]> 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
core: unified cgroup hierarchy support
[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 typedef struct ExposePort {
106 int protocol;
107 uint16_t host_port;
108 uint16_t container_port;
109 LIST_FIELDS(struct ExposePort, ports);
110 } ExposePort;
111
112 typedef enum ContainerStatus {
113 CONTAINER_TERMINATED,
114 CONTAINER_REBOOTED
115 } ContainerStatus;
116
117 typedef enum LinkJournal {
118 LINK_NO,
119 LINK_AUTO,
120 LINK_HOST,
121 LINK_GUEST
122 } LinkJournal;
123
124 typedef enum Volatile {
125 VOLATILE_NO,
126 VOLATILE_YES,
127 VOLATILE_STATE,
128 } Volatile;
129
130 typedef enum CustomMountType {
131 CUSTOM_MOUNT_BIND,
132 CUSTOM_MOUNT_TMPFS,
133 CUSTOM_MOUNT_OVERLAY,
134 } CustomMountType;
135
136 typedef struct CustomMount {
137 CustomMountType type;
138 bool read_only;
139 char *source; /* for overlayfs this is the upper directory */
140 char *destination;
141 char *options;
142 char *work_dir;
143 char **lower;
144 } CustomMount;
145
146 static char *arg_directory = NULL;
147 static char *arg_template = NULL;
148 static char *arg_user = NULL;
149 static sd_id128_t arg_uuid = {};
150 static char *arg_machine = NULL;
151 static const char *arg_selinux_context = NULL;
152 static const char *arg_selinux_apifs_context = NULL;
153 static const char *arg_slice = NULL;
154 static bool arg_private_network = false;
155 static bool arg_read_only = false;
156 static bool arg_boot = false;
157 static bool arg_ephemeral = false;
158 static LinkJournal arg_link_journal = LINK_AUTO;
159 static bool arg_link_journal_try = false;
160 static uint64_t arg_retain =
161 (1ULL << CAP_CHOWN) |
162 (1ULL << CAP_DAC_OVERRIDE) |
163 (1ULL << CAP_DAC_READ_SEARCH) |
164 (1ULL << CAP_FOWNER) |
165 (1ULL << CAP_FSETID) |
166 (1ULL << CAP_IPC_OWNER) |
167 (1ULL << CAP_KILL) |
168 (1ULL << CAP_LEASE) |
169 (1ULL << CAP_LINUX_IMMUTABLE) |
170 (1ULL << CAP_NET_BIND_SERVICE) |
171 (1ULL << CAP_NET_BROADCAST) |
172 (1ULL << CAP_NET_RAW) |
173 (1ULL << CAP_SETGID) |
174 (1ULL << CAP_SETFCAP) |
175 (1ULL << CAP_SETPCAP) |
176 (1ULL << CAP_SETUID) |
177 (1ULL << CAP_SYS_ADMIN) |
178 (1ULL << CAP_SYS_CHROOT) |
179 (1ULL << CAP_SYS_NICE) |
180 (1ULL << CAP_SYS_PTRACE) |
181 (1ULL << CAP_SYS_TTY_CONFIG) |
182 (1ULL << CAP_SYS_RESOURCE) |
183 (1ULL << CAP_SYS_BOOT) |
184 (1ULL << CAP_AUDIT_WRITE) |
185 (1ULL << CAP_AUDIT_CONTROL) |
186 (1ULL << CAP_MKNOD);
187 static CustomMount *arg_custom_mounts = NULL;
188 static unsigned arg_n_custom_mounts = 0;
189 static char **arg_setenv = NULL;
190 static bool arg_quiet = false;
191 static bool arg_share_system = false;
192 static bool arg_register = true;
193 static bool arg_keep_unit = false;
194 static char **arg_network_interfaces = NULL;
195 static char **arg_network_macvlan = NULL;
196 static char **arg_network_ipvlan = NULL;
197 static bool arg_network_veth = false;
198 static const char *arg_network_bridge = NULL;
199 static unsigned long arg_personality = PERSONALITY_INVALID;
200 static char *arg_image = NULL;
201 static Volatile arg_volatile = VOLATILE_NO;
202 static ExposePort *arg_expose_ports = NULL;
203 static char **arg_property = NULL;
204 static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
205 static bool arg_userns = false;
206 static int arg_kill_signal = 0;
207 static bool arg_unified_cgroup_hierarchy = false;
208
209 static void help(void) {
210 printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
211 "Spawn a minimal namespace container for debugging, testing and building.\n\n"
212 " -h --help Show this help\n"
213 " --version Print version string\n"
214 " -q --quiet Do not show status information\n"
215 " -D --directory=PATH Root directory for the container\n"
216 " --template=PATH Initialize root directory from template directory,\n"
217 " if missing\n"
218 " -x --ephemeral Run container with snapshot of root directory, and\n"
219 " remove it after exit\n"
220 " -i --image=PATH File system device or disk image for the container\n"
221 " -b --boot Boot up full system (i.e. invoke init)\n"
222 " -u --user=USER Run the command under specified user or uid\n"
223 " -M --machine=NAME Set the machine name for the container\n"
224 " --uuid=UUID Set a specific machine UUID for the container\n"
225 " -S --slice=SLICE Place the container in the specified slice\n"
226 " --property=NAME=VALUE Set scope unit property\n"
227 " --private-users[=UIDBASE[:NUIDS]]\n"
228 " Run within user namespace\n"
229 " --private-network Disable network in container\n"
230 " --network-interface=INTERFACE\n"
231 " Assign an existing network interface to the\n"
232 " container\n"
233 " --network-macvlan=INTERFACE\n"
234 " Create a macvlan network interface based on an\n"
235 " existing network interface to the container\n"
236 " --network-ipvlan=INTERFACE\n"
237 " Create a ipvlan network interface based on an\n"
238 " existing network interface to the container\n"
239 " -n --network-veth Add a virtual ethernet connection between host\n"
240 " and container\n"
241 " --network-bridge=INTERFACE\n"
242 " Add a virtual ethernet connection between host\n"
243 " and container and add it to an existing bridge on\n"
244 " the host\n"
245 " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
246 " Expose a container IP port on the host\n"
247 " -Z --selinux-context=SECLABEL\n"
248 " Set the SELinux security context to be used by\n"
249 " processes in the container\n"
250 " -L --selinux-apifs-context=SECLABEL\n"
251 " Set the SELinux security context to be used by\n"
252 " API/tmpfs file systems in the container\n"
253 " --capability=CAP In addition to the default, retain specified\n"
254 " capability\n"
255 " --drop-capability=CAP Drop the specified capability from the default set\n"
256 " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
257 " --link-journal=MODE Link up guest journal, one of no, auto, guest, host,\n"
258 " try-guest, try-host\n"
259 " -j Equivalent to --link-journal=try-guest\n"
260 " --read-only Mount the root directory read-only\n"
261 " --bind=PATH[:PATH[:OPTIONS]]\n"
262 " Bind mount a file or directory from the host into\n"
263 " the container\n"
264 " --bind-ro=PATH[:PATH[:OPTIONS]\n"
265 " Similar, but creates a read-only bind mount\n"
266 " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
267 " --overlay=PATH[:PATH...]:PATH\n"
268 " Create an overlay mount from the host to \n"
269 " the container\n"
270 " --overlay-ro=PATH[:PATH...]:PATH\n"
271 " Similar, but creates a read-only overlay mount\n"
272 " --setenv=NAME=VALUE Pass an environment variable to PID 1\n"
273 " --share-system Share system namespaces with host\n"
274 " --register=BOOLEAN Register container as machine\n"
275 " --keep-unit Do not register a scope for the machine, reuse\n"
276 " the service unit nspawn is running in\n"
277 " --volatile[=MODE] Run the system in volatile mode\n"
278 , program_invocation_short_name);
279 }
280
281 static CustomMount* custom_mount_add(CustomMountType t) {
282 CustomMount *c, *ret;
283
284 c = realloc(arg_custom_mounts, (arg_n_custom_mounts + 1) * sizeof(CustomMount));
285 if (!c)
286 return NULL;
287
288 arg_custom_mounts = c;
289 ret = arg_custom_mounts + arg_n_custom_mounts;
290 arg_n_custom_mounts++;
291
292 *ret = (CustomMount) { .type = t };
293
294 return ret;
295 }
296
297 static void custom_mount_free_all(void) {
298 unsigned i;
299
300 for (i = 0; i < arg_n_custom_mounts; i++) {
301 CustomMount *m = &arg_custom_mounts[i];
302
303 free(m->source);
304 free(m->destination);
305 free(m->options);
306
307 if (m->work_dir) {
308 (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
309 free(m->work_dir);
310 }
311
312 strv_free(m->lower);
313 }
314
315 arg_custom_mounts = mfree(arg_custom_mounts);
316 arg_n_custom_mounts = 0;
317 }
318
319 static int custom_mount_compare(const void *a, const void *b) {
320 const CustomMount *x = a, *y = b;
321 int r;
322
323 r = path_compare(x->destination, y->destination);
324 if (r != 0)
325 return r;
326
327 if (x->type < y->type)
328 return -1;
329 if (x->type > y->type)
330 return 1;
331
332 return 0;
333 }
334
335 static int custom_mounts_prepare(void) {
336 unsigned i;
337 int r;
338
339 /* Ensure the mounts are applied prefix first. */
340 qsort_safe(arg_custom_mounts, arg_n_custom_mounts, sizeof(CustomMount), custom_mount_compare);
341
342 /* Allocate working directories for the overlay file systems that need it */
343 for (i = 0; i < arg_n_custom_mounts; i++) {
344 CustomMount *m = &arg_custom_mounts[i];
345
346 if (arg_userns && arg_uid_shift == UID_INVALID && path_equal(m->destination, "/")) {
347 log_error("--private-users with automatic UID shift may not be combined with custom root mounts.");
348 return -EINVAL;
349 }
350
351 if (m->type != CUSTOM_MOUNT_OVERLAY)
352 continue;
353
354 if (m->work_dir)
355 continue;
356
357 if (m->read_only)
358 continue;
359
360 r = tempfn_random(m->source, NULL, &m->work_dir);
361 if (r < 0)
362 return log_error_errno(r, "Failed to generate work directory from %s: %m", m->source);
363 }
364
365 return 0;
366 }
367
368 static int set_sanitized_path(char **b, const char *path) {
369 char *p;
370
371 assert(b);
372 assert(path);
373
374 p = canonicalize_file_name(path);
375 if (!p) {
376 if (errno != ENOENT)
377 return -errno;
378
379 p = path_make_absolute_cwd(path);
380 if (!p)
381 return -ENOMEM;
382 }
383
384 free(*b);
385 *b = path_kill_slashes(p);
386 return 0;
387 }
388
389 static int detect_unified_cgroup_hierarchy(void) {
390 const char *e;
391 int r;
392
393 /* Allow the user to control whether the unified hierarchy is used */
394 e = getenv("UNIFIED_CGROUP_HIERARCHY");
395 if (e) {
396 r = parse_boolean(e);
397 if (r < 0)
398 return log_error_errno(r, "Failed to parse $UNIFIED_CGROUP_HIERARCHY.");
399
400 arg_unified_cgroup_hierarchy = r;
401 return 0;
402 }
403
404 /* Otherwise inherit the default from the host system */
405 r = cg_unified();
406 if (r < 0)
407 return log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
408
409 arg_unified_cgroup_hierarchy = r;
410 return 0;
411 }
412
413 static int parse_argv(int argc, char *argv[]) {
414
415 enum {
416 ARG_VERSION = 0x100,
417 ARG_PRIVATE_NETWORK,
418 ARG_UUID,
419 ARG_READ_ONLY,
420 ARG_CAPABILITY,
421 ARG_DROP_CAPABILITY,
422 ARG_LINK_JOURNAL,
423 ARG_BIND,
424 ARG_BIND_RO,
425 ARG_TMPFS,
426 ARG_OVERLAY,
427 ARG_OVERLAY_RO,
428 ARG_SETENV,
429 ARG_SHARE_SYSTEM,
430 ARG_REGISTER,
431 ARG_KEEP_UNIT,
432 ARG_NETWORK_INTERFACE,
433 ARG_NETWORK_MACVLAN,
434 ARG_NETWORK_IPVLAN,
435 ARG_NETWORK_BRIDGE,
436 ARG_PERSONALITY,
437 ARG_VOLATILE,
438 ARG_TEMPLATE,
439 ARG_PROPERTY,
440 ARG_PRIVATE_USERS,
441 ARG_KILL_SIGNAL,
442 };
443
444 static const struct option options[] = {
445 { "help", no_argument, NULL, 'h' },
446 { "version", no_argument, NULL, ARG_VERSION },
447 { "directory", required_argument, NULL, 'D' },
448 { "template", required_argument, NULL, ARG_TEMPLATE },
449 { "ephemeral", no_argument, NULL, 'x' },
450 { "user", required_argument, NULL, 'u' },
451 { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
452 { "boot", no_argument, NULL, 'b' },
453 { "uuid", required_argument, NULL, ARG_UUID },
454 { "read-only", no_argument, NULL, ARG_READ_ONLY },
455 { "capability", required_argument, NULL, ARG_CAPABILITY },
456 { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
457 { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
458 { "bind", required_argument, NULL, ARG_BIND },
459 { "bind-ro", required_argument, NULL, ARG_BIND_RO },
460 { "tmpfs", required_argument, NULL, ARG_TMPFS },
461 { "overlay", required_argument, NULL, ARG_OVERLAY },
462 { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
463 { "machine", required_argument, NULL, 'M' },
464 { "slice", required_argument, NULL, 'S' },
465 { "setenv", required_argument, NULL, ARG_SETENV },
466 { "selinux-context", required_argument, NULL, 'Z' },
467 { "selinux-apifs-context", required_argument, NULL, 'L' },
468 { "quiet", no_argument, NULL, 'q' },
469 { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM },
470 { "register", required_argument, NULL, ARG_REGISTER },
471 { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
472 { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
473 { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
474 { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
475 { "network-veth", no_argument, NULL, 'n' },
476 { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
477 { "personality", required_argument, NULL, ARG_PERSONALITY },
478 { "image", required_argument, NULL, 'i' },
479 { "volatile", optional_argument, NULL, ARG_VOLATILE },
480 { "port", required_argument, NULL, 'p' },
481 { "property", required_argument, NULL, ARG_PROPERTY },
482 { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
483 { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
484 {}
485 };
486
487 int c, r;
488 uint64_t plus = 0, minus = 0;
489
490 assert(argc >= 0);
491 assert(argv);
492
493 while ((c = getopt_long(argc, argv, "+hD:u:bL:M:jS:Z:qi:xp:n", options, NULL)) >= 0)
494
495 switch (c) {
496
497 case 'h':
498 help();
499 return 0;
500
501 case ARG_VERSION:
502 puts(PACKAGE_STRING);
503 puts(SYSTEMD_FEATURES);
504 return 0;
505
506 case 'D':
507 r = set_sanitized_path(&arg_directory, optarg);
508 if (r < 0)
509 return log_error_errno(r, "Invalid root directory: %m");
510
511 break;
512
513 case ARG_TEMPLATE:
514 r = set_sanitized_path(&arg_template, optarg);
515 if (r < 0)
516 return log_error_errno(r, "Invalid template directory: %m");
517
518 break;
519
520 case 'i':
521 r = set_sanitized_path(&arg_image, optarg);
522 if (r < 0)
523 return log_error_errno(r, "Invalid image path: %m");
524
525 break;
526
527 case 'x':
528 arg_ephemeral = true;
529 break;
530
531 case 'u':
532 r = free_and_strdup(&arg_user, optarg);
533 if (r < 0)
534 return log_oom();
535
536 break;
537
538 case ARG_NETWORK_BRIDGE:
539 arg_network_bridge = optarg;
540
541 /* fall through */
542
543 case 'n':
544 arg_network_veth = true;
545 arg_private_network = true;
546 break;
547
548 case ARG_NETWORK_INTERFACE:
549 if (strv_extend(&arg_network_interfaces, optarg) < 0)
550 return log_oom();
551
552 arg_private_network = true;
553 break;
554
555 case ARG_NETWORK_MACVLAN:
556 if (strv_extend(&arg_network_macvlan, optarg) < 0)
557 return log_oom();
558
559 arg_private_network = true;
560 break;
561
562 case ARG_NETWORK_IPVLAN:
563 if (strv_extend(&arg_network_ipvlan, optarg) < 0)
564 return log_oom();
565
566 /* fall through */
567
568 case ARG_PRIVATE_NETWORK:
569 arg_private_network = true;
570 break;
571
572 case 'b':
573 arg_boot = true;
574 break;
575
576 case ARG_UUID:
577 r = sd_id128_from_string(optarg, &arg_uuid);
578 if (r < 0) {
579 log_error("Invalid UUID: %s", optarg);
580 return r;
581 }
582 break;
583
584 case 'S':
585 arg_slice = optarg;
586 break;
587
588 case 'M':
589 if (isempty(optarg))
590 arg_machine = mfree(arg_machine);
591 else {
592 if (!machine_name_is_valid(optarg)) {
593 log_error("Invalid machine name: %s", optarg);
594 return -EINVAL;
595 }
596
597 r = free_and_strdup(&arg_machine, optarg);
598 if (r < 0)
599 return log_oom();
600
601 break;
602 }
603
604 case 'Z':
605 arg_selinux_context = optarg;
606 break;
607
608 case 'L':
609 arg_selinux_apifs_context = optarg;
610 break;
611
612 case ARG_READ_ONLY:
613 arg_read_only = true;
614 break;
615
616 case ARG_CAPABILITY:
617 case ARG_DROP_CAPABILITY: {
618 const char *state, *word;
619 size_t length;
620
621 FOREACH_WORD_SEPARATOR(word, length, optarg, ",", state) {
622 _cleanup_free_ char *t;
623
624 t = strndup(word, length);
625 if (!t)
626 return log_oom();
627
628 if (streq(t, "all")) {
629 if (c == ARG_CAPABILITY)
630 plus = (uint64_t) -1;
631 else
632 minus = (uint64_t) -1;
633 } else {
634 int cap;
635
636 cap = capability_from_name(t);
637 if (cap < 0) {
638 log_error("Failed to parse capability %s.", t);
639 return -EINVAL;
640 }
641
642 if (c == ARG_CAPABILITY)
