]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
nspawn: split out mount related functions into a new nspawn-mount.c file
[thirdparty/systemd.git] / src / nspawn / nspawn.c
1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
2
3 /***
4 This file is part of systemd.
5
6 Copyright 2010 Lennart Poettering
7
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
12
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
17
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
20 ***/
21
22 #include <signal.h>
23 #include <sched.h>
24 #include <unistd.h>
25 #include <sys/types.h>
26 #include <sys/mount.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <stdio.h>
30 #include <errno.h>
31 #include <sys/prctl.h>
32 #include <getopt.h>
33 #include <grp.h>
34 #include <linux/fs.h>
35 #include <sys/socket.h>
36 #include <linux/netlink.h>
37 #include <net/if.h>
38 #include <linux/veth.h>
39 #include <sys/personality.h>
40 #include <linux/loop.h>
41 #include <sys/file.h>
42
43 #ifdef HAVE_SELINUX
44 #include <selinux/selinux.h>
45 #endif
46
47 #ifdef HAVE_SECCOMP
48 #include <seccomp.h>
49 #endif
50
51 #ifdef HAVE_BLKID
52 #include <blkid/blkid.h>
53 #endif
54
55 #include "sd-daemon.h"
56 #include "sd-bus.h"
57 #include "sd-id128.h"
58 #include "sd-netlink.h"
59 #include "random-util.h"
60 #include "log.h"
61 #include "util.h"
62 #include "mkdir.h"
63 #include "rm-rf.h"
64 #include "macro.h"
65 #include "missing.h"
66 #include "cgroup-util.h"
67 #include "strv.h"
68 #include "path-util.h"
69 #include "loopback-setup.h"
70 #include "dev-setup.h"
71 #include "fdset.h"
72 #include "build.h"
73 #include "fileio.h"
74 #include "bus-util.h"
75 #include "bus-error.h"
76 #include "ptyfwd.h"
77 #include "env-util.h"
78 #include "netlink-util.h"
79 #include "udev-util.h"
80 #include "blkid-util.h"
81 #include "gpt.h"
82 #include "siphash24.h"
83 #include "copy.h"
84 #include "base-filesystem.h"
85 #include "barrier.h"
86 #include "event-util.h"
87 #include "capability.h"
88 #include "cap-list.h"
89 #include "btrfs-util.h"
90 #include "machine-image.h"
91 #include "list.h"
92 #include "in-addr-util.h"
93 #include "firewall-util.h"
94 #include "local-addresses.h"
95 #include "formats-util.h"
96 #include "process-util.h"
97 #include "terminal-util.h"
98 #include "hostname-util.h"
99 #include "signal-util.h"
100
101 #ifdef HAVE_SECCOMP
102 #include "seccomp-util.h"
103 #endif
104
105 #include "nspawn.h"
106 #include "nspawn-settings.h"
107 #include "nspawn-mount.h"
108
109 typedef enum ContainerStatus {
110 CONTAINER_TERMINATED,
111 CONTAINER_REBOOTED
112 } ContainerStatus;
113
114 typedef enum LinkJournal {
115 LINK_NO,
116 LINK_AUTO,
117 LINK_HOST,
118 LINK_GUEST
119 } LinkJournal;
120
121 static char *arg_directory = NULL;
122 static char *arg_template = NULL;
123 static char *arg_user = NULL;
124 static sd_id128_t arg_uuid = {};
125 static char *arg_machine = NULL;
126 static const char *arg_selinux_context = NULL;
127 static const char *arg_selinux_apifs_context = NULL;
128 static const char *arg_slice = NULL;
129 static bool arg_private_network = false;
130 static bool arg_read_only = false;
131 static bool arg_boot = false;
132 static bool arg_ephemeral = false;
133 static LinkJournal arg_link_journal = LINK_AUTO;
134 static bool arg_link_journal_try = false;
135 static uint64_t arg_retain =
136 (1ULL << CAP_CHOWN) |
137 (1ULL << CAP_DAC_OVERRIDE) |
138 (1ULL << CAP_DAC_READ_SEARCH) |
139 (1ULL << CAP_FOWNER) |
140 (1ULL << CAP_FSETID) |
141 (1ULL << CAP_IPC_OWNER) |
142 (1ULL << CAP_KILL) |
143 (1ULL << CAP_LEASE) |
144 (1ULL << CAP_LINUX_IMMUTABLE) |
145 (1ULL << CAP_NET_BIND_SERVICE) |
146 (1ULL << CAP_NET_BROADCAST) |
147 (1ULL << CAP_NET_RAW) |
148 (1ULL << CAP_SETGID) |
149 (1ULL << CAP_SETFCAP) |
150 (1ULL << CAP_SETPCAP) |
151 (1ULL << CAP_SETUID) |
152 (1ULL << CAP_SYS_ADMIN) |
153 (1ULL << CAP_SYS_CHROOT) |
154 (1ULL << CAP_SYS_NICE) |
155 (1ULL << CAP_SYS_PTRACE) |
156 (1ULL << CAP_SYS_TTY_CONFIG) |
157 (1ULL << CAP_SYS_RESOURCE) |
158 (1ULL << CAP_SYS_BOOT) |
159 (1ULL << CAP_AUDIT_WRITE) |
160 (1ULL << CAP_AUDIT_CONTROL) |
161 (1ULL << CAP_MKNOD);
162 static CustomMount *arg_custom_mounts = NULL;
163 static unsigned arg_n_custom_mounts = 0;
164 static char **arg_setenv = NULL;
165 static bool arg_quiet = false;
166 static bool arg_share_system = false;
167 static bool arg_register = true;
168 static bool arg_keep_unit = false;
169 static char **arg_network_interfaces = NULL;
170 static char **arg_network_macvlan = NULL;
171 static char **arg_network_ipvlan = NULL;
172 static bool arg_network_veth = false;
173 static char *arg_network_bridge = NULL;
174 static unsigned long arg_personality = PERSONALITY_INVALID;
175 static char *arg_image = NULL;
176 static VolatileMode arg_volatile_mode = VOLATILE_NO;
177 static ExposePort *arg_expose_ports = NULL;
178 static char **arg_property = NULL;
179 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
180 static bool arg_userns = false;
181 static int arg_kill_signal = 0;
182 static bool arg_unified_cgroup_hierarchy = false;
183 static SettingsMask arg_settings_mask = 0;
184 static int arg_settings_trusted = -1;
185 static char **arg_parameters = NULL;
186
187 static void help(void) {
188 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
189 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
190 " -h --help Show this help\n"
191 " --version Print version string\n"
192 " -q --quiet Do not show status information\n"
193 " -D --directory=PATH Root directory for the container\n"
194 " --template=PATH Initialize root directory from template directory,\n"
195 " if missing\n"
196 " -x --ephemeral Run container with snapshot of root directory, and\n"
197 " remove it after exit\n"
198 " -i --image=PATH File system device or disk image for the container\n"
199 " -b --boot Boot up full system (i.e. invoke init)\n"
200 " -u --user=USER Run the command under specified user or uid\n"
201 " -M --machine=NAME Set the machine name for the container\n"
202 " --uuid=UUID Set a specific machine UUID for the container\n"
203 " -S --slice=SLICE Place the container in the specified slice\n"
204 " --property=NAME=VALUE Set scope unit property\n"
205 " --private-users[=UIDBASE[:NUIDS]]\n"
206 " Run within user namespace\n"
207 " --private-network Disable network in container\n"
208 " --network-interface=INTERFACE\n"
209 " Assign an existing network interface to the\n"
210 " container\n"
211 " --network-macvlan=INTERFACE\n"
212 " Create a macvlan network interface based on an\n"
213 " existing network interface to the container\n"
214 " --network-ipvlan=INTERFACE\n"
215 " Create a ipvlan network interface based on an\n"
216 " existing network interface to the container\n"
217 " -n --network-veth Add a virtual ethernet connection between host\n"
218 " and container\n"
219 " --network-bridge=INTERFACE\n"
220 " Add a virtual ethernet connection between host\n"
221 " and container and add it to an existing bridge on\n"
222 " the host\n"
223 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
224 " Expose a container IP port on the host\n"
225 " -Z --selinux-context=SECLABEL\n"
226 " Set the SELinux security context to be used by\n"
227 " processes in the container\n"
228 " -L --selinux-apifs-context=SECLABEL\n"
229 " Set the SELinux security context to be used by\n"
230 " API/tmpfs file systems in the container\n"
231 " --capability=CAP In addition to the default, retain specified\n"
232 " capability\n"
233 " --drop-capability=CAP Drop the specified capability from the default set\n"
234 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
235 " --link-journal=MODE Link up guest journal, one of no, auto, guest, host,\n"
236 " try-guest, try-host\n"
237 " -j Equivalent to --link-journal=try-guest\n"
238 " --read-only Mount the root directory read-only\n"
239 " --bind=PATH[:PATH[:OPTIONS]]\n"
240 " Bind mount a file or directory from the host into\n"
241 " the container\n"
242 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
243 " Similar, but creates a read-only bind mount\n"
244 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
245 " --overlay=PATH[:PATH...]:PATH\n"
246 " Create an overlay mount from the host to \n"
247 " the container\n"
248 " --overlay-ro=PATH[:PATH...]:PATH\n"
249 " Similar, but creates a read-only overlay mount\n"
250 " --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
251 " --share-system Share system namespaces with host\n"
252 " --register=BOOLEAN Register container as machine\n"
253 " --keep-unit Do not register a scope for the machine, reuse\n"
254 " the service unit nspawn is running in\n"
255 " --volatile[=MODE] Run the system in volatile mode\n"
256 " --settings=BOOLEAN Load additional settings from .nspawn file\n"
257 , program_invocation_short_name);
258 }
259
260
261 static int custom_mounts_prepare(void) {
262 unsigned i;
263 int r;
264
265 /* Ensure the mounts are applied prefix first. */
266 qsort_safe(arg_custom_mounts, arg_n_custom_mounts, sizeof(CustomMount), custom_mount_compare);
267
268 /* Allocate working directories for the overlay file systems that need it */
269 for (i = 0; i < arg_n_custom_mounts; i++) {
270 CustomMount *m = &arg_custom_mounts[i];
271
272 if (arg_userns && arg_uid_shift == UID_INVALID && path_equal(m->destination, "/")) {
273 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
274 return -EINVAL;
275 }
276
277 if (m->type != CUSTOM_MOUNT_OVERLAY)
278 continue;
279
280 if (m->work_dir)
281 continue;
282
283 if (m->read_only)
284 continue;
285
286 r = tempfn_random(m->source, NULL, &m->work_dir);
287 if (r < 0)
288 return log_error_errno(r, "Failed to generate work directory from %s: %m", m->source);
289 }
290
291 return 0;
292 }
293
294 static int set_sanitized_path(char **b, const char *path) {
295 char *p;
296
297 assert(b);
298 assert(path);
299
300 p = canonicalize_file_name(path);
301 if (!p) {
302 if (errno != ENOENT)
303 return -errno;
304
305 p = path_make_absolute_cwd(path);
306 if (!p)
307 return -ENOMEM;
308 }
309
310 free(*b);
311 *b = path_kill_slashes(p);
312 return 0;
313 }
314
315 static int detect_unified_cgroup_hierarchy(void) {
316 const char *e;
317 int r;
318
319 /* Allow the user to control whether the unified hierarchy is used */
320 e = getenv("UNIFIED_CGROUP_HIERARCHY");
321 if (e) {
322 r = parse_boolean(e);
323 if (r < 0)
324 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
325
326 arg_unified_cgroup_hierarchy = r;
327 return 0;
328 }
329
330 /* Otherwise inherit the default from the host system */
331 r = cg_unified();
332 if (r < 0)
333 return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
334
335 arg_unified_cgroup_hierarchy = r;
336 return 0;
337 }
338
339 int expose_port_parse(ExposePort **l, const char *s) {
340
341 const char *split, *e;
342 uint16_t container_port, host_port;
343 int protocol;
344 ExposePort *p;
345 int r;
346
347 if ((e = startswith(s, "tcp:")))
348 protocol = IPPROTO_TCP;
349 else if ((e = startswith(s, "udp:")))
350 protocol = IPPROTO_UDP;
351 else {
352 e = s;
353 protocol = IPPROTO_TCP;
354 }
355
356 split = strchr(e, ':');
357 if (split) {
358 char v[split - e + 1];
359
360 memcpy(v, e, split - e);
361 v[split - e] = 0;
362
363 r = safe_atou16(v, &host_port);
364 if (r < 0 || host_port <= 0)
365 return -EINVAL;
366
367 r = safe_atou16(split + 1, &container_port);
368 } else {
369 r = safe_atou16(e, &container_port);
370 host_port = container_port;
371 }
372
373 if (r < 0 || container_port <= 0)
374 return -EINVAL;
375
376 LIST_FOREACH(ports, p, arg_expose_ports)
377 if (p->protocol == protocol && p->host_port == host_port)
378 return -EEXIST;
379
380 p = new(ExposePort, 1);
381 if (!p)
382 return -ENOMEM;
383
384 p->protocol = protocol;
385 p->host_port = host_port;
386 p->container_port = container_port;
387
388 LIST_PREPEND(ports, *l, p);
389
390 return 0;
391 }
392
393 static int parse_argv(int argc, char *argv[]) {
394
395 enum {
396 ARG_VERSION = 0x100,
397 ARG_PRIVATE_NETWORK,
398 ARG_UUID,
399 ARG_READ_ONLY,
400 ARG_CAPABILITY,
401 ARG_DROP_CAPABILITY,
402 ARG_LINK_JOURNAL,
403 ARG_BIND,
404 ARG_BIND_RO,
405 ARG_TMPFS,
406 ARG_OVERLAY,
407 ARG_OVERLAY_RO,
408 ARG_SETENV,
409 ARG_SHARE_SYSTEM,
410 ARG_REGISTER,
411 ARG_KEEP_UNIT,
412 ARG_NETWORK_INTERFACE,
413 ARG_NETWORK_MACVLAN,
414 ARG_NETWORK_IPVLAN,
415 ARG_NETWORK_BRIDGE,
416 ARG_PERSONALITY,
417 ARG_VOLATILE,
418 ARG_TEMPLATE,
419 ARG_PROPERTY,
420 ARG_PRIVATE_USERS,
421 ARG_KILL_SIGNAL,
422 ARG_SETTINGS,
423 };
424
425 static const struct option options[] = {
426 { "help", no_argument, NULL, 'h' },
427 { "version", no_argument, NULL, ARG_VERSION },
428 { "directory", required_argument, NULL, 'D' },
429 { "template", required_argument, NULL, ARG_TEMPLATE },
430 { "ephemeral", no_argument, NULL, 'x' },
431 { "user", required_argument, NULL, 'u' },
432 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
433 { "boot", no_argument, NULL, 'b' },
434 { "uuid", required_argument, NULL, ARG_UUID },
435 { "read-only", no_argument, NULL, ARG_READ_ONLY },
436 { "capability", required_argument, NULL, ARG_CAPABILITY },
437 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
438 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
439 { "bind", required_argument, NULL, ARG_BIND },
440 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
441 { "tmpfs", required_argument, NULL, ARG_TMPFS },
442 { "overlay", required_argument, NULL, ARG_OVERLAY },
443 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
444 { "machine", required_argument, NULL, 'M' },
445 { "slice", required_argument, NULL, 'S' },
446 { "setenv", required_argument, NULL, ARG_SETENV },
447 { "selinux-context", required_argument, NULL, 'Z' },
448 { "selinux-apifs-context", required_argument, NULL, 'L' },
449 { "quiet", no_argument, NULL, 'q' },
450 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM },
451 { "register", required_argument, NULL, ARG_REGISTER },
452 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
453 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
454 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
455 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
456 { "network-veth", no_argument, NULL, 'n' },
457 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
458 { "personality", required_argument, NULL, ARG_PERSONALITY },
459 { "image", required_argument, NULL, 'i' },
460 { "volatile", optional_argument, NULL, ARG_VOLATILE },
461 { "port", required_argument, NULL, 'p' },
462 { "property", required_argument, NULL, ARG_PROPERTY },
463 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
464 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
465 { "settings", required_argument, NULL, ARG_SETTINGS },
466 {}
467 };
468
469 int c, r;
470 uint64_t plus = 0, minus = 0;
471 bool mask_all_settings = false, mask_no_settings = false;
472
473 assert(argc >= 0);
474 assert(argv);
475
476 while ((c = getopt_long(argc, argv, "+hD:u:bL:M:jS:Z:qi:xp:n", options, NULL)) >= 0)
477
478 switch (c) {
479
480 case 'h':
481 help();
482 return 0;
483
484 case ARG_VERSION:
485 puts(PACKAGE_STRING);
486 puts(SYSTEMD_FEATURES);
487 return 0;
488
489 case 'D':
490 r = set_sanitized_path(&arg_directory, optarg);
491 if (r < 0)
492 return log_error_errno(r, "Invalid root directory: %m");
493
494 break;
495
496 case ARG_TEMPLATE:
497 r = set_sanitized_path(&arg_template, optarg);
498 if (r < 0)
499 return log_error_errno(r, "Invalid template directory: %m");
500
501 break;
