]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/basic/virt.c
Merge pull request #4889 from evverx/fix-test-fs-util-memleak
[thirdparty/systemd.git] / src / basic / virt.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2011 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <errno.h>
21 #include <stdint.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <unistd.h>
25
26 #include "alloc-util.h"
27 #include "dirent-util.h"
28 #include "env-util.h"
29 #include "fd-util.h"
30 #include "fileio.h"
31 #include "fs-util.h"
32 #include "macro.h"
33 #include "process-util.h"
34 #include "stat-util.h"
35 #include "string-table.h"
36 #include "string-util.h"
37 #include "virt.h"
38
39 static int detect_vm_cpuid(void) {
40
41 /* CPUID is an x86 specific interface. */
42 #if defined(__i386__) || defined(__x86_64__)
43
44 static const struct {
45 const char *cpuid;
46 int id;
47 } cpuid_vendor_table[] = {
48 { "XenVMMXenVMM", VIRTUALIZATION_XEN },
49 { "KVMKVMKVM", VIRTUALIZATION_KVM },
50 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
51 { "VMwareVMware", VIRTUALIZATION_VMWARE },
52 /* http://msdn.microsoft.com/en-us/library/ff542428.aspx */
53 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
54 /* https://wiki.freebsd.org/bhyve */
55 { "bhyve bhyve ", VIRTUALIZATION_BHYVE },
56 };
57
58 uint32_t eax, ecx;
59 bool hypervisor;
60
61 /* http://lwn.net/Articles/301888/ */
62
63 #if defined (__i386__)
64 #define REG_a "eax"
65 #define REG_b "ebx"
66 #elif defined (__amd64__)
67 #define REG_a "rax"
68 #define REG_b "rbx"
69 #endif
70
71 /* First detect whether there is a hypervisor */
72 eax = 1;
73 __asm__ __volatile__ (
74 /* ebx/rbx is being used for PIC! */
75 " push %%"REG_b" \n\t"
76 " cpuid \n\t"
77 " pop %%"REG_b" \n\t"
78
79 : "=a" (eax), "=c" (ecx)
80 : "0" (eax)
81 );
82
83 hypervisor = !!(ecx & 0x80000000U);
84
85 if (hypervisor) {
86 union {
87 uint32_t sig32[3];
88 char text[13];
89 } sig = {};
90 unsigned j;
91
92 /* There is a hypervisor, see what it is */
93 eax = 0x40000000U;
94 __asm__ __volatile__ (
95 /* ebx/rbx is being used for PIC! */
96 " push %%"REG_b" \n\t"
97 " cpuid \n\t"
98 " mov %%ebx, %1 \n\t"
99 " pop %%"REG_b" \n\t"
100
101 : "=a" (eax), "=r" (sig.sig32[0]), "=c" (sig.sig32[1]), "=d" (sig.sig32[2])
102 : "0" (eax)
103 );
104
105 log_debug("Virtualization found, CPUID=%s", sig.text);
106
107 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
108 if (streq(sig.text, cpuid_vendor_table[j].cpuid))
109 return cpuid_vendor_table[j].id;
110
111 return VIRTUALIZATION_VM_OTHER;
112 }
113 #endif
114 log_debug("No virtualization found in CPUID");
115
116 return VIRTUALIZATION_NONE;
117 }
118
119 static int detect_vm_device_tree(void) {
120 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
121 _cleanup_free_ char *hvtype = NULL;
122 int r;
123
124 r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
125 if (r == -ENOENT) {
126 _cleanup_closedir_ DIR *dir = NULL;
127 struct dirent *dent;
128
129 dir = opendir("/proc/device-tree");
130 if (!dir) {
131 if (errno == ENOENT) {
132 log_debug_errno(errno, "/proc/device-tree: %m");
133 return VIRTUALIZATION_NONE;
134 }
135 return -errno;
136 }
137
138 FOREACH_DIRENT(dent, dir, return -errno)
139 if (strstr(dent->d_name, "fw-cfg")) {
140 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
141 return VIRTUALIZATION_QEMU;
142 }
143
144 log_debug("No virtualization found in /proc/device-tree/*");
145 return VIRTUALIZATION_NONE;
146 } else if (r < 0)
147 return r;