643 plus |= 1ULL << (uint64_t) cap;
644 else
645 minus |= 1ULL << (uint64_t) cap;
646 }
647 }
648
649 break;
650 }
651
652 case 'j':
653 arg_link_journal = LINK_GUEST;
654 arg_link_journal_try = true;
655 break;
656
657 case ARG_LINK_JOURNAL:
658 if (streq(optarg, "auto")) {
659 arg_link_journal = LINK_AUTO;
660 arg_link_journal_try = false;
661 } else if (streq(optarg, "no")) {
662 arg_link_journal = LINK_NO;
663 arg_link_journal_try = false;
664 } else if (streq(optarg, "guest")) {
665 arg_link_journal = LINK_GUEST;
666 arg_link_journal_try = false;
667 } else if (streq(optarg, "host")) {
668 arg_link_journal = LINK_HOST;
669 arg_link_journal_try = false;
670 } else if (streq(optarg, "try-guest")) {
671 arg_link_journal = LINK_GUEST;
672 arg_link_journal_try = true;
673 } else if (streq(optarg, "try-host")) {
674 arg_link_journal = LINK_HOST;
675 arg_link_journal_try = true;
676 } else {
677 log_error("Failed to parse link journal mode %s", optarg);
678 return -EINVAL;
679 }
680
681 break;
682
683 case ARG_BIND:
684 case ARG_BIND_RO: {
685 const char *current = optarg;
686 _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL;
687 CustomMount *m;
688
689 r = extract_many_words(&current, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, &opts, NULL);
690 switch (r) {
691 case 1:
692 destination = strdup(source);
693 case 2:
694 case 3:
695 break;
696 case -ENOMEM:
697 return log_oom();
698 default:
699 log_error("Invalid bind mount specification: %s", optarg);
700 return -EINVAL;
701 }
702
703 if (!source || !destination)
704 return log_oom();
705
706 if (!path_is_absolute(source) || !path_is_absolute(destination)) {
707 log_error("Invalid bind mount specification: %s", optarg);
708 return -EINVAL;
709 }
710
711 m = custom_mount_add(CUSTOM_MOUNT_BIND);
712 if (!m)
713 return log_oom();
714
715 m->source = source;
716 m->destination = destination;
717 m->read_only = c == ARG_BIND_RO;
718 m->options = opts;
719
720 source = destination = opts = NULL;
721
722 break;
723 }
724
725 case ARG_TMPFS: {
726 const char *current = optarg;
727 _cleanup_free_ char *path = NULL, *opts = NULL;
728 CustomMount *m;
729
730 r = extract_first_word(&current, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
731 if (r == -ENOMEM)
732 return log_oom();
733 else if (r < 0) {
734 log_error("Invalid tmpfs specification: %s", optarg);
735 return r;
736 }
737 if (r)
738 opts = strdup(current);
739 else
740 opts = strdup("mode=0755");
741
742 if (!path || !opts)
743 return log_oom();
744
745 if (!path_is_absolute(path)) {
746 log_error("Invalid tmpfs specification: %s", optarg);
747 return -EINVAL;
748 }
749
750 m = custom_mount_add(CUSTOM_MOUNT_TMPFS);
751 if (!m)
752 return log_oom();
753
754 m->destination = path;
755 m->options = opts;
756
757 path = opts = NULL;
758
759 break;
760 }
761
762 case ARG_OVERLAY:
763 case ARG_OVERLAY_RO: {
764 _cleanup_free_ char *upper = NULL, *destination = NULL;
765 _cleanup_strv_free_ char **lower = NULL;
766 CustomMount *m;
767 unsigned n = 0;
768 char **i;
769
770 r = strv_split_extract(&lower, optarg, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
771 if (r == -ENOMEM)
772 return log_oom();
773 else if (r < 0) {
774 log_error("Invalid overlay specification: %s", optarg);
775 return r;
776 }
777
778 STRV_FOREACH(i, lower) {
779 if (!path_is_absolute(*i)) {
780 log_error("Overlay path %s is not absolute.", *i);
781 return -EINVAL;
782 }
783
784 n++;
785 }
786
787 if (n < 2) {
788 log_error("--overlay= needs at least two colon-separated directories specified.");
789 return -EINVAL;
790 }
791
792 if (n == 2) {
793 /* If two parameters are specified,
794 * the first one is the lower, the
795 * second one the upper directory. And
796 * we'll also define the destination
797 * mount point the same as the upper. */
798 upper = lower[1];
799 lower[1] = NULL;
800
801 destination = strdup(upper);
802 if (!destination)
803 return log_oom();
804
805 } else {
806 upper = lower[n - 2];
807 destination = lower[n - 1];
808 lower[n - 2] = NULL;
809 }
810
811 m = custom_mount_add(CUSTOM_MOUNT_OVERLAY);
812 if (!m)
813 return log_oom();
814
815 m->destination = destination;
816 m->source = upper;
817 m->lower = lower;
818 m->read_only = c == ARG_OVERLAY_RO;
819
820 upper = destination = NULL;
821 lower = NULL;
822
823 break;
824 }
825
826 case ARG_SETENV: {
827 char **n;
828
829 if (!env_assignment_is_valid(optarg)) {
830 log_error("Environment variable assignment '%s' is not valid.", optarg);
831 return -EINVAL;
832 }
833
834 n = strv_env_set(arg_setenv, optarg);
835 if (!n)
836 return log_oom();
837
838 strv_free(arg_setenv);
839 arg_setenv = n;
840 break;
841 }
842
843 case 'q':
844 arg_quiet = true;
845 break;
846
847 case ARG_SHARE_SYSTEM:
848 arg_share_system = true;
849 break;
850
851 case ARG_REGISTER:
852 r = parse_boolean(optarg);
853 if (r < 0) {
854 log_error("Failed to parse --register= argument: %s", optarg);
855 return r;
856 }
857
858 arg_register = r;
859 break;
860
861 case ARG_KEEP_UNIT:
862 arg_keep_unit = true;
863 break;
864
865 case ARG_PERSONALITY:
866
867 arg_personality = personality_from_string(optarg);
868 if (arg_personality == PERSONALITY_INVALID) {
869 log_error("Unknown or unsupported personality '%s'.", optarg);
870 return -EINVAL;
871 }
872
873 break;
874
875 case ARG_VOLATILE:
876
877 if (!optarg)
878 arg_volatile = VOLATILE_YES;
879 else {
880 r = parse_boolean(optarg);
881 if (r < 0) {
882 if (streq(optarg, "state"))
883 arg_volatile = VOLATILE_STATE;
884 else {
885 log_error("Failed to parse --volatile= argument: %s", optarg);
886 return r;
887 }
888 } else
889 arg_volatile = r ? VOLATILE_YES : VOLATILE_NO;
890 }
891
892 break;
893
894 case 'p': {
895 const char *split, *e;
896 uint16_t container_port, host_port;
897 int protocol;
898 ExposePort *p;
899
900 if ((e = startswith(optarg, "tcp:")))
901 protocol = IPPROTO_TCP;
902 else if ((e = startswith(optarg, "udp:")))
903 protocol = IPPROTO_UDP;
904 else {
905 e = optarg;
906 protocol = IPPROTO_TCP;
907 }
908
909 split = strchr(e, ':');
910 if (split) {
911 char v[split - e + 1];
912
913 memcpy(v, e, split - e);
914 v[split - e] = 0;
915
916 r = safe_atou16(v, &host_port);
917 if (r < 0 || host_port <= 0) {
918 log_error("Failed to parse host port: %s", optarg);
919 return -EINVAL;
920 }
921
922 r = safe_atou16(split + 1, &container_port);
923 } else {
924 r = safe_atou16(e, &container_port);
925 host_port = container_port;
926 }
927
928 if (r < 0 || container_port <= 0) {
929 log_error("Failed to parse host port: %s", optarg);
930 return -EINVAL;
931 }
932
933 LIST_FOREACH(ports, p, arg_expose_ports) {
934 if (p->protocol == protocol && p->host_port == host_port) {
935 log_error("Duplicate port specification: %s", optarg);
936 return -EINVAL;
937 }
938 }
939
940 p = new(ExposePort, 1);
941 if (!p)
942 return log_oom();
943
944 p->protocol = protocol;
945 p->host_port = host_port;
946 p->container_port = container_port;
947
948 LIST_PREPEND(ports, arg_expose_ports, p);
949
950 break;
951 }
952
953 case ARG_PROPERTY:
954 if (strv_extend(&arg_property, optarg) < 0)
955 return log_oom();
956
957 break;
958
959 case ARG_PRIVATE_USERS:
960 if (optarg) {
961 _cleanup_free_ char *buffer = NULL;
962 const char *range, *shift;
963
964 range = strchr(optarg, ':');
965 if (range) {
966 buffer = strndup(optarg, range - optarg);
967 if (!buffer)
968 return log_oom();
969 shift = buffer;
970
971 range++;
972 if (safe_atou32(range, &arg_uid_range) < 0 || arg_uid_range <= 0) {
973 log_error("Failed to parse UID range: %s", range);
974 return -EINVAL;
975 }
976 } else
977 shift = optarg;
978
979 if (parse_uid(shift, &arg_uid_shift) < 0) {
980 log_error("Failed to parse UID: %s", optarg);
981 return -EINVAL;
982 }
983 }
984
985 arg_userns = true;
986 break;
987
988 case ARG_KILL_SIGNAL:
989 arg_kill_signal = signal_from_string_try_harder(optarg);
990 if (arg_kill_signal < 0) {
991 log_error("Cannot parse signal: %s", optarg);
992 return -EINVAL;
993 }
994
995 break;
996
997 case '?':
998 return -EINVAL;
999
1000 default:
1001 assert_not_reached("Unhandled option");
1002 }
1003
1004 if (arg_share_system)
1005 arg_register = false;
1006
1007 if (arg_boot && arg_share_system) {
1008 log_error("--boot and --share-system may not be combined.");
1009 return -EINVAL;
1010 }
1011
1012 if (arg_keep_unit && cg_pid_get_owner_uid(0, NULL) >= 0) {
1013 log_error("--keep-unit may not be used when invoked from a user session.");
1014 return -EINVAL;
1015 }
1016
1017 if (arg_directory && arg_image) {
1018 log_error("--directory= and --image= may not be combined.");
1019 return -EINVAL;
1020 }
1021
1022 if (arg_template && arg_image) {
1023 log_error("--template= and --image= may not be combined.");
1024 return -EINVAL;
1025 }
1026
1027 if (arg_template && !(arg_directory || arg_machine)) {
1028 log_error("--template= needs --directory= or --machine=.");
1029 return -EINVAL;
1030 }
1031
1032 if (arg_ephemeral && arg_template) {
1033 log_error("--ephemeral and --template= may not be combined.");
1034 return -EINVAL;
1035 }
1036
1037 if (arg_ephemeral && arg_image) {
1038 log_error("--ephemeral and --image= may not be combined.");
1039 return -EINVAL;
1040 }
1041
1042 if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO)) {
1043 log_error("--ephemeral and --link-journal= may not be combined.");
1044 return -EINVAL;
1045 }
1046
1047 if (arg_volatile != VOLATILE_NO && arg_read_only) {
1048 log_error("Cannot combine --read-only with --volatile. Note that --volatile already implies a read-only base hierarchy.");
1049 return -EINVAL;
1050 }
1051
1052 if (arg_expose_ports && !arg_private_network) {
1053 log_error("Cannot use --port= without private networking.");
1054 return -EINVAL;
1055 }
1056
1057 if (arg_userns && access("/proc/self/uid_map", F_OK) < 0)
1058 return log_error_errno(EOPNOTSUPP, "--private-users= is not supported, kernel compiled without user namespace support.");
1059
1060 arg_retain = (arg_retain | plus | (arg_private_network ? 1ULL << CAP_NET_ADMIN : 0)) & ~minus;
1061
1062 if (arg_boot && arg_kill_signal <= 0)
1063 arg_kill_signal = SIGRTMIN+3;
1064
1065 r = detect_unified_cgroup_hierarchy();
1066 if (r < 0)
1067 return r;
1068
1069 return 1;
1070 }
1071
1072 static int tmpfs_patch_options(const char *options, char **ret) {
1073 char *buf = NULL;
1074
1075 if (arg_userns && arg_uid_shift != 0) {
1076 assert(arg_uid_shift != UID_INVALID);
1077
1078 if (options)
1079 (void) asprintf(&buf, "%s,uid=" UID_FMT ",gid=" UID_FMT, options, arg_uid_shift, arg_uid_shift);
1080 else
1081 (void) asprintf(&buf, "uid=" UID_FMT ",gid=" UID_FMT, arg_uid_shift, arg_uid_shift);
1082 if (!buf)
1083 return -ENOMEM;
1084
1085 options = buf;
1086 }
1087
1088 #ifdef HAVE_SELINUX
1089 if (arg_selinux_apifs_context) {
1090 char *t;
1091
1092 if (options)
1093 t = strjoin(options, ",context=\"", arg_selinux_apifs_context, "\"", NULL);
1094 else
1095 t = strjoin("context=\"", arg_selinux_apifs_context, "\"", NULL);
1096 if (!t) {
1097 free(buf);
1098 return -ENOMEM;
1099 }
1100
1101 free(buf);
1102 buf = t;
1103 }
1104 #endif
1105
1106 *ret = buf;
1107 return !!buf;
1108 }
1109
1110 static int mount_all(const char *dest, bool userns) {
1111
1112 typedef struct MountPoint {
1113 const char *what;
1114 const char *where;
1115 const char *type;
1116 const char *options;
1117 unsigned long flags;
1118 bool fatal;
1119 bool userns;
1120 } MountPoint;
1121
1122 static const MountPoint mount_table[] = {
1123 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true, true },
1124 { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND, true, true }, /* Bind mount first */
1125 { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, true, true }, /* Then, make it r/o */
1126 { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, true, false },
1127 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME, true, false },
1128 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true, false },
1129 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true, false },
1130 { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME, true, false },
1131 #ifdef HAVE_SELINUX
1132 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND, false, false }, /* Bind mount first */
1133 { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, false, false }, /* Then, make it r/o */
1134 #endif
1135 };
1136
1137 unsigned k;
1138 int r;
1139
1140 for (k = 0; k < ELEMENTSOF(mount_table); k++) {
1141 _cleanup_free_ char *where = NULL, *options = NULL;
1142 const char *o;
1143
1144 if (userns != mount_table[k].userns)
1145 continue;
1146
1147 where = prefix_root(dest, mount_table[k].where);
1148 if (!where)
1149 return log_oom();
1150
1151 r = path_is_mount_point(where, AT_SYMLINK_FOLLOW);
1152 if (r < 0 && r != -ENOENT)
1153 return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
1154
1155 /* Skip this entry if it is not a remount. */
1156 if (mount_table[k].what && r > 0)
1157 continue;
1158
1159 r = mkdir_p(where, 0755);
1160 if (r < 0) {
1161 if (mount_table[k].fatal)
1162 return log_error_errno(r, "Failed to create directory %s: %m", where);
1163
1164 log_warning_errno(r, "Failed to create directory %s: %m", where);
1165 continue;
1166 }
1167
1168 o = mount_table[k].options;
1169 if (streq_ptr(mount_table[k].type, "tmpfs")) {
1170 r = tmpfs_patch_options(o, &options);
1171 if (r < 0)
1172 return log_oom();
1173 if (r > 0)
1174 o = options;
1175 }
1176
1177 if (mount(mount_table[k].what,
1178 where,
1179 mount_table[k].type,
1180 mount_table[k].flags,
1181 o) < 0) {
1182
1183 if (mount_table[k].fatal)
1184 return log_error_errno(errno, "mount(%s) failed: %m", where);
1185
1186 log_warning_errno(errno, "mount(%s) failed, ignoring: %m", where);
1187 }
1188 }
1189
1190 return 0;
1191 }
1192
1193 static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts) {
1194 const char *p = options;
1195 unsigned long flags = *mount_flags;
1196 char *opts = NULL;
1197
1198 assert(options);
1199
1200 for (;;) {
1201 _cleanup_free_ char *word = NULL;
1202 int r = extract_first_word(&p, &word, ",", 0);
1203 if (r < 0)
1204 return log_error_errno(r, "Failed to extract mount option: %m");
1205 if (r == 0)
1206 break;
1207
1208 if (streq(word, "rbind"))
1209 flags |= MS_REC;
1210 else if (streq(word, "norbind"))
1211 flags &= ~MS_REC;
1212 else {
1213 log_error("Invalid bind mount option: %s", word);
1214 return -EINVAL;
1215 }
1216 }
1217
1218 *mount_flags = flags;
1219 /* in the future mount_opts will hold string options for mount(2) */
1220 *mount_opts = opts;
1221
1222 return 0;
1223 }
1224
1225 static int mount_bind(const char *dest, CustomMount *m) {
1226 struct stat source_st, dest_st;
1227 const char *where;
1228 unsigned long mount_flags = MS_BIND | MS_REC;
1229 _cleanup_free_ char *mount_opts = NULL;
1230 int r;
1231
1232 assert(m);
1233
1234 if (m->options) {
1235 r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts);
1236 if (r < 0)
1237 return r;
1238 }
1239
1240 if (stat(m->source, &source_st) < 0)
1241 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
1242
1243 where = prefix_roota(dest, m->destination);
1244
1245 if (stat(where, &dest_st) >= 0) {
1246 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) {
1247 log_error("Cannot bind mount directory %s on file %s.", m->source, where);
1248 return -EINVAL;
1249 }
1250
1251 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) {
1252 log_error("Cannot bind mount file %s on directory %s.", m->source, where);
1253 return -EINVAL;
1254 }
1255
1256 } else if (errno == ENOENT) {
1257 r = mkdir_parents_label(where, 0755);
1258 if (r < 0)
1259 return log_error_errno(r, "Failed to make parents of %s: %m", where);
1260 } else {
1261 log_error_errno(errno, "Failed to stat %s: %m", where);
1262 return -errno;
1263 }
1264
1265 /* Create the mount point. Any non-directory file can be
1266 * mounted on any non-directory file (regular, fifo, socket,
1267 * char, block).