502
503 case 'i':
504 r = set_sanitized_path(&arg_image, optarg);
505 if (r < 0)
506 return log_error_errno(r, "Invalid image path: %m");
507
508 break;
509
510 case 'x':
511 arg_ephemeral = true;
512 break;
513
514 case 'u':
515 r = free_and_strdup(&arg_user, optarg);
516 if (r < 0)
517 return log_oom();
518
519 arg_settings_mask |= SETTING_USER;
520 break;
521
522 case ARG_NETWORK_BRIDGE:
523 r = free_and_strdup(&arg_network_bridge, optarg);
524 if (r < 0)
525 return log_oom();
526
527 /* fall through */
528
529 case 'n':
530 arg_network_veth = true;
531 arg_private_network = true;
532 arg_settings_mask |= SETTING_NETWORK;
533 break;
534
535 case ARG_NETWORK_INTERFACE:
536 if (strv_extend(&arg_network_interfaces, optarg) < 0)
537 return log_oom();
538
539 arg_private_network = true;
540 arg_settings_mask |= SETTING_NETWORK;
541 break;
542
543 case ARG_NETWORK_MACVLAN:
544 if (strv_extend(&arg_network_macvlan, optarg) < 0)
545 return log_oom();
546
547 arg_private_network = true;
548 arg_settings_mask |= SETTING_NETWORK;
549 break;
550
551 case ARG_NETWORK_IPVLAN:
552 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
553 return log_oom();
554
555 /* fall through */
556
557 case ARG_PRIVATE_NETWORK:
558 arg_private_network = true;
559 arg_settings_mask |= SETTING_NETWORK;
560 break;
561
562 case 'b':
563 arg_boot = true;
564 arg_settings_mask |= SETTING_BOOT;
565 break;
566
567 case ARG_UUID:
568 r = sd_id128_from_string(optarg, &arg_uuid);
569 if (r < 0) {
570 log_error("Invalid UUID: %s", optarg);
571 return r;
572 }
573
574 arg_settings_mask |= SETTING_MACHINE_ID;
575 break;
576
577 case 'S':
578 arg_slice = optarg;
579 break;
580
581 case 'M':
582 if (isempty(optarg))
583 arg_machine = mfree(arg_machine);
584 else {
585 if (!machine_name_is_valid(optarg)) {
586 log_error("Invalid machine name: %s", optarg);
587 return -EINVAL;
588 }
589
590 r = free_and_strdup(&arg_machine, optarg);
591 if (r < 0)
592 return log_oom();
593
594 break;
595 }
596
597 case 'Z':
598 arg_selinux_context = optarg;
599 break;
600
601 case 'L':
602 arg_selinux_apifs_context = optarg;
603 break;
604
605 case ARG_READ_ONLY:
606 arg_read_only = true;
607 arg_settings_mask |= SETTING_READ_ONLY;
608 break;
609
610 case ARG_CAPABILITY:
611 case ARG_DROP_CAPABILITY: {
612 const char *state, *word;
613 size_t length;
614
615 FOREACH_WORD_SEPARATOR(word, length, optarg, ",", state) {
616 _cleanup_free_ char *t;
617
618 t = strndup(word, length);
619 if (!t)
620 return log_oom();
621
622 if (streq(t, "all")) {
623 if (c == ARG_CAPABILITY)
624 plus = (uint64_t) -1;
625 else
626 minus = (uint64_t) -1;
627 } else {
628 int cap;
629
630 cap = capability_from_name(t);
631 if (cap < 0) {
632 log_error("Failed to parse capability %s.", t);
633 return -EINVAL;
634 }
635
636 if (c == ARG_CAPABILITY)
637 plus |= 1ULL << (uint64_t) cap;
638 else
639 minus |= 1ULL << (uint64_t) cap;
640 }
641 }
642
643 arg_settings_mask |= SETTING_CAPABILITY;
644 break;
645 }
646
647 case 'j':
648 arg_link_journal = LINK_GUEST;
649 arg_link_journal_try = true;
650 break;
651
652 case ARG_LINK_JOURNAL:
653 if (streq(optarg, "auto")) {
654 arg_link_journal = LINK_AUTO;
655 arg_link_journal_try = false;
656 } else if (streq(optarg, "no")) {
657 arg_link_journal = LINK_NO;
658 arg_link_journal_try = false;
659 } else if (streq(optarg, "guest")) {
660 arg_link_journal = LINK_GUEST;
661 arg_link_journal_try = false;
662 } else if (streq(optarg, "host")) {
663 arg_link_journal = LINK_HOST;
664 arg_link_journal_try = false;
665 } else if (streq(optarg, "try-guest")) {
666 arg_link_journal = LINK_GUEST;
667 arg_link_journal_try = true;
668 } else if (streq(optarg, "try-host")) {
669 arg_link_journal = LINK_HOST;
670 arg_link_journal_try = true;
671 } else {
672 log_error("Failed to parse link journal mode %s", optarg);
673 return -EINVAL;
674 }
675
676 break;
677
678 case ARG_BIND:
679 case ARG_BIND_RO:
680 r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
681 if (r < 0)
682 return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
683
684 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
685 break;
686
687 case ARG_TMPFS:
688 r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
689 if (r < 0)
690 return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
691
692 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
693 break;
694
695 case ARG_OVERLAY:
696 case ARG_OVERLAY_RO: {
697 _cleanup_free_ char *upper = NULL, *destination = NULL;
698 _cleanup_strv_free_ char **lower = NULL;
699 CustomMount *m;
700 unsigned n = 0;
701 char **i;
702
703 r = strv_split_extract(&lower, optarg, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
704 if (r == -ENOMEM)
705 return log_oom();
706 else if (r < 0) {
707 log_error("Invalid overlay specification: %s", optarg);
708 return r;
709 }
710
711 STRV_FOREACH(i, lower) {
712 if (!path_is_absolute(*i)) {
713 log_error("Overlay path %s is not absolute.", *i);
714 return -EINVAL;
715 }
716
717 n++;
718 }
719
720 if (n < 2) {
721 log_error("--overlay= needs at least two colon-separated directories specified.");
722 return -EINVAL;
723 }
724
725 if (n == 2) {
726 /* If two parameters are specified,
727 * the first one is the lower, the
728 * second one the upper directory. And
729 * we'll also define the destination
730 * mount point the same as the upper. */
731 upper = lower[1];
732 lower[1] = NULL;
733
734 destination = strdup(upper);
735 if (!destination)
736 return log_oom();
737
738 } else {
739 upper = lower[n - 2];
740 destination = lower[n - 1];
741 lower[n - 2] = NULL;
742 }
743
744 m = custom_mount_add(&arg_custom_mounts, &arg_n_custom_mounts, CUSTOM_MOUNT_OVERLAY);
745 if (!m)
746 return log_oom();
747
748 m->destination = destination;
749 m->source = upper;
750 m->lower = lower;
751 m->read_only = c == ARG_OVERLAY_RO;
752
753 upper = destination = NULL;
754 lower = NULL;
755
756 arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
757 break;
758 }
759
760 case ARG_SETENV: {
761 char **n;
762
763 if (!env_assignment_is_valid(optarg)) {
764 log_error("Environment variable assignment '%s' is not valid.", optarg);
765 return -EINVAL;
766 }
767
768 n = strv_env_set(arg_setenv, optarg);
769 if (!n)
770 return log_oom();
771
772 strv_free(arg_setenv);
773 arg_setenv = n;
774
775 arg_settings_mask |= SETTING_ENVIRONMENT;
776 break;
777 }
778
779 case 'q':
780 arg_quiet = true;
781 break;
782
783 case ARG_SHARE_SYSTEM:
784 arg_share_system = true;
785 break;
786
787 case ARG_REGISTER:
788 r = parse_boolean(optarg);
789 if (r < 0) {
790 log_error("Failed to parse --register= argument: %s", optarg);
791 return r;
792 }
793
794 arg_register = r;
795 break;
796
797 case ARG_KEEP_UNIT:
798 arg_keep_unit = true;
799 break;
800
801 case ARG_PERSONALITY:
802
803 arg_personality = personality_from_string(optarg);
804 if (arg_personality == PERSONALITY_INVALID) {
805 log_error("Unknown or unsupported personality '%s'.", optarg);
806 return -EINVAL;
807 }
808
809 arg_settings_mask |= SETTING_PERSONALITY;
810 break;
811
812 case ARG_VOLATILE:
813
814 if (!optarg)
815 arg_volatile_mode = VOLATILE_YES;
816 else {
817 VolatileMode m;
818
819 m = volatile_mode_from_string(optarg);
820 if (m < 0) {
821 log_error("Failed to parse --volatile= argument: %s", optarg);
822 return -EINVAL;
823 } else
824 arg_volatile_mode = m;
825 }
826
827 arg_settings_mask |= SETTING_VOLATILE_MODE;
828 break;
829
830 case 'p':
831 r = expose_port_parse(&arg_expose_ports, optarg);
832 if (r == -EEXIST)
833 return log_error_errno(r, "Duplicate port specification: %s", optarg);
834 if (r < 0)
835 return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
836
837 arg_settings_mask |= SETTING_EXPOSE_PORTS;
838 break;
839
840 case ARG_PROPERTY:
841 if (strv_extend(&arg_property, optarg) < 0)
842 return log_oom();
843
844 break;
845
846 case ARG_PRIVATE_USERS:
847 if (optarg) {
848 _cleanup_free_ char *buffer = NULL;
849 const char *range, *shift;
850
851 range = strchr(optarg, ':');
852 if (range) {
853 buffer = strndup(optarg, range - optarg);
854 if (!buffer)
855 return log_oom();
856 shift = buffer;
857
858 range++;
859 if (safe_atou32(range, &arg_uid_range) < 0 || arg_uid_range <= 0) {
860 log_error("Failed to parse UID range: %s", range);
861 return -EINVAL;
862 }
863 } else
864 shift = optarg;
865
866 if (parse_uid(shift, &arg_uid_shift) < 0) {
867 log_error("Failed to parse UID: %s", optarg);
868 return -EINVAL;
869 }
870 }
871
872 arg_userns = true;
873 break;
874
875 case ARG_KILL_SIGNAL:
876 arg_kill_signal = signal_from_string_try_harder(optarg);
877 if (arg_kill_signal < 0) {
878 log_error("Cannot parse signal: %s", optarg);
879 return -EINVAL;
880 }
881
882 arg_settings_mask |= SETTING_KILL_SIGNAL;
883 break;
884
885 case ARG_SETTINGS:
886
887 /* no → do not read files
888 * yes → read files, do not override cmdline, trust only subset
889 * override → read files, override cmdline, trust only subset
890 * trusted → read files, do not override cmdline, trust all
891 */
892
893 r = parse_boolean(optarg);
894 if (r < 0) {
895 if (streq(optarg, "trusted")) {
896 mask_all_settings = false;
897 mask_no_settings = false;
898 arg_settings_trusted = true;
899
900 } else if (streq(optarg, "override")) {
901 mask_all_settings = false;
902 mask_no_settings = true;
903 arg_settings_trusted = -1;
904 } else
905 return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
906 } else if (r > 0) {
907 /* yes */
908 mask_all_settings = false;
909 mask_no_settings = false;
910 arg_settings_trusted = -1;
911 } else {
912 /* no */
913 mask_all_settings = true;
914 mask_no_settings = false;
915 arg_settings_trusted = false;
916 }
917
918 break;
919
920 case '?':
921 return -EINVAL;
922
923 default:
924 assert_not_reached("Unhandled option");
925 }
926
927 if (arg_share_system)
928 arg_register = false;
929
930 if (arg_boot && arg_share_system) {
931 log_error("--boot and --share-system may not be combined.");
932 return -EINVAL;
933 }
934
935 if (arg_keep_unit && cg_pid_get_owner_uid(0, NULL) >= 0) {
936 log_error("--keep-unit may not be used when invoked from a user session.");
937 return -EINVAL;
938 }
939
940 if (arg_directory && arg_image) {
941 log_error("--directory= and --image= may not be combined.");
942 return -EINVAL;
943 }
944
945 if (arg_template && arg_image) {
946 log_error("--template= and --image= may not be combined.");
947 return -EINVAL;
948 }
949
950 if (arg_template && !(arg_directory || arg_machine)) {
951 log_error("--template= needs --directory= or --machine=.");
952 return -EINVAL;
953 }
954
955 if (arg_ephemeral && arg_template) {
956 log_error("--ephemeral and --template= may not be combined.");
957 return -EINVAL;
958 }
959
960 if (arg_ephemeral && arg_image) {
961 log_error("--ephemeral and --image= may not be combined.");
962 return -EINVAL;
963 }
964
965 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
966 log_error("--ephemeral and --link-journal= may not be combined.");
967 return -EINVAL;
968 }
969
970 if (arg_userns && access("/proc/self/uid_map", F_OK) < 0)
971 return log_error_errno(EOPNOTSUPP, "--private-users= is not supported, kernel compiled without user namespace support.");
972
973 if (argc > optind) {
974 arg_parameters = strv_copy(argv + optind);
975 if (!arg_parameters)
976 return log_oom();
977
978 arg_settings_mask |= SETTING_BOOT;
979 }
980
981 /* Load all settings from .nspawn files */
982 if (mask_no_settings)
983 arg_settings_mask = 0;
984
985 /* Don't load any settings from .nspawn files */
986 if (mask_all_settings)
987 arg_settings_mask = _SETTINGS_MASK_ALL;
988
989 arg_retain = (arg_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
990
991 r = detect_unified_cgroup_hierarchy();
992 if (r < 0)
993 return r;
994
995 return 1;
996 }
997
998 static int verify_arguments(void) {
999
1000 if (arg_volatile_mode != VOLATILE_NO && arg_read_only) {
1001 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1002 return -EINVAL;
1003 }
1004
1005 if (arg_expose_ports && !arg_private_network) {
1006 log_error("Cannot use --port= without private networking.");
1007 return -EINVAL;
1008 }
1009
1010 if (arg_boot && arg_kill_signal <= 0)
1011 arg_kill_signal = SIGRTMIN+3;
1012
1013 return 0;
1014 }
1015
1016 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1017 assert(p);
1018
1019 if (!arg_userns)
1020 return 0;
1021
1022 if (uid == UID_INVALID && gid == GID_INVALID)
1023 return 0;
1024
1025 if (uid != UID_INVALID) {
1026 uid += arg_uid_shift;
1027
1028 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1029 return -EOVERFLOW;
1030 }
1031
1032 if (gid != GID_INVALID) {
1033 gid += (gid_t) arg_uid_shift;
1034
1035 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1036 return -EOVERFLOW;
1037 }
1038
1039 if (lchown(p, uid, gid) < 0)
1040 return -errno;
1041
1042 return 0;
1043 }
1044
1045 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1046 const char *q;
1047
1048 q = prefix_roota(root, path);
1049 if (mkdir(q, mode) < 0) {
1050 if (errno == EEXIST)
1051 return 0;
1052 return -errno;
1053 }
1054
1055 return userns_lchown(q, uid, gid);
1056 }
1057
1058 static int setup_timezone(const char *dest) {
1059 _cleanup_free_ char *p = NULL, *q = NULL;
1060 const char *where, *check, *what;
1061 char *z, *y;
1062 int r;
1063
1064 assert(dest);
1065
1066 /* Fix the timezone, if possible */
1067 r = readlink_malloc("/etc/localtime", &p);
1068 if (r < 0) {
1069 log_warning("/etc/localtime is not a symlink, not updating container timezone.");
1070 return 0;
1071 }
1072
1073 z = path_startswith(p, "../usr/share/zoneinfo/");
1074 if (!z)
1075 z = path_startswith(p, "/usr/share/zoneinfo/");
1076 if (!z) {
1077 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1078 return 0;
1079 }
1080
1081 where = prefix_roota(dest, "/etc/localtime");
1082 r = readlink_malloc(where, &q);
1083 if (r >= 0) {
1084 y = path_startswith(q, "../usr/share/zoneinfo/");
1085 if (!y)
1086 y = path_startswith(q, "/usr/share/zoneinfo/");
1087
1088 /* Already pointing to the right place? Then do nothing .. */
1089 if (y && streq(y, z))
1090 return 0;
1091 }
1092
1093 check = strjoina("/usr/share/zoneinfo/", z);
1094 check = prefix_root(dest, check);
1095 if (laccess(check, F_OK) < 0) {
1096 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1097 return 0;
1098 }
1099
1100 r = unlink(where);
1101 if (r < 0 && errno != ENOENT) {
1102 log_error_errno(errno, "Failed to remove existing timezone info %s in container: %m", where);
1103 return 0;
1104 }
1105
1106 what = strjoina("../usr/share/zoneinfo/", z);
1107 if (symlink(what, where) < 0) {
1108 log_error_errno(errno, "Failed to correct timezone of container: %m");
1109 return 0;
1110 }
1111
1112 r = userns_lchown(where, 0, 0);
1113 if (r < 0)
1114 return log_warning_errno(r, "Failed to chown /etc/localtime: %m");
1115
1116 return 0;
1117 }
1118
1119 static int setup_resolv_conf(const char *dest) {
1120 const char *where = NULL;
1121 int r;
1122
1123 assert(dest);
1124
1125 if (arg_private_network)
1126 return 0;
1127
1128 /* Fix resolv.conf, if possible */
1129 where = prefix_roota(dest, "/etc/resolv.conf");
1130
1131 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0);
1132 if (r < 0) {
1133 /* If the file already exists as symlink, let's
1134 * suppress the warning, under the assumption that
1135 * resolved or something similar runs inside and the
1136 * symlink points there.
1137 *
1138 * If the disk image is read-only, there's also no
1139 * point in complaining.
