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