1268 */
1269 if (S_ISDIR(source_st.st_mode))
1270 r = mkdir_label(where, 0755);
1271 else
1272 r = touch(where);
1273 if (r < 0 && r != -EEXIST)
1274 return log_error_errno(r, "Failed to create mount point %s: %m", where);
1275
1276 if (mount(m->source, where, NULL, mount_flags, mount_opts) < 0)
1277 return log_error_errno(errno, "mount(%s) failed: %m", where);
1278
1279 if (m->read_only) {
1280 r = bind_remount_recursive(where, true);
1281 if (r < 0)
1282 return log_error_errno(r, "Read-only bind mount failed: %m");
1283 }
1284
1285 return 0;
1286 }
1287
1288 static int mount_tmpfs(const char *dest, CustomMount *m) {
1289 const char *where, *options;
1290 _cleanup_free_ char *buf = NULL;
1291 int r;
1292
1293 assert(dest);
1294 assert(m);
1295
1296 where = prefix_roota(dest, m->destination);
1297
1298 r = mkdir_p_label(where, 0755);
1299 if (r < 0 && r != -EEXIST)
1300 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
1301
1302 r = tmpfs_patch_options(m->options, &buf);
1303 if (r < 0)
1304 return log_oom();
1305 options = r > 0 ? buf : m->options;
1306
1307 if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options) < 0)
1308 return log_error_errno(errno, "tmpfs mount to %s failed: %m", where);
1309
1310 return 0;
1311 }
1312
1313 static char *joined_and_escaped_lower_dirs(char * const *lower) {
1314 _cleanup_strv_free_ char **sv = NULL;
1315
1316 sv = strv_copy(lower);
1317 if (!sv)
1318 return NULL;
1319
1320 strv_reverse(sv);
1321
1322 if (!strv_shell_escape(sv, ",:"))
1323 return NULL;
1324
1325 return strv_join(sv, ":");
1326 }
1327
1328 static int mount_overlay(const char *dest, CustomMount *m) {
1329 _cleanup_free_ char *lower = NULL;
1330 const char *where, *options;
1331 int r;
1332
1333 assert(dest);
1334 assert(m);
1335
1336 where = prefix_roota(dest, m->destination);
1337
1338 r = mkdir_label(where, 0755);
1339 if (r < 0 && r != -EEXIST)
1340 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
1341
1342 (void) mkdir_p_label(m->source, 0755);
1343
1344 lower = joined_and_escaped_lower_dirs(m->lower);
1345 if (!lower)
1346 return log_oom();
1347
1348 if (m->read_only) {
1349 _cleanup_free_ char *escaped_source = NULL;
1350
1351 escaped_source = shell_escape(m->source, ",:");
1352 if (!escaped_source)
1353 return log_oom();
1354
1355 options = strjoina("lowerdir=", escaped_source, ":", lower);
1356 } else {
1357 _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL;
1358
1359 assert(m->work_dir);
1360 (void) mkdir_label(m->work_dir, 0700);
1361
1362 escaped_source = shell_escape(m->source, ",:");
1363 if (!escaped_source)
1364 return log_oom();
1365 escaped_work_dir = shell_escape(m->work_dir, ",:");
1366 if (!escaped_work_dir)
1367 return log_oom();
1368
1369 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
1370 }
1371
1372 if (mount("overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options) < 0)
1373 return log_error_errno(errno, "overlay mount to %s failed: %m", where);
1374
1375 return 0;
1376 }
1377
1378 static int mount_custom(const char *dest) {
1379 unsigned i;
1380 int r;
1381
1382 assert(dest);
1383
1384 for (i = 0; i < arg_n_custom_mounts; i++) {
1385 CustomMount *m = &arg_custom_mounts[i];
1386
1387 switch (m->type) {
1388
1389 case CUSTOM_MOUNT_BIND:
1390 r = mount_bind(dest, m);
1391 break;
1392
1393 case CUSTOM_MOUNT_TMPFS:
1394 r = mount_tmpfs(dest, m);
1395 break;
1396
1397 case CUSTOM_MOUNT_OVERLAY:
1398 r = mount_overlay(dest, m);
1399 break;
1400
1401 default:
1402 assert_not_reached("Unknown custom mount type");
1403 }
1404
1405 if (r < 0)
1406 return r;
1407 }
1408
1409 return 0;
1410 }
1411
1412 static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, bool read_only) {
1413 char *to;
1414 int r;
1415
1416 to = strjoina(dest, "/sys/fs/cgroup/", hierarchy);
1417
1418 r = path_is_mount_point(to, 0);
1419 if (r < 0 && r != -ENOENT)
1420 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
1421 if (r > 0)
1422 return 0;
1423
1424 mkdir_p(to, 0755);
1425
1426 /* The superblock mount options of the mount point need to be
1427 * identical to the hosts', and hence writable... */
1428 if (mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller) < 0)
1429 return log_error_errno(errno, "Failed to mount to %s: %m", to);
1430
1431 /* ... hence let's only make the bind mount read-only, not the
1432 * superblock. */
1433 if (read_only) {
1434 if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
1435 return log_error_errno(errno, "Failed to remount %s read-only: %m", to);
1436 }
1437 return 1;
1438 }
1439
1440 static int mount_legacy_cgroups(const char *dest) {
1441 _cleanup_set_free_free_ Set *controllers = NULL;
1442 const char *cgroup_root;
1443 int r;
1444
1445 cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
1446
1447 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
1448 r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW);
1449 if (r < 0)
1450 return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
1451 if (r == 0) {
1452 _cleanup_free_ char *options = NULL;
1453
1454 r = tmpfs_patch_options("mode=755", &options);
1455 if (r < 0)
1456 return log_oom();
1457
1458 if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0)
1459 return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m");
1460 }
1461
1462 if (cg_unified() > 0)
1463 goto skip_controllers;
1464
1465 controllers = set_new(&string_hash_ops);
1466 if (!controllers)
1467 return log_oom();
1468
1469 r = cg_kernel_controllers(controllers);
1470 if (r < 0)
1471 return log_error_errno(r, "Failed to determine cgroup controllers: %m");
1472
1473 for (;;) {
1474 _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
1475
1476 controller = set_steal_first(controllers);
1477 if (!controller)
1478 break;
1479
1480 origin = prefix_root("/sys/fs/cgroup/", controller);
1481 if (!origin)
1482 return log_oom();
1483
1484 r = readlink_malloc(origin, &combined);
1485 if (r == -EINVAL) {
1486 /* Not a symbolic link, but directly a single cgroup hierarchy */
1487
1488 r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
1489 if (r < 0)
1490 return r;
1491
1492 } else if (r < 0)
1493 return log_error_errno(r, "Failed to read link %s: %m", origin);
1494 else {
1495 _cleanup_free_ char *target = NULL;
1496
1497 target = prefix_root(dest, origin);
1498 if (!target)
1499 return log_oom();
1500
1501 /* A symbolic link, a combination of controllers in one hierarchy */
1502
1503 if (!filename_is_valid(combined)) {
1504 log_warning("Ignoring invalid combined hierarchy %s.", combined);
1505 continue;
1506 }
1507
1508 r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
1509 if (r < 0)
1510 return r;
1511
1512 r = symlink_idempotent(combined, target);
1513 if (r == -EINVAL) {
1514 log_error("Invalid existing symlink for combined hierarchy");
1515 return r;
1516 }
1517 if (r < 0)
1518 return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
1519 }
1520 }
1521
1522 skip_controllers:
1523 r = mount_legacy_cgroup_hierarchy(dest, "none,name=systemd,xattr", "systemd", false);
1524 if (r < 0)
1525 return r;
1526
1527 if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0)
1528 return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root);
1529
1530 return 0;
1531 }
1532
1533 static int mount_unified_cgroups(const char *dest) {
1534 const char *p;
1535 int r;
1536
1537 assert(dest);
1538
1539 p = strjoina(dest, "/sys/fs/cgroup");
1540
1541 r = path_is_mount_point(p, AT_SYMLINK_FOLLOW);
1542 if (r < 0)
1543 return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
1544 if (r > 0) {
1545 p = strjoina(dest, "/sys/fs/cgroup/cgroup.procs");
1546 if (access(p, F_OK) >= 0)
1547 return 0;
1548 if (errno != ENOENT)
1549 return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
1550
1551 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
1552 return -EINVAL;
1553 }
1554
1555 if (mount("cgroup", p, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "__DEVEL__sane_behavior") < 0)
1556 return log_error_errno(errno, "Failed to mount unified cgroup hierarchy to %s: %m", p);
1557
1558 return 0;
1559 }
1560
1561 static int mount_cgroups(const char *dest) {
1562 if (arg_unified_cgroup_hierarchy)
1563 return mount_unified_cgroups(dest);
1564 else
1565 return mount_legacy_cgroups(dest);
1566 }
1567
1568 static int mount_systemd_cgroup_writable(const char *dest) {
1569 _cleanup_free_ char *own_cgroup_path = NULL;
1570 const char *systemd_root, *systemd_own;
1571 int r;
1572
1573 assert(dest);
1574
1575 r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
1576 if (r < 0)
1577 return log_error_errno(r, "Failed to determine our own cgroup path: %m");
1578
1579 /* If we are living in the top-level, then there's nothing to do... */
1580 if (path_equal(own_cgroup_path, "/"))
1581 return 0;
1582
1583 if (arg_unified_cgroup_hierarchy) {
1584 systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path);
1585 systemd_root = prefix_roota(dest, "/sys/fs/cgroup");
1586 } else {
1587 systemd_own = strjoina(dest, "/sys/fs/cgroup/systemd", own_cgroup_path);
1588 systemd_root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
1589 }
1590
1591 /* Make our own cgroup a (writable) bind mount */
1592 if (mount(systemd_own, systemd_own, NULL, MS_BIND, NULL) < 0)
1593 return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path);
1594
1595 /* And then remount the systemd cgroup root read-only */
1596 if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0)
1597 return log_error_errno(errno, "Failed to mount cgroup root read-only: %m");
1598
1599 return 0;
1600 }
1601
1602 static int userns_lchown(const char *p, uid_t uid, gid_t gid) {
1603 assert(p);
1604
1605 if (!arg_userns)
1606 return 0;
1607
1608 if (uid == UID_INVALID && gid == GID_INVALID)
1609 return 0;
1610
1611 if (uid != UID_INVALID) {
1612 uid += arg_uid_shift;
1613
1614 if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
1615 return -EOVERFLOW;
1616 }
1617
1618 if (gid != GID_INVALID) {
1619 gid += (gid_t) arg_uid_shift;
1620
1621 if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
1622 return -EOVERFLOW;
1623 }
1624
1625 if (lchown(p, uid, gid) < 0)
1626 return -errno;
1627
1628 return 0;
1629 }
1630
1631 static int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
1632 const char *q;
1633
1634 q = prefix_roota(root, path);
1635 if (mkdir(q, mode) < 0) {
1636 if (errno == EEXIST)
1637 return 0;
1638 return -errno;
1639 }
1640
1641 return userns_lchown(q, uid, gid);
1642 }
1643
1644 static int setup_timezone(const char *dest) {
1645 _cleanup_free_ char *p = NULL, *q = NULL;
1646 const char *where, *check, *what;
1647 char *z, *y;
1648 int r;
1649
1650 assert(dest);
1651
1652 /* Fix the timezone, if possible */
1653 r = readlink_malloc("/etc/localtime", &p);
1654 if (r < 0) {
1655 log_warning("/etc/localtime is not a symlink, not updating container timezone.");
1656 return 0;
1657 }
1658
1659 z = path_startswith(p, "../usr/share/zoneinfo/");
1660 if (!z)
1661 z = path_startswith(p, "/usr/share/zoneinfo/");
1662 if (!z) {
1663 log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
1664 return 0;
1665 }
1666
1667 where = prefix_roota(dest, "/etc/localtime");
1668 r = readlink_malloc(where, &q);
1669 if (r >= 0) {
1670 y = path_startswith(q, "../usr/share/zoneinfo/");
1671 if (!y)
1672 y = path_startswith(q, "/usr/share/zoneinfo/");
1673
1674 /* Already pointing to the right place? Then do nothing .. */
1675 if (y && streq(y, z))
1676 return 0;
1677 }
1678
1679 check = strjoina("/usr/share/zoneinfo/", z);
1680 check = prefix_root(dest, check);
1681 if (laccess(check, F_OK) < 0) {
1682 log_warning("Timezone %s does not exist in container, not updating container timezone.", z);
1683 return 0;
1684 }
1685
1686 r = unlink(where);
1687 if (r < 0 && errno != ENOENT) {
1688 log_error_errno(errno, "Failed to remove existing timezone info %s in container: %m", where);
1689 return 0;
1690 }
1691
1692 what = strjoina("../usr/share/zoneinfo/", z);
1693 if (symlink(what, where) < 0) {
1694 log_error_errno(errno, "Failed to correct timezone of container: %m");
1695 return 0;
1696 }
1697
1698 r = userns_lchown(where, 0, 0);
1699 if (r < 0)
1700 return log_warning_errno(r, "Failed to chown /etc/localtime: %m");
1701
1702 return 0;
1703 }
1704
1705 static int setup_resolv_conf(const char *dest) {
1706 const char *where = NULL;
1707 int r;
1708
1709 assert(dest);
1710
1711 if (arg_private_network)
1712 return 0;
1713
1714 /* Fix resolv.conf, if possible */
1715 where = prefix_roota(dest, "/etc/resolv.conf");
1716
1717 r = copy_file("/etc/resolv.conf", where, O_TRUNC|O_NOFOLLOW, 0644, 0);
1718 if (r < 0) {
1719 /* If the file already exists as symlink, let's
1720 * suppress the warning, under the assumption that
1721 * resolved or something similar runs inside and the
1722 * symlink points there.
1723 *
1724 * If the disk image is read-only, there's also no
1725 * point in complaining.
1726 */
1727 log_full_errno(IN_SET(r, -ELOOP, -EROFS) ? LOG_DEBUG : LOG_WARNING, r,
1728 "Failed to copy /etc/resolv.conf to %s: %m", where);
1729 return 0;
1730 }
1731
1732 r = userns_lchown(where, 0, 0);
1733 if (r < 0)
1734 log_warning_errno(r, "Failed to chown /etc/resolv.conf: %m");
1735
1736 return 0;
1737 }
1738
1739 static int setup_volatile_state(const char *directory) {
1740 _cleanup_free_ char *buf = NULL;
1741 const char *p, *options;
1742 int r;
1743
1744 assert(directory);
1745
1746 if (arg_volatile != VOLATILE_STATE)
1747 return 0;
1748
1749 /* --volatile=state means we simply overmount /var
1750 with a tmpfs, and the rest read-only. */
1751
1752 r = bind_remount_recursive(directory, true);
1753 if (r < 0)
1754 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1755
1756 p = prefix_roota(directory, "/var");
1757 r = mkdir(p, 0755);
1758 if (r < 0 && errno != EEXIST)
1759 return log_error_errno(errno, "Failed to create %s: %m", directory);
1760
1761 options = "mode=755";
1762 r = tmpfs_patch_options(options, &buf);
1763 if (r < 0)
1764 return log_oom();
1765 if (r > 0)
1766 options = buf;
1767
1768 if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, options) < 0)
1769 return log_error_errno(errno, "Failed to mount tmpfs to /var: %m");
1770
1771 return 0;
1772 }
1773
1774 static int setup_volatile(const char *directory) {
1775 bool tmpfs_mounted = false, bind_mounted = false;
1776 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1777 _cleanup_free_ char *buf = NULL;
1778 const char *f, *t, *options;
1779 int r;
1780
1781 assert(directory);
1782
1783 if (arg_volatile != VOLATILE_YES)
1784 return 0;
1785
1786 /* --volatile=yes means we mount a tmpfs to the root dir, and
1787 the original /usr to use inside it, and that read-only. */
1788
1789 if (!mkdtemp(template))
1790 return log_error_errno(errno, "Failed to create temporary directory: %m");
1791
1792 options = "mode=755";
1793 r = tmpfs_patch_options(options, &buf);
1794 if (r < 0)
1795 return log_oom();
1796 if (r > 0)
1797 options = buf;
1798
1799 if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, options) < 0) {
1800 r = log_error_errno(errno, "Failed to mount tmpfs for root directory: %m");
1801 goto fail;
1802 }
1803
1804 tmpfs_mounted = true;
1805
1806 f = prefix_roota(directory, "/usr");
1807 t = prefix_roota(template, "/usr");
1808
1809 r = mkdir(t, 0755);
1810 if (r < 0 && errno != EEXIST) {
1811 r = log_error_errno(errno, "Failed to create %s: %m", t);
1812 goto fail;
1813 }
1814
1815 if (mount(f, t, NULL, MS_BIND|MS_REC, NULL) < 0) {
1816 r = log_error_errno(errno, "Failed to create /usr bind mount: %m");
1817 goto fail;
1818 }
1819
1820 bind_mounted = true;
1821
1822 r = bind_remount_recursive(t, true);
1823 if (r < 0) {
1824 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1825 goto fail;
1826 }
1827
1828 if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) {
1829 r = log_error_errno(errno, "Failed to move root mount: %m");
1830 goto fail;
1831 }
1832
1833 (void) rmdir(template);
1834
1835 return 0;
1836
1837 fail:
1838 if (bind_mounted)
1839 (void) umount(t);
1840
1841 if (tmpfs_mounted)
1842 (void) umount(template);
1843 (void) rmdir(template);
1844 return r;
1845 }
1846
1847 static char* id128_format_as_uuid(sd_id128_t id, char s[37]) {
1848 assert(s);
1849
1850 snprintf(s, 37,
1851 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-%02x%02x%02x%02x%02x%02x",
1852 SD_ID128_FORMAT_VAL(id));
1853
1854 return s;
1855 }
1856
1857 static int setup_boot_id(const char *dest) {
1858 const char *from, *to;
1859 sd_id128_t rnd = {};
1860 char as_uuid[37];
1861 int r;
1862
1863 if (arg_share_system)
1864 return 0;
1865
1866 /* Generate a new randomized boot ID, so that each boot-up of
1867 * the container gets a new one */
1868
1869 from = prefix_roota(dest, "/run/proc-sys-kernel-random-boot-id");
1870 to = prefix_roota(dest, "/proc/sys/kernel/random/boot_id");
1871
1872 r = sd_id128_randomize(&rnd);
1873 if (r < 0)
1874 return log_error_errno(r, "Failed to generate random boot id: %m");
1875
1876 id128_format_as_uuid(rnd, as_uuid);
1877
1878 r = write_string_file(from, as_uuid, WRITE_STRING_FILE_CREATE);
1879 if (r < 0)
1880 return log_error_errno(r, "Failed to write boot id: %m");
1881
1882 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
1883 r = log_error_errno(errno, "Failed to bind mount boot id: %m");
1884 else if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL) < 0)
1885 log_warning_errno(errno, "Failed to make boot id read-only: %m");
1886
1887 unlink(from);
1888 return r;
1889 }
1890
1891 static int copy_devnodes(const char *dest) {
1892
1893 static const char devnodes[] =
1894 "null\0"
1895 "zero\0"
1896 "full\0"
1897 "random\0"
1898 "urandom\0"
1899 "tty\0"
1900 "net/tun\0";
1901
1902 const char *d;
1903 int r = 0;
1904 _cleanup_umask_ mode_t u;
1905
1906 assert(dest);
1907
1908 u = umask(0000);
1909
1910 /* Create /dev/net, so that we can create /dev/net/tun in it */
1911 if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
1912 return log_error_errno(r, "Failed to create /dev/net directory: %m");
1913
1914 NULSTR_FOREACH(d, devnodes) {
1915 _cleanup_free_ char *from = NULL, *to = NULL;
1916 struct stat st;
1917
1918 from = strappend("/dev/", d);
1919 to = prefix_root(dest, from);
1920
1921 if (stat(from, &st) < 0) {
1922
1923 if (errno != ENOENT)
1924 return log_error_errno(errno, "Failed to stat %s: %m", from);
1925
1926 } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
1927
1928 log_error("%s is not a char or block device, cannot copy.", from);
1929 return -EIO;
1930
1931 } else {
1932 if (mknod(to, st.st_mode, st.st_rdev) < 0) {
1933 if (errno != EPERM)
1934 return log_error_errno(errno, "mknod(%s) failed: %m", to);
1935
1936 /* Some systems abusively restrict mknod but
1937 * allow bind mounts. */
1938 r = touch(to);
1939 if (r < 0)
1940 return log_error_errno(r, "touch (%s) failed: %m", to);
1941 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
1942 return log_error_errno(errno, "Both mknod and bind mount (%s) failed: %m", to);
1943 }
1944
1945 r = userns_lchown(to, 0, 0);
1946 if (r < 0)
1947 return log_error_errno(r, "chown() of device node %s failed: %m", to);
1948 }
1949 }
1950
1951 return r;
1952 }
1953
1954 static int setup_pts(const char *dest) {
1955 _cleanup_free_ char *options = NULL;
1956 const char *p;
1957
1958 #ifdef HAVE_SELINUX
1959 if (arg_selinux_apifs_context)
1960 (void) asprintf(&options,
1961 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
1962 arg_uid_shift + TTY_GID,
1963 arg_selinux_apifs_context);
1964 else
1965 #endif
1966 (void) asprintf(&options,
1967 "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
1968 arg_uid_shift + TTY_GID);
1969
1970 if (!options)
1971 return log_oom();
1972
1973 /* Mount /dev/pts itself */
1974 p = prefix_roota(dest, "/dev/pts");
1975 if (mkdir(p, 0755) < 0)
1976 return log_error_errno(errno, "Failed to create /dev/pts: %m");
1977 if (mount("devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options) < 0)
1978 return log_error_errno(errno, "Failed to mount /dev/pts: %m");
