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         int r = 0;
 332
 333         assert(m);
 334         assert(m->class != MACHINE_HOST);
 335
 336         if (!m->unit) {
 337                 _cleanup_free_ char *escaped = NULL;
 338                 char *scope, *description, *job = NULL;
 339
 340                 escaped = unit_name_escape(m->name);
 341                 if (!escaped)
 342                         return log_oom();
 343
 344                 scope = strjoin("machine-", escaped, ".scope");
 345                 if (!scope)
 346                         return log_oom();
 347
 348                 description = strjoina(m->class == MACHINE_VM ? "Virtual Machine " : "Container ", m->name);
 349
 350                 r = manager_start_scope(m->manager, scope, m->leader, SPECIAL_MACHINE_SLICE, description, properties, error, &job);
 351                 if (r < 0) {
 352                         log_error("Failed to start machine scope: %s", bus_error_message(error, r));
 353                         free(scope);
 354                         return r;
 355                 } else {
 356                         m->unit = scope;
 357
 358                         free(m->scope_job);
 359                         m->scope_job = job;
 360                 }
 361         }
 362
 363         if (m->unit)
 364                 hashmap_put(m->manager->machine_units, m->unit, m);
 365
 366         return r;
 367 }
 368
 369 int machine_start(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
 370         int r;
 371
 372         assert(m);
 373
 374         if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
 375                 return -EOPNOTSUPP;
 376
 377         if (m->started)
 378                 return 0;
 379
 380         r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
 381         if (r < 0)
 382                 return r;
 383
 384         /* Create cgroup */
 385         r = machine_start_scope(m, properties, error);
 386         if (r < 0)
 387                 return r;
 388
 389         log_struct(LOG_INFO,
 390                    "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
 391                    "NAME=%s", m->name,
 392                    "LEADER="PID_FMT, m->leader,
 393                    LOG_MESSAGE("New machine %s.", m->name));
 394
 395         if (!dual_timestamp_is_set(&m->timestamp))
 396                 dual_timestamp_get(&m->timestamp);
 397
 398         m->started = true;
 399
 400         /* Save new machine data */
 401         machine_save(m);
 402
 403         machine_send_signal(m, true);
 404
 405         return 0;
 406 }
 407
 408 static int machine_stop_scope(Machine *m) {
 409         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
 410         char *job = NULL;
 411         int r;
 412
 413         assert(m);
 414         assert(m->class != MACHINE_HOST);
 415
 416         if (!m->unit)
 417                 return 0;
 418
 419         r = manager_stop_unit(m->manager, m->unit, &error, &job);
 420         if (r < 0) {
 421                 log_error("Failed to stop machine scope: %s", bus_error_message(&error, r));
 422                 return r;
 423         }
 424
 425         free(m->scope_job);
 426         m->scope_job = job;
 427
 428         return 0;
 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
 444         return r;
 445 }
 446
 447 int machine_finalize(Machine *m) {
 448         assert(m);
 449
 450         if (m->started)
 451                 log_struct(LOG_INFO,
 452                            "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
 453                            "NAME=%s", m->name,
 454                            "LEADER="PID_FMT, m->leader,
 455                            LOG_MESSAGE("Machine %s terminated.", m->name));
 456
 457         machine_unlink(m);
 458         machine_add_to_gc_queue(m);
 459
 460         if (m->started) {
 461                 machine_send_signal(m, false);
 462                 m->started = false;
 463         }
 464
 465         return 0;
 466 }
 467
 468 bool machine_may_gc(Machine *m, bool drop_not_started) {
 469         assert(m);
 470
 471         if (m->class == MACHINE_HOST)
 472                 return false;
 473
 474         if (drop_not_started && !m->started)
 475                 return true;
 476
 477         if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
 478                 return false;
 479
 480         if (m->unit && manager_unit_is_active(m->manager, m->unit))
 481                 return false;
 482
 483         return true;
 484 }
 485
 486 void machine_add_to_gc_queue(Machine *m) {
 487         assert(m);
 488
 489         if (m->in_gc_queue)
 490                 return;
 491
 492         LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
 493         m->in_gc_queue = true;
 494 }
 495
 496 MachineState machine_get_state(Machine *s) {
 497         assert(s);
 498
 499         if (s->class == MACHINE_HOST)
 500                 return MACHINE_RUNNING;
 501
 502         if (s->stopping)
 503                 return MACHINE_CLOSING;
 504
 505         if (s->scope_job)
 506                 return MACHINE_OPENING;
 507
 508         return MACHINE_RUNNING;
 509 }
 510
 511 int machine_kill(Machine *m, KillWho who, int signo) {
 512         assert(m);
 513
 514         if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
 515                 return -EOPNOTSUPP;
 516
 517         if (!m->unit)
 518                 return -ESRCH;
 519
 520         if (who == KILL_LEADER) {
 521                 /* If we shall simply kill the leader, do so directly */
 522
 523                 if (kill(m->leader, signo) < 0)
 524                         return -errno;
 525
 526                 return 0;
