src/machine/machine.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2011 Lennart Poettering
   6 ***/
   7
   8 #include <errno.h>
   9 #include <string.h>
  10 #include <unistd.h>
  11 #include <stdio_ext.h>
  12
  13 #include "sd-messages.h"
  14
  15 #include "alloc-util.h"
  16 #include "bus-error.h"
  17 #include "bus-util.h"
  18 #include "escape.h"
  19 #include "extract-word.h"
  20 #include "fd-util.h"
  21 #include "fileio.h"
  22 #include "format-util.h"
  23 #include "hashmap.h"
  24 #include "machine-dbus.h"
  25 #include "machine.h"
  26 #include "mkdir.h"
  27 #include "parse-util.h"
  28 #include "process-util.h"
  29 #include "special.h"
  30 #include "stdio-util.h"
  31 #include "string-table.h"
  32 #include "terminal-util.h"
  33 #include "unit-name.h"
  34 #include "user-util.h"
  35 #include "util.h"
  36
  37 Machine* machine_new(Manager *manager, MachineClass class, const char *name) {
  38         Machine *m;
  39
  40         assert(manager);
  41         assert(class < _MACHINE_CLASS_MAX);
  42         assert(name);
  43
  44         /* Passing class == _MACHINE_CLASS_INVALID here is fine. It
  45          * means as much as "we don't know yet", and that we'll figure
  46          * it out later when loading the state file. */
  47
  48         m = new0(Machine, 1);
  49         if (!m)
  50                 return NULL;
  51
  52         m->name = strdup(name);
  53         if (!m->name)
  54                 goto fail;
  55
  56         if (class != MACHINE_HOST) {
  57                 m->state_file = strappend("/run/systemd/machines/", m->name);
  58                 if (!m->state_file)
  59                         goto fail;
  60         }
  61
  62         m->class = class;
  63
  64         if (hashmap_put(manager->machines, m->name, m) < 0)
  65                 goto fail;
  66
  67         m->manager = manager;
  68
  69         return m;
  70
  71 fail:
  72         free(m->state_file);
  73         free(m->name);
  74         return mfree(m);
  75 }
  76
  77 void machine_free(Machine *m) {
  78         assert(m);
  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         free(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                 unsigned 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
 240                 char *sl;
 241
 242                 sl = strjoina("/run/systemd/machines/unit:", m->unit);
 243                 (void) unlink(sl);
 244         }
 245
 246         if (m->state_file)
 247                 (void) unlink(m->state_file);
 248 }
 249
 250 int machine_load(Machine *m) {
 251         _cleanup_free_ char *realtime = NULL, *monotonic = NULL, *id = NULL, *leader = NULL, *class = NULL, *netif = NULL;
 252         int r;
 253
 254         assert(m);
 255
 256         if (!m->state_file)
 257                 return 0;
 258
 259         r = parse_env_file(m->state_file, NEWLINE,
 260                            "SCOPE",     &m->unit,
 261                            "SCOPE_JOB", &m->scope_job,
 262                            "SERVICE",   &m->service,
 263                            "ROOT",      &m->root_directory,
 264                            "ID",        &id,
 265                            "LEADER",    &leader,
 266                            "CLASS",     &class,
 267                            "REALTIME",  &realtime,
 268                            "MONOTONIC", &monotonic,
 269                            "NETIF",     &netif,
 270                            NULL);
 271         if (r < 0) {
 272                 if (r == -ENOENT)
 273                         return 0;
 274
 275                 return log_error_errno(r, "Failed to read %s: %m", m->state_file);
 276         }
 277
 278         if (id)
 279                 sd_id128_from_string(id, &m->id);
 280
 281         if (leader)
 282                 parse_pid(leader, &m->leader);
 283
 284         if (class) {
 285                 MachineClass c;
 286
 287                 c = machine_class_from_string(class);
 288                 if (c >= 0)
 289                         m->class = c;
 290         }
 291
 292         if (realtime)
 293                 timestamp_deserialize(realtime, &m->timestamp.realtime);
 294         if (monotonic)
 295                 timestamp_deserialize(monotonic, &m->timestamp.monotonic);
 296
 297         if (netif) {
 298                 size_t allocated = 0, nr = 0;
 299                 const char *p;
 300                 int *ni = NULL;
 301
 302                 p = netif;
 303                 for (;;) {
 304                         _cleanup_free_ char *word = NULL;
 305                         int ifi;
 306
 307                         r = extract_first_word(&p, &word, NULL, 0);
 308                         if (r == 0)