1979 if (userns_lchown(p, 0, 0) < 0)
1980 return log_error_errno(errno, "Failed to chown /dev/pts: %m");
1981
1982 /* Create /dev/ptmx symlink */
1983 p = prefix_roota(dest, "/dev/ptmx");
1984 if (symlink("pts/ptmx", p) < 0)
1985 return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
1986 if (userns_lchown(p, 0, 0) < 0)
1987 return log_error_errno(errno, "Failed to chown /dev/ptmx: %m");
1988
1989 /* And fix /dev/pts/ptmx ownership */
1990 p = prefix_roota(dest, "/dev/pts/ptmx");
1991 if (userns_lchown(p, 0, 0) < 0)
1992 return log_error_errno(errno, "Failed to chown /dev/pts/ptmx: %m");
1993
1994 return 0;
1995 }
1996
1997 static int setup_dev_console(const char *dest, const char *console) {
1998 _cleanup_umask_ mode_t u;
1999 const char *to;
2000 int r;
2001
2002 assert(dest);
2003 assert(console);
2004
2005 u = umask(0000);
2006
2007 r = chmod_and_chown(console, 0600, arg_uid_shift, arg_uid_shift);
2008 if (r < 0)
2009 return log_error_errno(r, "Failed to correct access mode for TTY: %m");
2010
2011 /* We need to bind mount the right tty to /dev/console since
2012 * ptys can only exist on pts file systems. To have something
2013 * to bind mount things on we create a empty regular file. */
2014
2015 to = prefix_roota(dest, "/dev/console");
2016 r = touch(to);
2017 if (r < 0)
2018 return log_error_errno(r, "touch() for /dev/console failed: %m");
2019
2020 if (mount(console, to, NULL, MS_BIND, NULL) < 0)
2021 return log_error_errno(errno, "Bind mount for /dev/console failed: %m");
2022
2023 return 0;
2024 }
2025
2026 static int setup_kmsg(const char *dest, int kmsg_socket) {
2027 const char *from, *to;
2028 _cleanup_umask_ mode_t u;
2029 int fd, k;
2030 union {
2031 struct cmsghdr cmsghdr;
2032 uint8_t buf[CMSG_SPACE(sizeof(int))];
2033 } control = {};
2034 struct msghdr mh = {
2035 .msg_control = &control,
2036 .msg_controllen = sizeof(control),
2037 };
2038 struct cmsghdr *cmsg;
2039
2040 assert(kmsg_socket >= 0);
2041
2042 u = umask(0000);
2043
2044 /* We create the kmsg FIFO as /run/kmsg, but immediately
2045 * delete it after bind mounting it to /proc/kmsg. While FIFOs
2046 * on the reading side behave very similar to /proc/kmsg,
2047 * their writing side behaves differently from /dev/kmsg in
2048 * that writing blocks when nothing is reading. In order to
2049 * avoid any problems with containers deadlocking due to this
2050 * we simply make /dev/kmsg unavailable to the container. */
2051 from = prefix_roota(dest, "/run/kmsg");
2052 to = prefix_roota(dest, "/proc/kmsg");
2053
2054 if (mkfifo(from, 0600) < 0)
2055 return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
2056 if (mount(from, to, NULL, MS_BIND, NULL) < 0)
2057 return log_error_errno(errno, "Bind mount for /proc/kmsg failed: %m");
2058
2059 fd = open(from, O_RDWR|O_NDELAY|O_CLOEXEC);
2060 if (fd < 0)
2061 return log_error_errno(errno, "Failed to open fifo: %m");
2062
2063 cmsg = CMSG_FIRSTHDR(&mh);
2064 cmsg->cmsg_level = SOL_SOCKET;
2065 cmsg->cmsg_type = SCM_RIGHTS;
2066 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
2067 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
2068
2069 mh.msg_controllen = cmsg->cmsg_len;
2070
2071 /* Store away the fd in the socket, so that it stays open as
2072 * long as we run the child */
2073 k = sendmsg(kmsg_socket, &mh, MSG_NOSIGNAL);
2074 safe_close(fd);
2075
2076 if (k < 0)
2077 return log_error_errno(errno, "Failed to send FIFO fd: %m");
2078
2079 /* And now make the FIFO unavailable as /run/kmsg... */
2080 (void) unlink(from);
2081
2082 return 0;
2083 }
2084
2085 static int send_rtnl(int send_fd) {
2086 union {
2087 struct cmsghdr cmsghdr;
2088 uint8_t buf[CMSG_SPACE(sizeof(int))];
2089 } control = {};
2090 struct msghdr mh = {
2091 .msg_control = &control,
2092 .msg_controllen = sizeof(control),
2093 };
2094 struct cmsghdr *cmsg;
2095 _cleanup_close_ int fd = -1;
2096 ssize_t k;
2097
2098 assert(send_fd >= 0);
2099
2100 if (!arg_expose_ports)
2101 return 0;
2102
2103 fd = socket(PF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_ROUTE);
2104 if (fd < 0)
2105 return log_error_errno(errno, "Failed to allocate container netlink: %m");
2106
2107 cmsg = CMSG_FIRSTHDR(&mh);
2108 cmsg->cmsg_level = SOL_SOCKET;
2109 cmsg->cmsg_type = SCM_RIGHTS;
2110 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
2111 memcpy(CMSG_DATA(cmsg), &fd, sizeof(int));
2112
2113 mh.msg_controllen = cmsg->cmsg_len;
2114
2115 /* Store away the fd in the socket, so that it stays open as
2116 * long as we run the child */
2117 k = sendmsg(send_fd, &mh, MSG_NOSIGNAL);
2118 if (k < 0)
2119 return log_error_errno(errno, "Failed to send netlink fd: %m");
2120
2121 return 0;
2122 }
2123
2124 static int flush_ports(union in_addr_union *exposed) {
2125 ExposePort *p;
2126 int r, af = AF_INET;
2127
2128 assert(exposed);
2129
2130 if (!arg_expose_ports)
2131 return 0;
2132
2133 if (in_addr_is_null(af, exposed))
2134 return 0;
2135
2136 log_debug("Lost IP address.");
2137
2138 LIST_FOREACH(ports, p, arg_expose_ports) {
2139 r = fw_add_local_dnat(false,
2140 af,
2141 p->protocol,
2142 NULL,
2143 NULL, 0,
2144 NULL, 0,
2145 p->host_port,
2146 exposed,
2147 p->container_port,
2148 NULL);
2149 if (r < 0)
2150 log_warning_errno(r, "Failed to modify firewall: %m");
2151 }
2152
2153 *exposed = IN_ADDR_NULL;
2154 return 0;
2155 }
2156
2157 static int expose_ports(sd_netlink *rtnl, union in_addr_union *exposed) {
2158 _cleanup_free_ struct local_address *addresses = NULL;
2159 _cleanup_free_ char *pretty = NULL;
2160 union in_addr_union new_exposed;
2161 ExposePort *p;
2162 bool add;
2163 int af = AF_INET, r;
2164
2165 assert(exposed);
2166
2167 /* Invoked each time an address is added or removed inside the
2168 * container */
2169
2170 if (!arg_expose_ports)
2171 return 0;
2172
2173 r = local_addresses(rtnl, 0, af, &addresses);
2174 if (r < 0)
2175 return log_error_errno(r, "Failed to enumerate local addresses: %m");
2176
2177 add = r > 0 &&
2178 addresses[0].family == af &&
2179 addresses[0].scope < RT_SCOPE_LINK;
2180
2181 if (!add)
2182 return flush_ports(exposed);
2183
2184 new_exposed = addresses[0].address;
2185 if (in_addr_equal(af, exposed, &new_exposed))
2186 return 0;
2187
2188 in_addr_to_string(af, &new_exposed, &pretty);
2189 log_debug("New container IP is %s.", strna(pretty));
2190
2191 LIST_FOREACH(ports, p, arg_expose_ports) {
2192
2193 r = fw_add_local_dnat(true,
2194 af,
2195 p->protocol,
2196 NULL,
2197 NULL, 0,
2198 NULL, 0,
2199 p->host_port,
2200 &new_exposed,
2201 p->container_port,
2202 in_addr_is_null(af, exposed) ? NULL : exposed);
2203 if (r < 0)
2204 log_warning_errno(r, "Failed to modify firewall: %m");
2205 }
2206
2207 *exposed = new_exposed;
2208 return 0;
2209 }
2210
2211 static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
2212 union in_addr_union *exposed = userdata;
2213
2214 assert(rtnl);
2215 assert(m);
2216 assert(exposed);
2217
2218 expose_ports(rtnl, exposed);
2219 return 0;
2220 }
2221
2222 static int watch_rtnl(sd_event *event, int recv_fd, union in_addr_union *exposed, sd_netlink **ret) {
2223 union {
2224 struct cmsghdr cmsghdr;
2225 uint8_t buf[CMSG_SPACE(sizeof(int))];
2226 } control = {};
2227 struct msghdr mh = {
2228 .msg_control = &control,
2229 .msg_controllen = sizeof(control),
2230 };
2231 struct cmsghdr *cmsg;
2232 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2233 int fd, r;
2234 ssize_t k;
2235
2236 assert(event);
2237 assert(recv_fd >= 0);
2238 assert(ret);
2239
2240 if (!arg_expose_ports)
2241 return 0;
2242
2243 k = recvmsg(recv_fd, &mh, MSG_NOSIGNAL);
2244 if (k < 0)
2245 return log_error_errno(errno, "Failed to recv netlink fd: %m");
2246
2247 cmsg = CMSG_FIRSTHDR(&mh);
2248 assert(cmsg->cmsg_level == SOL_SOCKET);
2249 assert(cmsg->cmsg_type == SCM_RIGHTS);
2250 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
2251 memcpy(&fd, CMSG_DATA(cmsg), sizeof(int));
2252
2253 r = sd_netlink_open_fd(&rtnl, fd);
2254 if (r < 0) {
2255 safe_close(fd);
2256 return log_error_errno(r, "Failed to create rtnl object: %m");
2257 }
2258
2259 r = sd_netlink_add_match(rtnl, RTM_NEWADDR, on_address_change, exposed);
2260 if (r < 0)
2261 return log_error_errno(r, "Failed to subscribe to RTM_NEWADDR messages: %m");
2262
2263 r = sd_netlink_add_match(rtnl, RTM_DELADDR, on_address_change, exposed);
2264 if (r < 0)
2265 return log_error_errno(r, "Failed to subscribe to RTM_DELADDR messages: %m");
2266
2267 r = sd_netlink_attach_event(rtnl, event, 0);
2268 if (r < 0)
2269 return log_error_errno(r, "Failed to add to even loop: %m");
2270
2271 *ret = rtnl;
2272 rtnl = NULL;
2273
2274 return 0;
2275 }
2276
2277 static int setup_hostname(void) {
2278
2279 if (arg_share_system)
2280 return 0;
2281
2282 if (sethostname_idempotent(arg_machine) < 0)
2283 return -errno;
2284
2285 return 0;
2286 }
2287
2288 static int setup_journal(const char *directory) {
2289 sd_id128_t machine_id, this_id;
2290 _cleanup_free_ char *b = NULL, *d = NULL;
2291 const char *etc_machine_id, *p, *q;
2292 char *id;
2293 int r;
2294
2295 /* Don't link journals in ephemeral mode */
2296 if (arg_ephemeral)
2297 return 0;
2298
2299 etc_machine_id = prefix_roota(directory, "/etc/machine-id");
2300
2301 r = read_one_line_file(etc_machine_id, &b);
2302 if (r == -ENOENT && arg_link_journal == LINK_AUTO)
2303 return 0;
2304 else if (r < 0)
2305 return log_error_errno(r, "Failed to read machine ID from %s: %m", etc_machine_id);
2306
2307 id = strstrip(b);
2308 if (isempty(id) && arg_link_journal == LINK_AUTO)
2309 return 0;
2310
2311 /* Verify validity */
2312 r = sd_id128_from_string(id, &machine_id);
2313 if (r < 0)
2314 return log_error_errno(r, "Failed to parse machine ID from %s: %m", etc_machine_id);
2315
2316 r = sd_id128_get_machine(&this_id);
2317 if (r < 0)
2318 return log_error_errno(r, "Failed to retrieve machine ID: %m");
2319
2320 if (sd_id128_equal(machine_id, this_id)) {
2321 log_full(arg_link_journal == LINK_AUTO ? LOG_WARNING : LOG_ERR,
2322 "Host and machine ids are equal (%s): refusing to link journals", id);
2323 if (arg_link_journal == LINK_AUTO)
2324 return 0;
2325 return -EEXIST;
2326 }
2327
2328 if (arg_link_journal == LINK_NO)
2329 return 0;
2330
2331 r = userns_mkdir(directory, "/var", 0755, 0, 0);
2332 if (r < 0)
2333 return log_error_errno(r, "Failed to create /var: %m");
2334
2335 r = userns_mkdir(directory, "/var/log", 0755, 0, 0);
2336 if (r < 0)
2337 return log_error_errno(r, "Failed to create /var/log: %m");
2338
2339 r = userns_mkdir(directory, "/var/log/journal", 0755, 0, 0);
2340 if (r < 0)
2341 return log_error_errno(r, "Failed to create /var/log/journal: %m");
2342
2343 p = strjoina("/var/log/journal/", id);
2344 q = prefix_roota(directory, p);
2345
2346 if (path_is_mount_point(p, 0) > 0) {
2347 if (arg_link_journal != LINK_AUTO) {
2348 log_error("%s: already a mount point, refusing to use for journal", p);
2349 return -EEXIST;
2350 }
2351
2352 return 0;
2353 }
2354
2355 if (path_is_mount_point(q, 0) > 0) {
2356 if (arg_link_journal != LINK_AUTO) {
2357 log_error("%s: already a mount point, refusing to use for journal", q);
2358 return -EEXIST;
2359 }
2360
2361 return 0;
2362 }
2363
2364 r = readlink_and_make_absolute(p, &d);
2365 if (r >= 0) {
2366 if ((arg_link_journal == LINK_GUEST ||
2367 arg_link_journal == LINK_AUTO) &&
2368 path_equal(d, q)) {
2369
2370 r = userns_mkdir(directory, p, 0755, 0, 0);
2371 if (r < 0)
2372 log_warning_errno(errno, "Failed to create directory %s: %m", q);
2373 return 0;
2374 }
2375
2376 if (unlink(p) < 0)
2377 return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
2378 } else if (r == -EINVAL) {
2379
2380 if (arg_link_journal == LINK_GUEST &&
2381 rmdir(p) < 0) {
2382
2383 if (errno == ENOTDIR) {
2384 log_error("%s already exists and is neither a symlink nor a directory", p);
2385 return r;
2386 } else {
2387 log_error_errno(errno, "Failed to remove %s: %m", p);
2388 return -errno;
2389 }
2390 }
2391 } else if (r != -ENOENT) {
2392 log_error_errno(errno, "readlink(%s) failed: %m", p);
2393 return r;
2394 }
2395
2396 if (arg_link_journal == LINK_GUEST) {
2397
2398 if (symlink(q, p) < 0) {
2399 if (arg_link_journal_try) {
2400 log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
2401 return 0;
2402 } else {
2403 log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
2404 return -errno;
2405 }
2406 }
2407
2408 r = userns_mkdir(directory, p, 0755, 0, 0);
2409 if (r < 0)
2410 log_warning_errno(errno, "Failed to create directory %s: %m", q);
2411 return 0;
2412 }
2413
2414 if (arg_link_journal == LINK_HOST) {
2415 /* don't create parents here -- if the host doesn't have
2416 * permanent journal set up, don't force it here */
2417 r = mkdir(p, 0755);
2418 if (r < 0) {
2419 if (arg_link_journal_try) {
2420 log_debug_errno(errno, "Failed to create %s, skipping journal setup: %m", p);
2421 return 0;
2422 } else {
2423 log_error_errno(errno, "Failed to create %s: %m", p);
2424 return r;
2425 }
2426 }
2427
2428 } else if (access(p, F_OK) < 0)
2429 return 0;
2430
2431 if (dir_is_empty(q) == 0)
2432 log_warning("%s is not empty, proceeding anyway.", q);
2433
2434 r = userns_mkdir(directory, p, 0755, 0, 0);
2435 if (r < 0) {
2436 log_error_errno(errno, "Failed to create %s: %m", q);
2437 return r;
2438 }
2439
2440 if (mount(p, q, NULL, MS_BIND, NULL) < 0)
2441 return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
2442
2443 return 0;
2444 }
2445
2446 static int drop_capabilities(void) {
2447 return capability_bounding_set_drop(~arg_retain, false);
2448 }
2449
2450 static int register_machine(pid_t pid, int local_ifindex) {
2451 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
2452 _cleanup_bus_flush_close_unref_ sd_bus *bus = NULL;
2453 int r;
2454
2455 if (!arg_register)
2456 return 0;
2457
2458 r = sd_bus_default_system(&bus);
2459 if (r < 0)
2460 return log_error_errno(r, "Failed to open system bus: %m");
2461
2462 if (arg_keep_unit) {
2463 r = sd_bus_call_method(
2464 bus,
2465 "org.freedesktop.machine1",
2466 "/org/freedesktop/machine1",
2467 "org.freedesktop.machine1.Manager",
2468 "RegisterMachineWithNetwork",
2469 &error,
2470 NULL,
2471 "sayssusai",
2472 arg_machine,
2473 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
2474 "nspawn",
2475 "container",
2476 (uint32_t) pid,
2477 strempty(arg_directory),
2478 local_ifindex > 0 ? 1 : 0, local_ifindex);
2479 } else {
2480 _cleanup_bus_message_unref_ sd_bus_message *m = NULL;
2481 char **i;
2482 unsigned j;
2483
2484 r = sd_bus_message_new_method_call(
2485 bus,
2486 &m,
2487 "org.freedesktop.machine1",
2488 "/org/freedesktop/machine1",
2489 "org.freedesktop.machine1.Manager",
2490 "CreateMachineWithNetwork");
2491 if (r < 0)
2492 return bus_log_create_error(r);
2493
2494 r = sd_bus_message_append(
2495 m,
2496 "sayssusai",
2497 arg_machine,
2498 SD_BUS_MESSAGE_APPEND_ID128(arg_uuid),
2499 "nspawn",
2500 "container",
2501 (uint32_t) pid,
2502 strempty(arg_directory),
2503 local_ifindex > 0 ? 1 : 0, local_ifindex);
2504 if (r < 0)
2505 return bus_log_create_error(r);
2506
2507 r = sd_bus_message_open_container(m, 'a', "(sv)");
2508 if (r < 0)
2509 return bus_log_create_error(r);
2510
2511 if (!isempty(arg_slice)) {
2512 r = sd_bus_message_append(m, "(sv)", "Slice", "s", arg_slice);
2513 if (r < 0)
2514 return bus_log_create_error(r);
2515 }
2516
2517 r = sd_bus_message_append(m, "(sv)", "DevicePolicy", "s", "strict");
2518 if (r < 0)
2519 return bus_log_create_error(r);
2520
2521 /* If you make changes here, also make sure to update
2522 * systemd-nspawn@.service, to keep the device
2523 * policies in sync regardless if we are run with or
2524 * without the --keep-unit switch. */
2525 r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 9,
2526 /* Allow the container to
2527 * access and create the API
2528 * device nodes, so that
2529 * PrivateDevices= in the
2530 * container can work
2531 * fine */
2532 "/dev/null", "rwm",
2533 "/dev/zero", "rwm",
2534 "/dev/full", "rwm",
2535 "/dev/random", "rwm",
2536 "/dev/urandom", "rwm",
2537 "/dev/tty", "rwm",
2538 "/dev/net/tun", "rwm",
2539 /* Allow the container
2540 * access to ptys. However,
2541 * do not permit the
2542 * container to ever create
2543 * these device nodes. */
2544 "/dev/pts/ptmx", "rw",
2545 "char-pts", "rw");
2546 if (r < 0)
2547 return bus_log_create_error(r);
2548
2549 for (j = 0; j < arg_n_custom_mounts; j++) {
2550 CustomMount *cm = &arg_custom_mounts[j];
2551
2552 if (cm->type != CUSTOM_MOUNT_BIND)
2553 continue;
2554
2555 r = is_device_node(cm->source);
2556 if (r < 0)
2557 return log_error_errno(r, "Failed to stat %s: %m", cm->source);
2558
2559 if (r) {
2560 r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 1,
2561 cm->source, cm->read_only ? "r" : "rw");
2562 if (r < 0)
2563 return log_error_errno(r, "Failed to append message arguments: %m");
2564 }
2565 }
2566
2567 if (arg_kill_signal != 0) {
2568 r = sd_bus_message_append(m, "(sv)", "KillSignal", "i", arg_kill_signal);
2569 if (r < 0)
2570 return bus_log_create_error(r);
2571
2572 r = sd_bus_message_append(m, "(sv)", "KillMode", "s", "mixed");
2573 if (r < 0)
2574 return bus_log_create_error(r);
2575 }
2576
2577 STRV_FOREACH(i, arg_property) {
2578 r = sd_bus_message_open_container(m, 'r', "sv");
2579 if (r < 0)
2580 return bus_log_create_error(r);
2581
2582 r = bus_append_unit_property_assignment(m, *i);
2583 if (r < 0)
2584 return r;
2585
2586 r = sd_bus_message_close_container(m);
2587 if (r < 0)
2588 return bus_log_create_error(r);
2589 }
2590
2591 r = sd_bus_message_close_container(m);
2592 if (r < 0)
2593 return bus_log_create_error(r);
2594
2595 r = sd_bus_call(bus, m, 0, &error, NULL);
2596 }
2597
2598 if (r < 0) {
2599 log_error("Failed to register machine: %s", bus_error_message(&error, r));
2600 return r;
2601 }
2602
2603 return 0;
2604 }
2605
2606 static int terminate_machine(pid_t pid) {
2607 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
2608 _cleanup_bus_message_unref_ sd_bus_message *reply = NULL;
2609 _cleanup_bus_flush_close_unref_ sd_bus *bus = NULL;
2610 const char *path;
2611 int r;
2612
2613 if (!arg_register)
2614 return 0;
2615
2616 /* If we are reusing the unit, then just exit, systemd will do
2617 * the right thing when we exit. */
2618 if (arg_keep_unit)
2619 return 0;
2620
2621 r = sd_bus_default_system(&bus);
2622 if (r < 0)
2623 return log_error_errno(r, "Failed to open system bus: %m");
2624
2625 r = sd_bus_call_method(
2626 bus,
2627 "org.freedesktop.machine1",
2628 "/org/freedesktop/machine1",
2629 "org.freedesktop.machine1.Manager",
2630 "GetMachineByPID",
2631 &error,
2632 &reply,
2633 "u",
2634 (uint32_t) pid);
2635 if (r < 0) {
2636 /* Note that the machine might already have been
2637 * cleaned up automatically, hence don't consider it a
2638 * failure if we cannot get the machine object. */
2639 log_debug("Failed to get machine: %s", bus_error_message(&error, r));
2640 return 0;
2641 }
2642
2643 r = sd_bus_message_read(reply, "o", &path);
2644 if (r < 0)
2645 return bus_log_parse_error(r);
2646
2647 r = sd_bus_call_method(
2648 bus,
2649 "org.freedesktop.machine1",
2650 path,
2651 "org.freedesktop.machine1.Machine",
2652 "Terminate",
2653 &error,
2654 NULL,
2655 NULL);
2656 if (r < 0) {
2657 log_debug("Failed to terminate machine: %s", bus_error_message(&error, r));
2658 return 0;
2659 }
2660
2661 return 0;
2662 }
2663
2664 static int reset_audit_loginuid(void) {
2665 _cleanup_free_ char *p = NULL;
2666 int r;
2667
2668 if (arg_share_system)
2669 return 0;
2670
2671 r = read_one_line_file("/proc/self/loginuid", &p);
2672 if (r == -ENOENT)
2673 return 0;
2674 if (r < 0)
2675 return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
2676
2677 /* Already reset? */
2678 if (streq(p, "4294967295"))
2679 return 0;
2680
2681 r = write_string_file("/proc/self/loginuid", "4294967295", 0);
2682 if (r < 0) {
2683 log_error_errno(r,
2684 "Failed to reset audit login UID. This probably means that your kernel is too\n"
2685 "old and you have audit enabled. Note that the auditing subsystem is known to\n"
