]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/basic/virt.c
virt: if we detect Xen by DMI, trust that over CPUID
[thirdparty/systemd.git] / src / basic / virt.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2 /***
3 This file is part of systemd.
4
5 Copyright 2011 Lennart Poettering
6 ***/
7
8 #if defined(__i386__) || defined(__x86_64__)
9 #include <cpuid.h>
10 #endif
11 #include <errno.h>
12 #include <stdint.h>
13 #include <stdlib.h>
14 #include <string.h>
15 #include <unistd.h>
16
17 #include "alloc-util.h"
18 #include "dirent-util.h"
19 #include "env-util.h"
20 #include "fd-util.h"
21 #include "fileio.h"
22 #include "macro.h"
23 #include "process-util.h"
24 #include "stat-util.h"
25 #include "string-table.h"
26 #include "string-util.h"
27 #include "virt.h"
28
29 static int detect_vm_cpuid(void) {
30
31 /* CPUID is an x86 specific interface. */
32 #if defined(__i386__) || defined(__x86_64__)
33
34 static const struct {
35 const char *cpuid;
36 int id;
37 } cpuid_vendor_table[] = {
38 { "XenVMMXenVMM", VIRTUALIZATION_XEN },
39 { "KVMKVMKVM", VIRTUALIZATION_KVM },
40 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU },
41 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
42 { "VMwareVMware", VIRTUALIZATION_VMWARE },
43 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
44 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
45 /* https://wiki.freebsd.org/bhyve */
46 { "bhyve bhyve ", VIRTUALIZATION_BHYVE },
47 { "QNXQVMBSQG", VIRTUALIZATION_QNX },
48 };
49
50 uint32_t eax, ebx, ecx, edx;
51 bool hypervisor;
52
53 /* http://lwn.net/Articles/301888/ */
54
55 /* First detect whether there is a hypervisor */
56 if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) == 0)
57 return VIRTUALIZATION_NONE;
58
59 hypervisor = !!(ecx & 0x80000000U);
60
61 if (hypervisor) {
62 union {
63 uint32_t sig32[3];
64 char text[13];
65 } sig = {};
66 unsigned j;
67
68 /* There is a hypervisor, see what it is */
69 __cpuid(0x40000000U, eax, ebx, ecx, edx);
70
71 sig.sig32[0] = ebx;
72 sig.sig32[1] = ecx;
73 sig.sig32[2] = edx;
74
75 log_debug("Virtualization found, CPUID=%s", sig.text);
76
77 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
78 if (streq(sig.text, cpuid_vendor_table[j].cpuid))
79 return cpuid_vendor_table[j].id;
80
81 return VIRTUALIZATION_VM_OTHER;
82 }
83 #endif
84 log_debug("No virtualization found in CPUID");
85
86 return VIRTUALIZATION_NONE;
87 }
88
89 static int detect_vm_device_tree(void) {
90 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
91 _cleanup_free_ char *hvtype = NULL;
92 int r;
93
94 r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
95 if (r == -ENOENT) {
96 _cleanup_closedir_ DIR *dir = NULL;
97 struct dirent *dent;
98
99 dir = opendir("/proc/device-tree");
100 if (!dir) {
101 if (errno == ENOENT) {
102 log_debug_errno(errno, "/proc/device-tree: %m");
103 return VIRTUALIZATION_NONE;
104 }
105 return -errno;
106 }
107
108 FOREACH_DIRENT(dent, dir, return -errno)
109 if (strstr(dent->d_name, "fw-cfg")) {
110 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
111 return VIRTUALIZATION_QEMU;
112 }
113
114 log_debug("No virtualization found in /proc/device-tree/*");
115 return VIRTUALIZATION_NONE;
116 } else if (r < 0)
117 return r;
118
119 log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
120 if (streq(hvtype, "linux,kvm"))