 309                                 break;
 310                         if (r == -ENOMEM)
 311                                 return log_oom();
 312                         if (r < 0) {
 313                                 log_warning_errno(r, "Failed to parse NETIF: %s", netif);
 314                                 break;
 315                         }
 316
 317                         if (parse_ifindex(word, &ifi) < 0)
 318                                 continue;
 319
 320                         if (!GREEDY_REALLOC(ni, allocated, nr+1)) {
 321                                 free(ni);
 322                                 return log_oom();
 323                         }
 324
 325                         ni[nr++] = ifi;
 326                 }
 327
 328                 free(m->netif);
 329                 m->netif = ni;
 330                 m->n_netif = nr;
 331         }
 332
 333         return r;
 334 }
 335
 336 static int machine_start_scope(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 337         int r = 0;
 338
 339         assert(m);
 340         assert(m->class != MACHINE_HOST);
 341
 342         if (!m->unit) {
 343                 _cleanup_free_ char *escaped = NULL;
 344                 char *scope, *description, *job = NULL;
 345
 346                 escaped = unit_name_escape(m->name);
 347                 if (!escaped)
 348                         return log_oom();
 349
 350                 scope = strjoin("machine-", escaped, ".scope");
 351                 if (!scope)
 352                         return log_oom();
 353
 354                 description = strjoina(m->class == MACHINE_VM ? "Virtual Machine " : "Container ", m->name);
 355
 356                 r = manager_start_scope(m->manager, scope, m->leader, SPECIAL_MACHINE_SLICE, description, properties, error, &job);
 357                 if (r < 0) {
 358                         log_error("Failed to start machine scope: %s", bus_error_message(error, r));
 359                         free(scope);
 360                         return r;
 361                 } else {
 362                         m->unit = scope;
 363
 364                         free(m->scope_job);
 365                         m->scope_job = job;
 366                 }
 367         }
 368
 369         if (m->unit)
 370                 hashmap_put(m->manager->machine_units, m->unit, m);
 371
 372         return r;
 373 }
 374
 375 int machine_start(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 376         int r;
 377
 378         assert(m);
 379
 380         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 381                 return -EOPNOTSUPP;
 382
 383         if (m->started)
 384                 return 0;
 385
 386         r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
 387         if (r < 0)
 388                 return r;
 389
 390         /* Create cgroup */
 391         r = machine_start_scope(m, properties, error);
 392         if (r < 0)
 393                 return r;
 394
 395         log_struct(LOG_INFO,
 396                    "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
 397                    "NAME=%s", m->name,
 398                    "LEADER="PID_FMT, m->leader,
 399                    LOG_MESSAGE("New machine %s.", m->name),
 400                    NULL);
 401
 402         if (!dual_timestamp_is_set(&m->timestamp))
 403                 dual_timestamp_get(&m->timestamp);
 404
 405         m->started = true;
 406
 407         /* Save new machine data */
 408         machine_save(m);
 409
 410         machine_send_signal(m, true);
 411
 412         return 0;
 413 }
 414
 415 static int machine_stop_scope(Machine *m) {
 416         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
 417         char *job = NULL;
 418         int r;
 419
 420         assert(m);
 421         assert(m->class != MACHINE_HOST);
 422
 423         if (!m->unit)
 424                 return 0;
 425
 426         r = manager_stop_unit(m->manager, m->unit, &error, &job);
 427         if (r < 0) {
 428                 log_error("Failed to stop machine scope: %s", bus_error_message(&error, r));
 429                 return r;
 430         }
 431
 432         free(m->scope_job);
 433         m->scope_job = job;
 434
 435         return 0;
 436 }
 437
 438 int machine_stop(Machine *m) {
 439         int r;
 440         assert(m);
 441
 442         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 443                 return -EOPNOTSUPP;
 444
 445         r = machine_stop_scope(m);
 446
 447         m->stopping = true;
 448
 449         machine_save(m);
 450
 451         return r;
 452 }
 453
 454 int machine_finalize(Machine *m) {
 455         assert(m);
 456
 457         if (m->started)
