]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/machine/machine.c
8154c42ed774b8ea46fc1dfcb57711f61f1b9e9c
[thirdparty/systemd.git] / src / machine / machine.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <unistd.h>
5 #include <sys/stat.h>
6
7 #include "sd-messages.h"
8
9 #include "alloc-util.h"
10 #include "bus-error.h"
11 #include "bus-util.h"
12 #include "env-file.h"
13 #include "errno-util.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 "path-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 = path_join("/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 _cleanup_free_ int *ni = NULL;
298
299 p = netif;
300 for (;;) {
301 _cleanup_free_ char *word = NULL;
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 r = parse_ifindex(word);
314 if (r < 0)
315 continue;
316
317 if (!GREEDY_REALLOC(ni, allocated, nr + 1))
318 return log_oom();
319
320 ni[nr++] = r;
321 }
322
323 free(m->netif);
324 m->netif = TAKE_PTR(ni);
325 m->n_netif = nr;
326 }
327
328 return r;
329 }
330
331 static int machine_start_scope(
332 Machine *machine,
333 sd_bus_message *more_properties,
334 sd_bus_error *error) {
335
336 _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL, *reply = NULL;
337 _cleanup_free_ char *escaped = NULL, *unit = NULL;
338 const char *description;
339 int r;
340
341 assert(machine);
342 assert(machine->leader > 0);
343 assert(!machine->unit);
344
345 escaped = unit_name_escape(machine->name);
346 if (!escaped)
347 return log_oom();
348
349 unit = strjoin("machine-", escaped, ".scope");
350 if (!unit)
351 return log_oom();
352
353 r = sd_bus_message_new_method_call(
354 machine->manager->bus,
355 &m,
356 "org.freedesktop.systemd1",
357 "/org/freedesktop/systemd1",
358 "org.freedesktop.systemd1.Manager",
359 "StartTransientUnit");
360 if (r < 0)
361 return r;
362
363 r = sd_bus_message_append(m, "ss", unit, "fail");
364 if (r < 0)
365 return r;
366
367 r = sd_bus_message_open_container(m, 'a', "(sv)");
368 if (r < 0)
369 return r;
370
371 r = sd_bus_message_append(m, "(sv)", "Slice", "s", SPECIAL_MACHINE_SLICE);
372 if (r < 0)
373 return r;
374
375 description = strjoina(machine->class == MACHINE_VM ? "Virtual Machine " : "Container ", machine->name);
376 r = sd_bus_message_append(m, "(sv)", "Description", "s", description);
377 if (r < 0)
378 return r;
379
380 r = sd_bus_message_append(m, "(sv)(sv)(sv)(sv)(sv)",
381 "PIDs", "au", 1, machine->leader,
382 "Delegate", "b", 1,
383 "CollectMode", "s", "inactive-or-failed",
384 "AddRef", "b", 1,
385 "TasksMax", "t", UINT64_C(16384));
386 if (r < 0)
387 return r;
388
389 if (more_properties) {
390 r = sd_bus_message_copy(m, more_properties, true);
391 if (r < 0)
392 return r;
393 }
394
395 r = sd_bus_message_close_container(m);
396 if (r < 0)
397 return r;
398
399 r = sd_bus_message_append(m, "a(sa(sv))", 0);
400 if (r < 0)
401 return r;
402
403 r = sd_bus_call(NULL, m, 0, error, &reply);
404 if (r < 0)
405 return r;
406
407 machine->unit = TAKE_PTR(unit);
408 machine->referenced = true;
409
410 const char *job;
411 r = sd_bus_message_read(reply, "o", &job);
412 if (r < 0)
413 return r;
414
415 return free_and_strdup(&machine->scope_job, job);
416 }
417
418 static int machine_ensure_scope(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
419 int r;
420
421 assert(m);
422 assert(m->class != MACHINE_HOST);
423
424 if (!m->unit) {
425 r = machine_start_scope(m, properties, error);
426 if (r < 0)
427 return log_error_errno(r, "Failed to start machine scope: %s", bus_error_message(error, r));
428 }
429
430 assert(m->unit);
431 hashmap_put(m->manager->machine_units, m->unit, m);
432
433 return 0;
434 }
435
436 int machine_start(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
437 int r;
438
439 assert(m);
440
441 if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
442 return -EOPNOTSUPP;
443
444 if (m->started)
445 return 0;
446
