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