1140 */
1141 log_full_errno(IN_SET(r, -ELOOP, -EROFS) ? LOG_DEBUG : LOG_WARNING, r,
1142 "Failed to copy /etc/resolv.conf to %s: %m", where);
1143 return 0;
1144 }
1145
1146 r = userns_lchown(where, 0, 0);
1147 if (r < 0)
1148 log_warning_errno(r, "Failed to chown /etc/resolv.conf: %m");
1149
1150 return 0;
1151 }
1152
1153 static char* id128_format_as_uuid(sd_id128_t id, char s[37]) {
1154 assert(s);
1155
1156 snprintf(s, 37,
1157 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
1158 SD_ID128_FORMAT_VAL(id));
1159
1160 return s;
1161 }
1162
1163 static int setup_boot_id(const char *dest) {
1164 const char *from, *to;
1165 sd_id128_t rnd = {};
1166 char as_uuid[37];
1167 int r;
1168
1169 if (arg_share_system)
1170 return 0;
1171
1172 /* Generate a new randomized boot ID, so that each boot-up of
1173 * the container gets a new one */
1174
1175 from = prefix_roota(dest, "/run/proc-sys-kernel-random-boot-id");
1176 to = prefix_roota(dest, "/proc/sys/kernel/random/boot_id");
1177
1178 r = sd_id128_randomize(&rnd);
1179 if (r < 0)
1180 return log_error_errno(r, "Failed to generate random boot id: %m");
1181
1182 id128_format_as_uuid(rnd, as_uuid);
1183
1184 r = write_string_file(from, as_uuid, WRITE_STRING_FILE_CREATE);
1185 if (r < 0)
1186 return log_error_errno(r, "Failed to write boot id: %m");
1187
1188 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
1189 r = log_error_errno(errno, "Failed to bind mount boot id: %m");
1190 else if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL) < 0)
1191 log_warning_errno(errno, "Failed to make boot id read-only: %m");
1192
1193 unlink(from);
1194 return r;
1195 }
1196
1197 static int copy_devnodes(const char *dest) {
1198
1199 static const char devnodes[] =
1200 "null\0"
1201 "zero\0"
1202 "full\0"
1203 "random\0"
1204 "urandom\0"
1205 "tty\0"
1206 "net/tun\0";
1207
1208 const char *d;
1209 int r = 0;
1210 _cleanup_umask_ mode_t u;
1211
1212 assert(dest);
1213
1214 u = umask(0000);
1215
1216 /* Create /dev/net, so that we can create /dev/net/tun in it */
1217 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1218 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1219
1220 NULSTR_FOREACH(d, devnodes) {
1221 _cleanup_free_ char *from = NULL, *to = NULL;
1222 struct stat st;
1223
1224 from = strappend("/dev/", d);
1225 to = prefix_root(dest, from);
1226
1227 if (stat(from, &st) < 0) {
1228
1229 if (errno != ENOENT)
1230 return log_error_errno(errno, "Failed to stat %s: %m", from);
1231
1232 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1233
1234 log_error("%s is not a char or block device, cannot copy.", from);
1235 return -EIO;
1236
1237 } else {
1238 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1239 if (errno != EPERM)
1240 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1241
1242 /* Some systems abusively restrict mknod but
1243 * allow bind mounts. */
1244 r = touch(to);
1245 if (r < 0)
1246 return log_error_errno(r, "touch (%s) failed: %m", to);
1247 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
1248 return log_error_errno(errno, "Both mknod and bind mount (%s) failed: %m", to);
1249 }
1250
1251 r = userns_lchown(to, 0, 0);
1252 if (r < 0)
1253 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1254 }
1255 }
1256
1257 return r;
1258 }
1259
1260 static int setup_pts(const char *dest) {
1261 _cleanup_free_ char *options = NULL;
1262 const char *p;
1263
1264 #ifdef HAVE_SELINUX
1265 if (arg_selinux_apifs_context)
1266 (void) asprintf(&options,
1267 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1268 arg_uid_shift + TTY_GID,
1269 arg_selinux_apifs_context);
1270 else
1271 #endif
1272 (void) asprintf(&options,
1273 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1274 arg_uid_shift + TTY_GID);
1275
1276 if (!options)
1277 return log_oom();
1278
1279 /* Mount /dev/pts itself */
1280 p = prefix_roota(dest, "/dev/pts");
1281 if (mkdir(p, 0755) < 0)
1282 return log_error_errno(errno, "Failed to create /dev/pts: %m");
1283 if (mount("devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options) < 0)
1284 return log_error_errno(errno, "Failed to mount /dev/pts: %m");
1285 if (userns_lchown(p, 0, 0) < 0)
1286 return log_error_errno(errno, "Failed to chown /dev/pts: %m");
1287
1288 /* Create /dev/ptmx symlink */
1289 p = prefix_roota(dest, "/dev/ptmx");
1290 if (symlink("pts/ptmx", p) < 0)
1291 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1292 if (userns_lchown(p, 0, 0) < 0)
1293 return log_error_errno(errno, "Failed to chown /dev/ptmx: %m");
1294
1295 /* And fix /dev/pts/ptmx ownership */
1296 p = prefix_roota(dest, "/dev/pts/ptmx");
1297 if (userns_lchown(p, 0, 0) < 0)
1298 return log_error_errno(errno, "Failed to chown /dev/pts/ptmx: %m");
1299
1300 return 0;
1301 }
1302
1303 static int setup_dev_console(const char *dest, const char *console) {
1304 _cleanup_umask_ mode_t u;
1305 const char *to;
1306 int r;
1307
1308 assert(dest);
1309 assert(console);
1310
1311 u = umask(0000);
1312
1313 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
1314 if (r < 0)
1315 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
1316
1317 /* We need to bind mount the right tty to /dev/console since
1318 * ptys can only exist on pts file systems. To have something
1319 * to bind mount things on we create a empty regular file. */
1320
1321 to = prefix_roota(dest, "/dev/console");
1322 r = touch(to);
1323 if (r < 0)
1324 return log_error_errno(r, "touch() for /dev/console failed: %m");
1325
1326 if (mount(console, to, NULL, MS_BIND, NULL) < 0)
1327 return log_error_errno(errno, "Bind mount for /dev/console failed: %m");
1328
1329 return 0;
1330 }
1331
1332 static int setup_kmsg(const char *dest, int kmsg_socket) {
1333 const char *from, *to;
1334 _cleanup_umask_ mode_t u;
1335 int fd, k;
1336 union {
1337 struct cmsghdr cmsghdr;
1338 uint8_t buf[CMSG_SPACE(sizeof(int))];
1339 } control = {};
1340 struct msghdr mh = {
1341 .msg_control = &control,
1342 .msg_controllen = sizeof(control),
1343 };
1344 struct cmsghdr *cmsg;
1345
1346 assert(kmsg_socket >= 0);
1347
1348 u = umask(0000);
1349
1350 /* We create the kmsg FIFO as /run/kmsg, but immediately
1351 * delete it after bind mounting it to /proc/kmsg. While FIFOs
1352 * on the reading side behave very similar to /proc/kmsg,
1353 * their writing side behaves differently from /dev/kmsg in
1354 * that writing blocks when nothing is reading. In order to
1355 * avoid any problems with containers deadlocking due to this
1356 * we simply make /dev/kmsg unavailable to the container. */
1357 from = prefix_roota(dest, "/run/kmsg");
1358 to = prefix_roota(dest, "/proc/kmsg");
1359
1360 if (mkfifo(from, 0600) < 0)
1361 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
1362 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
1363 return log_error_errno(errno, "Bind mount for /proc/kmsg failed: %m");
1364
1365 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
1366 if (fd < 0)
1367 return log_error_errno(errno, "Failed to open fifo: %m");
1368
1369 cmsg = CMSG_FIRSTHDR(&mh);
1370 cmsg->cmsg_level = SOL_SOCKET;
1371 cmsg->cmsg_type = SCM_RIGHTS;
1372 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1373 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1374
1375 mh.msg_controllen = cmsg->cmsg_len;
1376
1377 /* Store away the fd in the socket, so that it stays open as
1378 * long as we run the child */
1379 k = sendmsg(kmsg_socket, &mh, MSG_NOSIGNAL);
1380 safe_close(fd);
1381
1382 if (k < 0)
1383 return log_error_errno(errno, "Failed to send FIFO fd: %m");
1384
1385 /* And now make the FIFO unavailable as /run/kmsg... */
1386 (void) unlink(from);
1387
1388 return 0;
1389 }
1390
1391 static int send_rtnl(int send_fd) {
1392 union {
1393 struct cmsghdr cmsghdr;
1394 uint8_t buf[CMSG_SPACE(sizeof(int))];
1395 } control = {};
1396 struct msghdr mh = {
1397 .msg_control = &control,
1398 .msg_controllen = sizeof(control),
1399 };
1400 struct cmsghdr *cmsg;
1401 _cleanup_close_ int fd = -1;
1402 ssize_t k;
1403
1404 assert(send_fd >= 0);
1405
1406 if (!arg_expose_ports)
1407 return 0;
1408
1409 fd = socket(PF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_ROUTE);
1410 if (fd < 0)
1411 return log_error_errno(errno, "Failed to allocate container netlink: %m");
1412
1413 cmsg = CMSG_FIRSTHDR(&mh);
1414 cmsg->cmsg_level = SOL_SOCKET;
1415 cmsg->cmsg_type = SCM_RIGHTS;
1416 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
1417 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
1418
1419 mh.msg_controllen = cmsg->cmsg_len;
1420
1421 /* Store away the fd in the socket, so that it stays open as
1422 * long as we run the child */
1423 k = sendmsg(send_fd, &mh, MSG_NOSIGNAL);
1424 if (k < 0)
1425 return log_error_errno(errno, "Failed to send netlink fd: %m");
1426
1427 return 0;
1428 }
1429
1430 static int flush_ports(union in_addr_union *exposed) {
1431 ExposePort *p;
1432 int r, af = AF_INET;
1433
1434 assert(exposed);
1435
1436 if (!arg_expose_ports)
1437 return 0;
1438
1439 if (in_addr_is_null(af, exposed))
1440 return 0;
1441
1442 log_debug("Lost IP address.");
1443
1444 LIST_FOREACH(ports, p, arg_expose_ports) {
1445 r = fw_add_local_dnat(false,
1446 af,
1447 p->protocol,
1448 NULL,
1449 NULL, 0,
1450 NULL, 0,
1451 p->host_port,
1452 exposed,
1453 p->container_port,
1454 NULL);
1455 if (r < 0)
1456 log_warning_errno(r, "Failed to modify firewall: %m");
1457 }
1458
1459 *exposed = IN_ADDR_NULL;
1460 return 0;
1461 }
1462
1463 static int expose_ports(sd_netlink *rtnl, union in_addr_union *exposed) {
1464 _cleanup_free_ struct local_address *addresses = NULL;
1465 _cleanup_free_ char *pretty = NULL;
1466 union in_addr_union new_exposed;
1467 ExposePort *p;
1468 bool add;
1469 int af = AF_INET, r;
1470
1471 assert(exposed);
1472
1473 /* Invoked each time an address is added or removed inside the
1474 * container */
1475
1476 if (!arg_expose_ports)
1477 return 0;
1478
1479 r = local_addresses(rtnl, 0, af, &addresses);
1480 if (r < 0)
1481 return log_error_errno(r, "Failed to enumerate local addresses: %m");
1482
1483 add = r > 0 &&
1484 addresses[0].family == af &&
1485 addresses[0].scope < RT_SCOPE_LINK;
1486
1487 if (!add)
1488 return flush_ports(exposed);
1489
1490 new_exposed = addresses[0].address;
1491 if (in_addr_equal(af, exposed, &new_exposed))
1492 return 0;
1493
1494 in_addr_to_string(af, &new_exposed, &pretty);
1495 log_debug("New container IP is %s.", strna(pretty));
1496
1497 LIST_FOREACH(ports, p, arg_expose_ports) {
1498
1499 r = fw_add_local_dnat(true,
1500 af,
1501 p->protocol,
1502 NULL,
1503 NULL, 0,
1504 NULL, 0,
1505 p->host_port,
1506 &new_exposed,
1507 p->container_port,
1508 in_addr_is_null(af, exposed) ? NULL : exposed);
1509 if (r < 0)
1510 log_warning_errno(r, "Failed to modify firewall: %m");
1511 }
1512
1513 *exposed = new_exposed;
1514 return 0;
1515 }
1516
1517 void expose_port_free_all(ExposePort *p) {
1518
1519 while (p) {
1520 ExposePort *q = p;
1521 LIST_REMOVE(ports, p, q);
1522 free(q);
1523 }
1524 }
1525
1526 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
1527 union in_addr_union *exposed = userdata;
1528
1529 assert(rtnl);
1530 assert(m);
1531 assert(exposed);
1532
1533 expose_ports(rtnl, exposed);
1534 return 0;
1535 }
1536
1537 static int watch_rtnl(sd_event *event, int recv_fd, union in_addr_union *exposed, sd_netlink **ret) {
1538 union {
1539 struct cmsghdr cmsghdr;
1540 uint8_t buf[CMSG_SPACE(sizeof(int))];
1541 } control = {};
1542 struct msghdr mh = {
1543 .msg_control = &control,
1544 .msg_controllen = sizeof(control),
1545 };
1546 struct cmsghdr *cmsg;
1547 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
1548 int fd, r;
1549 ssize_t k;
1550
1551 assert(event);
1552 assert(recv_fd >= 0);
1553 assert(ret);
1554
1555 if (!arg_expose_ports)
1556 return 0;
1557
1558 k = recvmsg(recv_fd, &mh, MSG_NOSIGNAL);
1559 if (k < 0)
1560 return log_error_errno(errno, "Failed to recv netlink fd: %m");
1561
1562 cmsg = CMSG_FIRSTHDR(&mh);
1563 assert(cmsg->cmsg_level == SOL_SOCKET);
1564 assert(cmsg->cmsg_type == SCM_RIGHTS);
1565 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
1566 memcpy(&fd, CMSG_DATA(cmsg), sizeof(int));
1567
1568 r = sd_netlink_open_fd(&rtnl, fd);
1569 if (r < 0) {
1570 safe_close(fd);
1571 return log_error_errno(r, "Failed to create rtnl object: %m");
1572 }
1573
1574 r = sd_netlink_add_match(rtnl, RTM_NEWADDR, on_address_change, exposed);
1575 if (r < 0)
1576 return log_error_errno(r, "Failed to subscribe to RTM_NEWADDR messages: %m");
1577
1578 r = sd_netlink_add_match(rtnl, RTM_DELADDR, on_address_change, exposed);
1579 if (r < 0)
1580 return log_error_errno(r, "Failed to subscribe to RTM_DELADDR messages: %m");
1581
1582 r = sd_netlink_attach_event(rtnl, event, 0);
1583 if (r < 0)
1584 return log_error_errno(r, "Failed to add to even loop: %m");
1585
1586 *ret = rtnl;
1587 rtnl = NULL;
1588
1589 return 0;
1590 }
1591
1592 static int setup_hostname(void) {
1593
1594 if (arg_share_system)
1595 return 0;
1596
1597 if (sethostname_idempotent(arg_machine) < 0)
1598 return -errno;
1599
1600 return 0;
1601 }
1602
1603 static int setup_journal(const char *directory) {
1604 sd_id128_t machine_id, this_id;
1605 _cleanup_free_ char *b = NULL, *d = NULL;
1606 const char *etc_machine_id, *p, *q;
1607 char *id;
1608 int r;
1609
1610 /* Don't link journals in ephemeral mode */
1611 if (arg_ephemeral)
1612 return 0;
1613
1614 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
1615
1616 r = read_one_line_file(etc_machine_id, &b);
1617 if (r == -ENOENT && arg_link_journal == LINK_AUTO)
1618 return 0;
1619 else if (r < 0)
1620 return log_error_errno(r, "Failed to read machine ID from %s: %m", etc_machine_id);
1621
1622 id = strstrip(b);
1623 if (isempty(id) && arg_link_journal == LINK_AUTO)
1624 return 0;
1625
1626 /* Verify validity */
1627 r = sd_id128_from_string(id, &machine_id);
1628 if (r < 0)
1629 return log_error_errno(r, "Failed to parse machine ID from %s: %m", etc_machine_id);
1630
1631 r = sd_id128_get_machine(&this_id);
1632 if (r < 0)
1633 return log_error_errno(r, "Failed to retrieve machine ID: %m");
1634
1635 if (sd_id128_equal(machine_id, this_id)) {
1636 log_full(arg_link_journal == LINK_AUTO ? LOG_WARNING : LOG_ERR,
1637 "Host and machine ids are equal (%s): refusing to link journals", id);
1638 if (arg_link_journal == LINK_AUTO)
1639 return 0;
1640 return -EEXIST;
1641 }
1642
1643 if (arg_link_journal == LINK_NO)
1644 return 0;
1645
1646 r = userns_mkdir(directory, "/var", 0755, 0, 0);
1647 if (r < 0)
1648 return log_error_errno(r, "Failed to create /var: %m");
1649
1650 r = userns_mkdir(directory, "/var/log", 0755, 0, 0);
1651 if (r < 0)
1652 return log_error_errno(r, "Failed to create /var/log: %m");
1653
1654 r = userns_mkdir(directory, "/var/log/journal", 0755, 0, 0);
1655 if (r < 0)
1656 return log_error_errno(r, "Failed to create /var/log/journal: %m");
1657
1658 p = strjoina("/var/log/journal/", id);
1659 q = prefix_roota(directory, p);
1660
1661 if (path_is_mount_point(p, 0) > 0) {
1662 if (arg_link_journal != LINK_AUTO) {
1663 log_error("%s: already a mount point, refusing to use for journal", p);
1664 return -EEXIST;
1665 }
1666
1667 return 0;
1668 }
1669
1670 if (path_is_mount_point(q, 0) > 0) {
1671 if (arg_link_journal != LINK_AUTO) {
1672 log_error("%s: already a mount point, refusing to use for journal", q);
1673 return -EEXIST;
1674 }
1675
1676 return 0;
1677 }
1678
1679 r = readlink_and_make_absolute(p, &d);
1680 if (r >= 0) {
1681 if ((arg_link_journal == LINK_GUEST ||
1682 arg_link_journal == LINK_AUTO) &&
1683 path_equal(d, q)) {
1684
1685 r = userns_mkdir(directory, p, 0755, 0, 0);
1686 if (r < 0)
1687 log_warning_errno(errno, "Failed to create directory %s: %m", q);
1688 return 0;
1689 }
1690
1691 if (unlink(p) < 0)
1692 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
1693 } else if (r == -EINVAL) {
1694
1695 if (arg_link_journal == LINK_GUEST &&
1696 rmdir(p) < 0) {
1697
1698 if (errno == ENOTDIR) {
1699 log_error("%s already exists and is neither a symlink nor a directory", p);
1700 return r;
1701 } else {
1702 log_error_errno(errno, "Failed to remove %s: %m", p);
1703 return -errno;
1704 }
1705 }
1706 } else if (r != -ENOENT) {
1707 log_error_errno(errno, "readlink(%s) failed: %m", p);
1708 return r;
1709 }
1710
1711 if (arg_link_journal == LINK_GUEST) {
1712
1713 if (symlink(q, p) < 0) {
1714 if (arg_link_journal_try) {
1715 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
1716 return 0;
1717 } else {
1718 log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
1719 return -errno;
1720 }
1721 }
1722
1723 r = userns_mkdir(directory, p, 0755, 0, 0);
1724 if (r < 0)
1725 log_warning_errno(errno, "Failed to create directory %s: %m", q);
1726 return 0;
1727 }
1728
1729 if (arg_link_journal == LINK_HOST) {
1730 /* don't create parents here -- if the host doesn't have
1731 * permanent journal set up, don't force it here */
1732 r = mkdir(p, 0755);
1733 if (r < 0) {
1734 if (arg_link_journal_try) {
1735 log_debug_errno(errno, "Failed to create %s, skipping journal setup: %m", p);
1736 return 0;
1737 } else {
1738 log_error_errno(errno, "Failed to create %s: %m", p);
1739 return r;
1740 }
1741 }
1742
1743 } else if (access(p, F_OK) < 0)
1744 return 0;
1745
1746 if (dir_is_empty(q) == 0)
1747 log_warning("%s is not empty, proceeding anyway.", q);
1748
1749 r = userns_mkdir(directory, p, 0755, 0, 0);
1750 if (r < 0) {
1751 log_error_errno(errno, "Failed to create %s: %m", q);
1752 return r;
1753 }
1754
1755 if (mount(p, q, NULL, MS_BIND, NULL) < 0)
1756 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
1757
1758 return 0;
1759 }
1760
1761 static int drop_capabilities(void) {
1762 return capability_bounding_set_drop(~arg_retain, false);
1763 }
1764
1765 static int register_machine(pid_t pid, int local_ifindex) {
1766 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1767 _cleanup_bus_flush_close_unref_ sd_bus *bus = NULL;
1768 int r;
1769
1770 if (!arg_register)
1771 return 0;
1772
1773 r = sd_bus_default_system(&bus);
1774 if (r < 0)
1775 return log_error_errno(r, "Failed to open system bus: %m");
1776
1777 if (arg_keep_unit) {
1778 r = sd_bus_call_method(
1779 bus,
1780 "org.freedesktop.machine1",
1781 "/org/freedesktop/machine1",
1782 "org.freedesktop.machine1.Manager",
1783 "RegisterMachineWithNetwork",
1784 &error,
1785 NULL,
1786 "sayssusai",
1787 arg_machine,
1788 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
1789 "nspawn",
1790 "container",
1791 (uint32_t) pid,
1792 strempty(arg_directory),
1793 local_ifindex > 0 ? 1 : 0, local_ifindex);
1794 } else {
1795 _cleanup_bus_message_unref_ sd_bus_message *m = NULL;
1796 char **i;
1797 unsigned j;
1798
1799 r = sd_bus_message_new_method_call(
1800 bus,
1801 &m,
1802 "org.freedesktop.machine1",
1803 "/org/freedesktop/machine1",
1804 "org.freedesktop.machine1.Manager",
1805 "CreateMachineWithNetwork");
1806 if (r < 0)
1807 return bus_log_create_error(r);
1808
1809 r = sd_bus_message_append(
1810 m,
1811 "sayssusai",
1812 arg_machine,
1813 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
1814 "nspawn",
1815 "container",
1816 (uint32_t) pid,
1817 strempty(arg_directory),
1818 local_ifindex > 0 ? 1 : 0, local_ifindex);
1819 if (r < 0)
1820 return bus_log_create_error(r);
1821
1822 r = sd_bus_message_open_container(m, 'a', "(sv)");
1823 if (r < 0)
1824 return bus_log_create_error(r);
1825
1826 if (!isempty(arg_slice)) {
1827 r = sd_bus_message_append(m, "(sv)", "Slice", "s", arg_slice);
1828 if (r < 0)
1829 return bus_log_create_error(r);
1830 }
1831
1832 r = sd_bus_message_append(m, "(sv)", "DevicePolicy", "s", "strict");
1833 if (r < 0)
