]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/basic/virt.c
shared/virt: update link to vmware article
[thirdparty/systemd.git] / src / basic / virt.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #if defined(__i386__) || defined(__x86_64__)
4 #include <cpuid.h>
5 #endif
6 #include <errno.h>
7 #include <stdint.h>
8 #include <stdlib.h>
9 #include <string.h>
10 #include <unistd.h>
11
12 #include "alloc-util.h"
13 #include "dirent-util.h"
14 #include "env-util.h"
15 #include "fd-util.h"
16 #include "fileio.h"
17 #include "macro.h"
18 #include "process-util.h"
19 #include "stat-util.h"
20 #include "string-table.h"
21 #include "string-util.h"
22 #include "virt.h"
23
24 static int detect_vm_cpuid(void) {
25
26 /* CPUID is an x86 specific interface. */
27 #if defined(__i386__) || defined(__x86_64__)
28
29 static const struct {
30 const char *cpuid;
31 int id;
32 } cpuid_vendor_table[] = {
33 { "XenVMMXenVMM", VIRTUALIZATION_XEN },
34 { "KVMKVMKVM", VIRTUALIZATION_KVM },
35 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU },
36 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
37 { "VMwareVMware", VIRTUALIZATION_VMWARE },
38 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
39 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
40 /* https://wiki.freebsd.org/bhyve */
41 { "bhyve bhyve ", VIRTUALIZATION_BHYVE },
42 { "QNXQVMBSQG", VIRTUALIZATION_QNX },
43 /* https://projectacrn.org */
44 { "ACRNACRNACRN", VIRTUALIZATION_ACRN },
45 };
46
47 uint32_t eax, ebx, ecx, edx;
48 bool hypervisor;
49
50 /* http://lwn.net/Articles/301888/ */
51
52 /* First detect whether there is a hypervisor */
53 if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) == 0)
54 return VIRTUALIZATION_NONE;
55
56 hypervisor = ecx & 0x80000000U;
57
58 if (hypervisor) {
59 union {
60 uint32_t sig32[3];
61 char text[13];
62 } sig = {};
63 unsigned j;
64
65 /* There is a hypervisor, see what it is */
66 __cpuid(0x40000000U, eax, ebx, ecx, edx);
67
68 sig.sig32[0] = ebx;
69 sig.sig32[1] = ecx;
70 sig.sig32[2] = edx;
71
72 log_debug("Virtualization found, CPUID=%s", sig.text);
73
74 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
75 if (streq(sig.text, cpuid_vendor_table[j].cpuid))
76 return cpuid_vendor_table[j].id;
77
78 return VIRTUALIZATION_VM_OTHER;
79 }
80 #endif
81 log_debug("No virtualization found in CPUID");
82
83 return VIRTUALIZATION_NONE;
84 }
85
86 static int detect_vm_device_tree(void) {
87 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
88 _cleanup_free_ char *hvtype = NULL;
89 int r;
90
91 r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
92 if (r == -ENOENT) {
93 _cleanup_closedir_ DIR *dir = NULL;
94 struct dirent *dent;
95
96 dir = opendir("/proc/device-tree");
97 if (!dir) {
98 if (errno == ENOENT) {
99 log_debug_errno(errno, "/proc/device-tree: %m");
100 return VIRTUALIZATION_NONE;
101 }
102 return -errno;
103 }
104
105 FOREACH_DIRENT(dent, dir, return -errno)
106 if (strstr(dent->d_name, "fw-cfg")) {
107 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
108 return VIRTUALIZATION_QEMU;
109 }
110
111 log_debug("No virtualization found in /proc/device-tree/*");
112 return VIRTUALIZATION_NONE;
113 } else if (r < 0)
114 return r;
115
116 log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
117 if (streq(hvtype, "linux,kvm"))
118 return VIRTUALIZATION_KVM;
119 else if (strstr(hvtype, "xen"))
120 return VIRTUALIZATION_XEN;
121 else
122 return VIRTUALIZATION_VM_OTHER;