2686 "be incompatible with containers on old kernels. Please make sure to upgrade\n"
2687 "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
2688 "using systemd-nspawn. Sleeping for 5s... (%m)");
2689
2690 sleep(5);
2691 }
2692
2693 return 0;
2694 }
2695
2696 #define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
2697 #define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
2698 #define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
2699
2700 static int generate_mac(struct ether_addr *mac, sd_id128_t hash_key, uint64_t idx) {
2701 uint8_t result[8];
2702 size_t l, sz;
2703 uint8_t *v, *i;
2704 int r;
2705
2706 l = strlen(arg_machine);
2707 sz = sizeof(sd_id128_t) + l;
2708 if (idx > 0)
2709 sz += sizeof(idx);
2710
2711 v = alloca(sz);
2712
2713 /* fetch some persistent data unique to the host */
2714 r = sd_id128_get_machine((sd_id128_t*) v);
2715 if (r < 0)
2716 return r;
2717
2718 /* combine with some data unique (on this host) to this
2719 * container instance */
2720 i = mempcpy(v + sizeof(sd_id128_t), arg_machine, l);
2721 if (idx > 0) {
2722 idx = htole64(idx);
2723 memcpy(i, &idx, sizeof(idx));
2724 }
2725
2726 /* Let's hash the host machine ID plus the container name. We
2727 * use a fixed, but originally randomly created hash key here. */
2728 siphash24(result, v, sz, hash_key.bytes);
2729
2730 assert_cc(ETH_ALEN <= sizeof(result));
2731 memcpy(mac->ether_addr_octet, result, ETH_ALEN);
2732
2733 /* see eth_random_addr in the kernel */
2734 mac->ether_addr_octet[0] &= 0xfe; /* clear multicast bit */
2735 mac->ether_addr_octet[0] |= 0x02; /* set local assignment bit (IEEE802) */
2736
2737 return 0;
2738 }
2739
2740 static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ], int *ifi) {
2741 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2742 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2743 struct ether_addr mac_host, mac_container;
2744 int r, i;
2745
2746 if (!arg_private_network)
2747 return 0;
2748
2749 if (!arg_network_veth)
2750 return 0;
2751
2752 /* Use two different interface name prefixes depending whether
2753 * we are in bridge mode or not. */
2754 snprintf(iface_name, IFNAMSIZ - 1, "%s-%s",
2755 arg_network_bridge ? "vb" : "ve", arg_machine);
2756
2757 r = generate_mac(&mac_container, CONTAINER_HASH_KEY, 0);
2758 if (r < 0)
2759 return log_error_errno(r, "Failed to generate predictable MAC address for container side: %m");
2760
2761 r = generate_mac(&mac_host, HOST_HASH_KEY, 0);
2762 if (r < 0)
2763 return log_error_errno(r, "Failed to generate predictable MAC address for host side: %m");
2764
2765 r = sd_netlink_open(&rtnl);
2766 if (r < 0)
2767 return log_error_errno(r, "Failed to connect to netlink: %m");
2768
2769 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2770 if (r < 0)
2771 return log_error_errno(r, "Failed to allocate netlink message: %m");
2772
2773 r = sd_netlink_message_append_string(m, IFLA_IFNAME, iface_name);
2774 if (r < 0)
2775 return log_error_errno(r, "Failed to add netlink interface name: %m");
2776
2777 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac_host);
2778 if (r < 0)
2779 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2780
2781 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
2782 if (r < 0)
2783 return log_error_errno(r, "Failed to open netlink container: %m");
2784
2785 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "veth");
2786 if (r < 0)
2787 return log_error_errno(r, "Failed to open netlink container: %m");
2788
2789 r = sd_netlink_message_open_container(m, VETH_INFO_PEER);
2790 if (r < 0)
2791 return log_error_errno(r, "Failed to open netlink container: %m");
2792
2793 r = sd_netlink_message_append_string(m, IFLA_IFNAME, "host0");
2794 if (r < 0)
2795 return log_error_errno(r, "Failed to add netlink interface name: %m");
2796
2797 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac_container);
2798 if (r < 0)
2799 return log_error_errno(r, "Failed to add netlink MAC address: %m");
2800
2801 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
2802 if (r < 0)
2803 return log_error_errno(r, "Failed to add netlink namespace field: %m");
2804
2805 r = sd_netlink_message_close_container(m);
2806 if (r < 0)
2807 return log_error_errno(r, "Failed to close netlink container: %m");
2808
2809 r = sd_netlink_message_close_container(m);
2810 if (r < 0)
2811 return log_error_errno(r, "Failed to close netlink container: %m");
2812
2813 r = sd_netlink_message_close_container(m);
2814 if (r < 0)
2815 return log_error_errno(r, "Failed to close netlink container: %m");
2816
2817 r = sd_netlink_call(rtnl, m, 0, NULL);
2818 if (r < 0)
2819 return log_error_errno(r, "Failed to add new veth interfaces (host0, %s): %m", iface_name);
2820
2821 i = (int) if_nametoindex(iface_name);
2822 if (i <= 0)
2823 return log_error_errno(errno, "Failed to resolve interface %s: %m", iface_name);
2824
2825 *ifi = i;
2826
2827 return 0;
2828 }
2829
2830 static int setup_bridge(const char veth_name[], int *ifi) {
2831 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2832 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2833 int r, bridge;
2834
2835 if (!arg_private_network)
2836 return 0;
2837
2838 if (!arg_network_veth)
2839 return 0;
2840
2841 if (!arg_network_bridge)
2842 return 0;
2843
2844 bridge = (int) if_nametoindex(arg_network_bridge);
2845 if (bridge <= 0)
2846 return log_error_errno(errno, "Failed to resolve interface %s: %m", arg_network_bridge);
2847
2848 *ifi = bridge;
2849
2850 r = sd_netlink_open(&rtnl);
2851 if (r < 0)
2852 return log_error_errno(r, "Failed to connect to netlink: %m");
2853
2854 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, 0);
2855 if (r < 0)
2856 return log_error_errno(r, "Failed to allocate netlink message: %m");
2857
2858 r = sd_rtnl_message_link_set_flags(m, IFF_UP, IFF_UP);
2859 if (r < 0)
2860 return log_error_errno(r, "Failed to set IFF_UP flag: %m");
2861
2862 r = sd_netlink_message_append_string(m, IFLA_IFNAME, veth_name);
2863 if (r < 0)
2864 return log_error_errno(r, "Failed to add netlink interface name field: %m");
2865
2866 r = sd_netlink_message_append_u32(m, IFLA_MASTER, bridge);
2867 if (r < 0)
2868 return log_error_errno(r, "Failed to add netlink master field: %m");
2869
2870 r = sd_netlink_call(rtnl, m, 0, NULL);
2871 if (r < 0)
2872 return log_error_errno(r, "Failed to add veth interface to bridge: %m");
2873
2874 return 0;
2875 }
2876
2877 static int parse_interface(struct udev *udev, const char *name) {
2878 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
2879 char ifi_str[2 + DECIMAL_STR_MAX(int)];
2880 int ifi;
2881
2882 ifi = (int) if_nametoindex(name);
2883 if (ifi <= 0)
2884 return log_error_errno(errno, "Failed to resolve interface %s: %m", name);
2885
2886 sprintf(ifi_str, "n%i", ifi);
2887 d = udev_device_new_from_device_id(udev, ifi_str);
2888 if (!d)
2889 return log_error_errno(errno, "Failed to get udev device for interface %s: %m", name);
2890
2891 if (udev_device_get_is_initialized(d) <= 0) {
2892 log_error("Network interface %s is not initialized yet.", name);
2893 return -EBUSY;
2894 }
2895
2896 return ifi;
2897 }
2898
2899 static int move_network_interfaces(pid_t pid) {
2900 _cleanup_udev_unref_ struct udev *udev = NULL;
2901 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2902 char **i;
2903 int r;
2904
2905 if (!arg_private_network)
2906 return 0;
2907
2908 if (strv_isempty(arg_network_interfaces))
2909 return 0;
2910
2911 r = sd_netlink_open(&rtnl);
2912 if (r < 0)
2913 return log_error_errno(r, "Failed to connect to netlink: %m");
2914
2915 udev = udev_new();
2916 if (!udev) {
2917 log_error("Failed to connect to udev.");
2918 return -ENOMEM;
2919 }
2920
2921 STRV_FOREACH(i, arg_network_interfaces) {
2922 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2923 int ifi;
2924
2925 ifi = parse_interface(udev, *i);
2926 if (ifi < 0)
2927 return ifi;
2928
2929 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, ifi);
2930 if (r < 0)
2931 return log_error_errno(r, "Failed to allocate netlink message: %m");
2932
2933 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
2934 if (r < 0)
2935 return log_error_errno(r, "Failed to append namespace PID to netlink message: %m");
2936
2937 r = sd_netlink_call(rtnl, m, 0, NULL);
2938 if (r < 0)
2939 return log_error_errno(r, "Failed to move interface %s to namespace: %m", *i);
2940 }
2941
2942 return 0;
2943 }
2944
2945 static int setup_macvlan(pid_t pid) {
2946 _cleanup_udev_unref_ struct udev *udev = NULL;
2947 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
2948 unsigned idx = 0;
2949 char **i;
2950 int r;
2951
2952 if (!arg_private_network)
2953 return 0;
2954
2955 if (strv_isempty(arg_network_macvlan))
2956 return 0;
2957
2958 r = sd_netlink_open(&rtnl);
2959 if (r < 0)
2960 return log_error_errno(r, "Failed to connect to netlink: %m");
2961
2962 udev = udev_new();
2963 if (!udev) {
2964 log_error("Failed to connect to udev.");
2965 return -ENOMEM;
2966 }
2967
2968 STRV_FOREACH(i, arg_network_macvlan) {
2969 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
2970 _cleanup_free_ char *n = NULL;
2971 struct ether_addr mac;
2972 int ifi;
2973
2974 ifi = parse_interface(udev, *i);
2975 if (ifi < 0)
2976 return ifi;
2977
2978 r = generate_mac(&mac, MACVLAN_HASH_KEY, idx++);
2979 if (r < 0)
2980 return log_error_errno(r, "Failed to create MACVLAN MAC address: %m");
2981
2982 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
2983 if (r < 0)
2984 return log_error_errno(r, "Failed to allocate netlink message: %m");
2985
2986 r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
2987 if (r < 0)
2988 return log_error_errno(r, "Failed to add netlink interface index: %m");
2989
2990 n = strappend("mv-", *i);
2991 if (!n)
2992 return log_oom();
2993
2994 strshorten(n, IFNAMSIZ-1);
2995
2996 r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
2997 if (r < 0)
2998 return log_error_errno(r, "Failed to add netlink interface name: %m");
2999
3000 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac);
3001 if (r < 0)
3002 return log_error_errno(r, "Failed to add netlink MAC address: %m");
3003
3004 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
3005 if (r < 0)
3006 return log_error_errno(r, "Failed to add netlink namespace field: %m");
3007
3008 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
3009 if (r < 0)
3010 return log_error_errno(r, "Failed to open netlink container: %m");
3011
3012 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "macvlan");
3013 if (r < 0)
3014 return log_error_errno(r, "Failed to open netlink container: %m");
3015
3016 r = sd_netlink_message_append_u32(m, IFLA_MACVLAN_MODE, MACVLAN_MODE_BRIDGE);
3017 if (r < 0)
3018 return log_error_errno(r, "Failed to append macvlan mode: %m");
3019
3020 r = sd_netlink_message_close_container(m);
3021 if (r < 0)
3022 return log_error_errno(r, "Failed to close netlink container: %m");
3023
3024 r = sd_netlink_message_close_container(m);
3025 if (r < 0)
3026 return log_error_errno(r, "Failed to close netlink container: %m");
3027
3028 r = sd_netlink_call(rtnl, m, 0, NULL);
3029 if (r < 0)
3030 return log_error_errno(r, "Failed to add new macvlan interfaces: %m");
3031 }
3032
3033 return 0;
3034 }
3035
3036 static int setup_ipvlan(pid_t pid) {
3037 _cleanup_udev_unref_ struct udev *udev = NULL;
3038 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
3039 char **i;
3040 int r;
3041
3042 if (!arg_private_network)
3043 return 0;
3044
3045 if (strv_isempty(arg_network_ipvlan))
3046 return 0;
3047
3048 r = sd_netlink_open(&rtnl);
3049 if (r < 0)
3050 return log_error_errno(r, "Failed to connect to netlink: %m");
3051
3052 udev = udev_new();
3053 if (!udev) {
3054 log_error("Failed to connect to udev.");
3055 return -ENOMEM;
3056 }
3057
3058 STRV_FOREACH(i, arg_network_ipvlan) {
3059 _cleanup_netlink_message_unref_ sd_netlink_message *m = NULL;
3060 _cleanup_free_ char *n = NULL;
3061 int ifi;
3062
3063 ifi = parse_interface(udev, *i);
3064 if (ifi < 0)
3065 return ifi;
3066
3067 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
3068 if (r < 0)
3069 return log_error_errno(r, "Failed to allocate netlink message: %m");
3070
3071 r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
3072 if (r < 0)
3073 return log_error_errno(r, "Failed to add netlink interface index: %m");
3074
3075 n = strappend("iv-", *i);
3076 if (!n)
3077 return log_oom();
3078
3079 strshorten(n, IFNAMSIZ-1);
3080
3081 r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
3082 if (r < 0)
3083 return log_error_errno(r, "Failed to add netlink interface name: %m");
3084
3085 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
3086 if (r < 0)
3087 return log_error_errno(r, "Failed to add netlink namespace field: %m");
3088
3089 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
3090 if (r < 0)
3091 return log_error_errno(r, "Failed to open netlink container: %m");
3092
3093 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "ipvlan");
3094 if (r < 0)
3095 return log_error_errno(r, "Failed to open netlink container: %m");
3096
3097 r = sd_netlink_message_append_u16(m, IFLA_IPVLAN_MODE, IPVLAN_MODE_L2);
3098 if (r < 0)
3099 return log_error_errno(r, "Failed to add ipvlan mode: %m");
3100
3101 r = sd_netlink_message_close_container(m);
3102 if (r < 0)
3103 return log_error_errno(r, "Failed to close netlink container: %m");
3104
3105 r = sd_netlink_message_close_container(m);
3106 if (r < 0)
3107 return log_error_errno(r, "Failed to close netlink container: %m");
3108
3109 r = sd_netlink_call(rtnl, m, 0, NULL);
3110 if (r < 0)
3111 return log_error_errno(r, "Failed to add new ipvlan interfaces: %m");
3112 }
3113
3114 return 0;
3115 }
3116
3117 static int setup_seccomp(void) {
3118
3119 #ifdef HAVE_SECCOMP
3120 static const struct {
3121 uint64_t capability;
3122 int syscall_num;
3123 } blacklist[] = {
3124 { CAP_SYS_RAWIO, SCMP_SYS(iopl) },
3125 { CAP_SYS_RAWIO, SCMP_SYS(ioperm) },
3126 { CAP_SYS_BOOT, SCMP_SYS(kexec_load) },
3127 { CAP_SYS_ADMIN, SCMP_SYS(swapon) },
3128 { CAP_SYS_ADMIN, SCMP_SYS(swapoff) },
3129 { CAP_SYS_ADMIN, SCMP_SYS(open_by_handle_at) },
3130 { CAP_SYS_MODULE, SCMP_SYS(init_module) },
3131 { CAP_SYS_MODULE, SCMP_SYS(finit_module) },
3132 { CAP_SYS_MODULE, SCMP_SYS(delete_module) },
3133 { CAP_SYSLOG, SCMP_SYS(syslog) },
3134 };
3135
3136 scmp_filter_ctx seccomp;
3137 unsigned i;
3138 int r;
3139
3140 seccomp = seccomp_init(SCMP_ACT_ALLOW);
3141 if (!seccomp)
3142 return log_oom();
3143
3144 r = seccomp_add_secondary_archs(seccomp);
3145 if (r < 0) {
3146 log_error_errno(r, "Failed to add secondary archs to seccomp filter: %m");
3147 goto finish;
3148 }
3149
3150 for (i = 0; i < ELEMENTSOF(blacklist); i++) {
3151 if (arg_retain & (1ULL << blacklist[i].capability))
3152 continue;
3153
3154 r = seccomp_rule_add(seccomp, SCMP_ACT_ERRNO(EPERM), blacklist[i].syscall_num, 0);
3155 if (r == -EFAULT)
3156 continue; /* unknown syscall */
3157 if (r < 0) {
3158 log_error_errno(r, "Failed to block syscall: %m");
3159 goto finish;
3160 }
3161 }
3162
3163
3164 /*
3165 Audit is broken in containers, much of the userspace audit
3166 hookup will fail if running inside a container. We don't
3167 care and just turn off creation of audit sockets.
3168
3169 This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail
3170 with EAFNOSUPPORT which audit userspace uses as indication
3171 that audit is disabled in the kernel.
3172 */
3173
3174 r = seccomp_rule_add(
3175 seccomp,
3176 SCMP_ACT_ERRNO(EAFNOSUPPORT),
3177 SCMP_SYS(socket),
3178 2,
3179 SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
3180 SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
3181 if (r < 0) {
3182 log_error_errno(r, "Failed to add audit seccomp rule: %m");
3183 goto finish;
3184 }
3185
3186 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
3187 if (r < 0) {
3188 log_error_errno(r, "Failed to unset NO_NEW_PRIVS: %m");
3189 goto finish;
3190 }
3191
3192 r = seccomp_load(seccomp);
3193 if (r == -EINVAL) {
3194 log_debug_errno(r, "Kernel is probably not configured with CONFIG_SECCOMP. Disabling seccomp audit filter: %m");
3195 r = 0;
3196 goto finish;
3197 }
3198 if (r < 0) {
3199 log_error_errno(r, "Failed to install seccomp audit filter: %m");
3200 goto finish;
3201 }
3202
3203 finish:
3204 seccomp_release(seccomp);
3205 return r;
3206 #else
3207 return 0;
3208 #endif
3209
3210 }
3211
3212 static int setup_propagate(const char *root) {
3213 const char *p, *q;
3214
3215 (void) mkdir_p("/run/systemd/nspawn/", 0755);
3216 (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
3217 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
3218 (void) mkdir_p(p, 0600);
3219
3220 if (userns_mkdir(root, "/run/systemd", 0755, 0, 0) < 0)
3221 return log_error_errno(errno, "Failed to create /run/systemd: %m");
3222
3223 if (userns_mkdir(root, "/run/systemd/nspawn", 0755, 0, 0) < 0)
3224 return log_error_errno(errno, "Failed to create /run/systemd/nspawn: %m");
3225
3226 if (userns_mkdir(root, "/run/systemd/nspawn/incoming", 0600, 0, 0) < 0)
3227 return log_error_errno(errno, "Failed to create /run/systemd/nspawn/incoming: %m");
3228
3229 q = prefix_roota(root, "/run/systemd/nspawn/incoming");
3230 if (mount(p, q, NULL, MS_BIND, NULL) < 0)
3231 return log_error_errno(errno, "Failed to install propagation bind mount.");
3232
3233 if (mount(NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0)
3234 return log_error_errno(errno, "Failed to make propagation mount read-only");
3235
3236 return 0;
3237 }
3238
3239 static int setup_image(char **device_path, int *loop_nr) {
3240 struct loop_info64 info = {
3241 .lo_flags = LO_FLAGS_AUTOCLEAR|LO_FLAGS_PARTSCAN
3242 };
3243 _cleanup_close_ int fd = -1, control = -1, loop = -1;
3244 _cleanup_free_ char* loopdev = NULL;
3245 struct stat st;
3246 int r, nr;
3247
3248 assert(device_path);
3249 assert(loop_nr);
3250 assert(arg_image);
3251
3252 fd = open(arg_image, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
3253 if (fd < 0)
3254 return log_error_errno(errno, "Failed to open %s: %m", arg_image);
3255
3256 if (fstat(fd, &st) < 0)
3257 return log_error_errno(errno, "Failed to stat %s: %m", arg_image);
3258
3259 if (S_ISBLK(st.st_mode)) {
3260 char *p;
3261
3262 p = strdup(arg_image);
3263 if (!p)
3264 return log_oom();
3265
3266 *device_path = p;
3267
3268 *loop_nr = -1;
3269
3270 r = fd;
3271 fd = -1;
3272
3273 return r;
3274 }
3275
3276 if (!S_ISREG(st.st_mode)) {
3277 log_error_errno(errno, "%s is not a regular file or block device: %m", arg_image);
3278 return -EINVAL;
3279 }
3280
3281 control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
3282 if (control < 0)
3283 return log_error_errno(errno, "Failed to open /dev/loop-control: %m");
3284
3285 nr = ioctl(control, LOOP_CTL_GET_FREE);
3286 if (nr < 0)
3287 return log_error_errno(errno, "Failed to allocate loop device: %m");
3288
3289 if (asprintf(&loopdev, "/dev/loop%i", nr) < 0)
3290 return log_oom();
3291
3292 loop = open(loopdev, O_CLOEXEC|(arg_read_only ? O_RDONLY : O_RDWR)|O_NONBLOCK|O_NOCTTY);
3293 if (loop < 0)
3294 return log_error_errno(errno, "Failed to open loop device %s: %m", loopdev);
3295
3296 if (ioctl(loop, LOOP_SET_FD, fd) < 0)
3297 return log_error_errno(errno, "Failed to set loopback file descriptor on %s: %m", loopdev);
3298
3299 if (arg_read_only)
3300 info.lo_flags |= LO_FLAGS_READ_ONLY;
3301
3302 if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0)
3303 return log_error_errno(errno, "Failed to set loopback settings on %s: %m", loopdev);
3304
3305 *device_path = loopdev;
3306 loopdev = NULL;
3307
3308 *loop_nr = nr;
3309
3310 r = loop;
3311 loop = -1;
3312
3313 return r;
3314 }
3315
3316 #define PARTITION_TABLE_BLURB \
3317 "Note that the disk image needs to either contain only a single MBR partition of\n" \
3318 "type 0x83 that is marked bootable, or a single GPT partition of type " \
3319 "0FC63DAF-8483-4772-8E79-3D69D8477DE4 or follow\n" \
3320 " http://www.freedesktop.org/wiki/Specifications/DiscoverablePartitionsSpec/\n" \
3321 "to be bootable with systemd-nspawn."