1834 return bus_log_create_error(r);
1835
1836 /* If you make changes here, also make sure to update
1837 * systemd-nspawn@.service, to keep the device
1838 * policies in sync regardless if we are run with or
1839 * without the --keep-unit switch. */
1840 r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 9,
1841 /* Allow the container to
1842 * access and create the API
1843 * device nodes, so that
1844 * PrivateDevices= in the
1845 * container can work
1846 * fine */
1847 "/dev/null", "rwm",
1848 "/dev/zero", "rwm",
1849 "/dev/full", "rwm",
1850 "/dev/random", "rwm",
1851 "/dev/urandom", "rwm",
1852 "/dev/tty", "rwm",
1853 "/dev/net/tun", "rwm",
1854 /* Allow the container
1855 * access to ptys. However,
1856 * do not permit the
1857 * container to ever create
1858 * these device nodes. */
1859 "/dev/pts/ptmx", "rw",
1860 "char-pts", "rw");
1861 if (r < 0)
1862 return bus_log_create_error(r);
1863
1864 for (j = 0; j < arg_n_custom_mounts; j++) {
1865 CustomMount *cm = &arg_custom_mounts[j];
1866
1867 if (cm->type != CUSTOM_MOUNT_BIND)
1868 continue;
1869
1870 r = is_device_node(cm->source);
1871 if (r < 0)
1872 return log_error_errno(r, "Failed to stat %s: %m", cm->source);
1873
1874 if (r) {
1875 r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 1,
1876 cm->source, cm->read_only ? "r" : "rw");
1877 if (r < 0)
1878 return log_error_errno(r, "Failed to append message arguments: %m");
1879 }
1880 }
1881
1882 if (arg_kill_signal != 0) {
1883 r = sd_bus_message_append(m, "(sv)", "KillSignal", "i", arg_kill_signal);
1884 if (r < 0)
1885 return bus_log_create_error(r);
1886
1887 r = sd_bus_message_append(m, "(sv)", "KillMode", "s", "mixed");
1888 if (r < 0)
1889 return bus_log_create_error(r);
1890 }
1891
1892 STRV_FOREACH(i, arg_property) {
1893 r = sd_bus_message_open_container(m, 'r', "sv");
1894 if (r < 0)
1895 return bus_log_create_error(r);
1896
1897 r = bus_append_unit_property_assignment(m, *i);
1898 if (r < 0)
1899 return r;
1900
1901 r = sd_bus_message_close_container(m);
1902 if (r < 0)
1903 return bus_log_create_error(r);
1904 }
1905
1906 r = sd_bus_message_close_container(m);
1907 if (r < 0)
1908 return bus_log_create_error(r);
1909
1910 r = sd_bus_call(bus, m, 0, &error, NULL);
1911 }
1912
1913 if (r < 0) {
1914 log_error("Failed to register machine: %s", bus_error_message(&error, r));
1915 return r;
1916 }
1917
1918 return 0;
1919 }
1920
1921 static int terminate_machine(pid_t pid) {
1922 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1923 _cleanup_bus_message_unref_ sd_bus_message *reply = NULL;
1924 _cleanup_bus_flush_close_unref_ sd_bus *bus = NULL;
1925 const char *path;
1926 int r;
1927
1928 if (!arg_register)
1929 return 0;
1930
1931 /* If we are reusing the unit, then just exit, systemd will do
1932 * the right thing when we exit. */
1933 if (arg_keep_unit)
1934 return 0;
1935
1936 r = sd_bus_default_system(&bus);
1937 if (r < 0)
1938 return log_error_errno(r, "Failed to open system bus: %m");
1939
1940 r = sd_bus_call_method(
1941 bus,
1942 "org.freedesktop.machine1",
1943 "/org/freedesktop/machine1",
1944 "org.freedesktop.machine1.Manager",
1945 "GetMachineByPID",
1946 &error,
1947 &reply,
1948 "u",
1949 (uint32_t) pid);
1950 if (r < 0) {
1951 /* Note that the machine might already have been
1952 * cleaned up automatically, hence don't consider it a
1953 * failure if we cannot get the machine object. */
1954 log_debug("Failed to get machine: %s", bus_error_message(&error, r));
1955 return 0;
1956 }
1957
1958 r = sd_bus_message_read(reply, "o", &path);
1959 if (r < 0)
1960 return bus_log_parse_error(r);
1961
1962 r = sd_bus_call_method(
1963 bus,
1964 "org.freedesktop.machine1",
1965 path,
1966 "org.freedesktop.machine1.Machine",
1967 "Terminate",
1968 &error,
1969 NULL,
1970 NULL);
1971 if (r < 0) {
1972 log_debug("Failed to terminate machine: %s", bus_error_message(&error, r));
1973 return 0;
1974 }
1975
1976 return 0;
1977 }
1978
1979 static int reset_audit_loginuid(void) {
1980 _cleanup_free_ char *p = NULL;
1981 int r;
1982
1983 if (arg_share_system)
1984 return 0;
1985
1986 r = read_one_line_file("/proc/self/loginuid", &p);
1987 if (r == -ENOENT)
1988 return 0;
1989 if (r < 0)
1990 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
1991
1992 /* Already reset? */
1993 if (streq(p, "4294967295"))
1994 return 0;
1995
1996 r = write_string_file("/proc/self/loginuid", "4294967295", 0);
1997 if (r < 0) {
1998 log_error_errno(r,
1999 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2000 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2001 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2002 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2003 "using systemd-nspawn. Sleeping for 5s... (%m)");
2004
2005 sleep(5);
2006 }
2007
2008 return 0;
2009 }
2010
2011 #define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
2012 #define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
2013 #define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
2014
2015 static int generate_mac(struct ether_addr *mac, sd_id128_t hash_key, uint64_t idx) {
2016 uint8_t result[8];
2017 size_t l, sz;
2018 uint8_t *v, *i;
2019 int r;
2020
2021 l = strlen(arg_machine);
2022 sz = sizeof(sd_id128_t) + l;
2023 if (idx > 0)
2024 sz += sizeof(idx);
2025
2026 v = alloca(sz);
2027
2028 /* fetch some persistent data unique to the host */
2029 r = sd_id128_get_machine((sd_id128_t*) v);
2030 if (r < 0)
2031 return r;
2032
2033 /* combine with some data unique (on this host) to this
2034 * container instance */
2035 i = mempcpy(v + sizeof(sd_id128_t), arg_machine, l);
2036 if (idx > 0) {
2037 idx = htole64(idx);
2038 memcpy(i, &idx, sizeof(idx));
2039 }
2040
2041 /* Let's hash the host machine ID plus the container name. We
2042 * use a fixed, but originally randomly created hash key here. */
2043 siphash24(result, v, sz, hash_key.bytes);
2044
2045 assert_cc(ETH_ALEN <= sizeof(result));
2046 memcpy(mac->ether_addr_octet, result, ETH_ALEN);
2047
2048 /* see eth_random_addr in the kernel */
2049 mac->ether_addr_octet[0] &= 0xfe; /* clear multicast bit */
2050 mac->ether_addr_octet[0] |= 0x02; /* set local assignment bit (IEEE802) */
2051
2052 return 0;
2053 }
2054
2055 static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ], int *ifi) {
2056 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2057 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2058 struct ether_addr mac_host, mac_container;
2059 int r, i;
2060
2061 if (!arg_private_network)
2062 return 0;
2063
2064 if (!arg_network_veth)
2065 return 0;
2066
2067 /* Use two different interface name prefixes depending whether
2068 * we are in bridge mode or not. */
2069 snprintf(iface_name, IFNAMSIZ - 1, "%s-%s",
2070 arg_network_bridge ? "vb" : "ve", arg_machine);
2071
2072 r = generate_mac(&mac_container, CONTAINER_HASH_KEY, 0);
2073 if (r < 0)
2074 return log_error_errno(r, "Failed to generate predictable MAC address for container side: %m");
2075
2076 r = generate_mac(&mac_host, HOST_HASH_KEY, 0);
2077 if (r < 0)
2078 return log_error_errno(r, "Failed to generate predictable MAC address for host side: %m");
2079
2080 r = sd_netlink_open(&rtnl);
2081 if (r < 0)
2082 return log_error_errno(r, "Failed to connect to netlink: %m");
2083
2084 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2085 if (r < 0)
2086 return log_error_errno(r, "Failed to allocate netlink message: %m");
2087
2088 r = sd_netlink_message_append_string(m, IFLA_IFNAME, iface_name);
2089 if (r < 0)
2090 return log_error_errno(r, "Failed to add netlink interface name: %m");
2091
2092 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac_host);
2093 if (r < 0)
2094 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2095
2096 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
2097 if (r < 0)
2098 return log_error_errno(r, "Failed to open netlink container: %m");
2099
2100 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "veth");
2101 if (r < 0)
2102 return log_error_errno(r, "Failed to open netlink container: %m");
2103
2104 r = sd_netlink_message_open_container(m, VETH_INFO_PEER);
2105 if (r < 0)
2106 return log_error_errno(r, "Failed to open netlink container: %m");
2107
2108 r = sd_netlink_message_append_string(m, IFLA_IFNAME, "host0");
2109 if (r < 0)
2110 return log_error_errno(r, "Failed to add netlink interface name: %m");
2111
2112 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac_container);
2113 if (r < 0)
2114 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2115
2116 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
2117 if (r < 0)
2118 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2119
2120 r = sd_netlink_message_close_container(m);
2121 if (r < 0)
2122 return log_error_errno(r, "Failed to close netlink container: %m");
2123
2124 r = sd_netlink_message_close_container(m);
2125 if (r < 0)
2126 return log_error_errno(r, "Failed to close netlink container: %m");
2127
2128 r = sd_netlink_message_close_container(m);
2129 if (r < 0)
2130 return log_error_errno(r, "Failed to close netlink container: %m");
2131
2132 r = sd_netlink_call(rtnl, m, 0, NULL);
2133 if (r < 0)
2134 return log_error_errno(r, "Failed to add new veth interfaces (host0, %s): %m", iface_name);
2135
2136 i = (int) if_nametoindex(iface_name);
2137 if (i <= 0)
2138 return log_error_errno(errno, "Failed to resolve interface %s: %m", iface_name);
2139
2140 *ifi = i;
2141
2142 return 0;
2143 }
2144
2145 static int setup_bridge(const char veth_name[], int *ifi) {
2146 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2147 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2148 int r, bridge;
2149
2150 if (!arg_private_network)
2151 return 0;
2152
2153 if (!arg_network_veth)
2154 return 0;
2155
2156 if (!arg_network_bridge)
2157 return 0;
2158
2159 bridge = (int) if_nametoindex(arg_network_bridge);
2160 if (bridge <= 0)
2161 return log_error_errno(errno, "Failed to resolve interface %s: %m", arg_network_bridge);
2162
2163 *ifi = bridge;
2164
2165 r = sd_netlink_open(&rtnl);
2166 if (r < 0)
2167 return log_error_errno(r, "Failed to connect to netlink: %m");
2168
2169 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, 0);
2170 if (r < 0)
2171 return log_error_errno(r, "Failed to allocate netlink message: %m");
2172
2173 r = sd_rtnl_message_link_set_flags(m, IFF_UP, IFF_UP);
2174 if (r < 0)
2175 return log_error_errno(r, "Failed to set IFF_UP flag: %m");
2176
2177 r = sd_netlink_message_append_string(m, IFLA_IFNAME, veth_name);
2178 if (r < 0)
2179 return log_error_errno(r, "Failed to add netlink interface name field: %m");
2180
2181 r = sd_netlink_message_append_u32(m, IFLA_MASTER, bridge);
2182 if (r < 0)
2183 return log_error_errno(r, "Failed to add netlink master field: %m");
2184
2185 r = sd_netlink_call(rtnl, m, 0, NULL);
2186 if (r < 0)
2187 return log_error_errno(r, "Failed to add veth interface to bridge: %m");
2188
2189 return 0;
2190 }
2191
2192 static int parse_interface(struct udev *udev, const char *name) {
2193 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
2194 char ifi_str[2 + DECIMAL_STR_MAX(int)];
2195 int ifi;
2196
2197 ifi = (int) if_nametoindex(name);
2198 if (ifi <= 0)
2199 return log_error_errno(errno, "Failed to resolve interface %s: %m", name);
2200
2201 sprintf(ifi_str, "n%i", ifi);
2202 d = udev_device_new_from_device_id(udev, ifi_str);
2203 if (!d)
2204 return log_error_errno(errno, "Failed to get udev device for interface %s: %m", name);
2205
2206 if (udev_device_get_is_initialized(d) <= 0) {
2207 log_error("Network interface %s is not initialized yet.", name);
2208 return -EBUSY;
2209 }
2210
2211 return ifi;
2212 }
2213
2214 static int move_network_interfaces(pid_t pid) {
2215 _cleanup_udev_unref_ struct udev *udev = NULL;
2216 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2217 char **i;
2218 int r;
2219
2220 if (!arg_private_network)
2221 return 0;
2222
2223 if (strv_isempty(arg_network_interfaces))
2224 return 0;
2225
2226 r = sd_netlink_open(&rtnl);
2227 if (r < 0)
2228 return log_error_errno(r, "Failed to connect to netlink: %m");
2229
2230 udev = udev_new();
2231 if (!udev) {
2232 log_error("Failed to connect to udev.");
2233 return -ENOMEM;
2234 }
2235
2236 STRV_FOREACH(i, arg_network_interfaces) {
2237 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2238 int ifi;
2239
2240 ifi = parse_interface(udev, *i);
2241 if (ifi < 0)
2242 return ifi;
2243
2244 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, ifi);
2245 if (r < 0)
2246 return log_error_errno(r, "Failed to allocate netlink message: %m");
2247
2248 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
2249 if (r < 0)
2250 return log_error_errno(r, "Failed to append namespace PID to netlink message: %m");
2251
2252 r = sd_netlink_call(rtnl, m, 0, NULL);
2253 if (r < 0)
2254 return log_error_errno(r, "Failed to move interface %s to namespace: %m", *i);
2255 }
2256
2257 return 0;
2258 }
2259
2260 static int setup_macvlan(pid_t pid) {
2261 _cleanup_udev_unref_ struct udev *udev = NULL;
2262 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2263 unsigned idx = 0;
2264 char **i;
2265 int r;
2266
2267 if (!arg_private_network)
2268 return 0;
2269
2270 if (strv_isempty(arg_network_macvlan))
2271 return 0;
2272
2273 r = sd_netlink_open(&rtnl);
2274 if (r < 0)
2275 return log_error_errno(r, "Failed to connect to netlink: %m");
2276
2277 udev = udev_new();
2278 if (!udev) {
2279 log_error("Failed to connect to udev.");
2280 return -ENOMEM;
2281 }
2282
2283 STRV_FOREACH(i, arg_network_macvlan) {
2284 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2285 _cleanup_free_ char *n = NULL;
2286 struct ether_addr mac;
2287 int ifi;
2288
2289 ifi = parse_interface(udev, *i);
2290 if (ifi < 0)
2291 return ifi;
2292
2293 r = generate_mac(&mac, MACVLAN_HASH_KEY, idx++);
2294 if (r < 0)
2295 return log_error_errno(r, "Failed to create MACVLAN MAC address: %m");
2296
2297 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2298 if (r < 0)
2299 return log_error_errno(r, "Failed to allocate netlink message: %m");
2300
2301 r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
2302 if (r < 0)
2303 return log_error_errno(r, "Failed to add netlink interface index: %m");
2304
2305 n = strappend("mv-", *i);
2306 if (!n)
2307 return log_oom();
2308
2309 strshorten(n, IFNAMSIZ-1);
2310
2311 r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
2312 if (r < 0)
2313 return log_error_errno(r, "Failed to add netlink interface name: %m");
2314
2315 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac);
2316 if (r < 0)
2317 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2318
2319 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
2320 if (r < 0)
2321 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2322
2323 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
2324 if (r < 0)
2325 return log_error_errno(r, "Failed to open netlink container: %m");
2326
2327 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "macvlan");
2328 if (r < 0)
2329 return log_error_errno(r, "Failed to open netlink container: %m");
2330
2331 r = sd_netlink_message_append_u32(m, IFLA_MACVLAN_MODE, MACVLAN_MODE_BRIDGE);
2332 if (r < 0)
2333 return log_error_errno(r, "Failed to append macvlan mode: %m");
2334
2335 r = sd_netlink_message_close_container(m);
2336 if (r < 0)
2337 return log_error_errno(r, "Failed to close netlink container: %m");
2338
2339 r = sd_netlink_message_close_container(m);
2340 if (r < 0)
2341 return log_error_errno(r, "Failed to close netlink container: %m");
2342
2343 r = sd_netlink_call(rtnl, m, 0, NULL);
2344 if (r < 0)
2345 return log_error_errno(r, "Failed to add new macvlan interfaces: %m");
2346 }
2347
2348 return 0;
2349 }
2350
2351 static int setup_ipvlan(pid_t pid) {
2352 _cleanup_udev_unref_ struct udev *udev = NULL;
2353 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2354 char **i;
2355 int r;
2356
2357 if (!arg_private_network)
2358 return 0;
2359
2360 if (strv_isempty(arg_network_ipvlan))
2361 return 0;
2362
2363 r = sd_netlink_open(&rtnl);
2364 if (r < 0)
2365 return log_error_errno(r, "Failed to connect to netlink: %m");
2366
2367 udev = udev_new();
2368 if (!udev) {
2369 log_error("Failed to connect to udev.");
2370 return -ENOMEM;
2371 }
2372
2373 STRV_FOREACH(i, arg_network_ipvlan) {
2374 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2375 _cleanup_free_ char *n = NULL;
2376 int ifi;
2377
2378 ifi = parse_interface(udev, *i);
2379 if (ifi < 0)
2380 return ifi;
2381
2382 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2383 if (r < 0)
2384 return log_error_errno(r, "Failed to allocate netlink message: %m");
2385
2386 r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
2387 if (r < 0)
2388 return log_error_errno(r, "Failed to add netlink interface index: %m");
2389
2390 n = strappend("iv-", *i);
2391 if (!n)
2392 return log_oom();
2393
2394 strshorten(n, IFNAMSIZ-1);
2395
2396 r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
2397 if (r < 0)
2398 return log_error_errno(r, "Failed to add netlink interface name: %m");
2399
2400 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
2401 if (r < 0)
2402 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2403
2404 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
2405 if (r < 0)
2406 return log_error_errno(r, "Failed to open netlink container: %m");
2407
2408 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "ipvlan");
2409 if (r < 0)
2410 return log_error_errno(r, "Failed to open netlink container: %m");
2411
2412 r = sd_netlink_message_append_u16(m, IFLA_IPVLAN_MODE, IPVLAN_MODE_L2);
2413 if (r < 0)
2414 return log_error_errno(r, "Failed to add ipvlan mode: %m");
2415
2416 r = sd_netlink_message_close_container(m);
2417 if (r < 0)
2418 return log_error_errno(r, "Failed to close netlink container: %m");
2419
2420 r = sd_netlink_message_close_container(m);
2421 if (r < 0)
2422 return log_error_errno(r, "Failed to close netlink container: %m");
2423
2424 r = sd_netlink_call(rtnl, m, 0, NULL);
2425 if (r < 0)
2426 return log_error_errno(r, "Failed to add new ipvlan interfaces: %m");
2427 }
2428
2429 return 0;
2430 }
2431
2432 static int setup_seccomp(void) {
2433
2434 #ifdef HAVE_SECCOMP
2435 static const struct {
2436 uint64_t capability;
2437 int syscall_num;
2438 } blacklist[] = {
2439 { CAP_SYS_RAWIO, SCMP_SYS(iopl) },
2440 { CAP_SYS_RAWIO, SCMP_SYS(ioperm) },
2441 { CAP_SYS_BOOT, SCMP_SYS(kexec_load) },
2442 { CAP_SYS_ADMIN, SCMP_SYS(swapon) },
2443 { CAP_SYS_ADMIN, SCMP_SYS(swapoff) },
2444 { CAP_SYS_ADMIN, SCMP_SYS(open_by_handle_at) },
2445 { CAP_SYS_MODULE, SCMP_SYS(init_module) },
2446 { CAP_SYS_MODULE, SCMP_SYS(finit_module) },
2447 { CAP_SYS_MODULE, SCMP_SYS(delete_module) },
2448 { CAP_SYSLOG, SCMP_SYS(syslog) },
2449 };
2450
2451 scmp_filter_ctx seccomp;
2452 unsigned i;
2453 int r;
2454
2455 seccomp = seccomp_init(SCMP_ACT_ALLOW);
2456 if (!seccomp)
2457 return log_oom();
2458
2459 r = seccomp_add_secondary_archs(seccomp);
2460 if (r < 0) {
2461 log_error_errno(r, "Failed to add secondary archs to seccomp filter: %m");
2462 goto finish;
2463 }
2464
2465 for (i = 0; i < ELEMENTSOF(blacklist); i++) {
2466 if (arg_retain & (1ULL << blacklist[i].capability))
2467 continue;
2468
2469 r = seccomp_rule_add(seccomp, SCMP_ACT_ERRNO(EPERM), blacklist[i].syscall_num, 0);
2470 if (r == -EFAULT)
2471 continue; /* unknown syscall */
2472 if (r < 0) {
2473 log_error_errno(r, "Failed to block syscall: %m");
2474 goto finish;
2475 }
2476 }
2477
2478
2479 /*
2480 Audit is broken in containers, much of the userspace audit
2481 hookup will fail if running inside a container. We don't
2482 care and just turn off creation of audit sockets.