 458                 log_struct(LOG_INFO,
 459                            "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
 460                            "NAME=%s", m->name,
 461                            "LEADER="PID_FMT, m->leader,
 462                            LOG_MESSAGE("Machine %s terminated.", m->name),
 463                            NULL);
 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) {
 542         assert(m);
 543
 544         switch (m->class) {
 545
 546         case MACHINE_HOST: {
 547                 int fd;
 548
 549                 fd = posix_openpt(flags);
 550                 if (fd < 0)
 551                         return -errno;
 552
 553                 if (unlockpt(fd) < 0)
 554                         return -errno;
 555
 556                 return fd;
 557         }
 558
 559         case MACHINE_CONTAINER:
 560                 if (m->leader <= 0)
 561                         return -EINVAL;
 562
 563                 return openpt_in_namespace(m->leader, flags);
 564
 565         default:
 566                 return -EOPNOTSUPP;
 567         }
 568 }
 569
 570 int machine_open_terminal(Machine *m, const char *path, int mode) {
 571         assert(m);
 572
 573         switch (m->class) {
 574
 575         case MACHINE_HOST:
 576                 return open_terminal(path, mode);
 577
 578         case MACHINE_CONTAINER:
 579                 if (m->leader <= 0)
 580                         return -EINVAL;
 581
 582                 return open_terminal_in_namespace(m->leader, path, mode);
 583
 584         default:
 585                 return -EOPNOTSUPP;
 586         }
 587 }
 588
 589 void machine_release_unit(Machine *m) {
 590         assert(m);
 591
 592         if (!m->unit)
 593                 return;
 594
 595         (void) hashmap_remove(m->manager->machine_units, m->unit);
 596         m->unit = mfree(m->unit);
 597 }
 598
 599 int machine_get_uid_shift(Machine *m, uid_t *ret) {
 600         char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
 601         uid_t uid_base, uid_shift, uid_range;
 602         gid_t gid_base, gid_shift, gid_range;
 603         _cleanup_fclose_ FILE *f = NULL;
 604         int k;
 605
 606         assert(m);
 607         assert(ret);
 608
 609         /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
 610          * mappings. In most cases setups should be simple like this, and administrators should only care about the
 611          * basic offset a container has relative to the host. This is what this function exposes.
 612          *
 613          * If we encounter any more complex mappings we politely refuse this with ENXIO. */
 614
 615         if (m->class == MACHINE_HOST) {
 616                 *ret = 0;
 617                 return 0;
 618         }
 619
 620         if (m->class != MACHINE_CONTAINER)
 621                 return -EOPNOTSUPP;
 622
 623         xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
 624         f = fopen(p, "re");
 625         if (!f) {
 626                 if (errno == ENOENT) {
 627                         /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
 628                         *ret = 0;
 629                         return 0;
 630                 }
 631
 632                 return -errno;
 633         }
 634
 635         /* Read the first line. There's at least one. */
 636         errno = 0;
 637         k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
 638         if (k != 3) {
 639                 if (ferror(f))
 640                         return -errno;
 641
 642                 return -EBADMSG;
 643         }
 644
 645         /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
 646         if (uid_base != 0)
 647                 return -ENXIO;
 648         /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
 649         if (uid_range < UID_NOBODY)
 650                 return -ENXIO;
 651
 652         /* If there's more than one line, then we don't support this mapping. */
 653         if (fgetc(f) != EOF)
 654                 return -ENXIO;
 655
 656         fclose(f);
 657
 658         xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
 659         f = fopen(p, "re");
 660         if (!f)
 661                 return -errno;
 662
 663         /* Read the first line. There's at least one. */
 664         errno = 0;
 665         k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
 666         if (k != 3) {
 667                 if (ferror(f))
 668                         return -errno;
 669
 670                 return -EBADMSG;
 671         }
 672
 673         /* If there's more than one line, then we don't support this file. */
 674         if (fgetc(f) != 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);