3322
3323 static int dissect_image(
3324 int fd,
3325 char **root_device, bool *root_device_rw,
3326 char **home_device, bool *home_device_rw,
3327 char **srv_device, bool *srv_device_rw,
3328 bool *secondary) {
3329
3330 #ifdef HAVE_BLKID
3331 int home_nr = -1, srv_nr = -1;
3332 #ifdef GPT_ROOT_NATIVE
3333 int root_nr = -1;
3334 #endif
3335 #ifdef GPT_ROOT_SECONDARY
3336 int secondary_root_nr = -1;
3337 #endif
3338 _cleanup_free_ char *home = NULL, *root = NULL, *secondary_root = NULL, *srv = NULL, *generic = NULL;
3339 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
3340 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
3341 _cleanup_blkid_free_probe_ blkid_probe b = NULL;
3342 _cleanup_udev_unref_ struct udev *udev = NULL;
3343 struct udev_list_entry *first, *item;
3344 bool home_rw = true, root_rw = true, secondary_root_rw = true, srv_rw = true, generic_rw = true;
3345 bool is_gpt, is_mbr, multiple_generic = false;
3346 const char *pttype = NULL;
3347 blkid_partlist pl;
3348 struct stat st;
3349 unsigned i;
3350 int r;
3351
3352 assert(fd >= 0);
3353 assert(root_device);
3354 assert(home_device);
3355 assert(srv_device);
3356 assert(secondary);
3357 assert(arg_image);
3358
3359 b = blkid_new_probe();
3360 if (!b)
3361 return log_oom();
3362
3363 errno = 0;
3364 r = blkid_probe_set_device(b, fd, 0, 0);
3365 if (r != 0) {
3366 if (errno == 0)
3367 return log_oom();
3368
3369 log_error_errno(errno, "Failed to set device on blkid probe: %m");
3370 return -errno;
3371 }
3372
3373 blkid_probe_enable_partitions(b, 1);
3374 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
3375
3376 errno = 0;
3377 r = blkid_do_safeprobe(b);
3378 if (r == -2 || r == 1) {
3379 log_error("Failed to identify any partition table on\n"
3380 " %s\n"
3381 PARTITION_TABLE_BLURB, arg_image);
3382 return -EINVAL;
3383 } else if (r != 0) {
3384 if (errno == 0)
3385 errno = EIO;
3386 log_error_errno(errno, "Failed to probe: %m");
3387 return -errno;
3388 }
3389
3390 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
3391
3392 is_gpt = streq_ptr(pttype, "gpt");
3393 is_mbr = streq_ptr(pttype, "dos");
3394
3395 if (!is_gpt && !is_mbr) {
3396 log_error("No GPT or MBR partition table discovered on\n"
3397 " %s\n"
3398 PARTITION_TABLE_BLURB, arg_image);
3399 return -EINVAL;
3400 }
3401
3402 errno = 0;
3403 pl = blkid_probe_get_partitions(b);
3404 if (!pl) {
3405 if (errno == 0)
3406 return log_oom();
3407
3408 log_error("Failed to list partitions of %s", arg_image);
3409 return -errno;
3410 }
3411
3412 udev = udev_new();
3413 if (!udev)
3414 return log_oom();
3415
3416 if (fstat(fd, &st) < 0)
3417 return log_error_errno(errno, "Failed to stat block device: %m");
3418
3419 d = udev_device_new_from_devnum(udev, 'b', st.st_rdev);
3420 if (!d)
3421 return log_oom();
3422
3423 for (i = 0;; i++) {
3424 int n, m;
3425
3426 if (i >= 10) {
3427 log_error("Kernel partitions never appeared.");
3428 return -ENXIO;
3429 }
3430
3431 e = udev_enumerate_new(udev);
3432 if (!e)
3433 return log_oom();
3434
3435 r = udev_enumerate_add_match_parent(e, d);
3436 if (r < 0)
3437 return log_oom();
3438
3439 r = udev_enumerate_scan_devices(e);
3440 if (r < 0)
3441 return log_error_errno(r, "Failed to scan for partition devices of %s: %m", arg_image);
3442
3443 /* Count the partitions enumerated by the kernel */
3444 n = 0;
3445 first = udev_enumerate_get_list_entry(e);
3446 udev_list_entry_foreach(item, first)
3447 n++;
3448
3449 /* Count the partitions enumerated by blkid */
3450 m = blkid_partlist_numof_partitions(pl);
3451 if (n == m + 1)
3452 break;
3453 if (n > m + 1) {
3454 log_error("blkid and kernel partition list do not match.");
3455 return -EIO;
3456 }
3457 if (n < m + 1) {
3458 unsigned j;
3459
3460 /* The kernel has probed fewer partitions than
3461 * blkid? Maybe the kernel prober is still
3462 * running or it got EBUSY because udev
3463 * already opened the device. Let's reprobe
3464 * the device, which is a synchronous call
3465 * that waits until probing is complete. */
3466
3467 for (j = 0; j < 20; j++) {
3468
3469 r = ioctl(fd, BLKRRPART, 0);
3470 if (r < 0)
3471 r = -errno;
3472 if (r >= 0 || r != -EBUSY)
3473 break;
3474
3475 /* If something else has the device
3476 * open, such as an udev rule, the
3477 * ioctl will return EBUSY. Since
3478 * there's no way to wait until it
3479 * isn't busy anymore, let's just wait
3480 * a bit, and try again.
3481 *
3482 * This is really something they
3483 * should fix in the kernel! */
3484
3485 usleep(50 * USEC_PER_MSEC);
3486 }
3487
3488 if (r < 0)
3489 return log_error_errno(r, "Failed to reread partition table: %m");
3490 }
3491
3492 e = udev_enumerate_unref(e);
3493 }
3494
3495 first = udev_enumerate_get_list_entry(e);
3496 udev_list_entry_foreach(item, first) {
3497 _cleanup_udev_device_unref_ struct udev_device *q;
3498 const char *node;
3499 unsigned long long flags;
3500 blkid_partition pp;
3501 dev_t qn;
3502 int nr;
3503
3504 errno = 0;
3505 q = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
3506 if (!q) {
3507 if (!errno)
3508 errno = ENOMEM;
3509
3510 log_error_errno(errno, "Failed to get partition device of %s: %m", arg_image);
3511 return -errno;
3512 }
3513
3514 qn = udev_device_get_devnum(q);
3515 if (major(qn) == 0)
3516 continue;
3517
3518 if (st.st_rdev == qn)
3519 continue;
3520
3521 node = udev_device_get_devnode(q);
3522 if (!node)
3523 continue;
3524
3525 pp = blkid_partlist_devno_to_partition(pl, qn);
3526 if (!pp)
3527 continue;
3528
3529 flags = blkid_partition_get_flags(pp);
3530
3531 nr = blkid_partition_get_partno(pp);
3532 if (nr < 0)
3533 continue;
3534
3535 if (is_gpt) {
3536 sd_id128_t type_id;
3537 const char *stype;
3538
3539 if (flags & GPT_FLAG_NO_AUTO)
3540 continue;
3541
3542 stype = blkid_partition_get_type_string(pp);
3543 if (!stype)
3544 continue;
3545
3546 if (sd_id128_from_string(stype, &type_id) < 0)
3547 continue;
3548
3549 if (sd_id128_equal(type_id, GPT_HOME)) {
3550
3551 if (home && nr >= home_nr)
3552 continue;
3553
3554 home_nr = nr;
3555 home_rw = !(flags & GPT_FLAG_READ_ONLY);
3556
3557 r = free_and_strdup(&home, node);
3558 if (r < 0)
3559 return log_oom();
3560
3561 } else if (sd_id128_equal(type_id, GPT_SRV)) {
3562
3563 if (srv && nr >= srv_nr)
3564 continue;
3565
3566 srv_nr = nr;
3567 srv_rw = !(flags & GPT_FLAG_READ_ONLY);
3568
3569 r = free_and_strdup(&srv, node);
3570 if (r < 0)
3571 return log_oom();
3572 }
3573 #ifdef GPT_ROOT_NATIVE
3574 else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
3575
3576 if (root && nr >= root_nr)
3577 continue;
3578
3579 root_nr = nr;
3580 root_rw = !(flags & GPT_FLAG_READ_ONLY);
3581
3582 r = free_and_strdup(&root, node);
3583 if (r < 0)
3584 return log_oom();
3585 }
3586 #endif
3587 #ifdef GPT_ROOT_SECONDARY
3588 else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
3589
3590 if (secondary_root && nr >= secondary_root_nr)
3591 continue;
3592
3593 secondary_root_nr = nr;
3594 secondary_root_rw = !(flags & GPT_FLAG_READ_ONLY);
3595
3596 r = free_and_strdup(&secondary_root, node);
3597 if (r < 0)
3598 return log_oom();
3599 }
3600 #endif
3601 else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
3602
3603 if (generic)
3604 multiple_generic = true;
3605 else {
3606 generic_rw = !(flags & GPT_FLAG_READ_ONLY);
3607
3608 r = free_and_strdup(&generic, node);
3609 if (r < 0)
3610 return log_oom();
3611 }
3612 }
3613
3614 } else if (is_mbr) {
3615 int type;
3616
3617 if (flags != 0x80) /* Bootable flag */
3618 continue;
3619
3620 type = blkid_partition_get_type(pp);
3621 if (type != 0x83) /* Linux partition */
3622 continue;
3623
3624 if (generic)
3625 multiple_generic = true;
3626 else {
3627 generic_rw = true;
3628
3629 r = free_and_strdup(&root, node);
3630 if (r < 0)
3631 return log_oom();
3632 }
3633 }
3634 }
3635
3636 if (root) {
3637 *root_device = root;
3638 root = NULL;
3639
3640 *root_device_rw = root_rw;
3641 *secondary = false;
3642 } else if (secondary_root) {
3643 *root_device = secondary_root;
3644 secondary_root = NULL;
3645
3646 *root_device_rw = secondary_root_rw;
3647 *secondary = true;
3648 } else if (generic) {
3649
3650 /* There were no partitions with precise meanings
3651 * around, but we found generic partitions. In this
3652 * case, if there's only one, we can go ahead and boot
3653 * it, otherwise we bail out, because we really cannot
3654 * make any sense of it. */
3655
3656 if (multiple_generic) {
3657 log_error("Identified multiple bootable Linux partitions on\n"
3658 " %s\n"
3659 PARTITION_TABLE_BLURB, arg_image);
3660 return -EINVAL;
3661 }
3662
3663 *root_device = generic;
3664 generic = NULL;
3665
3666 *root_device_rw = generic_rw;
3667 *secondary = false;
3668 } else {
3669 log_error("Failed to identify root partition in disk image\n"
3670 " %s\n"
3671 PARTITION_TABLE_BLURB, arg_image);
3672 return -EINVAL;
3673 }
3674
3675 if (home) {
3676 *home_device = home;
3677 home = NULL;
3678
3679 *home_device_rw = home_rw;
3680 }
3681
3682 if (srv) {
3683 *srv_device = srv;
3684 srv = NULL;
3685
3686 *srv_device_rw = srv_rw;
3687 }
3688
3689 return 0;
3690 #else
3691 log_error("--image= is not supported, compiled without blkid support.");
3692 return -EOPNOTSUPP;
3693 #endif
3694 }
3695
3696 static int mount_device(const char *what, const char *where, const char *directory, bool rw) {
3697 #ifdef HAVE_BLKID
3698 _cleanup_blkid_free_probe_ blkid_probe b = NULL;
3699 const char *fstype, *p;
3700 int r;
3701
3702 assert(what);
3703 assert(where);
3704
3705 if (arg_read_only)
3706 rw = false;
3707
3708 if (directory)
3709 p = strjoina(where, directory);
3710 else
3711 p = where;
3712
3713 errno = 0;
3714 b = blkid_new_probe_from_filename(what);
3715 if (!b) {
3716 if (errno == 0)
3717 return log_oom();
3718 log_error_errno(errno, "Failed to allocate prober for %s: %m", what);
3719 return -errno;
3720 }
3721
3722 blkid_probe_enable_superblocks(b, 1);
3723 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
3724
3725 errno = 0;
3726 r = blkid_do_safeprobe(b);
3727 if (r == -1 || r == 1) {
3728 log_error("Cannot determine file system type of %s", what);
3729 return -EINVAL;
3730 } else if (r != 0) {
3731 if (errno == 0)
3732 errno = EIO;
3733 log_error_errno(errno, "Failed to probe %s: %m", what);
3734 return -errno;
3735 }
3736
3737 errno = 0;
3738 if (blkid_probe_lookup_value(b, "TYPE", &fstype, NULL) < 0) {
3739 if (errno == 0)
3740 errno = EINVAL;
3741 log_error("Failed to determine file system type of %s", what);
3742 return -errno;
3743 }
3744
3745 if (streq(fstype, "crypto_LUKS")) {
3746 log_error("nspawn currently does not support LUKS disk images.");
3747 return -EOPNOTSUPP;
3748 }
3749
3750 if (mount(what, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), NULL) < 0)
3751 return log_error_errno(errno, "Failed to mount %s: %m", what);
3752
3753 return 0;
3754 #else
3755 log_error("--image= is not supported, compiled without blkid support.");
3756 return -EOPNOTSUPP;
3757 #endif
3758 }
3759
3760 static int mount_devices(
3761 const char *where,
3762 const char *root_device, bool root_device_rw,
3763 const char *home_device, bool home_device_rw,
3764 const char *srv_device, bool srv_device_rw) {
3765 int r;
3766
3767 assert(where);
3768
3769 if (root_device) {
3770 r = mount_device(root_device, arg_directory, NULL, root_device_rw);
3771 if (r < 0)
3772 return log_error_errno(r, "Failed to mount root directory: %m");
3773 }
3774
3775 if (home_device) {
3776 r = mount_device(home_device, arg_directory, "/home", home_device_rw);
3777 if (r < 0)
3778 return log_error_errno(r, "Failed to mount home directory: %m");
3779 }
3780
3781 if (srv_device) {
3782 r = mount_device(srv_device, arg_directory, "/srv", srv_device_rw);
3783 if (r < 0)
3784 return log_error_errno(r, "Failed to mount server data directory: %m");
3785 }
3786
3787 return 0;
3788 }
3789
3790 static void loop_remove(int nr, int *image_fd) {
3791 _cleanup_close_ int control = -1;
3792 int r;
3793
3794 if (nr < 0)
3795 return;
3796
3797 if (image_fd && *image_fd >= 0) {
3798 r = ioctl(*image_fd, LOOP_CLR_FD);
3799 if (r < 0)
3800 log_debug_errno(errno, "Failed to close loop image: %m");
3801 *image_fd = safe_close(*image_fd);
3802 }
3803
3804 control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
3805 if (control < 0) {
3806 log_warning_errno(errno, "Failed to open /dev/loop-control: %m");
3807 return;
3808 }
3809
3810 r = ioctl(control, LOOP_CTL_REMOVE, nr);
3811 if (r < 0)
3812 log_debug_errno(errno, "Failed to remove loop %d: %m", nr);
3813 }
3814
3815 static int spawn_getent(const char *database, const char *key, pid_t *rpid) {
3816 int pipe_fds[2];
3817 pid_t pid;
3818
3819 assert(database);
3820 assert(key);
3821 assert(rpid);
3822
3823 if (pipe2(pipe_fds, O_CLOEXEC) < 0)
3824 return log_error_errno(errno, "Failed to allocate pipe: %m");
3825
3826 pid = fork();
3827 if (pid < 0)
3828 return log_error_errno(errno, "Failed to fork getent child: %m");
3829 else if (pid == 0) {
3830 int nullfd;
3831 char *empty_env = NULL;
3832
3833 if (dup3(pipe_fds[1], STDOUT_FILENO, 0) < 0)
3834 _exit(EXIT_FAILURE);
3835
3836 if (pipe_fds[0] > 2)
3837 safe_close(pipe_fds[0]);
3838 if (pipe_fds[1] > 2)
3839 safe_close(pipe_fds[1]);
3840
3841 nullfd = open("/dev/null", O_RDWR);
3842 if (nullfd < 0)
3843 _exit(EXIT_FAILURE);
3844
3845 if (dup3(nullfd, STDIN_FILENO, 0) < 0)
3846 _exit(EXIT_FAILURE);
3847
3848 if (dup3(nullfd, STDERR_FILENO, 0) < 0)
3849 _exit(EXIT_FAILURE);
3850
3851 if (nullfd > 2)
3852 safe_close(nullfd);
3853
3854 (void) reset_all_signal_handlers();
3855 (void) reset_signal_mask();
3856 close_all_fds(NULL, 0);
3857
3858 execle("/usr/bin/getent", "getent", database, key, NULL, &empty_env);
3859 execle("/bin/getent", "getent", database, key, NULL, &empty_env);
3860 _exit(EXIT_FAILURE);
3861 }
3862
3863 pipe_fds[1] = safe_close(pipe_fds[1]);
3864
3865 *rpid = pid;
3866
3867 return pipe_fds[0];
3868 }
3869
3870 static int change_uid_gid(char **_home) {
3871 char line[LINE_MAX], *x, *u, *g, *h;
3872 const char *word, *state;
3873 _cleanup_free_ uid_t *uids = NULL;
3874 _cleanup_free_ char *home = NULL;
3875 _cleanup_fclose_ FILE *f = NULL;
3876 _cleanup_close_ int fd = -1;
3877 unsigned n_uids = 0;
3878 size_t sz = 0, l;
3879 uid_t uid;
3880 gid_t gid;
3881 pid_t pid;
3882 int r;
3883
3884 assert(_home);
3885
3886 if (!arg_user || streq(arg_user, "root") || streq(arg_user, "0")) {
3887 /* Reset everything fully to 0, just in case */
3888
3889 r = reset_uid_gid();
3890 if (r < 0)
3891 return log_error_errno(r, "Failed to become root: %m");
3892
3893 *_home = NULL;
3894 return 0;
3895 }
3896
3897 /* First, get user credentials */
3898 fd = spawn_getent("passwd", arg_user, &pid);
3899 if (fd < 0)
3900 return fd;
3901
3902 f = fdopen(fd, "r");
3903 if (!f)
3904 return log_oom();
3905 fd = -1;
3906
3907 if (!fgets(line, sizeof(line), f)) {
3908
3909 if (!ferror(f)) {
3910 log_error("Failed to resolve user %s.", arg_user);
3911 return -ESRCH;
3912 }
3913
3914 log_error_errno(errno, "Failed to read from getent: %m");
3915 return -errno;
3916 }
3917
3918 truncate_nl(line);
3919
3920 wait_for_terminate_and_warn("getent passwd", pid, true);
3921
3922 x = strchr(line, ':');
3923 if (!x) {
3924 log_error("/etc/passwd entry has invalid user field.");
3925 return -EIO;
3926 }
3927
3928 u = strchr(x+1, ':');
3929 if (!u) {
3930 log_error("/etc/passwd entry has invalid password field.");
3931 return -EIO;
3932 }
3933
3934 u++;
3935 g = strchr(u, ':');
3936 if (!g) {
3937 log_error("/etc/passwd entry has invalid UID field.");
3938 return -EIO;
3939 }
3940
3941 *g = 0;
3942 g++;
3943 x = strchr(g, ':');
3944 if (!x) {
3945 log_error("/etc/passwd entry has invalid GID field.");
3946 return -EIO;
3947 }
3948
3949 *x = 0;
3950 h = strchr(x+1, ':');
3951 if (!h) {
3952 log_error("/etc/passwd entry has invalid GECOS field.");
3953 return -EIO;
3954 }
3955
3956 h++;
3957 x = strchr(h, ':');
3958 if (!x) {
3959 log_error("/etc/passwd entry has invalid home directory field.");
3960 return -EIO;
3961 }
3962
3963 *x = 0;
3964
3965 r = parse_uid(u, &uid);
3966 if (r < 0) {
3967 log_error("Failed to parse UID of user.");
3968 return -EIO;
3969 }
3970
3971 r = parse_gid(g, &gid);
3972 if (r < 0) {
3973 log_error("Failed to parse GID of user.");
3974 return -EIO;
3975 }
3976
3977 home = strdup(h);
3978 if (!home)
3979 return log_oom();
3980
3981 /* Second, get group memberships */
3982 fd = spawn_getent("initgroups", arg_user, &pid);
3983 if (fd < 0)
3984 return fd;
3985
3986 fclose(f);
3987 f = fdopen(fd, "r");
3988 if (!f)
3989 return log_oom();
3990 fd = -1;
3991
3992 if (!fgets(line, sizeof(line), f)) {
3993 if (!ferror(f)) {
3994 log_error("Failed to resolve user %s.", arg_user);
3995 return -ESRCH;
3996 }
3997
3998 log_error_errno(errno, "Failed to read from getent: %m");
3999 return -errno;
4000 }
4001
4002 truncate_nl(line);
4003
4004 wait_for_terminate_and_warn("getent initgroups", pid, true);
4005
4006 /* Skip over the username and subsequent separator whitespace */
4007 x = line;
4008 x += strcspn(x, WHITESPACE);
4009 x += strspn(x, WHITESPACE);
4010
4011 FOREACH_WORD(word, l, x, state) {
4012 char c[l+1];
4013
4014 memcpy(c, word, l);
4015 c[l] = 0;
4016
4017 if (!GREEDY_REALLOC(uids, sz, n_uids+1))
4018 return log_oom();
4019
4020 r = parse_uid(c, &uids[n_uids++]);
4021 if (r < 0) {
4022 log_error("Failed to parse group data from getent.");
4023 return -EIO;
4024 }
4025 }
4026
4027 r = mkdir_parents(home, 0775);
4028 if (r < 0)
4029 return log_error_errno(r, "Failed to make home root directory: %m");
4030
4031 r = mkdir_safe(home, 0755, uid, gid);
4032 if (r < 0 && r != -EEXIST)
4033 return log_error_errno(r, "Failed to make home directory: %m");
4034
4035 (void) fchown(STDIN_FILENO, uid, gid);
4036 (void) fchown(STDOUT_FILENO, uid, gid);
4037 (void) fchown(STDERR_FILENO, uid, gid);
4038
4039 if (setgroups(n_uids, uids) < 0)
4040 return log_error_errno(errno, "Failed to set auxiliary groups: %m");
4041
4042 if (setresgid(gid, gid, gid) < 0)
4043 return log_error_errno(errno, "setregid() failed: %m");
4044
4045 if (setresuid(uid, uid, uid) < 0)
4046 return log_error_errno(errno, "setreuid() failed: %m");
4047
4048 if (_home) {
4049 *_home = home;
4050 home = NULL;
4051 }
4052
4053 return 0;
4054 }
4055
4056 /*
4057 * Return values:
4058 * < 0 : wait_for_terminate() failed to get the state of the
4059 * container, the container was terminated by a signal, or
4060 * failed for an unknown reason. No change is made to the
4061 * container argument.
