]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/basic/virt.c
Merge pull request #10094 from keszybz/wants-loading
[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 "def.h"
15 #include "env-util.h"
16 #include "fd-util.h"
17 #include "fileio.h"
18 #include "macro.h"
19 #include "process-util.h"
20 #include "stat-util.h"
21 #include "string-table.h"
22 #include "string-util.h"
23 #include "virt.h"
24
25 static int detect_vm_cpuid(void) {
26
27 /* CPUID is an x86 specific interface. */
28 #if defined(__i386__) || defined(__x86_64__)
29
30 static const struct {
31 const char *cpuid;
32 int id;
33 } cpuid_vendor_table[] = {
34 { "XenVMMXenVMM", VIRTUALIZATION_XEN },
35 { "KVMKVMKVM", VIRTUALIZATION_KVM },
36 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU },
37 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
38 { "VMwareVMware", VIRTUALIZATION_VMWARE },
39 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
40 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
41 /* https://wiki.freebsd.org/bhyve */
42 { "bhyve bhyve ", VIRTUALIZATION_BHYVE },
43 { "QNXQVMBSQG", VIRTUALIZATION_QNX },
44 };
45
46 uint32_t eax, ebx, ecx, edx;
47 bool hypervisor;
48
49 /* http://lwn.net/Articles/301888/ */
50
51 /* First detect whether there is a hypervisor */
52 if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) == 0)
53 return VIRTUALIZATION_NONE;
54
55 hypervisor = ecx & 0x80000000U;
56
57 if (hypervisor) {
58 union {
59 uint32_t sig32[3];
60 char text[13];
61 } sig = {};
62 unsigned j;
63
64 /* There is a hypervisor, see what it is */
65 __cpuid(0x40000000U, eax, ebx, ecx, edx);
66
67 sig.sig32[0] = ebx;
68 sig.sig32[1] = ecx;
69 sig.sig32[2] = edx;
70
71 log_debug("Virtualization found, CPUID=%s", sig.text);
72
73 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
74 if (streq(sig.text, cpuid_vendor_table[j].cpuid))
75 return cpuid_vendor_table[j].id;
76
77 return VIRTUALIZATION_VM_OTHER;
78 }
79 #endif
80 log_debug("No virtualization found in CPUID");
81
82 return VIRTUALIZATION_NONE;
83 }
84
85 static int detect_vm_device_tree(void) {
86 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
87 _cleanup_free_ char *hvtype = NULL;
88 int r;
89
90 r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
91 if (r == -ENOENT) {
92 _cleanup_closedir_ DIR *dir = NULL;
93 struct dirent *dent;
94
95 dir = opendir("/proc/device-tree");
96 if (!dir) {
97 if (errno == ENOENT) {
98 log_debug_errno(errno, "/proc/device-tree: %m");
99 return VIRTUALIZATION_NONE;
100 }
101 return -errno;
102 }
103
104 FOREACH_DIRENT(dent, dir, return -errno)
105 if (strstr(dent->d_name, "fw-cfg")) {
106 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
107 return VIRTUALIZATION_QEMU;
108 }
109
110 log_debug("No virtualization found in /proc/device-tree/*");
111 return VIRTUALIZATION_NONE;
112 } else if (r < 0)
113 return r;
114
115 log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
116 if (streq(hvtype, "linux,kvm"))
117 return VIRTUALIZATION_KVM;
118 else if (strstr(hvtype, "xen"))
119 return VIRTUALIZATION_XEN;
120 else
121 return VIRTUALIZATION_VM_OTHER;
122 #else
123 log_debug("This platform does not support /proc/device-tree");
124 return VIRTUALIZATION_NONE;
125 #endif
126 }
127
128 static int detect_vm_dmi(void) {
129 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
130
131 static const char *const dmi_vendors[] = {
132 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
133 "/sys/class/dmi/id/sys_vendor",
134 "/sys/class/dmi/id/board_vendor",
135 "/sys/class/dmi/id/bios_vendor"
136 };
137
138 static const struct {
139 const char *vendor;
140 int id;
141 } dmi_vendor_table[] = {
142 { "KVM", VIRTUALIZATION_KVM },
143 { "QEMU", VIRTUALIZATION_QEMU },
144 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
145 { "VMware", VIRTUALIZATION_VMWARE },
146 { "VMW", VIRTUALIZATION_VMWARE },
147 { "innotek GmbH", VIRTUALIZATION_ORACLE },
148 { "Xen", VIRTUALIZATION_XEN },
149 { "Bochs", VIRTUALIZATION_BOCHS },
150 { "Parallels", VIRTUALIZATION_PARALLELS },
151 /* https://wiki.freebsd.org/bhyve */
152 { "BHYVE", VIRTUALIZATION_BHYVE },