123 #else
124 log_debug("This platform does not support /proc/device-tree");
125 return VIRTUALIZATION_NONE;
126 #endif
127 }
128
129 static int detect_vm_dmi(void) {
130 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
131
132 static const char *const dmi_vendors[] = {
133 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
134 "/sys/class/dmi/id/sys_vendor",
135 "/sys/class/dmi/id/board_vendor",
136 "/sys/class/dmi/id/bios_vendor"
137 };
138
139 static const struct {
140 const char *vendor;
141 int id;
142 } dmi_vendor_table[] = {
143 { "KVM", VIRTUALIZATION_KVM },
144 { "QEMU", VIRTUALIZATION_QEMU },
145 { "VMware", VIRTUALIZATION_VMWARE }, /* https://kb.vmware.com/s/article/1009458 */
146 { "VMW", VIRTUALIZATION_VMWARE },
147 { "innotek GmbH", VIRTUALIZATION_ORACLE },
148 { "Oracle Corporation", VIRTUALIZATION_ORACLE },
149 { "Xen", VIRTUALIZATION_XEN },
150 { "Bochs", VIRTUALIZATION_BOCHS },
151 { "Parallels", VIRTUALIZATION_PARALLELS },
152 /* https://wiki.freebsd.org/bhyve */
153 { "BHYVE", VIRTUALIZATION_BHYVE },
154 };
155 unsigned i;
156 int r;
157
158 for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
159 _cleanup_free_ char *s = NULL;
160 unsigned j;
161
162 r = read_one_line_file(dmi_vendors[i], &s);
163 if (r < 0) {
164 if (r == -ENOENT)
165 continue;
166
167 return r;
168 }
169
170 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
171 if (startswith(s, dmi_vendor_table[j].vendor)) {
172 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
173 return dmi_vendor_table[j].id;
174 }
175 }
176 #endif
177
178 log_debug("No virtualization found in DMI");
179
180 return VIRTUALIZATION_NONE;
181 }
182
183 static int detect_vm_xen(void) {
184
185 /* Check for Dom0 will be executed later in detect_vm_xen_dom0
186 The presence of /proc/xen indicates some form of a Xen domain */
187 if (access("/proc/xen", F_OK) < 0) {
188 log_debug("Virtualization XEN not found, /proc/xen does not exist");
189 return VIRTUALIZATION_NONE;
190 }
191
192 log_debug("Virtualization XEN found (/proc/xen exists)");
193 return VIRTUALIZATION_XEN;
194 }
195
196 #define XENFEAT_dom0 11 /* xen/include/public/features.h */
197 #define PATH_FEATURES "/sys/hypervisor/properties/features"
198 /* Returns -errno, or 0 for domU, or 1 for dom0 */
199 static int detect_vm_xen_dom0(void) {
200 _cleanup_free_ char *domcap = NULL;
201 int r;
202
203 r = read_one_line_file(PATH_FEATURES, &domcap);
204 if (r < 0 && r != -ENOENT)
205 return r;
206 if (r >= 0) {
207 unsigned long features;
208
209 /* Here, we need to use sscanf() instead of safe_atoul()
210 * as the string lacks the leading "0x". */
211 r = sscanf(domcap, "%lx", &features);
212 if (r == 1) {
213 r = !!(features & (1U << XENFEAT_dom0));
214 log_debug("Virtualization XEN, found %s with value %08lx, "
215 "XENFEAT_dom0 (indicating the 'hardware domain') is%s set.",
216 PATH_FEATURES, features, r ? "" : " not");
217 return r;
218 }
219 log_debug("Virtualization XEN, found %s, unhandled content '%s'",
220 PATH_FEATURES, domcap);
221 }
222
223 r = read_one_line_file("/proc/xen/capabilities", &domcap);
224 if (r == -ENOENT) {
225 log_debug("Virtualization XEN because /proc/xen/capabilities does not exist");
226 return 0;
227 }
228 if (r < 0)
229 return r;
230
231 for (const char *i = domcap;;) {
232 _cleanup_free_ char *cap = NULL;
233
234 r = extract_first_word(&i, &cap, ",", 0);
235 if (r < 0)
236 return r;
237 if (r == 0) {