4062 * > 0 : The program executed in the container terminated with an
4063 * error. The exit code of the program executed in the
4064 * container is returned. The container argument has been set
4065 * to CONTAINER_TERMINATED.
4066 * 0 : The container is being rebooted, has been shut down or exited
4067 * successfully. The container argument has been set to either
4068 * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
4069 *
4070 * That is, success is indicated by a return value of zero, and an
4071 * error is indicated by a non-zero value.
4072 */
4073 static int wait_for_container(pid_t pid, ContainerStatus *container) {
4074 siginfo_t status;
4075 int r;
4076
4077 r = wait_for_terminate(pid, &status);
4078 if (r < 0)
4079 return log_warning_errno(r, "Failed to wait for container: %m");
4080
4081 switch (status.si_code) {
4082
4083 case CLD_EXITED:
4084 if (status.si_status == 0) {
4085 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
4086
4087 } else
4088 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
4089
4090 *container = CONTAINER_TERMINATED;
4091 return status.si_status;
4092
4093 case CLD_KILLED:
4094 if (status.si_status == SIGINT) {
4095
4096 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
4097 *container = CONTAINER_TERMINATED;
4098 return 0;
4099
4100 } else if (status.si_status == SIGHUP) {
4101
4102 log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
4103 *container = CONTAINER_REBOOTED;
4104 return 0;
4105 }
4106
4107 /* CLD_KILLED fallthrough */
4108
4109 case CLD_DUMPED:
4110 log_error("Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
4111 return -EIO;
4112
4113 default:
4114 log_error("Container %s failed due to unknown reason.", arg_machine);
4115 return -EIO;
4116 }
4117
4118 return r;
4119 }
4120
4121 static void nop_handler(int sig) {}
4122
4123 static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
4124 pid_t pid;
4125
4126 pid = PTR_TO_UINT32(userdata);
4127 if (pid > 0) {
4128 if (kill(pid, arg_kill_signal) >= 0) {
4129 log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
4130 sd_event_source_set_userdata(s, NULL);
4131 return 0;
4132 }
4133 }
4134
4135 sd_event_exit(sd_event_source_get_event(s), 0);
4136 return 0;
4137 }
4138
4139 static int determine_names(void) {
4140 int r;
4141
4142 if (arg_template && !arg_directory && arg_machine) {
4143
4144 /* If --template= was specified then we should not
4145 * search for a machine, but instead create a new one
4146 * in /var/lib/machine. */
4147
4148 arg_directory = strjoin("/var/lib/machines/", arg_machine, NULL);
4149 if (!arg_directory)
4150 return log_oom();
4151 }
4152
4153 if (!arg_image && !arg_directory) {
4154 if (arg_machine) {
4155 _cleanup_(image_unrefp) Image *i = NULL;
4156
4157 r = image_find(arg_machine, &i);
4158 if (r < 0)
4159 return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
4160 else if (r == 0) {
4161 log_error("No image for machine '%s': %m", arg_machine);
4162 return -ENOENT;
4163 }
4164
4165 if (i->type == IMAGE_RAW)
4166 r = set_sanitized_path(&arg_image, i->path);
4167 else
4168 r = set_sanitized_path(&arg_directory, i->path);
4169 if (r < 0)
4170 return log_error_errno(r, "Invalid image directory: %m");
4171
4172 if (!arg_ephemeral)
4173 arg_read_only = arg_read_only || i->read_only;
4174 } else
4175 arg_directory = get_current_dir_name();
4176
4177 if (!arg_directory && !arg_machine) {
4178 log_error("Failed to determine path, please use -D or -i.");
4179 return -EINVAL;
4180 }
4181 }
4182
4183 if (!arg_machine) {
4184 if (arg_directory && path_equal(arg_directory, "/"))
4185 arg_machine = gethostname_malloc();
4186 else
4187 arg_machine = strdup(basename(arg_image ?: arg_directory));
4188
4189 if (!arg_machine)
4190 return log_oom();
4191
4192 hostname_cleanup(arg_machine);
4193 if (!machine_name_is_valid(arg_machine)) {
4194 log_error("Failed to determine machine name automatically, please use -M.");
4195 return -EINVAL;
4196 }
4197
4198 if (arg_ephemeral) {
4199 char *b;
4200
4201 /* Add a random suffix when this is an
4202 * ephemeral machine, so that we can run many
4203 * instances at once without manually having
4204 * to specify -M each time. */
4205
4206 if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
4207 return log_oom();
4208
4209 free(arg_machine);
4210 arg_machine = b;
4211 }
4212 }
4213
4214 return 0;
4215 }
4216
4217 static int determine_uid_shift(const char *directory) {
4218 int r;
4219
4220 if (!arg_userns) {
4221 arg_uid_shift = 0;
4222 return 0;
4223 }
4224
4225 if (arg_uid_shift == UID_INVALID) {
4226 struct stat st;
4227
4228 r = stat(directory, &st);
4229 if (r < 0)
4230 return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
4231
4232 arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
4233
4234 if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000))) {
4235 log_error("UID and GID base of %s don't match.", directory);
4236 return -EINVAL;
4237 }
4238
4239 arg_uid_range = UINT32_C(0x10000);
4240 }
4241
4242 if (arg_uid_shift > (uid_t) -1 - arg_uid_range) {
4243 log_error("UID base too high for UID range.");
4244 return -EINVAL;
4245 }
4246
4247 log_info("Using user namespaces with base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
4248 return 0;
4249 }
4250
4251 static int inner_child(
4252 Barrier *barrier,
4253 const char *directory,
4254 bool secondary,
4255 int kmsg_socket,
4256 int rtnl_socket,
4257 FDSet *fds,
4258 int argc,
4259 char *argv[]) {
4260
4261 _cleanup_free_ char *home = NULL;
4262 unsigned n_env = 2;
4263 const char *envp[] = {
4264 "PATH=" DEFAULT_PATH_SPLIT_USR,
4265 "container=systemd-nspawn", /* LXC sets container=lxc, so follow the scheme here */
4266 NULL, /* TERM */
4267 NULL, /* HOME */
4268 NULL, /* USER */
4269 NULL, /* LOGNAME */
4270 NULL, /* container_uuid */
4271 NULL, /* LISTEN_FDS */
4272 NULL, /* LISTEN_PID */
4273 NULL
4274 };
4275
4276 _cleanup_strv_free_ char **env_use = NULL;
4277 int r;
4278
4279 assert(barrier);
4280 assert(directory);
4281 assert(kmsg_socket >= 0);
4282
4283 cg_unified_flush();
4284
4285 if (arg_userns) {
4286 /* Tell the parent, that it now can write the UID map. */
4287 (void) barrier_place(barrier); /* #1 */
4288
4289 /* Wait until the parent wrote the UID map */
4290 if (!barrier_place_and_sync(barrier)) { /* #2 */
4291 log_error("Parent died too early");
4292 return -ESRCH;
4293 }
4294 }
4295
4296 r = mount_all(NULL, true);
4297 if (r < 0)
4298 return r;
4299
4300 /* Wait until we are cgroup-ified, so that we
4301 * can mount the right cgroup path writable */
4302 if (!barrier_place_and_sync(barrier)) { /* #3 */
4303 log_error("Parent died too early");
4304 return -ESRCH;
4305 }
4306
4307 r = mount_systemd_cgroup_writable("");
4308 if (r < 0)
4309 return r;
4310
4311 r = reset_uid_gid();
4312 if (r < 0)
4313 return log_error_errno(r, "Couldn't become new root: %m");
4314
4315 r = setup_boot_id(NULL);
4316 if (r < 0)
4317 return r;
4318
4319 r = setup_kmsg(NULL, kmsg_socket);
4320 if (r < 0)
4321 return r;
4322 kmsg_socket = safe_close(kmsg_socket);
4323
4324 umask(0022);
4325
4326 if (setsid() < 0)
4327 return log_error_errno(errno, "setsid() failed: %m");
4328
4329 if (arg_private_network)
4330 loopback_setup();
4331
4332 r = send_rtnl(rtnl_socket);
4333 if (r < 0)
4334 return r;
4335 rtnl_socket = safe_close(rtnl_socket);
4336
4337 if (drop_capabilities() < 0)
4338 return log_error_errno(errno, "drop_capabilities() failed: %m");
4339
4340 setup_hostname();
4341
4342 if (arg_personality != PERSONALITY_INVALID) {
4343 if (personality(arg_personality) < 0)
4344 return log_error_errno(errno, "personality() failed: %m");
4345 } else if (secondary) {
4346 if (personality(PER_LINUX32) < 0)
4347 return log_error_errno(errno, "personality() failed: %m");
4348 }
4349
4350 #ifdef HAVE_SELINUX
4351 if (arg_selinux_context)
4352 if (setexeccon((security_context_t) arg_selinux_context) < 0)
4353 return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
4354 #endif
4355
4356 r = change_uid_gid(&home);
4357 if (r < 0)
4358 return r;
4359
4360 envp[n_env] = strv_find_prefix(environ, "TERM=");
4361 if (envp[n_env])
4362 n_env ++;
4363
4364 if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) ||
4365 (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) ||
4366 (asprintf((char**)(envp + n_env++), "LOGNAME=%s", arg_user ? arg_user : "root") < 0))
4367 return log_oom();
4368
4369 if (!sd_id128_equal(arg_uuid, SD_ID128_NULL)) {
4370 char as_uuid[37];
4371
4372 if (asprintf((char**)(envp + n_env++), "container_uuid=%s", id128_format_as_uuid(arg_uuid, as_uuid)) < 0)
4373 return log_oom();
4374 }
4375
4376 if (fdset_size(fds) > 0) {
4377 r = fdset_cloexec(fds, false);
4378 if (r < 0)
4379 return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
4380
4381 if ((asprintf((char **)(envp + n_env++), "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
4382 (asprintf((char **)(envp + n_env++), "LISTEN_PID=1") < 0))
4383 return log_oom();
4384 }
4385
4386 env_use = strv_env_merge(2, envp, arg_setenv);
4387 if (!env_use)
4388 return log_oom();
4389
4390 /* Let the parent know that we are ready and
4391 * wait until the parent is ready with the
4392 * setup, too... */
4393 if (!barrier_place_and_sync(barrier)) { /* #4 */
4394 log_error("Parent died too early");
4395 return -ESRCH;
4396 }
4397
4398 /* Now, explicitly close the log, so that we
4399 * then can close all remaining fds. Closing
4400 * the log explicitly first has the benefit
4401 * that the logging subsystem knows about it,
4402 * and is thus ready to be reopened should we
4403 * need it again. Note that the other fds
4404 * closed here are at least the locking and
4405 * barrier fds. */
4406 log_close();
4407 (void) fdset_close_others(fds);
4408
4409 if (arg_boot) {
4410 char **a;
4411 size_t m;
4412
4413 /* Automatically search for the init system */
4414
4415 m = 1 + argc - optind;
4416 a = newa(char*, m + 1);
4417 memcpy(a + 1, argv + optind, m * sizeof(char*));
4418
4419 a[0] = (char*) "/usr/lib/systemd/systemd";
4420 execve(a[0], a, env_use);
4421
4422 a[0] = (char*) "/lib/systemd/systemd";
4423 execve(a[0], a, env_use);
4424
4425 a[0] = (char*) "/sbin/init";
4426 execve(a[0], a, env_use);
4427 } else if (argc > optind)
4428 execvpe(argv[optind], argv + optind, env_use);
4429 else {
4430 chdir(home ? home : "/root");
4431 execle("/bin/bash", "-bash", NULL, env_use);
4432 execle("/bin/sh", "-sh", NULL, env_use);
4433 }
4434
4435 (void) log_open();
4436 return log_error_errno(errno, "execv() failed: %m");
4437 }
4438
4439 static int outer_child(
4440 Barrier *barrier,
4441 const char *directory,
4442 const char *console,
4443 const char *root_device, bool root_device_rw,
4444 const char *home_device, bool home_device_rw,
4445 const char *srv_device, bool srv_device_rw,
4446 bool interactive,
4447 bool secondary,
4448 int pid_socket,
4449 int kmsg_socket,
4450 int rtnl_socket,
4451 int uid_shift_socket,
4452 FDSet *fds,
4453 int argc,
4454 char *argv[]) {
4455
4456 pid_t pid;
4457 ssize_t l;
4458 int r;
4459
4460 assert(barrier);
4461 assert(directory);
4462 assert(console);
4463 assert(pid_socket >= 0);
4464 assert(kmsg_socket >= 0);
4465
4466 cg_unified_flush();
4467
4468 if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
4469 return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
4470
4471 if (interactive) {
4472 close_nointr(STDIN_FILENO);
4473 close_nointr(STDOUT_FILENO);
4474 close_nointr(STDERR_FILENO);
4475
4476 r = open_terminal(console, O_RDWR);
4477 if (r != STDIN_FILENO) {
4478 if (r >= 0) {
4479 safe_close(r);
4480 r = -EINVAL;
4481 }
4482
4483 return log_error_errno(r, "Failed to open console: %m");
4484 }
4485
4486 if (dup2(STDIN_FILENO, STDOUT_FILENO) != STDOUT_FILENO ||
4487 dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
4488 return log_error_errno(errno, "Failed to duplicate console: %m");
4489 }
4490
4491 r = reset_audit_loginuid();
4492 if (r < 0)
4493 return r;
4494
4495 /* Mark everything as slave, so that we still
4496 * receive mounts from the real root, but don't
4497 * propagate mounts to the real root. */
4498 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
4499 return log_error_errno(errno, "MS_SLAVE|MS_REC failed: %m");
4500
4501 r = mount_devices(directory,
4502 root_device, root_device_rw,
4503 home_device, home_device_rw,
4504 srv_device, srv_device_rw);
4505 if (r < 0)
4506 return r;
4507
4508 r = determine_uid_shift(directory);
4509 if (r < 0)
4510 return r;
4511
4512 if (arg_userns) {
4513 l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
4514 if (l < 0)
4515 return log_error_errno(errno, "Failed to send UID shift: %m");
4516 if (l != sizeof(arg_uid_shift)) {
4517 log_error("Short write while sending UID shift.");
4518 return -EIO;
4519 }
4520 }
4521
4522 /* Turn directory into bind mount */
4523 if (mount(directory, directory, NULL, MS_BIND|MS_REC, NULL) < 0)
4524 return log_error_errno(errno, "Failed to make bind mount: %m");
4525
4526 r = setup_volatile(directory);
4527 if (r < 0)
4528 return r;
4529
4530 r = setup_volatile_state(directory);
4531 if (r < 0)
4532 return r;
4533
4534 r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
4535 if (r < 0)
4536 return r;
4537
4538 if (arg_read_only) {
4539 r = bind_remount_recursive(directory, true);
4540 if (r < 0)
4541 return log_error_errno(r, "Failed to make tree read-only: %m");
4542 }
4543
4544 r = mount_all(directory, false);
4545 if (r < 0)
4546 return r;
4547
4548 if (copy_devnodes(directory) < 0)
4549 return r;
4550
4551 dev_setup(directory, arg_uid_shift, arg_uid_shift);
4552
4553 if (setup_pts(directory) < 0)
4554 return r;
4555
4556 r = setup_propagate(directory);
4557 if (r < 0)
4558 return r;
4559
4560 r = setup_dev_console(directory, console);
4561 if (r < 0)
4562 return r;
4563
4564 r = setup_seccomp();
4565 if (r < 0)
4566 return r;
4567
4568 r = setup_timezone(directory);
4569 if (r < 0)
4570 return r;
4571
4572 r = setup_resolv_conf(directory);
4573 if (r < 0)
4574 return r;
4575
4576 r = setup_journal(directory);
4577 if (r < 0)
4578 return r;
4579
4580 r = mount_custom(directory);
4581 if (r < 0)
4582 return r;
4583
4584 r = mount_cgroups(directory);
4585 if (r < 0)
4586 return r;
4587
4588 r = mount_move_root(directory);
4589 if (r < 0)
4590 return log_error_errno(r, "Failed to move root directory: %m");
4591
4592 pid = raw_clone(SIGCHLD|CLONE_NEWNS|
4593 (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS) |
4594 (arg_private_network ? CLONE_NEWNET : 0) |
4595 (arg_userns ? CLONE_NEWUSER : 0),
4596 NULL);
4597 if (pid < 0)
4598 return log_error_errno(errno, "Failed to fork inner child: %m");
4599 if (pid == 0) {
4600 pid_socket = safe_close(pid_socket);
4601 uid_shift_socket = safe_close(uid_shift_socket);
4602
4603 /* The inner child has all namespaces that are
4604 * requested, so that we all are owned by the user if
4605 * user namespaces are turned on. */
4606
4607 r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, fds, argc, argv);
4608 if (r < 0)
4609 _exit(EXIT_FAILURE);
4610
4611 _exit(EXIT_SUCCESS);
4612 }
4613
4614 l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
4615 if (l < 0)
4616 return log_error_errno(errno, "Failed to send PID: %m");
4617 if (l != sizeof(pid)) {
4618 log_error("Short write while sending PID.");
4619 return -EIO;
4620 }
4621
4622 pid_socket = safe_close(pid_socket);
4623
4624 return 0;
4625 }
4626
4627 static int setup_uid_map(pid_t pid) {
4628 char uid_map[strlen("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1], line[DECIMAL_STR_MAX(uid_t)*3+3+1];
4629 int r;
4630
4631 assert(pid > 1);
4632
4633 xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
4634 xsprintf(line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, arg_uid_shift, arg_uid_range);
4635 r = write_string_file(uid_map, line, 0);
4636 if (r < 0)
4637 return log_error_errno(r, "Failed to write UID map: %m");
4638
4639 /* We always assign the same UID and GID ranges */
4640 xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
4641 r = write_string_file(uid_map, line, 0);
4642 if (r < 0)
4643 return log_error_errno(r, "Failed to write GID map: %m");
4644
4645 return 0;
4646 }
4647
4648 static int chown_cgroup(pid_t pid) {
4649 _cleanup_free_ char *path = NULL, *fs = NULL;
4650 _cleanup_close_ int fd = -1;
4651 const char *fn;
4652 int r;
4653
4654 r = cg_pid_get_path(NULL, pid, &path);
4655 if (r < 0)
4656 return log_error_errno(r, "Failed to get container cgroup path: %m");
4657
4658 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, NULL, &fs);
4659 if (r < 0)
4660 return log_error_errno(r, "Failed to get file system path for container cgroup: %m");
4661
4662 fd = open(fs, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
4663 if (fd < 0)
4664 return log_error_errno(errno, "Failed to open %s: %m", fs);
4665
4666 FOREACH_STRING(fn,
4667 ".",
4668 "tasks",
4669 "notify_on_release",
4670 "cgroup.procs",
4671 "cgroup.clone_children",
4672 "cgroup.controllers",
4673 "cgroup.subtree_control",
4674 "cgroup.populated")
4675 if (fchownat(fd, fn, arg_uid_shift, arg_uid_shift, 0) < 0)
4676 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
4677 "Failed to chown() cgroup file %s, ignoring: %m", fn);
4678
4679 return 0;
4680 }
4681
4682 static int sync_cgroup(pid_t pid) {
4683 _cleanup_free_ char *cgroup = NULL;
4684 char tree[] = "/tmp/unifiedXXXXXX", pid_string[DECIMAL_STR_MAX(pid) + 1];
4685 bool undo_mount = false;
4686 const char *fn;
4687 int unified, r;
4688
4689 unified = cg_unified();
4690 if (unified < 0)
4691 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
4692
4693 if ((unified > 0) == arg_unified_cgroup_hierarchy)
4694 return 0;
4695
4696 /* When the host uses the legacy cgroup setup, but the
4697 * container shall use the unified hierarchy, let's make sure
4698 * we copy the path from the name=systemd hierarchy into the
4699 * unified hierarchy. Similar for the reverse situation. */
4700
4701 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
4702 if (r < 0)
4703 return log_error_errno(r, "Failed to get control group of " PID_FMT ": %m", pid);
4704
4705 /* In order to access the unified hierarchy we need to mount it */
4706 if (!mkdtemp(tree))
4707 return log_error_errno(errno, "Failed to generate temporary mount point for unified hierarchy: %m");
4708
4709 if (unified)
4710 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
4711 else
4712 r = mount("cgroup", tree, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "__DEVEL__sane_behavior");