153 };
154 unsigned i;
155 int r;
156
157 for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
158 _cleanup_free_ char *s = NULL;
159 unsigned j;
160
161 r = read_one_line_file(dmi_vendors[i], &s);
162 if (r < 0) {
163 if (r == -ENOENT)
164 continue;
165
166 return r;
167 }
168
169 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
170 if (startswith(s, dmi_vendor_table[j].vendor)) {
171 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
172 return dmi_vendor_table[j].id;
173 }
174 }
175 #endif
176
177 log_debug("No virtualization found in DMI");
178
179 return VIRTUALIZATION_NONE;
180 }
181
182 static int detect_vm_xen(void) {
183
184 /* Check for Dom0 will be executed later in detect_vm_xen_dom0
185 The presence of /proc/xen indicates some form of a Xen domain */
186 if (access("/proc/xen", F_OK) < 0) {
187 log_debug("Virtualization XEN not found, /proc/xen does not exist");
188 return VIRTUALIZATION_NONE;
189 }
190
191 log_debug("Virtualization XEN found (/proc/xen exists)");
192 return VIRTUALIZATION_XEN;
193 }
194
195 #define XENFEAT_dom0 11 /* xen/include/public/features.h */
196 #define PATH_FEATURES "/sys/hypervisor/properties/features"
197 /* Returns -errno, or 0 for domU, or 1 for dom0 */
198 static int detect_vm_xen_dom0(void) {
199 _cleanup_free_ char *domcap = NULL;
200 char *cap, *i;
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 i = domcap;
232 while ((cap = strsep(&i, ",")))
233 if (streq(cap, "control_d"))
234 break;
235 if (!cap) {
236 log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
237 return 0;
238 }
239
240 log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
241 return 1;
242 }
243
244 static int detect_vm_hypervisor(void) {
245 _cleanup_free_ char *hvtype = NULL;
246 int r;
247
248 r = read_one_line_file("/sys/hypervisor/type", &hvtype);
249 if (r == -ENOENT)
250 return VIRTUALIZATION_NONE;
251 if (r < 0)
252 return r;
253
254 log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
255
256 if (streq(hvtype, "xen"))
257 return VIRTUALIZATION_XEN;
258 else
259 return VIRTUALIZATION_VM_OTHER;
260 }
261
262 static int detect_vm_uml(void) {
263 _cleanup_fclose_ FILE *f = NULL;
264 int r;
265
266 /* Detect User-Mode Linux by reading /proc/cpuinfo */
267 f = fopen("/proc/cpuinfo", "re");
268 if (!f) {
269 if (errno == ENOENT) {
270 log_debug("/proc/cpuinfo not found, assuming no UML virtualization.");
271 return VIRTUALIZATION_NONE;
272 }
273 return -errno;
274 }
275
276 for (;;) {
277 _cleanup_free_ char *line = NULL;
278 const char *t;
279
280 r = read_line(f, LONG_LINE_MAX, &line);
281 if (r < 0)
282 return r;
283 if (r == 0)
284 break;
285
286 t = startswith(line, "vendor_id\t: ");
287 if (t) {
288 if (startswith(t, "User Mode Linux")) {
289 log_debug("UML virtualization found in /proc/cpuinfo");
290 return VIRTUALIZATION_UML;
291 }
292
293 break;
294 }
295 }
296
297 log_debug("UML virtualization not found in /proc/cpuinfo.");
298 return VIRTUALIZATION_NONE;
299 }
300
301 static int detect_vm_zvm(void) {
302
303 #if defined(__s390__)
304 _cleanup_free_ char *t = NULL;
305 int r;
306
307 r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
308 if (r == -ENOENT)
309 return VIRTUALIZATION_NONE;
310 if (r < 0)
311 return r;
312
313 log_debug("Virtualization %s found in /proc/sysinfo", t);
314 if (streq(t, "z/VM"))
315 return VIRTUALIZATION_ZVM;
316 else
317 return VIRTUALIZATION_KVM;
318 #else
319 log_debug("This platform does not support /proc/sysinfo");
320 return VIRTUALIZATION_NONE;
321 #endif
322 }
323
324 /* Returns a short identifier for the various VM implementations */
325 int detect_vm(void) {
326 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
327 bool other = false;
328 int r, dmi;
329
330 if (cached_found >= 0)
331 return cached_found;
332
333 /* We have to use the correct order here:
334 *
335 * → First, try to detect Oracle Virtualbox, even if it uses KVM, as well as Xen even if it cloaks as Microsoft
336 * Hyper-V.
337 *
338 * → Second, try to detect from CPUID, this will report KVM for whatever software is used even if info in DMI is
339 * overwritten.