121 return VIRTUALIZATION_KVM;
122 else if (strstr(hvtype, "xen"))
123 return VIRTUALIZATION_XEN;
124 else
125 return VIRTUALIZATION_VM_OTHER;
126 #else
127 log_debug("This platform does not support /proc/device-tree");
128 return VIRTUALIZATION_NONE;
129 #endif
130 }
131
132 static int detect_vm_dmi(void) {
133 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
134
135 static const char *const dmi_vendors[] = {
136 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
137 "/sys/class/dmi/id/sys_vendor",
138 "/sys/class/dmi/id/board_vendor",
139 "/sys/class/dmi/id/bios_vendor"
140 };
141
142 static const struct {
143 const char *vendor;
144 int id;
145 } dmi_vendor_table[] = {
146 { "KVM", VIRTUALIZATION_KVM },
147 { "QEMU", VIRTUALIZATION_QEMU },
148 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
149 { "VMware", VIRTUALIZATION_VMWARE },
150 { "VMW", VIRTUALIZATION_VMWARE },
151 { "innotek GmbH", VIRTUALIZATION_ORACLE },
152 { "Xen", VIRTUALIZATION_XEN },
153 { "Bochs", VIRTUALIZATION_BOCHS },
154 { "Parallels", VIRTUALIZATION_PARALLELS },
155 /* https://wiki.freebsd.org/bhyve */
156 { "BHYVE", VIRTUALIZATION_BHYVE },
157 };
158 unsigned i;
159 int r;
160
161 for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
162 _cleanup_free_ char *s = NULL;
163 unsigned j;
164
165 r = read_one_line_file(dmi_vendors[i], &s);
166 if (r < 0) {
167 if (r == -ENOENT)
168 continue;
169
170 return r;
171 }
172
173 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
174 if (startswith(s, dmi_vendor_table[j].vendor)) {
175 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
176 return dmi_vendor_table[j].id;
177 }
178 }
179 #endif
180
181 log_debug("No virtualization found in DMI");
182
183 return VIRTUALIZATION_NONE;
184 }
185
186 static int detect_vm_xen(void) {
187
188 /* Check for Dom0 will be executed later in detect_vm_xen_dom0
189 The presence of /proc/xen indicates some form of a Xen domain */
190 if (access("/proc/xen", F_OK) < 0) {
191 log_debug("Virtualization XEN not found, /proc/xen does not exist");
192 return VIRTUALIZATION_NONE;
193 }
194
195 log_debug("Virtualization XEN found (/proc/xen exists)");
196 return VIRTUALIZATION_XEN;
197 }
198
199 #define XENFEAT_dom0 11 /* xen/include/public/features.h */
200 #define PATH_FEATURES "/sys/hypervisor/properties/features"
201 /* Returns -errno, or 0 for domU, or 1 for dom0 */
202 static int detect_vm_xen_dom0(void) {
203 _cleanup_free_ char *domcap = NULL;
204 char *cap, *i;
205 int r;
206
207 r = read_one_line_file(PATH_FEATURES, &domcap);
208 if (r < 0 && r != -ENOENT)
209 return r;
210 if (r == 0) {
211 unsigned long features;
212
213 /* Here, we need to use sscanf() instead of safe_atoul()
214 * as the string lacks the leading "0x". */
215 r = sscanf(domcap, "%lx", &features);
216 if (r == 1) {
217 r = !!(features & (1U << XENFEAT_dom0));
218 log_debug("Virtualization XEN, found %s with value %08lx, "
219 "XENFEAT_dom0 (indicating the 'hardware domain') is%s set.",
220 PATH_FEATURES, features, r ? "" : " not");
221 return r;
222 }
223 log_debug("Virtualization XEN, found %s, unhandled content '%s'",
224 PATH_FEATURES, domcap);
225 }
226
227 r = read_one_line_file("/proc/xen/capabilities", &domcap);
228 if (r == -ENOENT) {
229 log_debug("Virtualization XEN because /proc/xen/capabilities does not exist");