4713 if (r < 0) {
4714 r = log_error_errno(errno, "Failed to mount unified hierarchy: %m");
4715 goto finish;
4716 }
4717
4718 undo_mount = true;
4719
4720 fn = strjoina(tree, cgroup, "/cgroup.procs");
4721 (void) mkdir_parents(fn, 0755);
4722
4723 sprintf(pid_string, PID_FMT, pid);
4724 r = write_string_file(fn, pid_string, 0);
4725 if (r < 0)
4726 log_error_errno(r, "Failed to move process: %m");
4727
4728 finish:
4729 if (undo_mount)
4730 (void) umount(tree);
4731
4732 (void) rmdir(tree);
4733 return r;
4734 }
4735
4736 static int create_subcgroup(pid_t pid) {
4737 _cleanup_free_ char *cgroup = NULL;
4738 const char *child;
4739 int unified, r;
4740
4741 /* In the unified hierarchy inner nodes may only only contain
4742 * subgroups, but not processes. Hence, if we running in the
4743 * unified hierarchy and the container does the same, and we
4744 * did not create a scope unit for the container move us and
4745 * the container into two separate subcgroups. */
4746
4747 if (!arg_keep_unit)
4748 return 0;
4749
4750 if (!arg_unified_cgroup_hierarchy)
4751 return 0;
4752
4753 unified = cg_unified();
4754 if (unified < 0)
4755 return log_error_errno(unified, "Failed to determine whether the unified hierachy is used: %m");
4756 if (unified == 0)
4757 return 0;
4758
4759 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
4760 if (r < 0)
4761 return log_error_errno(r, "Failed to get our control group: %m");
4762
4763 child = strjoina(cgroup, "/payload");
4764 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, pid);
4765 if (r < 0)
4766 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
4767
4768 child = strjoina(cgroup, "/supervisor");
4769 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, child, 0);
4770 if (r < 0)
4771 return log_error_errno(r, "Failed to create %s subcgroup: %m", child);
4772
4773 return 0;
4774 }
4775
4776 int main(int argc, char *argv[]) {
4777
4778 _cleanup_free_ char *device_path = NULL, *root_device = NULL, *home_device = NULL, *srv_device = NULL, *console = NULL;
4779 bool root_device_rw = true, home_device_rw = true, srv_device_rw = true;
4780 _cleanup_close_ int master = -1, image_fd = -1;
4781 _cleanup_fdset_free_ FDSet *fds = NULL;
4782 int r, n_fd_passed, loop_nr = -1;
4783 char veth_name[IFNAMSIZ];
4784 bool secondary = false, remove_subvol = false;
4785 sigset_t mask_chld;
4786 pid_t pid = 0;
4787 int ret = EXIT_SUCCESS;
4788 union in_addr_union exposed = {};
4789 _cleanup_release_lock_file_ LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
4790 bool interactive;
4791
4792 log_parse_environment();
4793 log_open();
4794
4795 r = parse_argv(argc, argv);
4796 if (r <= 0)
4797 goto finish;
4798
4799 r = determine_names();
4800 if (r < 0)
4801 goto finish;
4802
4803 if (geteuid() != 0) {
4804 log_error("Need to be root.");
4805 r = -EPERM;
4806 goto finish;
4807 }
4808
4809 n_fd_passed = sd_listen_fds(false);
4810 if (n_fd_passed > 0) {
4811 r = fdset_new_listen_fds(&fds, false);
4812 if (r < 0) {
4813 log_error_errno(r, "Failed to collect file descriptors: %m");
4814 goto finish;
4815 }
4816 }
4817
4818 if (arg_directory) {
4819 assert(!arg_image);
4820
4821 if (path_equal(arg_directory, "/") && !arg_ephemeral) {
4822 log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
4823 r = -EINVAL;
4824 goto finish;
4825 }
4826
4827 if (arg_ephemeral) {
4828 _cleanup_free_ char *np = NULL;
4829
4830 /* If the specified path is a mount point we
4831 * generate the new snapshot immediately
4832 * inside it under a random name. However if
4833 * the specified is not a mount point we
4834 * create the new snapshot in the parent
4835 * directory, just next to it. */
4836 r = path_is_mount_point(arg_directory, 0);
4837 if (r < 0) {
4838 log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
4839 goto finish;
4840 }
4841 if (r > 0)
4842 r = tempfn_random_child(arg_directory, "machine.", &np);
4843 else
4844 r = tempfn_random(arg_directory, "machine.", &np);
4845 if (r < 0) {
4846 log_error_errno(r, "Failed to generate name for snapshot: %m");
4847 goto finish;
4848 }
4849
4850 r = image_path_lock(np, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4851 if (r < 0) {
4852 log_error_errno(r, "Failed to lock %s: %m", np);
4853 goto finish;
4854 }
4855
4856 r = btrfs_subvol_snapshot(arg_directory, np, (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | BTRFS_SNAPSHOT_FALLBACK_COPY | BTRFS_SNAPSHOT_RECURSIVE);
4857 if (r < 0) {
4858 log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
4859 goto finish;
4860 }
4861
4862 free(arg_directory);
4863 arg_directory = np;
4864 np = NULL;
4865
4866 remove_subvol = true;
4867
4868 } else {
4869 r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4870 if (r == -EBUSY) {
4871 log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
4872 goto finish;
4873 }
4874 if (r < 0) {
4875 log_error_errno(r, "Failed to lock %s: %m", arg_directory);
4876 return r;
4877 }
4878
4879 if (arg_template) {
4880 r = btrfs_subvol_snapshot(arg_template, arg_directory, (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | BTRFS_SNAPSHOT_FALLBACK_COPY | BTRFS_SNAPSHOT_RECURSIVE);
4881 if (r == -EEXIST) {
4882 if (!arg_quiet)
4883 log_info("Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
4884 } else if (r < 0) {
4885 log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
4886 goto finish;
4887 } else {
4888 if (!arg_quiet)
4889 log_info("Populated %s from template %s.", arg_directory, arg_template);
4890 }
4891 }
4892 }
4893
4894 if (arg_boot) {
4895 if (path_is_os_tree(arg_directory) <= 0) {
4896 log_error("Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", arg_directory);
4897 r = -EINVAL;
4898 goto finish;
4899 }
4900 } else {
4901 const char *p;
4902
4903 p = strjoina(arg_directory,
4904 argc > optind && path_is_absolute(argv[optind]) ? argv[optind] : "/usr/bin/");
4905 if (access(p, F_OK) < 0) {
4906 log_error("Directory %s lacks the binary to execute or doesn't look like a binary tree. Refusing.", arg_directory);
4907 r = -EINVAL;
4908 goto finish;
4909 }
4910 }
4911
4912 } else {
4913 char template[] = "/tmp/nspawn-root-XXXXXX";
4914
4915 assert(arg_image);
4916 assert(!arg_template);
4917
4918 r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
4919 if (r == -EBUSY) {
4920 r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
4921 goto finish;
4922 }
4923 if (r < 0) {
4924 r = log_error_errno(r, "Failed to create image lock: %m");
4925 goto finish;
4926 }
4927
4928 if (!mkdtemp(template)) {
4929 log_error_errno(errno, "Failed to create temporary directory: %m");
4930 r = -errno;
4931 goto finish;
4932 }
4933
4934 arg_directory = strdup(template);
4935 if (!arg_directory) {
4936 r = log_oom();
4937 goto finish;
4938 }
4939
4940 image_fd = setup_image(&device_path, &loop_nr);
4941 if (image_fd < 0) {
4942 r = image_fd;
4943 goto finish;
4944 }
4945
4946 r = dissect_image(image_fd,
4947 &root_device, &root_device_rw,
4948 &home_device, &home_device_rw,
4949 &srv_device, &srv_device_rw,
4950 &secondary);
4951 if (r < 0)
4952 goto finish;
4953 }
4954
4955 r = custom_mounts_prepare();
4956 if (r < 0)
4957 goto finish;
4958
4959 interactive =
4960 isatty(STDIN_FILENO) > 0 &&
4961 isatty(STDOUT_FILENO) > 0;
4962
4963 master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
4964 if (master < 0) {
4965 r = log_error_errno(errno, "Failed to acquire pseudo tty: %m");
4966 goto finish;
4967 }
4968
4969 r = ptsname_malloc(master, &console);
4970 if (r < 0) {
4971 r = log_error_errno(r, "Failed to determine tty name: %m");
4972 goto finish;
4973 }
4974
4975 if (unlockpt(master) < 0) {
4976 r = log_error_errno(errno, "Failed to unlock tty: %m");
4977 goto finish;
4978 }
4979
4980 if (!arg_quiet)
4981 log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
4982 arg_machine, arg_image ?: arg_directory);
4983
4984 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
4985
4986 assert_se(sigemptyset(&mask_chld) == 0);
4987 assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
4988
4989 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
4990 r = log_error_errno(errno, "Failed to become subreaper: %m");
4991 goto finish;
4992 }
4993
4994 for (;;) {
4995 _cleanup_close_pair_ int kmsg_socket_pair[2] = { -1, -1 }, rtnl_socket_pair[2] = { -1, -1 }, pid_socket_pair[2] = { -1, -1 },
4996 uid_shift_socket_pair[2] = { -1, -1 };
4997 ContainerStatus container_status;
4998 _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
4999 static const struct sigaction sa = {
5000 .sa_handler = nop_handler,
5001 .sa_flags = SA_NOCLDSTOP,
5002 };
5003 int ifi = 0;
5004 ssize_t l;
5005 _cleanup_event_unref_ sd_event *event = NULL;
5006 _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
5007 _cleanup_netlink_unref_ sd_netlink *rtnl = NULL;
5008 char last_char = 0;
5009
5010 r = barrier_create(&barrier);
5011 if (r < 0) {
5012 log_error_errno(r, "Cannot initialize IPC barrier: %m");
5013 goto finish;
5014 }
5015
5016 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0) {
5017 r = log_error_errno(errno, "Failed to create kmsg socket pair: %m");
5018 goto finish;
5019 }
5020
5021 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0) {
5022 r = log_error_errno(errno, "Failed to create rtnl socket pair: %m");
5023 goto finish;
5024 }
5025
5026 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, pid_socket_pair) < 0) {
5027 r = log_error_errno(errno, "Failed to create pid socket pair: %m");
5028 goto finish;
5029 }
5030
5031 if (arg_userns)
5032 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0) {
5033 r = log_error_errno(errno, "Failed to create uid shift socket pair: %m");
5034 goto finish;
5035 }
5036
5037 /* Child can be killed before execv(), so handle SIGCHLD
5038 * in order to interrupt parent's blocking calls and
5039 * give it a chance to call wait() and terminate. */
5040 r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
5041 if (r < 0) {
5042 r = log_error_errno(errno, "Failed to change the signal mask: %m");
5043 goto finish;
5044 }
5045
5046 r = sigaction(SIGCHLD, &sa, NULL);
5047 if (r < 0) {
5048 r = log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
5049 goto finish;
5050 }
5051
5052 pid = raw_clone(SIGCHLD|CLONE_NEWNS, NULL);
5053 if (pid < 0) {
5054 if (errno == EINVAL)
5055 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");
5056 else
5057 r = log_error_errno(errno, "clone() failed: %m");
5058
5059 goto finish;
5060 }
5061
5062 if (pid == 0) {
5063 /* The outer child only has a file system namespace. */
5064 barrier_set_role(&barrier, BARRIER_CHILD);
5065
5066 master = safe_close(master);
5067
5068 kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
5069 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
5070 pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
5071 uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
5072
5073 (void) reset_all_signal_handlers();
5074 (void) reset_signal_mask();
5075
5076 r = outer_child(&barrier,
5077 arg_directory,
5078 console,
5079 root_device, root_device_rw,
5080 home_device, home_device_rw,
5081 srv_device, srv_device_rw,
5082 interactive,
5083 secondary,
5084 pid_socket_pair[1],
5085 kmsg_socket_pair[1],
5086 rtnl_socket_pair[1],
5087 uid_shift_socket_pair[1],
5088 fds,
5089 argc, argv);
5090 if (r < 0)
5091 _exit(EXIT_FAILURE);
5092
5093 _exit(EXIT_SUCCESS);
5094 }
5095
5096 barrier_set_role(&barrier, BARRIER_PARENT);
5097
5098 fdset_free(fds);
5099 fds = NULL;
5100
5101 kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
5102 rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
5103 pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
5104
5105 /* Wait for the outer child. */
5106 r = wait_for_terminate_and_warn("namespace helper", pid, NULL);
5107 if (r < 0)
5108 goto finish;
5109 if (r != 0) {
5110 r = -EIO;
5111 goto finish;
5112 }
5113 pid = 0;
5114
5115 /* And now retrieve the PID of the inner child. */
5116 l = recv(pid_socket_pair[0], &pid, sizeof(pid), 0);
5117 if (l < 0) {
5118 r = log_error_errno(errno, "Failed to read inner child PID: %m");
5119 goto finish;
5120 }
5121 if (l != sizeof(pid)) {
5122 log_error("Short read while reading inner child PID: %m");
5123 r = EIO;
5124 goto finish;
5125 }
5126
5127 log_debug("Init process invoked as PID " PID_FMT, pid);
5128
5129 if (arg_userns) {
5130 if (!barrier_place_and_sync(&barrier)) { /* #1 */
5131 log_error("Child died too early.");
5132 r = -ESRCH;
5133 goto finish;
5134 }
5135
5136 l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof(arg_uid_shift), 0);
5137 if (l < 0) {
5138 r = log_error_errno(errno, "Failed to read UID shift: %m");
5139 goto finish;
5140 }
5141 if (l != sizeof(arg_uid_shift)) {
5142 log_error("Short read while reading UID shift: %m");
5143 r = EIO;
5144 goto finish;
5145 }
5146
5147 r = setup_uid_map(pid);
5148 if (r < 0)
5149 goto finish;
5150
5151 (void) barrier_place(&barrier); /* #2 */
5152 }
5153
5154 r = move_network_interfaces(pid);
5155 if (r < 0)
5156 goto finish;
5157
5158 r = setup_veth(pid, veth_name, &ifi);
5159 if (r < 0)
5160 goto finish;
5161
5162 r = setup_bridge(veth_name, &ifi);
5163 if (r < 0)
5164 goto finish;
5165
5166 r = setup_macvlan(pid);
5167 if (r < 0)
5168 goto finish;
5169
5170 r = setup_ipvlan(pid);
5171 if (r < 0)
5172 goto finish;
5173
5174 r = register_machine(pid, ifi);
5175 if (r < 0)
5176 goto finish;
5177
5178 r = sync_cgroup(pid);
5179 if (r < 0)
5180 goto finish;
5181
5182 r = create_subcgroup(pid);
5183 if (r < 0)
5184 goto finish;
5185
5186 r = chown_cgroup(pid);
5187 if (r < 0)
5188 goto finish;
5189
5190 /* Notify the child that the parent is ready with all
5191 * its setup (including cgroup-ification), and that
5192 * the child can now hand over control to the code to
5193 * run inside the container. */
5194 (void) barrier_place(&barrier); /* #3 */
5195
5196 /* Block SIGCHLD here, before notifying child.
5197 * process_pty() will handle it with the other signals. */
5198 assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
5199
5200 /* Reset signal to default */
5201 r = default_signals(SIGCHLD, -1);
5202 if (r < 0) {
5203 log_error_errno(r, "Failed to reset SIGCHLD: %m");
5204 goto finish;
5205 }
5206
5207 /* Let the child know that we are ready and wait that the child is completely ready now. */
5208 if (!barrier_place_and_sync(&barrier)) { /* #5 */
5209 log_error("Client died too early.");
5210 r = -ESRCH;
5211 goto finish;
5212 }
5213
5214 sd_notifyf(false,
5215 "READY=1\n"
5216 "STATUS=Container running.\n"
5217 "X_NSPAWN_LEADER_PID=" PID_FMT, pid);
5218
5219 r = sd_event_new(&event);
5220 if (r < 0) {
5221 log_error_errno(r, "Failed to get default event source: %m");
5222 goto finish;
5223 }
5224
5225 if (arg_kill_signal > 0) {
5226 /* Try to kill the init system on SIGINT or SIGTERM */
5227 sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, UINT32_TO_PTR(pid));
5228 sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, UINT32_TO_PTR(pid));
5229 } else {
5230 /* Immediately exit */
5231 sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
5232 sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
5233 }
5234
5235 /* simply exit on sigchld */
5236 sd_event_add_signal(event, NULL, SIGCHLD, NULL, NULL);
5237
5238 if (arg_expose_ports) {
5239 r = watch_rtnl(event, rtnl_socket_pair[0], &exposed, &rtnl);
5240 if (r < 0)
5241 goto finish;
5242
5243 (void) expose_ports(rtnl, &exposed);
5244 }
5245
5246 rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
5247
5248 r = pty_forward_new(event, master, true, !interactive, &forward);
5249 if (r < 0) {
5250 log_error_errno(r, "Failed to create PTY forwarder: %m");
5251 goto finish;
5252 }
5253
5254 r = sd_event_loop(event);
5255 if (r < 0) {
5256 log_error_errno(r, "Failed to run event loop: %m");
5257 goto finish;
5258 }
5259
5260 pty_forward_get_last_char(forward, &last_char);
5261
5262 forward = pty_forward_free(forward);
5263
5264 if (!arg_quiet && last_char != '\n')
5265 putc('\n', stdout);
5266
5267 /* Kill if it is not dead yet anyway */
5268 terminate_machine(pid);
5269
5270 /* Normally redundant, but better safe than sorry */
5271 kill(pid, SIGKILL);
5272
5273 r = wait_for_container(pid, &container_status);
5274 pid = 0;
5275
5276 if (r < 0)
5277 /* We failed to wait for the container, or the
5278 * container exited abnormally */
5279 goto finish;
5280 else if (r > 0 || container_status == CONTAINER_TERMINATED){
5281 /* The container exited with a non-zero
5282 * status, or with zero status and no reboot
5283 * was requested. */
5284 ret = r;
5285 break;
5286 }
5287
5288 /* CONTAINER_REBOOTED, loop again */
5289
5290 if (arg_keep_unit) {
5291 /* Special handling if we are running as a
5292 * service: instead of simply restarting the
5293 * machine we want to restart the entire
5294 * service, so let's inform systemd about this
5295 * with the special exit code 133. The service
5296 * file uses RestartForceExitStatus=133 so
5297 * that this results in a full nspawn
5298 * restart. This is necessary since we might
5299 * have cgroup parameters set we want to have
5300 * flushed out. */
5301 ret = 133;
5302 r = 0;
5303 break;
5304 }
5305
5306 flush_ports(&exposed);
5307 }
5308
5309 finish:
5310 sd_notify(false,
5311 "STOPPING=1\n"
5312 "STATUS=Terminating...");
5313
5314 if (pid > 0)
5315 kill(pid, SIGKILL);
5316
5317 /* Try to flush whatever is still queued in the pty */
5318 if (master >= 0)
5319 (void) copy_bytes(master, STDOUT_FILENO, (off_t) -1, false);
5320
5321 loop_remove(loop_nr, &image_fd);
5322
5323 if (remove_subvol && arg_directory) {
5324 int k;
5325
5326 k = btrfs_subvol_remove(arg_directory, true);
5327 if (k < 0)
5328 log_warning_errno(k, "Cannot remove subvolume '%s', ignoring: %m", arg_directory);
5329 }
5330
5331 if (arg_machine) {
5332 const char *p;
5333
5334 p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
5335 (void) rm_rf(p, REMOVE_ROOT);
5336 }
5337
5338 free(arg_directory);
5339 free(arg_template);
5340 free(arg_image);
5341 free(arg_machine);
5342 free(arg_user);
5343 strv_free(arg_setenv);
5344 strv_free(arg_network_interfaces);
5345 strv_free(arg_network_macvlan);
5346 strv_free(arg_network_ipvlan);
5347 custom_mount_free_all();
5348
5349 flush_ports(&exposed);
5350
5351 while (arg_expose_ports) {
5352 ExposePort *p = arg_expose_ports;
5353 LIST_REMOVE(ports, arg_expose_ports, p);
5354 free(p);
5355 }
5356
5357 return r < 0 ? EXIT_FAILURE : ret;
5358 }
Unnamed repository; edit this file 'description' to name the repository.