2483
2484 This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail
2485 with EAFNOSUPPORT which audit userspace uses as indication
2486 that audit is disabled in the kernel.
2487 */
2488
2489 r = seccomp_rule_add(
2490 seccomp,
2491 SCMP_ACT_ERRNO(EAFNOSUPPORT),
2492 SCMP_SYS(socket),
2493 2,
2494 SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
2495 SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
2496 if (r < 0) {
2497 log_error_errno(r, "Failed to add audit seccomp rule: %m");
2498 goto finish;
2499 }
2500
2501 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
2502 if (r < 0) {
2503 log_error_errno(r, "Failed to unset NO_NEW_PRIVS: %m");
2504 goto finish;
2505 }
2506
2507 r = seccomp_load(seccomp);
2508 if (r == -EINVAL) {
2509 log_debug_errno(r, "Kernel is probably not configured with CONFIG_SECCOMP. Disabling seccomp audit filter: %m");
2510 r = 0;
2511 goto finish;
2512 }
2513 if (r < 0) {
2514 log_error_errno(r, "Failed to install seccomp audit filter: %m");
2515 goto finish;
2516 }
2517
2518 finish:
2519 seccomp_release(seccomp);
2520 return r;
2521 #else
2522 return 0;
2523 #endif
2524
2525 }
2526
2527 static int setup_propagate(const char *root) {
2528 const char *p, *q;
2529
2530 (void) mkdir_p("/run/systemd/nspawn/", 0755);
2531 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
2532 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
2533 (void) mkdir_p(p, 0600);
2534
2535 if (userns_mkdir(root, "/run/systemd", 0755, 0, 0) < 0)
2536 return log_error_errno(errno, "Failed to create /run/systemd: %m");
2537
2538 if (userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0) < 0)
2539 return log_error_errno(errno, "Failed to create /run/systemd/nspawn: %m");
2540
2541 if (userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0) < 0)
2542 return log_error_errno(errno, "Failed to create /run/systemd/nspawn/incoming: %m");
2543
2544 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
2545 if (mount(p, q, NULL, MS_BIND, NULL) < 0)
2546 return log_error_errno(errno, "Failed to install propagation bind mount.");
2547
2548 if (mount(NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0)
2549 return log_error_errno(errno, "Failed to make propagation mount read-only");
2550
2551 return 0;
2552 }
2553
2554 static int setup_image(char **device_path, int *loop_nr) {
2555 struct loop_info64 info = {
2556 .lo_flags = LO_FLAGS_AUTOCLEAR|LO_FLAGS_PARTSCAN
2557 };
2558 _cleanup_close_ int fd = -1, control = -1, loop = -1;
2559 _cleanup_free_ char* loopdev = NULL;
2560 struct stat st;
2561 int r, nr;
2562
2563 assert(device_path);
2564 assert(loop_nr);
2565 assert(arg_image);
2566
2567 fd = open(arg_image, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
2568 if (fd < 0)
2569 return log_error_errno(errno, "Failed to open %s: %m", arg_image);
2570
2571 if (fstat(fd, &st) < 0)
2572 return log_error_errno(errno, "Failed to stat %s: %m", arg_image);
2573
2574 if (S_ISBLK(st.st_mode)) {
2575 char *p;
2576
2577 p = strdup(arg_image);
2578 if (!p)
2579 return log_oom();
2580
2581 *device_path = p;
2582
2583 *loop_nr = -1;
2584
2585 r = fd;
2586 fd = -1;
2587
2588 return r;
2589 }
2590
2591 if (!S_ISREG(st.st_mode)) {
2592 log_error_errno(errno, "%s is not a regular file or block device: %m", arg_image);
2593 return -EINVAL;
2594 }
2595
2596 control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
2597 if (control < 0)
2598 return log_error_errno(errno, "Failed to open /dev/loop-control: %m");
2599
2600 nr = ioctl(control, LOOP_CTL_GET_FREE);
2601 if (nr < 0)
2602 return log_error_errno(errno, "Failed to allocate loop device: %m");
2603
2604 if (asprintf(&loopdev, "/dev/loop%i", nr) < 0)
2605 return log_oom();
2606
2607 loop = open(loopdev, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
2608 if (loop < 0)
2609 return log_error_errno(errno, "Failed to open loop device %s: %m", loopdev);
2610
2611 if (ioctl(loop, LOOP_SET_FD, fd) < 0)
2612 return log_error_errno(errno, "Failed to set loopback file descriptor on %s: %m", loopdev);
2613
2614 if (arg_read_only)
2615 info.lo_flags |= LO_FLAGS_READ_ONLY;
2616
2617 if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0)
2618 return log_error_errno(errno, "Failed to set loopback settings on %s: %m", loopdev);
2619
2620 *device_path = loopdev;
2621 loopdev = NULL;
2622
2623 *loop_nr = nr;
2624
2625 r = loop;
2626 loop = -1;
2627
2628 return r;
2629 }
2630
2631 #define PARTITION_TABLE_BLURB \
2632 "Note that the disk image needs to either contain only a single MBR partition of\n" \
2633 "type 0x83 that is marked bootable, or a single GPT partition of type " \
2634 "0FC63DAF-8483-4772-8E79-3D69D8477DE4 or follow\n" \
2635 " http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n" \
2636 "to be bootable with systemd-nspawn."
2637
2638 static int dissect_image(
2639 int fd,
2640 char **root_device, bool *root_device_rw,
2641 char **home_device, bool *home_device_rw,
2642 char **srv_device, bool *srv_device_rw,
2643 bool *secondary) {
2644
2645 #ifdef HAVE_BLKID
2646 int home_nr = -1, srv_nr = -1;
2647 #ifdef GPT_ROOT_NATIVE
2648 int root_nr = -1;
2649 #endif
2650 #ifdef GPT_ROOT_SECONDARY
2651 int secondary_root_nr = -1;
2652 #endif
2653 _cleanup_free_ char *home = NULL, *root = NULL, *secondary_root = NULL, *srv = NULL, *generic = NULL;
2654 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
2655 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
2656 _cleanup_blkid_free_probe_ blkid_probe b = NULL;
2657 _cleanup_udev_unref_ struct udev *udev = NULL;
2658 struct udev_list_entry *first, *item;
2659 bool home_rw = true, root_rw = true, secondary_root_rw = true, srv_rw = true, generic_rw = true;
2660 bool is_gpt, is_mbr, multiple_generic = false;
2661 const char *pttype = NULL;
2662 blkid_partlist pl;
2663 struct stat st;
2664 unsigned i;
2665 int r;
2666
2667 assert(fd >= 0);
2668 assert(root_device);
2669 assert(home_device);
2670 assert(srv_device);
2671 assert(secondary);
2672 assert(arg_image);
2673
2674 b = blkid_new_probe();
2675 if (!b)
2676 return log_oom();
2677
2678 errno = 0;
2679 r = blkid_probe_set_device(b, fd, 0, 0);
2680 if (r != 0) {
2681 if (errno == 0)
2682 return log_oom();
2683
2684 log_error_errno(errno, "Failed to set device on blkid probe: %m");
2685 return -errno;
2686 }
2687
2688 blkid_probe_enable_partitions(b, 1);
2689 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
2690
2691 errno = 0;
2692 r = blkid_do_safeprobe(b);
2693 if (r == -2 || r == 1) {
2694 log_error("Failed to identify any partition table on\n"
2695 " %s\n"
2696 PARTITION_TABLE_BLURB, arg_image);
2697 return -EINVAL;
2698 } else if (r != 0) {
2699 if (errno == 0)
2700 errno = EIO;
2701 log_error_errno(errno, "Failed to probe: %m");
2702 return -errno;
2703 }
2704
2705 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
2706
2707 is_gpt = streq_ptr(pttype, "gpt");
2708 is_mbr = streq_ptr(pttype, "dos");
2709
2710 if (!is_gpt && !is_mbr) {
2711 log_error("No GPT or MBR partition table discovered on\n"
2712 " %s\n"
2713 PARTITION_TABLE_BLURB, arg_image);
2714 return -EINVAL;
2715 }
2716
2717 errno = 0;
2718 pl = blkid_probe_get_partitions(b);
2719 if (!pl) {
2720 if (errno == 0)
2721 return log_oom();
2722
2723 log_error("Failed to list partitions of %s", arg_image);
2724 return -errno;
2725 }
2726
2727 udev = udev_new();
2728 if (!udev)
2729 return log_oom();
2730
2731 if (fstat(fd, &st) < 0)
2732 return log_error_errno(errno, "Failed to stat block device: %m");
2733
2734 d = udev_device_new_from_devnum(udev, 'b', st.st_rdev);
2735 if (!d)
2736 return log_oom();
2737
2738 for (i = 0;; i++) {
2739 int n, m;
2740
2741 if (i >= 10) {
2742 log_error("Kernel partitions never appeared.");
2743 return -ENXIO;
2744 }
2745
2746 e = udev_enumerate_new(udev);
2747 if (!e)
2748 return log_oom();
2749
2750 r = udev_enumerate_add_match_parent(e, d);
2751 if (r < 0)
2752 return log_oom();
2753
2754 r = udev_enumerate_scan_devices(e);
2755 if (r < 0)
2756 return log_error_errno(r, "Failed to scan for partition devices of %s: %m", arg_image);
2757
2758 /* Count the partitions enumerated by the kernel */
2759 n = 0;
2760 first = udev_enumerate_get_list_entry(e);
2761 udev_list_entry_foreach(item, first)
2762 n++;
2763
2764 /* Count the partitions enumerated by blkid */
2765 m = blkid_partlist_numof_partitions(pl);
2766 if (n == m + 1)
2767 break;
2768 if (n > m + 1) {
2769 log_error("blkid and kernel partition list do not match.");
2770 return -EIO;
2771 }
2772 if (n < m + 1) {
2773 unsigned j;
2774
2775 /* The kernel has probed fewer partitions than
2776 * blkid? Maybe the kernel prober is still
2777 * running or it got EBUSY because udev
2778 * already opened the device. Let's reprobe
2779 * the device, which is a synchronous call
2780 * that waits until probing is complete. */
2781
2782 for (j = 0; j < 20; j++) {
2783
2784 r = ioctl(fd, BLKRRPART, 0);
2785 if (r < 0)
2786 r = -errno;
2787 if (r >= 0 || r != -EBUSY)
2788 break;
2789
2790 /* If something else has the device
2791 * open, such as an udev rule, the
2792 * ioctl will return EBUSY. Since
2793 * there's no way to wait until it
2794 * isn't busy anymore, let's just wait
2795 * a bit, and try again.
2796 *
2797 * This is really something they
2798 * should fix in the kernel! */
2799
2800 usleep(50 * USEC_PER_MSEC);
2801 }
2802
2803 if (r < 0)
2804 return log_error_errno(r, "Failed to reread partition table: %m");
2805 }
2806
2807 e = udev_enumerate_unref(e);
2808 }
2809
2810 first = udev_enumerate_get_list_entry(e);
2811 udev_list_entry_foreach(item, first) {
2812 _cleanup_udev_device_unref_ struct udev_device *q;
2813 const char *node;
2814 unsigned long long flags;
2815 blkid_partition pp;
2816 dev_t qn;
2817 int nr;
2818
2819 errno = 0;
2820 q = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
2821 if (!q) {
2822 if (!errno)
2823 errno = ENOMEM;
2824
2825 log_error_errno(errno, "Failed to get partition device of %s: %m", arg_image);
2826 return -errno;
2827 }
2828
2829 qn = udev_device_get_devnum(q);
2830 if (major(qn) == 0)
2831 continue;
2832
2833 if (st.st_rdev == qn)
2834 continue;
2835
2836 node = udev_device_get_devnode(q);
2837 if (!node)
2838 continue;
2839
2840 pp = blkid_partlist_devno_to_partition(pl, qn);
2841 if (!pp)
2842 continue;
2843
2844 flags = blkid_partition_get_flags(pp);
2845
2846 nr = blkid_partition_get_partno(pp);
2847 if (nr < 0)
2848 continue;
2849
2850 if (is_gpt) {
2851 sd_id128_t type_id;
2852 const char *stype;
2853
2854 if (flags & GPT_FLAG_NO_AUTO)
2855 continue;
2856
2857 stype = blkid_partition_get_type_string(pp);
2858 if (!stype)
2859 continue;
2860
2861 if (sd_id128_from_string(stype, &type_id) < 0)
2862 continue;
2863
2864 if (sd_id128_equal(type_id, GPT_HOME)) {
2865
2866 if (home && nr >= home_nr)
2867 continue;
2868
2869 home_nr = nr;
2870 home_rw = !(flags & GPT_FLAG_READ_ONLY);
2871
2872 r = free_and_strdup(&home, node);
2873 if (r < 0)
2874 return log_oom();
2875
2876 } else if (sd_id128_equal(type_id, GPT_SRV)) {
2877
2878 if (srv && nr >= srv_nr)
2879 continue;
2880
2881 srv_nr = nr;
2882 srv_rw = !(flags & GPT_FLAG_READ_ONLY);
2883
2884 r = free_and_strdup(&srv, node);
2885 if (r < 0)
2886 return log_oom();
2887 }
2888 #ifdef GPT_ROOT_NATIVE
2889 else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
2890
2891 if (root && nr >= root_nr)
2892 continue;
2893
2894 root_nr = nr;
2895 root_rw = !(flags & GPT_FLAG_READ_ONLY);
2896
2897 r = free_and_strdup(&root, node);
2898 if (r < 0)
2899 return log_oom();
2900 }
2901 #endif
2902 #ifdef GPT_ROOT_SECONDARY
2903 else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
2904
2905 if (secondary_root && nr >= secondary_root_nr)
2906 continue;
2907
2908 secondary_root_nr = nr;
2909 secondary_root_rw = !(flags & GPT_FLAG_READ_ONLY);
2910
2911 r = free_and_strdup(&secondary_root, node);
2912 if (r < 0)
2913 return log_oom();
2914 }
2915 #endif
2916 else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
2917
2918 if (generic)
2919 multiple_generic = true;
2920 else {
2921 generic_rw = !(flags & GPT_FLAG_READ_ONLY);
2922
2923 r = free_and_strdup(&generic, node);
2924 if (r < 0)
2925 return log_oom();
2926 }
2927 }
2928
2929 } else if (is_mbr) {
2930 int type;
2931
2932 if (flags != 0x80) /* Bootable flag */
2933 continue;
2934
2935 type = blkid_partition_get_type(pp);
2936 if (type != 0x83) /* Linux partition */
2937 continue;
2938
2939 if (generic)
2940 multiple_generic = true;
2941 else {
2942 generic_rw = true;
2943
2944 r = free_and_strdup(&root, node);
2945 if (r < 0)
2946 return log_oom();
2947 }
2948 }
2949 }
2950
2951 if (root) {
2952 *root_device = root;
2953 root = NULL;
2954
2955 *root_device_rw = root_rw;
2956 *secondary = false;
2957 } else if (secondary_root) {
2958 *root_device = secondary_root;
2959 secondary_root = NULL;
2960
2961 *root_device_rw = secondary_root_rw;
2962 *secondary = true;
2963 } else if (generic) {
2964
2965 /* There were no partitions with precise meanings
2966 * around, but we found generic partitions. In this
2967 * case, if there's only one, we can go ahead and boot
2968 * it, otherwise we bail out, because we really cannot
2969 * make any sense of it. */
2970
2971 if (multiple_generic) {
2972 log_error("Identified multiple bootable Linux partitions on\n"
2973 " %s\n"
2974 PARTITION_TABLE_BLURB, arg_image);
2975 return -EINVAL;
2976 }
2977
2978 *root_device = generic;
2979 generic = NULL;
2980
2981 *root_device_rw = generic_rw;
2982 *secondary = false;
2983 } else {
2984 log_error("Failed to identify root partition in disk image\n"
2985 " %s\n"
2986 PARTITION_TABLE_BLURB, arg_image);
2987 return -EINVAL;
2988 }
2989
2990 if (home) {
2991 *home_device = home;
2992 home = NULL;
2993
2994 *home_device_rw = home_rw;
2995 }
2996
2997 if (srv) {
2998 *srv_device = srv;
2999 srv = NULL;
3000
3001 *srv_device_rw = srv_rw;
3002 }
3003
3004 return 0;
3005 #else
3006 log_error("--image= is not supported, compiled without blkid support.");
3007 return -EOPNOTSUPP;
3008 #endif
3009 }
3010
3011 static int mount_device(const char *what, const char *where, const char *directory, bool rw) {
3012 #ifdef HAVE_BLKID
3013 _cleanup_blkid_free_probe_ blkid_probe b = NULL;
3014 const char *fstype, *p;
3015 int r;
3016
3017 assert(what);
3018 assert(where);
3019
3020 if (arg_read_only)
3021 rw = false;
3022
3023 if (directory)
3024 p = strjoina(where, directory);
3025 else
3026 p = where;
3027
3028 errno = 0;
3029 b = blkid_new_probe_from_filename(what);
3030 if (!b) {
3031 if (errno == 0)
3032 return log_oom();
3033 log_error_errno(errno, "Failed to allocate prober for %s: %m", what);
3034 return -errno;
3035 }
3036
3037 blkid_probe_enable_superblocks(b, 1);
3038 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
3039
3040 errno = 0;
3041 r = blkid_do_safeprobe(b);
3042 if (r == -1 || r == 1) {
3043 log_error("Cannot determine file system type of %s", what);
3044 return -EINVAL;
3045 } else if (r != 0) {
3046 if (errno == 0)
3047 errno = EIO;
3048 log_error_errno(errno, "Failed to probe %s: %m", what);
3049 return -errno;
3050 }
3051
3052 errno = 0;
3053 if (blkid_probe_lookup_value(b, "TYPE", &fstype, NULL) < 0) {
3054 if (errno == 0)
3055 errno = EINVAL;
3056 log_error("Failed to determine file system type of %s", what);
3057 return -errno;
3058 }
3059
3060 if (streq(fstype, "crypto_LUKS")) {
3061 log_error("nspawn currently does not support LUKS disk images.");
3062 return -EOPNOTSUPP;
3063 }
3064
3065 if (mount(what, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), NULL) < 0)
3066 return log_error_errno(errno, "Failed to mount %s: %m", what);
3067
3068 return 0;
3069 #else
3070 log_error("--image= is not supported, compiled without blkid support.");
3071 return -EOPNOTSUPP;
3072 #endif
3073 }
3074
3075 static int mount_devices(
3076 const char *where,
3077 const char *root_device, bool root_device_rw,
3078 const char *home_device, bool home_device_rw,
3079 const char *srv_device, bool srv_device_rw) {
3080 int r;
3081
3082 assert(where);
3083
3084 if (root_device) {
3085 r = mount_device(root_device, arg_directory, NULL, root_device_rw);
3086 if (r < 0)
3087 return log_error_errno(r, "Failed to mount root directory: %m");
3088 }
3089
3090 if (home_device) {
3091 r = mount_device(home_device, arg_directory, "/home", home_device_rw);
3092 if (r < 0)
3093 return log_error_errno(r, "Failed to mount home directory: %m");
3094 }
3095
3096 if (srv_device) {
3097 r = mount_device(srv_device, arg_directory, "/srv", srv_device_rw);
3098 if (r < 0)
3099 return log_error_errno(r, "Failed to mount server data directory: %m");
3100 }
3101
3102 return 0;
3103 }
3104