230 return 0;
231 }
232 if (r < 0)
233 return r;
234
235 i = domcap;
236 while ((cap = strsep(&i, ",")))
237 if (streq(cap, "control_d"))
238 break;
239 if (!cap) {
240 log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
241 return 0;
242 }
243
244 log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
245 return 1;
246 }
247
248 static int detect_vm_hypervisor(void) {
249 _cleanup_free_ char *hvtype = NULL;
250 int r;
251
252 r = read_one_line_file("/sys/hypervisor/type", &hvtype);
253 if (r == -ENOENT)
254 return VIRTUALIZATION_NONE;
255 if (r < 0)
256 return r;
257
258 log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
259
260 if (streq(hvtype, "xen"))
261 return VIRTUALIZATION_XEN;
262 else
263 return VIRTUALIZATION_VM_OTHER;
264 }
265
266 static int detect_vm_uml(void) {
267 _cleanup_free_ char *cpuinfo_contents = NULL;
268 int r;
269
270 /* Detect User-Mode Linux by reading /proc/cpuinfo */
271 r = read_full_file("/proc/cpuinfo", &cpuinfo_contents, NULL);
272 if (r == -ENOENT) {
273 log_debug("/proc/cpuinfo not found, assuming no UML virtualization.");
274 return VIRTUALIZATION_NONE;
275 }
276 if (r < 0)
277 return r;
278
279 if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
280 log_debug("UML virtualization found in /proc/cpuinfo");
281 return VIRTUALIZATION_UML;
282 }
283
284 log_debug("UML virtualization not found in /proc/cpuinfo.");
285 return VIRTUALIZATION_NONE;
286 }
287
288 static int detect_vm_zvm(void) {
289
290 #if defined(__s390__)
291 _cleanup_free_ char *t = NULL;
292 int r;
293
294 r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
295 if (r == -ENOENT)
296 return VIRTUALIZATION_NONE;
297 if (r < 0)
298 return r;
299
300 log_debug("Virtualization %s found in /proc/sysinfo", t);
301 if (streq(t, "z/VM"))
302 return VIRTUALIZATION_ZVM;
303 else
304 return VIRTUALIZATION_KVM;
305 #else
306 log_debug("This platform does not support /proc/sysinfo");
307 return VIRTUALIZATION_NONE;
308 #endif
309 }
310
311 /* Returns a short identifier for the various VM implementations */
312 int detect_vm(void) {
313 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
314 int r, dmi;
315 bool other = false;
316
317 if (cached_found >= 0)
318 return cached_found;
319
320 /* We have to use the correct order here:
321 *
322 * → First, try to detect Oracle Virtualbox, even if it uses KVM, as well as Xen even if it cloaks as Microsoft
323 * Hyper-V.
324 *
325 * → Second, try to detect from CPUID, this will report KVM for whatever software is used even if info in DMI is
326 * overwritten.
327 *
328 * → Third, try to detect from DMI. */
329
330 dmi = detect_vm_dmi();
331 if (IN_SET(dmi, VIRTUALIZATION_ORACLE, VIRTUALIZATION_XEN)) {
332 r = dmi;
333 goto finish;
334 }
335
336 r = detect_vm_cpuid();
337 if (r < 0)
338 return r;
339 if (r != VIRTUALIZATION_NONE) {
340 if (r == VIRTUALIZATION_VM_OTHER)
341 other = true;
342 else
343 goto finish;
344 }
345
346 r = dmi;
347 if (r < 0)
348 return r;
349 if (r != VIRTUALIZATION_NONE) {
350 if (r == VIRTUALIZATION_VM_OTHER)
351 other = true;
352 else
353 goto finish;
354 }
355
356 /* x86 xen will most likely be detected by cpuid. If not (most likely
357 * because we're not an x86 guest), then we should try the /proc/xen
358 * directory next. If that's not found, then we check for the high-level
359 * hypervisor sysfs file.