340 *
341 * → Third, try to detect from DMI. */
342
343 dmi = detect_vm_dmi();
344 if (IN_SET(dmi, VIRTUALIZATION_ORACLE, VIRTUALIZATION_XEN)) {
345 r = dmi;
346 goto finish;
347 }
348
349 r = detect_vm_cpuid();
350 if (r < 0)
351 return r;
352 if (r == VIRTUALIZATION_VM_OTHER)
353 other = true;
354 else if (r != VIRTUALIZATION_NONE)
355 goto finish;
356
357 /* Now, let's get back to DMI */
358 if (dmi < 0)
359 return dmi;
360 if (dmi == VIRTUALIZATION_VM_OTHER)
361 other = true;
362 else if (dmi != VIRTUALIZATION_NONE) {
363 r = dmi;
364 goto finish;
365 }
366
367 /* x86 xen will most likely be detected by cpuid. If not (most likely
368 * because we're not an x86 guest), then we should try the /proc/xen
369 * directory next. If that's not found, then we check for the high-level
370 * hypervisor sysfs file.
371 */
372
373 r = detect_vm_xen();
374 if (r < 0)
375 return r;
376 if (r == VIRTUALIZATION_VM_OTHER)
377 other = true;
378 else if (r != VIRTUALIZATION_NONE)
379 goto finish;
380
381 r = detect_vm_hypervisor();
382 if (r < 0)
383 return r;
384 if (r == VIRTUALIZATION_VM_OTHER)
385 other = true;
386 else if (r != VIRTUALIZATION_NONE)
387 goto finish;
388
389 r = detect_vm_device_tree();
390 if (r < 0)
391 return r;
392 if (r == VIRTUALIZATION_VM_OTHER)
393 other = true;
394 else if (r != VIRTUALIZATION_NONE)
395 goto finish;
396
397 r = detect_vm_uml();
398 if (r < 0)
399 return r;
400 if (r == VIRTUALIZATION_VM_OTHER)
401 other = true;
402 else if (r != VIRTUALIZATION_NONE)
403 goto finish;
404
405 r = detect_vm_zvm();
406 if (r < 0)
407 return r;
408
409 finish:
410 /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
411 * In order to detect the Dom0 as not virtualization we need to
412 * double-check it */
413 if (r == VIRTUALIZATION_XEN) {
414 int dom0;
415
416 dom0 = detect_vm_xen_dom0();
417 if (dom0 < 0)
418 return dom0;
419 if (dom0 > 0)
420 r = VIRTUALIZATION_NONE;
421 } else if (r == VIRTUALIZATION_NONE && other)
422 r = VIRTUALIZATION_VM_OTHER;
423
424 cached_found = r;
425 log_debug("Found VM virtualization %s", virtualization_to_string(r));
426 return r;
427 }
428
429 int detect_container(void) {
430
431 static const struct {
432 const char *value;
433 int id;
434 } value_table[] = {
435 { "lxc", VIRTUALIZATION_LXC },
436 { "lxc-libvirt", VIRTUALIZATION_LXC_LIBVIRT },
437 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
438 { "docker", VIRTUALIZATION_DOCKER },
439 { "rkt", VIRTUALIZATION_RKT },
440 };
441
442 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
443 _cleanup_free_ char *m = NULL;
444 const char *e = NULL;
445 unsigned j;
446 int r;
447
448 if (cached_found >= 0)
449 return cached_found;
450
451 /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
452 if (access("/proc/vz", F_OK) >= 0 &&
453 access("/proc/bc", F_OK) < 0) {
454 r = VIRTUALIZATION_OPENVZ;
455 goto finish;
456 }
457
458 if (getpid_cached() == 1) {
459 /* If we are PID 1 we can just check our own environment variable, and that's authoritative. */
460
461 e = getenv("container");
462 if (isempty(e)) {
463 r = VIRTUALIZATION_NONE;
464 goto finish;
465 }
466
467 goto translate_name;
468 }
469
470 /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
471 * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
472 r = read_one_line_file("/run/systemd/container", &m);
473 if (r >= 0) {
474 e = m;
475 goto translate_name;
476 }
477 if (r != -ENOENT)
478 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
479
480 /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
481 r = getenv_for_pid(1, "container", &m);
482 if (r > 0) {
483 e = m;
484 goto translate_name;
485 }
486 if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
487 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
488
489 /* Interestingly /proc/1/sched actually shows the host's PID for what we see as PID 1. Hence, if the PID shown
490 * there is not 1, we know we are in a PID namespace. and hence a container. */
491 r = read_one_line_file("/proc/1/sched", &m);
492 if (r >= 0) {
493 const char *t;
494
495 t = strrchr(m, '(');
496 if (!t)
497 return -EIO;