3105 static void loop_remove(int nr, int *image_fd) {
3106 _cleanup_close_ int control = -1;
3107 int r;
3108
3109 if (nr < 0)
3110 return;
3111
3112 if (image_fd && *image_fd >= 0) {
3113 r = ioctl(*image_fd, LOOP_CLR_FD);
3114 if (r < 0)
3115 log_debug_errno(errno, "Failed to close loop image: %m");
3116 *image_fd = safe_close(*image_fd);
3117 }
3118
3119 control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
3120 if (control < 0) {
3121 log_warning_errno(errno, "Failed to open /dev/loop-control: %m");
3122 return;
3123 }
3124
3125 r = ioctl(control, LOOP_CTL_REMOVE, nr);
3126 if (r < 0)
3127 log_debug_errno(errno, "Failed to remove loop %d: %m", nr);
3128 }
3129
3130 static int spawn_getent(const char *database, const char *key, pid_t *rpid) {
3131 int pipe_fds[2];
3132 pid_t pid;
3133
3134 assert(database);
3135 assert(key);
3136 assert(rpid);
3137
3138 if (pipe2(pipe_fds, O_CLOEXEC) < 0)
3139 return log_error_errno(errno, "Failed to allocate pipe: %m");
3140
3141 pid = fork();
3142 if (pid < 0)
3143 return log_error_errno(errno, "Failed to fork getent child: %m");
3144 else if (pid == 0) {
3145 int nullfd;
3146 char *empty_env = NULL;
3147
3148 if (dup3(pipe_fds[1], STDOUT_FILENO, 0) < 0)
3149 _exit(EXIT_FAILURE);
3150
3151 if (pipe_fds[0] > 2)
3152 safe_close(pipe_fds[0]);
3153 if (pipe_fds[1] > 2)
3154 safe_close(pipe_fds[1]);
3155
3156 nullfd = open("/dev/null", O_RDWR);
3157 if (nullfd < 0)
3158 _exit(EXIT_FAILURE);
3159
3160 if (dup3(nullfd, STDIN_FILENO, 0) < 0)
3161 _exit(EXIT_FAILURE);
3162
3163 if (dup3(nullfd, STDERR_FILENO, 0) < 0)
3164 _exit(EXIT_FAILURE);
3165
3166 if (nullfd > 2)
3167 safe_close(nullfd);
3168
3169 (void) reset_all_signal_handlers();
3170 (void) reset_signal_mask();
3171 close_all_fds(NULL, 0);
3172
3173 execle("/usr/bin/getent", "getent", database, key, NULL, &empty_env);
3174 execle("/bin/getent", "getent", database, key, NULL, &empty_env);
3175 _exit(EXIT_FAILURE);
3176 }
3177
3178 pipe_fds[1] = safe_close(pipe_fds[1]);
3179
3180 *rpid = pid;
3181
3182 return pipe_fds[0];
3183 }
3184
3185 static int change_uid_gid(char **_home) {
3186 char line[LINE_MAX], *x, *u, *g, *h;
3187 const char *word, *state;
3188 _cleanup_free_ uid_t *uids = NULL;
3189 _cleanup_free_ char *home = NULL;
3190 _cleanup_fclose_ FILE *f = NULL;
3191 _cleanup_close_ int fd = -1;
3192 unsigned n_uids = 0;
3193 size_t sz = 0, l;
3194 uid_t uid;
3195 gid_t gid;
3196 pid_t pid;
3197 int r;
3198
3199 assert(_home);
3200
3201 if (!arg_user || streq(arg_user, "root") || streq(arg_user, "0")) {
3202 /* Reset everything fully to 0, just in case */
3203
3204 r = reset_uid_gid();
3205 if (r < 0)
3206 return log_error_errno(r, "Failed to become root: %m");
3207
3208 *_home = NULL;
3209 return 0;
3210 }
3211
3212 /* First, get user credentials */
3213 fd = spawn_getent("passwd", arg_user, &pid);
3214 if (fd < 0)
3215 return fd;
3216
3217 f = fdopen(fd, "r");
3218 if (!f)
3219 return log_oom();
3220 fd = -1;
3221
3222 if (!fgets(line, sizeof(line), f)) {
3223
3224 if (!ferror(f)) {
3225 log_error("Failed to resolve user %s.", arg_user);
3226 return -ESRCH;
3227 }
3228
3229 log_error_errno(errno, "Failed to read from getent: %m");
3230 return -errno;
3231 }
3232
3233 truncate_nl(line);
3234
3235 wait_for_terminate_and_warn("getent passwd", pid, true);
3236
3237 x = strchr(line, ':');
3238 if (!x) {
3239 log_error("/etc/passwd entry has invalid user field.");
3240 return -EIO;
3241 }
3242
3243 u = strchr(x+1, ':');
3244 if (!u) {
3245 log_error("/etc/passwd entry has invalid password field.");
3246 return -EIO;
3247 }
3248
3249 u++;
3250 g = strchr(u, ':');
3251 if (!g) {
3252 log_error("/etc/passwd entry has invalid UID field.");
3253 return -EIO;
3254 }
3255
3256 *g = 0;
3257 g++;
3258 x = strchr(g, ':');
3259 if (!x) {
3260 log_error("/etc/passwd entry has invalid GID field.");
3261 return -EIO;
3262 }
3263
3264 *x = 0;
3265 h = strchr(x+1, ':');
3266 if (!h) {
3267 log_error("/etc/passwd entry has invalid GECOS field.");
3268 return -EIO;
3269 }
3270
3271 h++;
3272 x = strchr(h, ':');
3273 if (!x) {
3274 log_error("/etc/passwd entry has invalid home directory field.");
3275 return -EIO;
3276 }
3277
3278 *x = 0;
3279
3280 r = parse_uid(u, &uid);
3281 if (r < 0) {
3282 log_error("Failed to parse UID of user.");
3283 return -EIO;
3284 }
3285
3286 r = parse_gid(g, &gid);
3287 if (r < 0) {
3288 log_error("Failed to parse GID of user.");
3289 return -EIO;
3290 }
3291
3292 home = strdup(h);
3293 if (!home)
3294 return log_oom();
3295
3296 /* Second, get group memberships */
3297 fd = spawn_getent("initgroups", arg_user, &pid);
3298 if (fd < 0)
3299 return fd;
3300
3301 fclose(f);
3302 f = fdopen(fd, "r");
3303 if (!f)
3304 return log_oom();
3305 fd = -1;
3306
3307 if (!fgets(line, sizeof(line), f)) {
3308 if (!ferror(f)) {
3309 log_error("Failed to resolve user %s.", arg_user);
3310 return -ESRCH;
3311 }
3312
3313 log_error_errno(errno, "Failed to read from getent: %m");
3314 return -errno;
3315 }
3316
3317 truncate_nl(line);
3318
3319 wait_for_terminate_and_warn("getent initgroups", pid, true);
3320
3321 /* Skip over the username and subsequent separator whitespace */
3322 x = line;
3323 x += strcspn(x, WHITESPACE);
3324 x += strspn(x, WHITESPACE);
3325
3326 FOREACH_WORD(word, l, x, state) {
3327 char c[l+1];
3328
3329 memcpy(c, word, l);
3330 c[l] = 0;
3331
3332 if (!GREEDY_REALLOC(uids, sz, n_uids+1))
3333 return log_oom();
3334
3335 r = parse_uid(c, &uids[n_uids++]);
3336 if (r < 0) {
3337 log_error("Failed to parse group data from getent.");
3338 return -EIO;
3339 }
3340 }
3341
3342 r = mkdir_parents(home, 0775);
3343 if (r < 0)
3344 return log_error_errno(r, "Failed to make home root directory: %m");
3345
3346 r = mkdir_safe(home, 0755, uid, gid);
3347 if (r < 0 && r != -EEXIST)
3348 return log_error_errno(r, "Failed to make home directory: %m");
3349
3350 (void) fchown(STDIN_FILENO, uid, gid);
3351 (void) fchown(STDOUT_FILENO, uid, gid);
3352 (void) fchown(STDERR_FILENO, uid, gid);
3353
3354 if (setgroups(n_uids, uids) < 0)
3355 return log_error_errno(errno, "Failed to set auxiliary groups: %m");
3356
3357 if (setresgid(gid, gid, gid) < 0)
3358 return log_error_errno(errno, "setregid() failed: %m");
3359
3360 if (setresuid(uid, uid, uid) < 0)
3361 return log_error_errno(errno, "setreuid() failed: %m");
3362
3363 if (_home) {
3364 *_home = home;
3365 home = NULL;
3366 }
3367
3368 return 0;
3369 }
3370
3371 /*
3372 * Return values:
3373 * < 0 : wait_for_terminate() failed to get the state of the
3374 * container, the container was terminated by a signal, or
3375 * failed for an unknown reason. No change is made to the
3376 * container argument.
3377 * > 0 : The program executed in the container terminated with an
3378 * error. The exit code of the program executed in the
3379 * container is returned. The container argument has been set
3380 * to CONTAINER_TERMINATED.
3381 * 0 : The container is being rebooted, has been shut down or exited
3382 * successfully. The container argument has been set to either
3383 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
3384 *
3385 * That is, success is indicated by a return value of zero, and an
3386 * error is indicated by a non-zero value.
3387 */
3388 static int wait_for_container(pid_t pid, ContainerStatus *container) {
3389 siginfo_t status;
3390 int r;
3391
3392 r = wait_for_terminate(pid, &status);
3393 if (r < 0)
3394 return log_warning_errno(r, "Failed to wait for container: %m");
3395
3396 switch (status.si_code) {
3397
3398 case CLD_EXITED:
3399 if (status.si_status == 0) {
3400 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
3401
3402 } else
3403 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
3404
3405 *container = CONTAINER_TERMINATED;
3406 return status.si_status;
3407
3408 case CLD_KILLED:
3409 if (status.si_status == SIGINT) {
3410
3411 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
3412 *container = CONTAINER_TERMINATED;
3413 return 0;
3414
3415 } else if (status.si_status == SIGHUP) {
3416
3417 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
3418 *container = CONTAINER_REBOOTED;
3419 return 0;
3420 }
3421
3422 /* CLD_KILLED fallthrough */
3423
3424 case CLD_DUMPED:
3425 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
3426 return -EIO;
3427
3428 default:
3429 log_error("Container %s failed due to unknown reason.", arg_machine);
3430 return -EIO;
3431 }
3432
3433 return r;
3434 }
3435
3436 static void nop_handler(int sig) {}
3437
3438 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
3439 pid_t pid;
3440
3441 pid = PTR_TO_UINT32(userdata);
3442 if (pid > 0) {
3443 if (kill(pid, arg_kill_signal) >= 0) {
3444 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
3445 sd_event_source_set_userdata(s, NULL);
3446 return 0;
3447 }
3448 }
3449
3450 sd_event_exit(sd_event_source_get_event(s), 0);
3451 return 0;
3452 }
3453
3454 static int determine_names(void) {
3455 int r;
3456
3457 if (arg_template && !arg_directory && arg_machine) {
3458
3459 /* If --template= was specified then we should not
3460 * search for a machine, but instead create a new one
3461 * in /var/lib/machine. */
3462
3463 arg_directory = strjoin("/var/lib/machines/", arg_machine, NULL);
3464 if (!arg_directory)
3465 return log_oom();
3466 }
3467
3468 if (!arg_image && !arg_directory) {
3469 if (arg_machine) {
3470 _cleanup_(image_unrefp) Image *i = NULL;
3471
3472 r = image_find(arg_machine, &i);
3473 if (r < 0)
3474 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
3475 else if (r == 0) {
3476 log_error("No image for machine '%s': %m", arg_machine);
3477 return -ENOENT;
3478 }
3479
3480 if (i->type == IMAGE_RAW)
3481 r = set_sanitized_path(&arg_image, i->path);
3482 else
3483 r = set_sanitized_path(&arg_directory, i->path);
3484 if (r < 0)
3485 return log_error_errno(r, "Invalid image directory: %m");
3486
3487 if (!arg_ephemeral)
3488 arg_read_only = arg_read_only || i->read_only;
3489 } else
3490 arg_directory = get_current_dir_name();
3491
3492 if (!arg_directory && !arg_machine) {
3493 log_error("Failed to determine path, please use -D or -i.");
3494 return -EINVAL;
3495 }
3496 }
3497
3498 if (!arg_machine) {
3499 if (arg_directory && path_equal(arg_directory, "/"))
3500 arg_machine = gethostname_malloc();
3501 else
3502 arg_machine = strdup(basename(arg_image ?: arg_directory));
3503
3504 if (!arg_machine)
3505 return log_oom();
3506
3507 hostname_cleanup(arg_machine);
3508 if (!machine_name_is_valid(arg_machine)) {
3509 log_error("Failed to determine machine name automatically, please use -M.");
3510 return -EINVAL;
3511 }
3512
3513 if (arg_ephemeral) {
3514 char *b;
3515
3516 /* Add a random suffix when this is an
3517 * ephemeral machine, so that we can run many
3518 * instances at once without manually having
3519 * to specify -M each time. */
3520
3521 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
3522 return log_oom();
3523
3524 free(arg_machine);
3525 arg_machine = b;
3526 }
3527 }
3528
3529 return 0;
3530 }
3531
3532 static int determine_uid_shift(const char *directory) {
3533 int r;
3534
3535 if (!arg_userns) {
3536 arg_uid_shift = 0;
3537 return 0;
3538 }
3539
3540 if (arg_uid_shift == UID_INVALID) {
3541 struct stat st;
3542
3543 r = stat(directory, &st);
3544 if (r < 0)
3545 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
3546
3547 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
3548
3549 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
3550 log_error("UID and GID base of %s don't match.", directory);
3551 return -EINVAL;
3552 }
3553
3554 arg_uid_range = UINT32_C(0x10000);
3555 }
3556
3557 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
3558 log_error("UID base too high for UID range.");
3559 return -EINVAL;
3560 }
3561
3562 log_info("Using user namespaces with base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
3563 return 0;
3564 }
3565
3566 static int inner_child(
3567 Barrier *barrier,
3568 const char *directory,
3569 bool secondary,
3570 int kmsg_socket,
3571 int rtnl_socket,
3572 FDSet *fds) {
3573
3574 _cleanup_free_ char *home = NULL;
3575 unsigned n_env = 2;
3576 const char *envp[] = {
3577 "PATH=" DEFAULT_PATH_SPLIT_USR,
3578 "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
3579 NULL, /* TERM */
3580 NULL, /* HOME */
3581 NULL, /* USER */
3582 NULL, /* LOGNAME */
3583 NULL, /* container_uuid */
3584 NULL, /* LISTEN_FDS */
3585 NULL, /* LISTEN_PID */
3586 NULL
3587 };
3588
3589 _cleanup_strv_free_ char **env_use = NULL;
3590 int r;
3591
3592 assert(barrier);
3593 assert(directory);
3594 assert(kmsg_socket >= 0);
3595
3596 cg_unified_flush();
3597
3598 if (arg_userns) {
3599 /* Tell the parent, that it now can write the UID map. */
3600 (void) barrier_place(barrier); /* #1 */
3601
3602 /* Wait until the parent wrote the UID map */
3603 if (!barrier_place_and_sync(barrier)) { /* #2 */
3604 log_error("Parent died too early");
3605 return -ESRCH;
3606 }
3607 }
3608
3609 r = mount_all(NULL, true, arg_uid_shift, arg_uid_range, arg_selinux_apifs_context);
3610 if (r < 0)
3611 return r;
3612
3613 /* Wait until we are cgroup-ified, so that we
3614 * can mount the right cgroup path writable */
3615 if (!barrier_place_and_sync(barrier)) { /* #3 */
3616 log_error("Parent died too early");
3617 return -ESRCH;
3618 }
3619
3620 r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
3621 if (r < 0)
3622 return r;
3623
3624 r = reset_uid_gid();
3625 if (r < 0)
3626 return log_error_errno(r, "Couldn't become new root: %m");
3627
3628 r = setup_boot_id(NULL);
3629 if (r < 0)
3630 return r;
3631
3632 r = setup_kmsg(NULL, kmsg_socket);
3633 if (r < 0)
3634 return r;
3635 kmsg_socket = safe_close(kmsg_socket);
3636
3637 umask(0022);
3638
3639 if (setsid() < 0)
3640 return log_error_errno(errno, "setsid() failed: %m");
3641
3642 if (arg_private_network)
3643 loopback_setup();
3644
3645 r = send_rtnl(rtnl_socket);
3646 if (r < 0)
3647 return r;
3648 rtnl_socket = safe_close(rtnl_socket);
3649
3650 if (drop_capabilities() < 0)
3651 return log_error_errno(errno, "drop_capabilities() failed: %m");
3652
3653 setup_hostname();
3654
3655 if (arg_personality != PERSONALITY_INVALID) {
3656 if (personality(arg_personality) < 0)
3657 return log_error_errno(errno, "personality() failed: %m");
3658 } else if (secondary) {
3659 if (personality(PER_LINUX32) < 0)
3660 return log_error_errno(errno, "personality() failed: %m");
3661 }
3662
3663 #ifdef HAVE_SELINUX
3664 if (arg_selinux_context)
3665 if (setexeccon((security_context_t) arg_selinux_context) < 0)
3666 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
3667 #endif
3668
3669 r = change_uid_gid(&home);
3670 if (r < 0)
3671 return r;
3672
3673 envp[n_env] = strv_find_prefix(environ, "TERM=");
3674 if (envp[n_env])
3675 n_env ++;
3676
3677 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
3678 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
3679 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
3680 return log_oom();
3681
3682 if (!sd_id128_equal(arg_uuid, SD_ID128_NULL)) {
3683 char as_uuid[37];
3684
3685 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_format_as_uuid(arg_uuid, as_uuid)) < 0)
3686 return log_oom();
3687 }
3688
3689 if (fdset_size(fds) > 0) {
3690 r = fdset_cloexec(fds, false);
3691 if (r < 0)
3692 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
3693
3694 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
3695 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
3696 return log_oom();
3697 }
3698
3699 env_use = strv_env_merge(2, envp, arg_setenv);
3700 if (!env_use)
3701 return log_oom();
3702
3703 /* Let the parent know that we are ready and
3704 * wait until the parent is ready with the
3705 * setup, too... */
3706 if (!barrier_place_and_sync(barrier)) { /* #4 */
3707 log_error("Parent died too early");
3708 return -ESRCH;
3709 }
3710
3711 /* Now, explicitly close the log, so that we
3712 * then can close all remaining fds. Closing
3713 * the log explicitly first has the benefit
3714 * that the logging subsystem knows about it,
3715 * and is thus ready to be reopened should we
3716 * need it again. Note that the other fds
3717 * closed here are at least the locking and
3718 * barrier fds. */
3719 log_close();
3720 (void) fdset_close_others(fds);
3721
3722 if (arg_boot) {
3723 char **a;
3724 size_t m;
3725
3726 /* Automatically search for the init system */
3727
3728 m = 1 + strv_length(arg_parameters);
3729 a = newa(char*, m + 1);
3730 if (strv_isempty(arg_parameters))
3731 a[1] = NULL;
3732 else
3733 memcpy(a + 1, arg_parameters, m * sizeof(char*));
3734
3735 a[0] = (char*) "/usr/lib/systemd/systemd";