360 */
361
362 r = detect_vm_xen();
363 if (r < 0)
364 return r;
365 if (r != VIRTUALIZATION_NONE) {
366 if (r == VIRTUALIZATION_VM_OTHER)
367 other = true;
368 else
369 goto finish;
370 }
371
372 r = detect_vm_hypervisor();
373 if (r < 0)
374 return r;
375 if (r != VIRTUALIZATION_NONE) {
376 if (r == VIRTUALIZATION_VM_OTHER)
377 other = true;
378 else
379 goto finish;
380 }
381
382 r = detect_vm_device_tree();
383 if (r < 0)
384 return r;
385 if (r != VIRTUALIZATION_NONE) {
386 if (r == VIRTUALIZATION_VM_OTHER)
387 other = true;
388 else
389 goto finish;
390 }
391
392 r = detect_vm_uml();
393 if (r < 0)
394 return r;
395 if (r != VIRTUALIZATION_NONE) {
396 if (r == VIRTUALIZATION_VM_OTHER)
397 other = true;
398 else
399 goto finish;
400 }
401
402 r = detect_vm_zvm();
403 if (r < 0)
404 return r;
405
406 finish:
407 /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
408 * In order to detect the Dom0 as not virtualization we need to
409 * double-check it */
410 if (r == VIRTUALIZATION_XEN) {
411 int ret = detect_vm_xen_dom0();
412 if (ret < 0)
413 return ret;
414 if (ret > 0)
415 r = VIRTUALIZATION_NONE;
416 } else if (r == VIRTUALIZATION_NONE && other)
417 r = VIRTUALIZATION_VM_OTHER;
418
419 cached_found = r;
420 log_debug("Found VM virtualization %s", virtualization_to_string(r));
421 return r;
422 }
423
424 int detect_container(void) {
425
426 static const struct {
427 const char *value;
428 int id;
429 } value_table[] = {
430 { "lxc", VIRTUALIZATION_LXC },
431 { "lxc-libvirt", VIRTUALIZATION_LXC_LIBVIRT },
432 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
433 { "docker", VIRTUALIZATION_DOCKER },
434 { "rkt", VIRTUALIZATION_RKT },
435 };
436
437 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
438 _cleanup_free_ char *m = NULL;
439 const char *e = NULL;
440 unsigned j;
441 int r;
442
443 if (cached_found >= 0)
444 return cached_found;
445
446 /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
447 if (access("/proc/vz", F_OK) >= 0 &&
448 access("/proc/bc", F_OK) < 0) {
449 r = VIRTUALIZATION_OPENVZ;
450 goto finish;
451 }
452
453 if (getpid_cached() == 1) {
454 /* If we are PID 1 we can just check our own environment variable, and that's authoritative. */
455
456 e = getenv("container");
457 if (isempty(e)) {
458 r = VIRTUALIZATION_NONE;
459 goto finish;
460 }
461
462 goto translate_name;
463 }
464
465 /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
466 * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
467 r = read_one_line_file("/run/systemd/container", &m);
468 if (r >= 0) {
469 e = m;
470 goto translate_name;
471 }
472 if (r != -ENOENT)
473 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
474
475 /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
476 r = getenv_for_pid(1, "container", &m);
477 if (r > 0) {
478 e = m;
479 goto translate_name;
480 }
481 if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
482 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
483
484 /* Interestingly /proc/1/sched actually shows the host's PID for what we see as PID 1. Hence, if the PID shown
485 * there is not 1, we know we are in a PID namespace. and hence a container. */
486 r = read_one_line_file("/proc/1/sched", &m);
487 if (r >= 0) {
488 const char *t;
489
490 t = strrchr(m, '(');
491 if (!t)
492 return -EIO;
493
494 if (!startswith(t, "(1,")) {
495 r = VIRTUALIZATION_CONTAINER_OTHER;
496 goto finish;
497 }
498 } else if (r != -ENOENT)
499 return r;
500
501 /* If that didn't work, give up, assume no container manager. */
502 r = VIRTUALIZATION_NONE;
503 goto finish;
504
505 translate_name:
506 for (j = 0; j < ELEMENTSOF(value_table); j++)
507 if (streq(e, value_table[j].value)) {
508 r = value_table[j].id;