238 log_debug("Virtualization XEN DomU found (/proc/xen/capabilities)");
239 return 0;
240 }
241
242 if (streq(cap, "control_d")) {
243 log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
244 return 1;
245 }
246 }
247 }
248
249 static int detect_vm_hypervisor(void) {
250 _cleanup_free_ char *hvtype = NULL;
251 int r;
252
253 r = read_one_line_file("/sys/hypervisor/type", &hvtype);
254 if (r == -ENOENT)
255 return VIRTUALIZATION_NONE;
256 if (r < 0)
257 return r;
258
259 log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
260
261 if (streq(hvtype, "xen"))
262 return VIRTUALIZATION_XEN;
263 else
264 return VIRTUALIZATION_VM_OTHER;
265 }
266
267 static int detect_vm_uml(void) {
268 _cleanup_fclose_ FILE *f = NULL;
269 int r;
270
271 /* Detect User-Mode Linux by reading /proc/cpuinfo */
272 f = fopen("/proc/cpuinfo", "re");
273 if (!f) {
274 if (errno == ENOENT) {
275 log_debug("/proc/cpuinfo not found, assuming no UML virtualization.");
276 return VIRTUALIZATION_NONE;
277 }
278 return -errno;
279 }
280
281 for (;;) {
282 _cleanup_free_ char *line = NULL;
283 const char *t;
284
285 r = read_line(f, LONG_LINE_MAX, &line);
286 if (r < 0)
287 return r;
288 if (r == 0)
289 break;
290
291 t = startswith(line, "vendor_id\t: ");
292 if (t) {
293 if (startswith(t, "User Mode Linux")) {
294 log_debug("UML virtualization found in /proc/cpuinfo");
295 return VIRTUALIZATION_UML;
296 }
297
298 break;
299 }
300 }
301
302 log_debug("UML virtualization not found in /proc/cpuinfo.");
303 return VIRTUALIZATION_NONE;
304 }
305
306 static int detect_vm_zvm(void) {
307
308 #if defined(__s390__)
309 _cleanup_free_ char *t = NULL;
310 int r;
311
312 r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
313 if (r == -ENOENT)
314 return VIRTUALIZATION_NONE;
315 if (r < 0)
316 return r;
317
318 log_debug("Virtualization %s found in /proc/sysinfo", t);
319 if (streq(t, "z/VM"))
320 return VIRTUALIZATION_ZVM;
321 else
322 return VIRTUALIZATION_KVM;
323 #else
324 log_debug("This platform does not support /proc/sysinfo");
325 return VIRTUALIZATION_NONE;
326 #endif
327 }
328
329 /* Returns a short identifier for the various VM implementations */
330 int detect_vm(void) {
331 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
332 bool other = false;
333 int r, dmi;
334
335 if (cached_found >= 0)
336 return cached_found;
337
338 /* We have to use the correct order here:
339 *
340 * → First, try to detect Oracle Virtualbox, even if it uses KVM, as well as Xen even if it cloaks as Microsoft
341 * Hyper-V.
342 *
343 * → Second, try to detect from CPUID, this will report KVM for whatever software is used even if info in DMI is
344 * overwritten.
345 *
346 * → Third, try to detect from DMI. */
347
348 dmi = detect_vm_dmi();
349 if (IN_SET(dmi, VIRTUALIZATION_ORACLE, VIRTUALIZATION_XEN)) {
350 r = dmi;
351 goto finish;
352 }
353
354 r = detect_vm_cpuid();
355 if (r < 0)
356 return r;
357 if (r == VIRTUALIZATION_VM_OTHER)
358 other = true;
359 else if (r != VIRTUALIZATION_NONE)
360 goto finish;
361
362 /* Now, let's get back to DMI */
363 if (dmi < 0)
364 return dmi;
365 if (dmi == VIRTUALIZATION_VM_OTHER)
366 other = true;
367 else if (dmi != VIRTUALIZATION_NONE) {
368 r = dmi;
369 goto finish;
370 }
371
372 /* x86 xen will most likely be detected by cpuid. If not (most likely
373 * because we're not an x86 guest), then we should try the /proc/xen
374 * directory next. If that's not found, then we check for the high-level
375 * hypervisor sysfs file.