3736 execve(a[0], a, env_use);
3737
3738 a[0] = (char*) "/lib/systemd/systemd";
3739 execve(a[0], a, env_use);
3740
3741 a[0] = (char*) "/sbin/init";
3742 execve(a[0], a, env_use);
3743 } else if (!strv_isempty(arg_parameters))
3744 execvpe(arg_parameters[0], arg_parameters, env_use);
3745 else {
3746 chdir(home ?: "/root");
3747 execle("/bin/bash", "-bash", NULL, env_use);
3748 execle("/bin/sh", "-sh", NULL, env_use);
3749 }
3750
3751 (void) log_open();
3752 return log_error_errno(errno, "execv() failed: %m");
3753 }
3754
3755 static int outer_child(
3756 Barrier *barrier,
3757 const char *directory,
3758 const char *console,
3759 const char *root_device, bool root_device_rw,
3760 const char *home_device, bool home_device_rw,
3761 const char *srv_device, bool srv_device_rw,
3762 bool interactive,
3763 bool secondary,
3764 int pid_socket,
3765 int kmsg_socket,
3766 int rtnl_socket,
3767 int uid_shift_socket,
3768 FDSet *fds) {
3769
3770 pid_t pid;
3771 ssize_t l;
3772 int r;
3773
3774 assert(barrier);
3775 assert(directory);
3776 assert(console);
3777 assert(pid_socket >= 0);
3778 assert(kmsg_socket >= 0);
3779
3780 cg_unified_flush();
3781
3782 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
3783 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
3784
3785 if (interactive) {
3786 close_nointr(STDIN_FILENO);
3787 close_nointr(STDOUT_FILENO);
3788 close_nointr(STDERR_FILENO);
3789
3790 r = open_terminal(console, O_RDWR);
3791 if (r != STDIN_FILENO) {
3792 if (r >= 0) {
3793 safe_close(r);
3794 r = -EINVAL;
3795 }
3796
3797 return log_error_errno(r, "Failed to open console: %m");
3798 }
3799
3800 if (dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
3801 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
3802 return log_error_errno(errno, "Failed to duplicate console: %m");
3803 }
3804
3805 r = reset_audit_loginuid();
3806 if (r < 0)
3807 return r;
3808
3809 /* Mark everything as slave, so that we still
3810 * receive mounts from the real root, but don't
3811 * propagate mounts to the real root. */
3812 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
3813 return log_error_errno(errno, "MS_SLAVE|MS_REC failed: %m");
3814
3815 r = mount_devices(directory,
3816 root_device, root_device_rw,
3817 home_device, home_device_rw,
3818 srv_device, srv_device_rw);
3819 if (r < 0)
3820 return r;
3821
3822 r = determine_uid_shift(directory);
3823 if (r < 0)
3824 return r;
3825
3826 if (arg_userns) {
3827 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
3828 if (l < 0)
3829 return log_error_errno(errno, "Failed to send UID shift: %m");
3830 if (l != sizeof(arg_uid_shift)) {
3831 log_error("Short write while sending UID shift.");
3832 return -EIO;
3833 }
3834 }
3835
3836 /* Turn directory into bind mount */
3837 if (mount(directory, directory, NULL, MS_BIND|MS_REC, NULL) < 0)
3838 return log_error_errno(errno, "Failed to make bind mount: %m");
3839
3840 r = setup_volatile(directory, arg_volatile_mode, arg_userns, arg_uid_shift, arg_uid_range, arg_selinux_context);
3841 if (r < 0)
3842 return r;
3843
3844 r = setup_volatile_state(directory, arg_volatile_mode, arg_userns, arg_uid_shift, arg_uid_range, arg_selinux_context);
3845 if (r < 0)
3846 return r;
3847
3848 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
3849 if (r < 0)
3850 return r;
3851
3852 if (arg_read_only) {
3853 r = bind_remount_recursive(directory, true);
3854 if (r < 0)
3855 return log_error_errno(r, "Failed to make tree read-only: %m");
3856 }
3857
3858 r = mount_all(directory, false, arg_uid_shift, arg_uid_range, arg_selinux_apifs_context);
3859 if (r < 0)
3860 return r;
3861
3862 if (copy_devnodes(directory) < 0)
3863 return r;
3864
3865 dev_setup(directory, arg_uid_shift, arg_uid_shift);
3866
3867 if (setup_pts(directory) < 0)
3868 return r;
3869
3870 r = setup_propagate(directory);
3871 if (r < 0)
3872 return r;
3873
3874 r = setup_dev_console(directory, console);
3875 if (r < 0)
3876 return r;
3877
3878 r = setup_seccomp();
3879 if (r < 0)
3880 return r;
3881
3882 r = setup_timezone(directory);
3883 if (r < 0)
3884 return r;
3885
3886 r = setup_resolv_conf(directory);
3887 if (r < 0)
3888 return r;
3889
3890 r = setup_journal(directory);
3891 if (r < 0)
3892 return r;
3893
3894 r = mount_custom(directory, arg_custom_mounts, arg_n_custom_mounts, arg_userns, arg_uid_shift, arg_uid_range, arg_selinux_apifs_context);
3895 if (r < 0)
3896 return r;
3897
3898 r = mount_cgroups(directory, arg_unified_cgroup_hierarchy, arg_userns, arg_uid_shift, arg_uid_range, arg_selinux_apifs_context);
3899 if (r < 0)
3900 return r;
3901
3902 r = mount_move_root(directory);
3903 if (r < 0)
3904 return log_error_errno(r, "Failed to move root directory: %m");
3905
3906 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
3907 (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS) |
3908 (arg_private_network ? CLONE_NEWNET : 0) |
3909 (arg_userns ? CLONE_NEWUSER : 0),
3910 NULL);
3911 if (pid < 0)
3912 return log_error_errno(errno, "Failed to fork inner child: %m");
3913 if (pid == 0) {
3914 pid_socket = safe_close(pid_socket);
3915 uid_shift_socket = safe_close(uid_shift_socket);
3916
3917 /* The inner child has all namespaces that are
3918 * requested, so that we all are owned by the user if
3919 * user namespaces are turned on. */
3920
3921 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds);
3922 if (r < 0)
3923 _exit(EXIT_FAILURE);
3924
3925 _exit(EXIT_SUCCESS);
3926 }
3927
3928 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
3929 if (l < 0)
3930 return log_error_errno(errno, "Failed to send PID: %m");
3931 if (l != sizeof(pid)) {
3932 log_error("Short write while sending PID.");
3933 return -EIO;
3934 }
3935
3936 pid_socket = safe_close(pid_socket);
3937
3938 return 0;
3939 }
3940
3941 static int setup_uid_map(pid_t pid) {
3942 char uid_map[strlen("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
3943 int r;
3944
3945 assert(pid > 1);
3946
3947 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
3948 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
3949 r = write_string_file(uid_map, line, 0);
3950 if (r < 0)
3951 return log_error_errno(r, "Failed to write UID map: %m");
3952
3953 /* We always assign the same UID and GID ranges */
3954 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
3955 r = write_string_file(uid_map, line, 0);
3956 if (r < 0)
3957 return log_error_errno(r, "Failed to write GID map: %m");
3958
3959 return 0;
3960 }
3961
3962 static int chown_cgroup(pid_t pid) {
3963 _cleanup_free_ char *path = NULL, *fs = NULL;
3964 _cleanup_close_ int fd = -1;
3965 const char *fn;
3966 int r;
3967
3968 r = cg_pid_get_path(NULL, pid, &path);
3969 if (r < 0)
3970 return log_error_errno(r, "Failed to get container cgroup path: %m");
3971
3972 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, NULL, &fs);
3973 if (r < 0)
3974 return log_error_errno(r, "Failed to get file system path for container cgroup: %m");
3975
3976 fd = open(fs, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
3977 if (fd < 0)
3978 return log_error_errno(errno, "Failed to open %s: %m", fs);
3979
3980 FOREACH_STRING(fn,
3981 ".",
3982 "tasks",
3983 "notify_on_release",
3984 "cgroup.procs",
3985 "cgroup.clone_children",
3986 "cgroup.controllers",
3987 "cgroup.subtree_control",
3988 "cgroup.populated")
3989 if (fchownat(fd, fn, arg_uid_shift, arg_uid_shift, 0) < 0)
3990 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
3991 "Failed to chown() cgroup file %s, ignoring: %m", fn);
3992
3993 return 0;
3994 }
3995
3996 static int sync_cgroup(pid_t pid) {
3997 _cleanup_free_ char *cgroup = NULL;
3998 char tree[] = "/tmp/unifiedXXXXXX", pid_string[DECIMAL_STR_MAX(pid) + 1];
3999 bool undo_mount = false;
4000 const char *fn;
4001 int unified, r;
4002
4003 unified = cg_unified();
4004 if (unified < 0)
4005 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
4006
4007 if ((unified > 0) == arg_unified_cgroup_hierarchy)
4008 return 0;
4009
4010 /* When the host uses the legacy cgroup setup, but the
4011 * container shall use the unified hierarchy, let's make sure
4012 * we copy the path from the name=systemd hierarchy into the
4013 * unified hierarchy. Similar for the reverse situation. */
4014
4015 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
4016 if (r < 0)
4017 return log_error_errno(r, "Failed to get control group of " PID_FMT ": %m", pid);
4018
4019 /* In order to access the unified hierarchy we need to mount it */
4020 if (!mkdtemp(tree))
4021 return log_error_errno(errno, "Failed to generate temporary mount point for unified hierarchy: %m");
4022
4023 if (unified)
4024 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
4025 else
4026 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "__DEVEL__sane_behavior");
4027 if (r < 0) {
4028 r = log_error_errno(errno, "Failed to mount unified hierarchy: %m");
4029 goto finish;
4030 }
4031
4032 undo_mount = true;
4033
4034 fn = strjoina(tree, cgroup, "/cgroup.procs");
4035 (void) mkdir_parents(fn, 0755);
4036
4037 sprintf(pid_string, PID_FMT, pid);
4038 r = write_string_file(fn, pid_string, 0);
4039 if (r < 0)
4040 log_error_errno(r, "Failed to move process: %m");
4041
4042 finish:
4043 if (undo_mount)
4044 (void) umount(tree);
4045
4046 (void) rmdir(tree);
4047 return r;
4048 }
4049
4050 static int create_subcgroup(pid_t pid) {
4051 _cleanup_free_ char *cgroup = NULL;
4052 const char *child;
4053 int unified, r;
4054 CGroupMask supported;
4055
4056 /* In the unified hierarchy inner nodes may only only contain
4057 * subgroups, but not processes. Hence, if we running in the
4058 * unified hierarchy and the container does the same, and we
4059 * did not create a scope unit for the container move us and
4060 * the container into two separate subcgroups. */
4061
4062 if (!arg_keep_unit)
4063 return 0;
4064
4065 if (!arg_unified_cgroup_hierarchy)
4066 return 0;
4067
4068 unified = cg_unified();
4069 if (unified < 0)
4070 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
4071 if (unified == 0)
4072 return 0;
4073
4074 r = cg_mask_supported(&supported);
4075 if (r < 0)
4076 return log_error_errno(r, "Failed to determine supported controllers: %m");
4077
4078 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
4079 if (r < 0)
4080 return log_error_errno(r, "Failed to get our control group: %m");
4081
4082 child = strjoina(cgroup, "/payload");
4083 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, pid);
4084 if (r < 0)
4085 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
4086
4087 child = strjoina(cgroup, "/supervisor");
4088 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, 0);
4089 if (r < 0)
4090 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
4091
4092 /* Try to enable as many controllers as possible for the new payload. */
4093 (void) cg_enable_everywhere(supported, supported, cgroup);
4094 return 0;
4095 }
4096
4097 static int load_settings(void) {
4098 _cleanup_(settings_freep) Settings *settings = NULL;
4099 _cleanup_fclose_ FILE *f = NULL;
4100 _cleanup_free_ char *p = NULL;
4101 const char *fn, *i;
4102 int r;
4103
4104 /* If all settings are masked, there's no point in looking for
4105 * the settings file */
4106 if ((arg_settings_mask & _SETTINGS_MASK_ALL) == _SETTINGS_MASK_ALL)
4107 return 0;
4108
4109 fn = strjoina(arg_machine, ".nspawn");
4110
4111 /* We first look in the admin's directories in /etc and /run */
4112 FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
4113 _cleanup_free_ char *j = NULL;
4114
4115 j = strjoin(i, "/", fn, NULL);
4116 if (!j)
4117 return log_oom();
4118
4119 f = fopen(j, "re");
4120 if (f) {
4121 p = j;
4122 j = NULL;
4123
4124 /* By default we trust configuration from /etc and /run */
4125 if (arg_settings_trusted < 0)
4126 arg_settings_trusted = true;
4127
4128 break;
4129 }
4130
4131 if (errno != ENOENT)
4132 return log_error_errno(errno, "Failed to open %s: %m", j);
4133 }
4134
4135 if (!f) {
4136 /* After that, let's look for a file next to the
4137 * actual image we shall boot. */
4138
4139 if (arg_image) {
4140 p = file_in_same_dir(arg_image, fn);
4141 if (!p)
4142 return log_oom();
4143 } else if (arg_directory) {
4144 p = file_in_same_dir(arg_directory, fn);
4145 if (!p)
4146 return log_oom();
4147 }
4148
4149 if (p) {
4150 f = fopen(p, "re");
4151 if (!f && errno != ENOENT)
4152 return log_error_errno(errno, "Failed to open %s: %m", p);
4153
4154 /* By default we do not trust configuration from /var/lib/machines */
4155 if (arg_settings_trusted < 0)
4156 arg_settings_trusted = false;
4157 }
4158 }
4159
4160 if (!f)
4161 return 0;
4162
4163 log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
4164
4165 r = settings_load(f, p, &settings);
4166 if (r < 0)
4167 return r;
4168
4169 /* Copy over bits from the settings, unless they have been
4170 * explicitly masked by command line switches. */
4171
4172 if ((arg_settings_mask & SETTING_BOOT) == 0 &&
4173 settings->boot >= 0) {
4174 arg_boot = settings->boot;
4175
4176 strv_free(arg_parameters);
4177 arg_parameters = settings->parameters;
4178 settings->parameters = NULL;
4179 }
4180
4181 if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
4182 settings->environment) {
4183 strv_free(arg_setenv);
4184 arg_setenv = settings->environment;
4185 settings->environment = NULL;
4186 }
4187
4188 if ((arg_settings_mask & SETTING_USER) == 0 &&
4189 settings->user) {
4190 free(arg_user);
4191 arg_user = settings->user;
4192 settings->user = NULL;
4193 }
4194
4195 if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
4196
4197 if (!arg_settings_trusted && settings->capability != 0)
4198 log_warning("Ignoring Capability= setting, file %s is not trusted.", p);
4199 else
4200 arg_retain |= settings->capability;
4201
4202 arg_retain &= ~settings->drop_capability;
4203 }
4204
4205 if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
4206 settings->kill_signal > 0)
4207 arg_kill_signal = settings->kill_signal;
4208
4209 if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
4210 settings->personality != PERSONALITY_INVALID)
4211 arg_personality = settings->personality;
4212
4213 if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
4214 !sd_id128_is_null(settings->machine_id)) {
4215
4216 if (!arg_settings_trusted)
4217 log_warning("Ignoring MachineID= setting, file %s is not trusted.", p);
4218 else
4219 arg_uuid = settings->machine_id;
4220 }
4221
4222 if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
4223 settings->read_only >= 0)
4224 arg_read_only = settings->read_only;
4225
4226 if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
4227 settings->volatile_mode != _VOLATILE_MODE_INVALID)
4228 arg_volatile_mode = settings->volatile_mode;
4229
4230 if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
4231 settings->n_custom_mounts > 0) {
4232
4233 if (!arg_settings_trusted)
4234 log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", p);
4235 else {
4236 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4237 arg_custom_mounts = settings->custom_mounts;
4238 arg_n_custom_mounts = settings->n_custom_mounts;
4239
4240 settings->custom_mounts = NULL;
4241 settings->n_custom_mounts = 0;
4242 }
4243 }
4244
4245 if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
4246 (settings->private_network >= 0 ||
4247 settings->network_veth >= 0 ||
4248 settings->network_bridge ||
4249 settings->network_interfaces ||
4250 settings->network_macvlan ||
4251 settings->network_ipvlan)) {
4252
4253 if (!arg_settings_trusted)
4254 log_warning("Ignoring network settings, file %s is not trusted.", p);
4255 else {
4256 strv_free(arg_network_interfaces);
4257 arg_network_interfaces = settings->network_interfaces;
4258 settings->network_interfaces = NULL;
4259
4260 strv_free(arg_network_macvlan);
4261 arg_network_macvlan = settings->network_macvlan;
4262 settings->network_macvlan = NULL;
4263
4264 strv_free(arg_network_ipvlan);
4265 arg_network_ipvlan = settings->network_ipvlan;
4266 settings->network_ipvlan = NULL;
4267
4268 free(arg_network_bridge);
4269 arg_network_bridge = settings->network_bridge;
4270 settings->network_bridge = NULL;
4271
4272 arg_network_veth = settings->network_veth > 0 || settings->network_bridge;
4273
4274 arg_private_network = true; /* all these settings imply private networking */
4275 }
4276 }
4277
4278 if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
4279 settings->expose_ports) {
4280
4281 if (!arg_settings_trusted)
4282 log_warning("Ignoring Port= setting, file %s is not trusted.", p);
4283 else {
4284 expose_port_free_all(arg_expose_ports);
4285 arg_expose_ports = settings->expose_ports;
4286 settings->expose_ports = NULL;