509 goto finish;
510 }
511
512 r = VIRTUALIZATION_CONTAINER_OTHER;
513
514 finish:
515 log_debug("Found container virtualization %s.", virtualization_to_string(r));
516 cached_found = r;
517 return r;
518 }
519
520 int detect_virtualization(void) {
521 int r;
522
523 r = detect_container();
524 if (r == 0)
525 r = detect_vm();
526
527 return r;
528 }
529
530 static int userns_has_mapping(const char *name) {
531 _cleanup_fclose_ FILE *f = NULL;
532 _cleanup_free_ char *buf = NULL;
533 size_t n_allocated = 0;
534 ssize_t n;
535 uint32_t a, b, c;
536 int r;
537
538 f = fopen(name, "re");
539 if (!f) {
540 log_debug_errno(errno, "Failed to open %s: %m", name);
541 return errno == ENOENT ? false : -errno;
542 }
543
544 n = getline(&buf, &n_allocated, f);
545 if (n < 0) {
546 if (feof(f)) {
547 log_debug("%s is empty, we're in an uninitialized user namespace", name);
548 return true;
549 }
550
551 return log_debug_errno(errno, "Failed to read %s: %m", name);
552 }
553
554 r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
555 if (r < 3)
556 return log_debug_errno(errno, "Failed to parse %s: %m", name);
557
558 if (a == 0 && b == 0 && c == UINT32_MAX) {
559 /* The kernel calls mappings_overlap() and does not allow overlaps */
560 log_debug("%s has a full 1:1 mapping", name);
561 return false;
562 }
563
564 /* Anything else implies that we are in a user namespace */
565 log_debug("Mapping found in %s, we're in a user namespace", name);
566 return true;
567 }
568
569 int running_in_userns(void) {
570 _cleanup_free_ char *line = NULL;
571 int r;
572
573 r = userns_has_mapping("/proc/self/uid_map");
574 if (r != 0)
575 return r;
576
577 r = userns_has_mapping("/proc/self/gid_map");
578 if (r != 0)
579 return r;
580
581 /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
582 * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
583 * also does not exist. We cannot distinguish those two cases, so assume that
584 * we're running on a stripped-down recent kernel, rather than on an old one,
585 * and if the file is not found, return false.
586 */
587 r = read_one_line_file("/proc/self/setgroups", &line);
588 if (r < 0) {
589 log_debug_errno(r, "/proc/self/setgroups: %m");
590 return r == -ENOENT ? false : r;
591 }
592
593 truncate_nl(line);
594 r = streq(line, "deny");
595 /* See user_namespaces(7) for a description of this "setgroups" contents. */
596 log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
597 return r;
598 }
599
600 int running_in_chroot(void) {
601 int r;
602
603 if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
604 return 0;
605
606 r = files_same("/proc/1/root", "/", 0);
607 if (r < 0)
608 return r;
609
610 return r == 0;
611 }
612
613 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
614 [VIRTUALIZATION_NONE] = "none",
615 [VIRTUALIZATION_KVM] = "kvm",
616 [VIRTUALIZATION_QEMU] = "qemu",
617 [VIRTUALIZATION_BOCHS] = "bochs",
618 [VIRTUALIZATION_XEN] = "xen",
619 [VIRTUALIZATION_UML] = "uml",
620 [VIRTUALIZATION_VMWARE] = "vmware",
621 [VIRTUALIZATION_ORACLE] = "oracle",
622 [VIRTUALIZATION_MICROSOFT] = "microsoft",
623 [VIRTUALIZATION_ZVM] = "zvm",
624 [VIRTUALIZATION_PARALLELS] = "parallels",
625 [VIRTUALIZATION_BHYVE] = "bhyve",
626 [VIRTUALIZATION_QNX] = "qnx",
627 [VIRTUALIZATION_VM_OTHER] = "vm-other",
628
629 [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
630 [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
631 [VIRTUALIZATION_LXC] = "lxc",
632 [VIRTUALIZATION_OPENVZ] = "openvz",
633 [VIRTUALIZATION_DOCKER] = "docker",
634 [VIRTUALIZATION_RKT] = "rkt",
635 [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
636 };
637
638 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);