148
149 log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
150 if (streq(hvtype, "linux,kvm"))
151 return VIRTUALIZATION_KVM;
152 else if (strstr(hvtype, "xen"))
153 return VIRTUALIZATION_XEN;
154 else
155 return VIRTUALIZATION_VM_OTHER;
156 #else
157 log_debug("This platform does not support /proc/device-tree");
158 return VIRTUALIZATION_NONE;
159 #endif
160 }
161
162 static int detect_vm_dmi(void) {
163 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
164
165 static const char *const dmi_vendors[] = {
166 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
167 "/sys/class/dmi/id/sys_vendor",
168 "/sys/class/dmi/id/board_vendor",
169 "/sys/class/dmi/id/bios_vendor"
170 };
171
172 static const struct {
173 const char *vendor;
174 int id;
175 } dmi_vendor_table[] = {
176 { "KVM", VIRTUALIZATION_KVM },
177 { "QEMU", VIRTUALIZATION_QEMU },
178 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
179 { "VMware", VIRTUALIZATION_VMWARE },
180 { "VMW", VIRTUALIZATION_VMWARE },
181 { "innotek GmbH", VIRTUALIZATION_ORACLE },
182 { "Xen", VIRTUALIZATION_XEN },
183 { "Bochs", VIRTUALIZATION_BOCHS },
184 { "Parallels", VIRTUALIZATION_PARALLELS },
185 /* https://wiki.freebsd.org/bhyve */
186 { "BHYVE", VIRTUALIZATION_BHYVE },
187 };
188 unsigned i;
189 int r;
190
191 for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
192 _cleanup_free_ char *s = NULL;
193 unsigned j;
194
195 r = read_one_line_file(dmi_vendors[i], &s);
196 if (r < 0) {
197 if (r == -ENOENT)
198 continue;
199
200 return r;
201 }
202
203
204
205 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
206 if (startswith(s, dmi_vendor_table[j].vendor)) {
207 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
208 return dmi_vendor_table[j].id;
209 }
210 }
211 #endif
212
213 log_debug("No virtualization found in DMI");
214
215 return VIRTUALIZATION_NONE;
216 }
217
218 static int detect_vm_xen(void) {
219 /* Check for Dom0 will be executed later in detect_vm_xen_dom0
220 Thats why we dont check the content of /proc/xen/capabilities here. */
221 if (access("/proc/xen/capabilities", F_OK) < 0) {
222 log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
223 return VIRTUALIZATION_NONE;
224 }
225
226 log_debug("Virtualization XEN found (/proc/xen/capabilities exists)");
227 return VIRTUALIZATION_XEN;
228
229 }
230
231 static bool detect_vm_xen_dom0(void) {
232 _cleanup_free_ char *domcap = NULL;
233 char *cap, *i;
234 int r;
235
236 r = read_one_line_file("/proc/xen/capabilities", &domcap);
237 if (r == -ENOENT) {
238 log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
239 return false;
240 }
241 if (r < 0)
242 return r;
243
244 i = domcap;
245 while ((cap = strsep(&i, ",")))
246 if (streq(cap, "control_d"))
247 break;
248 if (!cap) {
249 log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
250 return false;
251 }
252
253 log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
254 return true;
255 }
256
257 static int detect_vm_hypervisor(void) {
258 _cleanup_free_ char *hvtype = NULL;
259 int r;
260
261 r = read_one_line_file("/sys/hypervisor/type", &hvtype);
262 if (r == -ENOENT)
263 return VIRTUALIZATION_NONE;
264 if (r < 0)
265 return r;
266
267 log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
268
269 if (streq(hvtype, "xen"))
270 return VIRTUALIZATION_XEN;
271 else
272 return VIRTUALIZATION_VM_OTHER;
273 }
274
275 static int detect_vm_uml(void) {
276 _cleanup_free_ char *cpuinfo_contents = NULL;
277 int r;
278
279 /* Detect User-Mode Linux by reading /proc/cpuinfo */