447 r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
448 if (r < 0)
449 return r;
450
451 /* Create cgroup */
452 r = machine_ensure_scope(m, properties, error);
453 if (r < 0)
454 return r;
455
456 log_struct(LOG_INFO,
457 "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
458 "NAME=%s", m->name,
459 "LEADER="PID_FMT, m->leader,
460 LOG_MESSAGE("New machine %s.", m->name));
461
462 if (!dual_timestamp_is_set(&m->timestamp))
463 dual_timestamp_get(&m->timestamp);
464
465 m->started = true;
466
467 /* Save new machine data */
468 machine_save(m);
469
470 machine_send_signal(m, true);
471 (void) manager_enqueue_nscd_cache_flush(m->manager);
472
473 return 0;
474 }
475
476 int machine_stop(Machine *m) {
477 int r;
478
479 assert(m);
480
481 if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
482 return -EOPNOTSUPP;
483
484 if (m->unit) {
485 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
486 char *job = NULL;
487
488 r = manager_stop_unit(m->manager, m->unit, &error, &job);
489 if (r < 0)
490 return log_error_errno(r, "Failed to stop machine scope: %s", bus_error_message(&error, r));
491
492 free_and_replace(m->scope_job, job);
493 }
494
495 m->stopping = true;
496
497 machine_save(m);
498 (void) manager_enqueue_nscd_cache_flush(m->manager);
499
500 return 0;
501 }
502
503 int machine_finalize(Machine *m) {
504 assert(m);
505
506 if (m->started) {
507 log_struct(LOG_INFO,
508 "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
509 "NAME=%s", m->name,
510 "LEADER="PID_FMT, m->leader,
511 LOG_MESSAGE("Machine %s terminated.", m->name));
512
513 m->stopping = true; /* The machine is supposed to be going away. Don't try to kill it. */
514 }
515
516 machine_unlink(m);
517 machine_add_to_gc_queue(m);
518
519 if (m->started) {
520 machine_send_signal(m, false);
521 m->started = false;
522 }
523
524 return 0;
525 }
526
527 bool machine_may_gc(Machine *m, bool drop_not_started) {
528 assert(m);
529
530 if (m->class == MACHINE_HOST)
531 return false;
532
533 if (drop_not_started && !m->started)
534 return true;
535
536 if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
537 return false;
538
539 if (m->unit && manager_unit_is_active(m->manager, m->unit))
540 return false;
541
542 return true;
543 }
544
545 void machine_add_to_gc_queue(Machine *m) {
546 assert(m);
547
548 if (m->in_gc_queue)
549 return;
550
551 LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
552 m->in_gc_queue = true;
553 }
554
555 MachineState machine_get_state(Machine *s) {
556 assert(s);
557
558 if (s->class == MACHINE_HOST)
559 return MACHINE_RUNNING;
560
561 if (s->stopping)
562 return MACHINE_CLOSING;
563
564 if (s->scope_job)
565 return MACHINE_OPENING;
566
567 return MACHINE_RUNNING;
568 }
569
570 int machine_kill(Machine *m, KillWho who, int signo) {
571 assert(m);
572
573 if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
574 return -EOPNOTSUPP;
575
576 if (!m->unit)
577 return -ESRCH;
578
579 if (who == KILL_LEADER) {
580 /* If we shall simply kill the leader, do so directly */
581
582 if (kill(m->leader, signo) < 0)
583 return -errno;
584
585 return 0;
586 }
587
588 /* Otherwise, make PID 1 do it for us, for the entire cgroup */
589 return manager_kill_unit(m->manager, m->unit, signo, NULL);
590 }
591
592 int machine_openpt(Machine *m, int flags, char **ret_slave) {
593 assert(m);
594
595 switch (m->class) {
596
597 case MACHINE_HOST:
598
599 return openpt_allocate(flags, ret_slave);
600
601 case MACHINE_CONTAINER:
602 if (m->leader <= 0)
603 return -EINVAL;
604
605 return openpt_allocate_in_namespace(m->leader, flags, ret_slave);
606
607 default:
608 return -EOPNOTSUPP;
609 }
610 }
611
612 int machine_open_terminal(Machine *m, const char *path, int mode) {
613 assert(m);
614
615 switch (m->class) {
616
617 case MACHINE_HOST:
618 return open_terminal(path, mode);
619
620 case MACHINE_CONTAINER:
621 if (m->leader <= 0)