376 */
377
378 r = detect_vm_xen();
379 if (r < 0)
380 return r;
381 if (r == VIRTUALIZATION_VM_OTHER)
382 other = true;
383 else if (r != VIRTUALIZATION_NONE)
384 goto finish;
385
386 r = detect_vm_hypervisor();
387 if (r < 0)
388 return r;
389 if (r == VIRTUALIZATION_VM_OTHER)
390 other = true;
391 else if (r != VIRTUALIZATION_NONE)
392 goto finish;
393
394 r = detect_vm_device_tree();
395 if (r < 0)
396 return r;
397 if (r == VIRTUALIZATION_VM_OTHER)
398 other = true;
399 else if (r != VIRTUALIZATION_NONE)
400 goto finish;
401
402 r = detect_vm_uml();
403 if (r < 0)
404 return r;
405 if (r == VIRTUALIZATION_VM_OTHER)
406 other = true;
407 else if (r != VIRTUALIZATION_NONE)
408 goto finish;
409
410 r = detect_vm_zvm();
411 if (r < 0)
412 return r;
413
414 finish:
415 /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
416 * In order to detect the Dom0 as not virtualization we need to
417 * double-check it */
418 if (r == VIRTUALIZATION_XEN) {
419 int dom0;
420
421 dom0 = detect_vm_xen_dom0();
422 if (dom0 < 0)
423 return dom0;
424 if (dom0 > 0)
425 r = VIRTUALIZATION_NONE;
426 } else if (r == VIRTUALIZATION_NONE && other)
427 r = VIRTUALIZATION_VM_OTHER;
428
429 cached_found = r;
430 log_debug("Found VM virtualization %s", virtualization_to_string(r));
431 return r;
432 }
433
434 int detect_container(void) {
435 static const struct {
436 const char *value;
437 int id;
438 } value_table[] = {
439 { "lxc", VIRTUALIZATION_LXC },
440 { "lxc-libvirt", VIRTUALIZATION_LXC_LIBVIRT },
441 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
442 { "docker", VIRTUALIZATION_DOCKER },
443 { "podman", VIRTUALIZATION_PODMAN },
444 { "rkt", VIRTUALIZATION_RKT },
445 { "wsl", VIRTUALIZATION_WSL },
446 };
447
448 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
449 _cleanup_free_ char *m = NULL;
450 _cleanup_free_ char *o = NULL;
451 const char *e = NULL;
452 unsigned j;
453 int r;
454
455 if (cached_found >= 0)
456 return cached_found;
457
458 /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
459 if (access("/proc/vz", F_OK) >= 0 &&
460 access("/proc/bc", F_OK) < 0) {
461 r = VIRTUALIZATION_OPENVZ;
462 goto finish;
463 }
464
465 /* "Official" way of detecting WSL https://github.com/Microsoft/WSL/issues/423#issuecomment-221627364 */
466 r = read_one_line_file("/proc/sys/kernel/osrelease", &o);
467 if (r >= 0) {
468 if (strstr(o, "Microsoft") || strstr(o, "WSL")) {
469 r = VIRTUALIZATION_WSL;
470 goto finish;
471 }
472 }
473
474 if (getpid_cached() == 1) {
475 /* If we are PID 1 we can just check our own environment variable, and that's authoritative.
476 * We distinguish three cases:
477 * - the variable is not defined → we jump to other checks
478 * - the variable is defined to an empty value → we are not in a container
479 * - anything else → some container, either one of the known ones or "container-other"
480 */
481 e = getenv("container");
482 if (!e)
483 goto check_sched;
484 if (isempty(e)) {
485 r = VIRTUALIZATION_NONE;
486 goto finish;
487 }
488
489 goto translate_name;
490 }
491
492 /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
493 * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
494 r = read_one_line_file("/run/systemd/container", &m);
495 if (r > 0) {
496 e = m;
497 goto translate_name;
498 }
499 if (!IN_SET(r, -ENOENT, 0))
500 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
501
502 /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
503 r = getenv_for_pid(1, "container", &m);
504 if (r > 0) {
505 e = m;
506 goto translate_name;
507 }
508 if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
509 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
510
511 /* Interestingly /proc/1/sched actually shows the host's PID for what we see as PID 1. If the PID
512 * shown there is not 1, we know we are in a PID namespace and hence a container. */
513 check_sched:
514 r = read_one_line_file("/proc/1/sched", &m);
515 if (r >= 0) {
516 const char *t;
517
518 t = strrchr(m, '(');
519 if (!t)
520 return -EIO;
521
522 if (!startswith(t, "(1,")) {
523 r = VIRTUALIZATION_CONTAINER_OTHER;
524 goto finish;
525 }
