m->unit = mfree(m->unit);
 581 }
 582
 583 int machine_get_uid_shift(Machine *m, uid_t *ret) {
 584         char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
 585         uid_t uid_base, uid_shift, uid_range;
 586         gid_t gid_base, gid_shift, gid_range;
 587         _cleanup_fclose_ FILE *f = NULL;
 588         int k, r;
 589
 590         assert(m);
 591         assert(ret);
 592
 593         /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
 594          * mappings. In most cases setups should be simple like this, and administrators should only care about the
 595          * basic offset a container has relative to the host. This is what this function exposes.
 596          *
 597          * If we encounter any more complex mappings we politely refuse this with ENXIO. */
 598
 599         if (m->class == MACHINE_HOST) {
 600                 *ret = 0;
 601                 return 0;
 602         }
 603
 604         if (m->class != MACHINE_CONTAINER)
 605                 return -EOPNOTSUPP;
 606
 607         xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
 608         f = fopen(p, "re");
 609         if (!f) {
 610                 if (errno == ENOENT) {
 611                         /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
 612                         *ret = 0;
 613                         return 0;
 614                 }
 615
 616                 return -errno;
 617         }
 618
 619         /* Read the first line. There's at least one. */
 620         errno = 0;
 621         k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
 622         if (k != 3) {
 623                 if (ferror(f))
 624                         return errno_or_else(EIO);
 625
 626                 return -EBADMSG;
 627         }
 628
 629         /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
 630         if (uid_base != 0)
 631                 return -ENXIO;
 632         /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
 633         if (uid_range < UID_NOBODY)
 634                 return -ENXIO;
 635
 636         /* If there's more than one line, then we don't support this mapping. */
 637         r = safe_fgetc(f, NULL);
 638         if (r < 0)
 639                 return r;
 640         if (r != 0) /* Insist on EOF */
 641                 return -ENXIO;
 642
 643         fclose(f);
 644
 645         xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
 646         f = fopen(p, "re");
 647         if (!f)
 648                 return -errno;
 649
 650         /* Read the first line. There's at least one. */
 651         errno = 0;
 652         k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
 653         if (k != 3) {
 654                 if (ferror(f))
 655                         return errno_or_else(EIO);
 656
 657                 return -EBADMSG;
 658         }
 659
 660         /* If there's more than one line, then we don't support this file. */
 661         r = safe_fgetc(f, NULL);
 662         if (r < 0)
 663                 return r;
 664         if (r != 0) /* Insist on EOF */
 665                 return -ENXIO;
 666
 667         /* If the UID and GID mapping doesn't match, we don't support this mapping. */
 668         if (uid_base != (uid_t) gid_base)
 669                 return -ENXIO;
 670         if (uid_shift != (uid_t) gid_shift)
 671                 return -ENXIO;
 672         if (uid_range != (uid_t) gid_range)
 673                 return -ENXIO;
 674
 675         *ret = uid_shift;
 676         return 0;
 677 }
 678
 679 static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
 680         [MACHINE_CONTAINER] = "container",
 681         [MACHINE_VM] = "vm",
 682         [MACHINE_HOST] = "host",
 683 };
 684
 685 DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);
 686
 687 static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
 688         [MACHINE_OPENING] = "opening",
 689         [MACHINE_RUNNING] = "running",
 690         [MACHINE_CLOSING] = "closing"
 691 };
 692
 693 DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);
 694
 695 static const char* const kill_who_table[_KILL_WHO_MAX] = {
 696         [KILL_LEADER] = "leader",
 697         [KILL_ALL] = "all"
 698 };
 699
 700 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);