622 return -EINVAL;
623
624 return open_terminal_in_namespace(m->leader, path, mode);
625
626 default:
627 return -EOPNOTSUPP;
628 }
629 }
630
631 void machine_release_unit(Machine *m) {
632 assert(m);
633
634 if (!m->unit)
635 return;
636
637 if (m->referenced) {
638 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
639 int r;
640
641 r = manager_unref_unit(m->manager, m->unit, &error);
642 if (r < 0)
643 log_warning_errno(r, "Failed to drop reference to machine scope, ignoring: %s",
644 bus_error_message(&error, r));
645
646 m->referenced = false;
647 }
648
649 (void) hashmap_remove(m->manager->machine_units, m->unit);
650 m->unit = mfree(m->unit);
651 }
652
653 int machine_get_uid_shift(Machine *m, uid_t *ret) {
654 char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
655 uid_t uid_base, uid_shift, uid_range;
656 gid_t gid_base, gid_shift, gid_range;
657 _cleanup_fclose_ FILE *f = NULL;
658 int k, r;
659
660 assert(m);
661 assert(ret);
662
663 /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
664 * mappings. In most cases setups should be simple like this, and administrators should only care about the
665 * basic offset a container has relative to the host. This is what this function exposes.
666 *
667 * If we encounter any more complex mappings we politely refuse this with ENXIO. */
668
669 if (m->class == MACHINE_HOST) {
670 *ret = 0;
671 return 0;
672 }
673
674 if (m->class != MACHINE_CONTAINER)
675 return -EOPNOTSUPP;
676
677 xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
678 f = fopen(p, "re");
679 if (!f) {
680 if (errno == ENOENT) {
681 /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
682 *ret = 0;
683 return 0;
684 }
685
686 return -errno;
687 }
688
689 /* Read the first line. There's at least one. */
690 errno = 0;
691 k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
692 if (k != 3) {
693 if (ferror(f))
694 return errno_or_else(EIO);
695
696 return -EBADMSG;
697 }
698
699 /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
700 if (uid_base != 0)
701 return -ENXIO;
702 /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
703 if (uid_range < UID_NOBODY)
704 return -ENXIO;
705
706 /* If there's more than one line, then we don't support this mapping. */
707 r = safe_fgetc(f, NULL);
708 if (r < 0)
709 return r;
710 if (r != 0) /* Insist on EOF */
711 return -ENXIO;
712
713 fclose(f);
714
715 xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
716 f = fopen(p, "re");
717 if (!f)
718 return -errno;
719
720 /* Read the first line. There's at least one. */
721 errno = 0;
722 k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
723 if (k != 3) {
724 if (ferror(f))
725 return errno_or_else(EIO);
726
727 return -EBADMSG;
728 }
729
730 /* If there's more than one line, then we don't support this file. */
731 r = safe_fgetc(f, NULL);
732 if (r < 0)
733 return r;
734 if (r != 0) /* Insist on EOF */
735 return -ENXIO;
736
737 /* If the UID and GID mapping doesn't match, we don't support this mapping. */
738 if (uid_base != (uid_t) gid_base)
739 return -ENXIO;
740 if (uid_shift != (uid_t) gid_shift)
741 return -ENXIO;
742 if (uid_range != (uid_t) gid_range)
743 return -ENXIO;
744
745 *ret = uid_shift;
746 return 0;
747 }
748
749 static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
750 [MACHINE_CONTAINER] = "container",
751 [MACHINE_VM] = "vm",
752 [MACHINE_HOST] = "host",
753 };
754
755 DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);
756
757 static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
758 [MACHINE_OPENING] = "opening",
759 [MACHINE_RUNNING] = "running",
760 [MACHINE_CLOSING] = "closing"
761 };
762
763 DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);
764
765 static const char* const kill_who_table[_KILL_WHO_MAX] = {
766 [KILL_LEADER] = "leader",
767 [KILL_ALL] = "all"
768 };
769
770 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);