 527         }
 528
 529         /* Otherwise, make PID 1 do it for us, for the entire cgroup */
 530         return manager_kill_unit(m->manager, m->unit, signo, NULL);
 531 }
 532
 533 int machine_openpt(Machine *m, int flags) {
 534         assert(m);
 535
 536         switch (m->class) {
 537
 538         case MACHINE_HOST: {
 539                 int fd;
 540
 541                 fd = posix_openpt(flags);
 542                 if (fd < 0)
 543                         return -errno;
 544
 545                 if (unlockpt(fd) < 0)
 546                         return -errno;
 547
 548                 return fd;
 549         }
 550
 551         case MACHINE_CONTAINER:
 552                 if (m->leader <= 0)
 553                         return -EINVAL;
 554
 555                 return openpt_in_namespace(m->leader, flags);
 556
 557         default:
 558                 return -EOPNOTSUPP;
 559         }
 560 }
 561
 562 int machine_open_terminal(Machine *m, const char *path, int mode) {
 563         assert(m);
 564
 565         switch (m->class) {
 566
 567         case MACHINE_HOST:
 568                 return open_terminal(path, mode);
 569
 570         case MACHINE_CONTAINER:
 571                 if (m->leader <= 0)
 572                         return -EINVAL;
 573
 574                 return open_terminal_in_namespace(m->leader, path, mode);
 575
 576         default:
 577                 return -EOPNOTSUPP;
 578         }
 579 }
 580
 581 void machine_release_unit(Machine *m) {
 582         assert(m);
 583
 584         if (!m->unit)
 585                 return;
 586
 587         (void) hashmap_remove(m->manager->machine_units, m->unit);
 588         m->unit = mfree(m->unit);
 589 }
 590
 591 int machine_get_uid_shift(Machine *m, uid_t *ret) {
 592         char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
 593         uid_t uid_base, uid_shift, uid_range;
 594         gid_t gid_base, gid_shift, gid_range;
 595         _cleanup_fclose_ FILE *f = NULL;
 596         int k;
 597
 598         assert(m);
 599         assert(ret);
 600
 601         /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
 602          * mappings. In most cases setups should be simple like this, and administrators should only care about the
 603          * basic offset a container has relative to the host. This is what this function exposes.
 604          *
 605          * If we encounter any more complex mappings we politely refuse this with ENXIO. */
 606
 607         if (m->class == MACHINE_HOST) {
 608                 *ret = 0;
 609                 return 0;
 610         }
 611
 612         if (m->class != MACHINE_CONTAINER)
 613                 return -EOPNOTSUPP;
 614
 615         xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
 616         f = fopen(p, "re");
 617         if (!f) {
 618                 if (errno == ENOENT) {
 619                         /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
 620                         *ret = 0;
 621                         return 0;
 622                 }
 623
 624                 return -errno;
 625         }
 626
 627         /* Read the first line. There's at least one. */
 628         errno = 0;
 629         k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
 630         if (k != 3) {
 631                 if (ferror(f))
 632                         return -errno;
 633
 634                 return -EBADMSG;
 635         }
 636
 637         /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
 638         if (uid_base != 0)
 639                 return -ENXIO;
 640         /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
 641         if (uid_range < UID_NOBODY)
 642                 return -ENXIO;
 643
 644         /* If there's more than one line, then we don't support this mapping. */
 645         if (fgetc(f) != EOF)
 646                 return -ENXIO;
 647
 648         fclose(f);
 649
 650         xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
 651         f = fopen(p, "re");
 652         if (!f)
 653                 return -errno;
 654
 655         /* Read the first line. There's at least one. */
 656         errno = 0;
 657         k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
 658         if (k != 3) {
 659                 if (ferror(f))
 660                         return -errno;
 661
 662                 return -EBADMSG;
 663         }
 664
 665         /* If there's more than one line, then we don't support this file. */
 666         if (fgetc(f) != EOF)
 667                 return -ENXIO;
 668
 669         /* If the UID and GID mapping doesn't match, we don't support this mapping. */
 670         if (uid_base != (uid_t) gid_base)
 671                 return -ENXIO;
 672         if (uid_shift != (uid_t) gid_shift)
 673                 return -ENXIO;
 674         if (uid_range != (uid_t) gid_range)
 675                 return -ENXIO;
 676
 677         *ret = uid_shift;
 678         return 0;
 679 }
 680
 681 static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
 682         [MACHINE_CONTAINER] = "container",
 683         [MACHINE_VM] = "vm",
 684         [MACHINE_HOST] = "host",
 685 };
 686
 687 DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);
 688
 689 static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
 690         [MACHINE_OPENING] = "opening",
 691         [MACHINE_RUNNING] = "running",
 692         [MACHINE_CLOSING] = "closing"
 693 };
 694
 695 DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);
 696
 697 static const char* const kill_who_table[_KILL_WHO_MAX] = {
 698         [KILL_LEADER] = "leader",
 699         [KILL_ALL] = "all"
 700 };
 701
 702 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);