4287 }
4288 }
4289
4290 return 0;
4291 }
4292
4293 int main(int argc, char *argv[]) {
4294
4295 _cleanup_free_ char *device_path = NULL, *root_device = NULL, *home_device = NULL, *srv_device = NULL, *console = NULL;
4296 bool root_device_rw = true, home_device_rw = true, srv_device_rw = true;
4297 _cleanup_close_ int master = -1, image_fd = -1;
4298 _cleanup_fdset_free_ FDSet *fds = NULL;
4299 int r, n_fd_passed, loop_nr = -1;
4300 char veth_name[IFNAMSIZ];
4301 bool secondary = false, remove_subvol = false;
4302 sigset_t mask_chld;
4303 pid_t pid = 0;
4304 int ret = EXIT_SUCCESS;
4305 union in_addr_union exposed = {};
4306 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4307 bool interactive;
4308
4309 log_parse_environment();
4310 log_open();
4311
4312 r = parse_argv(argc, argv);
4313 if (r <= 0)
4314 goto finish;
4315
4316 if (geteuid() != 0) {
4317 log_error("Need to be root.");
4318 r = -EPERM;
4319 goto finish;
4320 }
4321 r = determine_names();
4322 if (r < 0)
4323 goto finish;
4324
4325 r = load_settings();
4326 if (r < 0)
4327 goto finish;
4328
4329 r = verify_arguments();
4330 if (r < 0)
4331 goto finish;
4332
4333 n_fd_passed = sd_listen_fds(false);
4334 if (n_fd_passed > 0) {
4335 r = fdset_new_listen_fds(&fds, false);
4336 if (r < 0) {
4337 log_error_errno(r, "Failed to collect file descriptors: %m");
4338 goto finish;
4339 }
4340 }
4341
4342 if (arg_directory) {
4343 assert(!arg_image);
4344
4345 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
4346 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4347 r = -EINVAL;
4348 goto finish;
4349 }
4350
4351 if (arg_ephemeral) {
4352 _cleanup_free_ char *np = NULL;
4353
4354 /* If the specified path is a mount point we
4355 * generate the new snapshot immediately
4356 * inside it under a random name. However if
4357 * the specified is not a mount point we
4358 * create the new snapshot in the parent
4359 * directory, just next to it. */
4360 r = path_is_mount_point(arg_directory, 0);
4361 if (r < 0) {
4362 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4363 goto finish;
4364 }
4365 if (r > 0)
4366 r = tempfn_random_child(arg_directory, "machine.", &np);
4367 else
4368 r = tempfn_random(arg_directory, "machine.", &np);
4369 if (r < 0) {
4370 log_error_errno(r, "Failed to generate name for snapshot: %m");
4371 goto finish;
4372 }
4373
4374 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4375 if (r < 0) {
4376 log_error_errno(r, "Failed to lock %s: %m", np);
4377 goto finish;
4378 }
4379
4380 r = btrfs_subvol_snapshot(arg_directory, np, (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | BTRFS_SNAPSHOT_FALLBACK_COPY | BTRFS_SNAPSHOT_RECURSIVE);
4381 if (r < 0) {
4382 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4383 goto finish;
4384 }
4385
4386 free(arg_directory);
4387 arg_directory = np;
4388 np = NULL;
4389
4390 remove_subvol = true;
4391
4392 } else {
4393 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4394 if (r == -EBUSY) {
4395 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4396 goto finish;
4397 }
4398 if (r < 0) {
4399 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4400 return r;
4401 }
4402
4403 if (arg_template) {
4404 r = btrfs_subvol_snapshot(arg_template, arg_directory, (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | BTRFS_SNAPSHOT_FALLBACK_COPY | BTRFS_SNAPSHOT_RECURSIVE);
4405 if (r == -EEXIST) {
4406 if (!arg_quiet)
4407 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4408 } else if (r < 0) {
4409 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4410 goto finish;
4411 } else {
4412 if (!arg_quiet)
4413 log_info("Populated %s from template %s.", arg_directory, arg_template);
4414 }
4415 }
4416 }
4417
4418 if (arg_boot) {
4419 if (path_is_os_tree(arg_directory) <= 0) {
4420 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
4421 r = -EINVAL;
4422 goto finish;
4423 }
4424 } else {
4425 const char *p;
4426
4427 p = strjoina(arg_directory,
4428 argc > optind && path_is_absolute(argv[optind]) ? argv[optind] : "/usr/bin/");
4429 if (access(p, F_OK) < 0) {
4430 log_error("Directory %s lacks the binary to execute or doesn't look like a binary tree. Refusing.", arg_directory);
4431 r = -EINVAL;
4432 goto finish;
4433 }
4434 }
4435
4436 } else {
4437 char template[] = "/tmp/nspawn-root-XXXXXX";
4438
4439 assert(arg_image);
4440 assert(!arg_template);
4441
4442 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4443 if (r == -EBUSY) {
4444 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4445 goto finish;
4446 }
4447 if (r < 0) {
4448 r = log_error_errno(r, "Failed to create image lock: %m");
4449 goto finish;
4450 }
4451
4452 if (!mkdtemp(template)) {
4453 log_error_errno(errno, "Failed to create temporary directory: %m");
4454 r = -errno;
4455 goto finish;
4456 }
4457
4458 arg_directory = strdup(template);
4459 if (!arg_directory) {
4460 r = log_oom();
4461 goto finish;
4462 }
4463
4464 image_fd = setup_image(&device_path, &loop_nr);
4465 if (image_fd < 0) {
4466 r = image_fd;
4467 goto finish;
4468 }
4469
4470 r = dissect_image(image_fd,
4471 &root_device, &root_device_rw,
4472 &home_device, &home_device_rw,
4473 &srv_device, &srv_device_rw,
4474 &secondary);
4475 if (r < 0)
4476 goto finish;
4477 }
4478
4479 r = custom_mounts_prepare();
4480 if (r < 0)
4481 goto finish;
4482
4483 interactive =
4484 isatty(STDIN_FILENO) > 0 &&
4485 isatty(STDOUT_FILENO) > 0;
4486
4487 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
4488 if (master < 0) {
4489 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4490 goto finish;
4491 }
4492
4493 r = ptsname_malloc(master, &console);
4494 if (r < 0) {
4495 r = log_error_errno(r, "Failed to determine tty name: %m");
4496 goto finish;
4497 }
4498
4499 if (unlockpt(master) < 0) {
4500 r = log_error_errno(errno, "Failed to unlock tty: %m");
4501 goto finish;
4502 }
4503
4504 if (!arg_quiet)
4505 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4506 arg_machine, arg_image ?: arg_directory);
4507
4508 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
4509
4510 assert_se(sigemptyset(&mask_chld) == 0);
4511 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4512
4513 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
4514 r = log_error_errno(errno, "Failed to become subreaper: %m");
4515 goto finish;
4516 }
4517
4518 for (;;) {
4519 _cleanup_close_pair_ int kmsg_socket_pair[2] = { -1, -1 }, rtnl_socket_pair[2] = { -1, -1 }, pid_socket_pair[2] = { -1, -1 },
4520 uid_shift_socket_pair[2] = { -1, -1 };
4521 ContainerStatus container_status;
4522 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4523 static const struct sigaction sa = {
4524 .sa_handler = nop_handler,
4525 .sa_flags = SA_NOCLDSTOP,
4526 };
4527 int ifi = 0;
4528 ssize_t l;
4529 _cleanup_event_unref_ sd_event *event = NULL;
4530 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
4531 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
4532 char last_char = 0;
4533
4534 r = barrier_create(&barrier);
4535 if (r < 0) {
4536 log_error_errno(r, "Cannot initialize IPC barrier: %m");
4537 goto finish;
4538 }
4539
4540 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
4541 r = log_error_errno(errno, "Failed to create kmsg socket pair: %m");
4542 goto finish;
4543 }
4544
4545 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0) {
4546 r = log_error_errno(errno, "Failed to create rtnl socket pair: %m");
4547 goto finish;
4548 }
4549
4550 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, pid_socket_pair) < 0) {
4551 r = log_error_errno(errno, "Failed to create pid socket pair: %m");
4552 goto finish;
4553 }
4554
4555 if (arg_userns)
4556 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0) {
4557 r = log_error_errno(errno, "Failed to create uid shift socket pair: %m");
4558 goto finish;
4559 }
4560
4561 /* Child can be killed before execv(), so handle SIGCHLD
4562 * in order to interrupt parent's blocking calls and
4563 * give it a chance to call wait() and terminate. */
4564 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
4565 if (r < 0) {
4566 r = log_error_errno(errno, "Failed to change the signal mask: %m");
4567 goto finish;
4568 }
4569
4570 r = sigaction(SIGCHLD, &sa, NULL);
4571 if (r < 0) {
4572 r = log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
4573 goto finish;
4574 }
4575
4576 pid = raw_clone(SIGCHLD|CLONE_NEWNS, NULL);
4577 if (pid < 0) {
4578 if (errno == EINVAL)
4579 r = log_error_errno(errno, "clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
4580 else
4581 r = log_error_errno(errno, "clone() failed: %m");
4582
4583 goto finish;
4584 }
4585
4586 if (pid == 0) {
4587 /* The outer child only has a file system namespace. */
4588 barrier_set_role(&barrier, BARRIER_CHILD);
4589
4590 master = safe_close(master);
4591
4592 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
4593 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4594 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
4595 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
4596
4597 (void) reset_all_signal_handlers();
4598 (void) reset_signal_mask();
4599
4600 r = outer_child(&barrier,
4601 arg_directory,
4602 console,
4603 root_device, root_device_rw,
4604 home_device, home_device_rw,
4605 srv_device, srv_device_rw,
4606 interactive,
4607 secondary,
4608 pid_socket_pair[1],
4609 kmsg_socket_pair[1],
4610 rtnl_socket_pair[1],
4611 uid_shift_socket_pair[1],
4612 fds);
4613 if (r < 0)
4614 _exit(EXIT_FAILURE);
4615
4616 _exit(EXIT_SUCCESS);
4617 }
4618
4619 barrier_set_role(&barrier, BARRIER_PARENT);
4620
4621 fdset_free(fds);
4622 fds = NULL;
4623
4624 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
4625 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
4626 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
4627
4628 /* Wait for the outer child. */
4629 r = wait_for_terminate_and_warn("namespace helper", pid, NULL);
4630 if (r < 0)
4631 goto finish;
4632 if (r != 0) {
4633 r = -EIO;
4634 goto finish;
4635 }
4636 pid = 0;
4637
4638 /* And now retrieve the PID of the inner child. */
4639 l = recv(pid_socket_pair[0], &pid, sizeof(pid), 0);
4640 if (l < 0) {
4641 r = log_error_errno(errno, "Failed to read inner child PID: %m");
4642 goto finish;
4643 }
4644 if (l != sizeof(pid)) {
4645 log_error("Short read while reading inner child PID: %m");
4646 r = EIO;
4647 goto finish;
4648 }
4649
4650 log_debug("Init process invoked as PID " PID_FMT, pid);
4651
4652 if (arg_userns) {
4653 if (!barrier_place_and_sync(&barrier)) { /* #1 */
4654 log_error("Child died too early.");
4655 r = -ESRCH;
4656 goto finish;
4657 }
4658
4659 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof(arg_uid_shift), 0);
4660 if (l < 0) {
4661 r = log_error_errno(errno, "Failed to read UID shift: %m");
4662 goto finish;
4663 }
4664 if (l != sizeof(arg_uid_shift)) {
4665 log_error("Short read while reading UID shift: %m");
4666 r = EIO;
4667 goto finish;
4668 }
4669
4670 r = setup_uid_map(pid);
4671 if (r < 0)
4672 goto finish;
4673
4674 (void) barrier_place(&barrier); /* #2 */
4675 }
4676
4677 r = move_network_interfaces(pid);
4678 if (r < 0)
4679 goto finish;
4680
4681 r = setup_veth(pid, veth_name, &ifi);
4682 if (r < 0)
4683 goto finish;
4684
4685 r = setup_bridge(veth_name, &ifi);
4686 if (r < 0)
4687 goto finish;
4688
4689 r = setup_macvlan(pid);
4690 if (r < 0)
4691 goto finish;
4692
4693 r = setup_ipvlan(pid);
4694 if (r < 0)
4695 goto finish;
4696
4697 r = register_machine(pid, ifi);
4698 if (r < 0)
4699 goto finish;
4700
4701 r = sync_cgroup(pid);
4702 if (r < 0)
4703 goto finish;
4704
4705 r = create_subcgroup(pid);
4706 if (r < 0)
4707 goto finish;
4708
4709 r = chown_cgroup(pid);
4710 if (r < 0)
4711 goto finish;
4712
4713 /* Notify the child that the parent is ready with all
4714 * its setup (including cgroup-ification), and that
4715 * the child can now hand over control to the code to
4716 * run inside the container. */
4717 (void) barrier_place(&barrier); /* #3 */
4718
4719 /* Block SIGCHLD here, before notifying child.
4720 * process_pty() will handle it with the other signals. */
4721 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
4722
4723 /* Reset signal to default */
4724 r = default_signals(SIGCHLD, -1);
4725 if (r < 0) {
4726 log_error_errno(r, "Failed to reset SIGCHLD: %m");
4727 goto finish;
4728 }
4729
4730 /* Let the child know that we are ready and wait that the child is completely ready now. */
4731 if (!barrier_place_and_sync(&barrier)) { /* #5 */
4732 log_error("Client died too early.");
4733 r = -ESRCH;
4734 goto finish;
4735 }
4736
4737 sd_notifyf(false,
4738 "READY=1\n"
4739 "STATUS=Container running.\n"
4740 "X_NSPAWN_LEADER_PID=" PID_FMT, pid);
4741
4742 r = sd_event_new(&event);
4743 if (r < 0) {
4744 log_error_errno(r, "Failed to get default event source: %m");
4745 goto finish;
4746 }
4747
4748 if (arg_kill_signal > 0) {
4749 /* Try to kill the init system on SIGINT or SIGTERM */
4750 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, UINT32_TO_PTR(pid));
4751 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, UINT32_TO_PTR(pid));
4752 } else {
4753 /* Immediately exit */
4754 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
4755 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
4756 }
4757
4758 /* simply exit on sigchld */
4759 sd_event_add_signal(event, NULL, SIGCHLD, NULL, NULL);
4760
4761 if (arg_expose_ports) {
4762 r = watch_rtnl(event, rtnl_socket_pair[0], &exposed, &rtnl);
4763 if (r < 0)
4764 goto finish;
4765
4766 (void) expose_ports(rtnl, &exposed);
4767 }
4768
4769 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
4770
4771 r = pty_forward_new(event, master, true, !interactive, &forward);
4772 if (r < 0) {
4773 log_error_errno(r, "Failed to create PTY forwarder: %m");
4774 goto finish;
4775 }
4776
4777 r = sd_event_loop(event);
4778 if (r < 0) {
4779 log_error_errno(r, "Failed to run event loop: %m");
4780 goto finish;
4781 }
4782
4783 pty_forward_get_last_char(forward, &last_char);
4784
4785 forward = pty_forward_free(forward);
4786
4787 if (!arg_quiet && last_char != '\n')
4788 putc('\n', stdout);
4789
4790 /* Kill if it is not dead yet anyway */
4791 terminate_machine(pid);
4792
4793 /* Normally redundant, but better safe than sorry */
4794 kill(pid, SIGKILL);
4795
4796 r = wait_for_container(pid, &container_status);
4797 pid = 0;
4798
4799 if (r < 0)
4800 /* We failed to wait for the container, or the
4801 * container exited abnormally */
4802 goto finish;
4803 else if (r > 0 || container_status == CONTAINER_TERMINATED){
4804 /* The container exited with a non-zero
4805 * status, or with zero status and no reboot
4806 * was requested. */
4807 ret = r;
4808 break;
4809 }
4810
4811 /* CONTAINER_REBOOTED, loop again */
4812
4813 if (arg_keep_unit) {
4814 /* Special handling if we are running as a
4815 * service: instead of simply restarting the
4816 * machine we want to restart the entire
4817 * service, so let's inform systemd about this
4818 * with the special exit code 133. The service
4819 * file uses RestartForceExitStatus=133 so
4820 * that this results in a full nspawn
4821 * restart. This is necessary since we might
4822 * have cgroup parameters set we want to have
4823 * flushed out. */
4824 ret = 133;
4825 r = 0;
4826 break;
4827 }
4828
4829 flush_ports(&exposed);
4830 }
4831
4832 finish:
4833 sd_notify(false,
4834 "STOPPING=1\n"
4835 "STATUS=Terminating...");
4836
4837 if (pid > 0)
4838 kill(pid, SIGKILL);
4839
4840 /* Try to flush whatever is still queued in the pty */
4841 if (master >= 0)
4842 (void) copy_bytes(master, STDOUT_FILENO, (off_t) -1, false);
4843
4844 loop_remove(loop_nr, &image_fd);
4845
4846 if (remove_subvol && arg_directory) {
4847 int k;
4848
4849 k = btrfs_subvol_remove(arg_directory, true);
4850 if (k < 0)
4851 log_warning_errno(k, "Cannot remove subvolume '%s', ignoring: %m", arg_directory);
4852 }
4853
4854 if (arg_machine) {
4855 const char *p;
4856
4857 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
4858 (void) rm_rf(p, REMOVE_ROOT);
4859 }
4860
4861 flush_ports(&exposed);
4862
4863 free(arg_directory);
4864 free(arg_template);
4865 free(arg_image);
4866 free(arg_machine);
4867 free(arg_user);
4868 strv_free(arg_setenv);
4869 free(arg_network_bridge);
4870 strv_free(arg_network_interfaces);
4871 strv_free(arg_network_macvlan);
4872 strv_free(arg_network_ipvlan);
4873 strv_free(arg_parameters);
4874 custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
4875 expose_port_free_all(arg_expose_ports);
4876
4877 return r < 0 ? EXIT_FAILURE : ret;
4878 }
Unnamed repository; edit this file 'description' to name the repository.