526 } else if (r != -ENOENT)
527 return r;
528
529 /* If that didn't work, give up, assume no container manager. */
530 r = VIRTUALIZATION_NONE;
531 goto finish;
532
533 translate_name:
534 for (j = 0; j < ELEMENTSOF(value_table); j++)
535 if (streq(e, value_table[j].value)) {
536 r = value_table[j].id;
537 goto finish;
538 }
539
540 r = VIRTUALIZATION_CONTAINER_OTHER;
541
542 finish:
543 log_debug("Found container virtualization %s.", virtualization_to_string(r));
544 cached_found = r;
545 return r;
546 }
547
548 int detect_virtualization(void) {
549 int r;
550
551 r = detect_container();
552 if (r == 0)
553 r = detect_vm();
554
555 return r;
556 }
557
558 static int userns_has_mapping(const char *name) {
559 _cleanup_fclose_ FILE *f = NULL;
560 _cleanup_free_ char *buf = NULL;
561 size_t n_allocated = 0;
562 ssize_t n;
563 uint32_t a, b, c;
564 int r;
565
566 f = fopen(name, "re");
567 if (!f) {
568 log_debug_errno(errno, "Failed to open %s: %m", name);
569 return errno == ENOENT ? false : -errno;
570 }
571
572 n = getline(&buf, &n_allocated, f);
573 if (n < 0) {
574 if (feof(f)) {
575 log_debug("%s is empty, we're in an uninitialized user namespace", name);
576 return true;
577 }
578
579 return log_debug_errno(errno, "Failed to read %s: %m", name);
580 }
581
582 r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
583 if (r < 3)
584 return log_debug_errno(errno, "Failed to parse %s: %m", name);
585
586 if (a == 0 && b == 0 && c == UINT32_MAX) {
587 /* The kernel calls mappings_overlap() and does not allow overlaps */
588 log_debug("%s has a full 1:1 mapping", name);
589 return false;
590 }
591
592 /* Anything else implies that we are in a user namespace */
593 log_debug("Mapping found in %s, we're in a user namespace", name);
594 return true;
595 }
596
597 int running_in_userns(void) {
598 _cleanup_free_ char *line = NULL;
599 int r;
600
601 r = userns_has_mapping("/proc/self/uid_map");
602 if (r != 0)
603 return r;
604
605 r = userns_has_mapping("/proc/self/gid_map");
606 if (r != 0)
607 return r;
608
609 /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
610 * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
611 * also does not exist. We cannot distinguish those two cases, so assume that
612 * we're running on a stripped-down recent kernel, rather than on an old one,
613 * and if the file is not found, return false.
614 */
615 r = read_one_line_file("/proc/self/setgroups", &line);
616 if (r < 0) {
617 log_debug_errno(r, "/proc/self/setgroups: %m");
618 return r == -ENOENT ? false : r;
619 }
620
621 truncate_nl(line);
622 r = streq(line, "deny");
623 /* See user_namespaces(7) for a description of this "setgroups" contents. */
624 log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
625 return r;
626 }
627
628 int running_in_chroot(void) {
629 int r;
630
631 if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
632 return 0;
633
634 r = files_same("/proc/1/root", "/", 0);
635 if (r < 0)
636 return r;
637
638 return r == 0;
639 }
640
641 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
642 [VIRTUALIZATION_NONE] = "none",
643 [VIRTUALIZATION_KVM] = "kvm",
644 [VIRTUALIZATION_QEMU] = "qemu",
645 [VIRTUALIZATION_BOCHS] = "bochs",
646 [VIRTUALIZATION_XEN] = "xen",
647 [VIRTUALIZATION_UML] = "uml",
648 [VIRTUALIZATION_VMWARE] = "vmware",
649 [VIRTUALIZATION_ORACLE] = "oracle",
650 [VIRTUALIZATION_MICROSOFT] = "microsoft",
651 [VIRTUALIZATION_ZVM] = "zvm",
652 [VIRTUALIZATION_PARALLELS] = "parallels",
653 [VIRTUALIZATION_BHYVE] = "bhyve",
654 [VIRTUALIZATION_QNX] = "qnx",
655 [VIRTUALIZATION_ACRN] = "acrn",
656 [VIRTUALIZATION_VM_OTHER] = "vm-other",
657
658 [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
659 [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
660 [VIRTUALIZATION_LXC] = "lxc",
661 [VIRTUALIZATION_OPENVZ] = "openvz",
662 [VIRTUALIZATION_DOCKER] = "docker",
663 [VIRTUALIZATION_PODMAN] = "podman",
664 [VIRTUALIZATION_RKT] = "rkt",
665 [VIRTUALIZATION_WSL] = "wsl",
666 [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
667 };
668
669 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);