498
499 if (!startswith(t, "(1,")) {
500 r = VIRTUALIZATION_CONTAINER_OTHER;
501 goto finish;
502 }
503 } else if (r != -ENOENT)
504 return r;
505
506 /* If that didn't work, give up, assume no container manager. */
507 r = VIRTUALIZATION_NONE;
508 goto finish;
509
510 translate_name:
511 for (j = 0; j < ELEMENTSOF(value_table); j++)
512 if (streq(e, value_table[j].value)) {
513 r = value_table[j].id;
514 goto finish;
515 }
516
517 r = VIRTUALIZATION_CONTAINER_OTHER;
518
519 finish:
520 log_debug("Found container virtualization %s.", virtualization_to_string(r));
521 cached_found = r;
522 return r;
523 }
524
525 int detect_virtualization(void) {
526 int r;
527
528 r = detect_container();
529 if (r == 0)
530 r = detect_vm();
531
532 return r;
533 }
534
535 static int userns_has_mapping(const char *name) {
536 _cleanup_fclose_ FILE *f = NULL;
537 _cleanup_free_ char *buf = NULL;
538 size_t n_allocated = 0;
539 ssize_t n;
540 uint32_t a, b, c;
541 int r;
542
543 f = fopen(name, "re");
544 if (!f) {
545 log_debug_errno(errno, "Failed to open %s: %m", name);
546 return errno == ENOENT ? false : -errno;
547 }
548
549 n = getline(&buf, &n_allocated, f);
550 if (n < 0) {
551 if (feof(f)) {
552 log_debug("%s is empty, we're in an uninitialized user namespace", name);
553 return true;
554 }
555
556 return log_debug_errno(errno, "Failed to read %s: %m", name);
557 }
558
559 r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
560 if (r < 3)
561 return log_debug_errno(errno, "Failed to parse %s: %m", name);
562
563 if (a == 0 && b == 0 && c == UINT32_MAX) {
564 /* The kernel calls mappings_overlap() and does not allow overlaps */
565 log_debug("%s has a full 1:1 mapping", name);
566 return false;
567 }
568
569 /* Anything else implies that we are in a user namespace */
570 log_debug("Mapping found in %s, we're in a user namespace", name);
571 return true;
572 }
573
574 int running_in_userns(void) {
575 _cleanup_free_ char *line = NULL;
576 int r;
577
578 r = userns_has_mapping("/proc/self/uid_map");
579 if (r != 0)
580 return r;
581
582 r = userns_has_mapping("/proc/self/gid_map");
583 if (r != 0)
584 return r;
585
586 /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
587 * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
588 * also does not exist. We cannot distinguish those two cases, so assume that
589 * we're running on a stripped-down recent kernel, rather than on an old one,
590 * and if the file is not found, return false.
591 */
592 r = read_one_line_file("/proc/self/setgroups", &line);
593 if (r < 0) {
594 log_debug_errno(r, "/proc/self/setgroups: %m");
595 return r == -ENOENT ? false : r;
596 }
597
598 truncate_nl(line);
599 r = streq(line, "deny");
600 /* See user_namespaces(7) for a description of this "setgroups" contents. */
601 log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
602 return r;
603 }
604
605 int running_in_chroot(void) {
606 int r;
607
608 if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
609 return 0;
610
611 r = files_same("/proc/1/root", "/", 0);
612 if (r < 0)
613 return r;
614
615 return r == 0;
616 }
617
618 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
619 [VIRTUALIZATION_NONE] = "none",
620 [VIRTUALIZATION_KVM] = "kvm",
621 [VIRTUALIZATION_QEMU] = "qemu",
622 [VIRTUALIZATION_BOCHS] = "bochs",
623 [VIRTUALIZATION_XEN] = "xen",
624 [VIRTUALIZATION_UML] = "uml",
625 [VIRTUALIZATION_VMWARE] = "vmware",
626 [VIRTUALIZATION_ORACLE] = "oracle",
627 [VIRTUALIZATION_MICROSOFT] = "microsoft",
628 [VIRTUALIZATION_ZVM] = "zvm",
629 [VIRTUALIZATION_PARALLELS] = "parallels",
630 [VIRTUALIZATION_BHYVE] = "bhyve",
631 [VIRTUALIZATION_QNX] = "qnx",
632 [VIRTUALIZATION_VM_OTHER] = "vm-other",
633
634 [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
635 [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
636 [VIRTUALIZATION_LXC] = "lxc",
637 [VIRTUALIZATION_OPENVZ] = "openvz",
638 [VIRTUALIZATION_DOCKER] = "docker",
639 [VIRTUALIZATION_RKT] = "rkt",
640 [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
641 };
642
643 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);