280 r = read_full_file("/proc/cpuinfo", &cpuinfo_contents, NULL);
281 if (r < 0)
282 return r;
283
284 if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
285 log_debug("UML virtualization found in /proc/cpuinfo");
286 return VIRTUALIZATION_UML;
287 }
288
289 log_debug("No virtualization found in /proc/cpuinfo.");
290 return VIRTUALIZATION_NONE;
291 }
292
293 static int detect_vm_zvm(void) {
294
295 #if defined(__s390__)
296 _cleanup_free_ char *t = NULL;
297 int r;
298
299 r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
300 if (r == -ENOENT)
301 return VIRTUALIZATION_NONE;
302 if (r < 0)
303 return r;
304
305 log_debug("Virtualization %s found in /proc/sysinfo", t);
306 if (streq(t, "z/VM"))
307 return VIRTUALIZATION_ZVM;
308 else
309 return VIRTUALIZATION_KVM;
310 #else
311 log_debug("This platform does not support /proc/sysinfo");
312 return VIRTUALIZATION_NONE;
313 #endif
314 }
315
316 /* Returns a short identifier for the various VM implementations */
317 int detect_vm(void) {
318 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
319 int r;
320
321 if (cached_found >= 0)
322 return cached_found;
323
324 /* We have to use the correct order here:
325 * Some virtualization technologies do use KVM hypervisor but are
326 * expected to be detected as something else. So detect DMI first.
327 *
328 * An example is Virtualbox since version 5.0, which uses KVM backend.
329 * Detection via DMI works corretly, the CPU ID would find KVM
330 * only. */
331 r = detect_vm_dmi();
332 if (r < 0)
333 return r;
334 if (r != VIRTUALIZATION_NONE)
335 goto finish;
336
337 r = detect_vm_cpuid();
338 if (r < 0)
339 return r;
340 if (r != VIRTUALIZATION_NONE)
341 goto finish;
342
343 /* x86 xen will most likely be detected by cpuid. If not (most likely
344 * because we're not an x86 guest), then we should try the xen capabilities
345 * file next. If that's not found, then we check for the high-level
346 * hypervisor sysfs file:
347 *
348 * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
349
350 r = detect_vm_xen();
351 if (r < 0)
352 return r;
353 if (r != VIRTUALIZATION_NONE)
354 goto finish;
355
356 r = detect_vm_hypervisor();
357 if (r < 0)
358 return r;
359 if (r != VIRTUALIZATION_NONE)
360 goto finish;
361
362 r = detect_vm_device_tree();
363 if (r < 0)
364 return r;
365 if (r != VIRTUALIZATION_NONE)
366 goto finish;
367
368 r = detect_vm_uml();
369 if (r < 0)
370 return r;
371 if (r != VIRTUALIZATION_NONE)
372 goto finish;
373
374 r = detect_vm_zvm();
375 if (r < 0)
376 return r;
377
378 finish:
379 /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
380 * In order to detect the Dom0 as not virtualization we need to
381 * double-check it */
382 if (r == VIRTUALIZATION_XEN && detect_vm_xen_dom0())
383 r = VIRTUALIZATION_NONE;
384
385 cached_found = r;
386 log_debug("Found VM virtualization %s", virtualization_to_string(r));
387 return r;
388 }
389
390 int detect_container(void) {
391
392 static const struct {
393 const char *value;
394 int id;
395 } value_table[] = {
396 { "lxc", VIRTUALIZATION_LXC },
397 { "lxc-libvirt", VIRTUALIZATION_LXC_LIBVIRT },
398 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
399 { "docker", VIRTUALIZATION_DOCKER },
400 { "rkt", VIRTUALIZATION_RKT },
401 };
402
403 static thread_local int cached_found = _VIRTUALIZATION_INVALID;
404 _cleanup_free_ char *m = NULL;
405 const char *e = NULL;
406 unsigned j;
407 int r;
408
409 if (cached_found >= 0)
410 return cached_found;
411
412 /* /proc/vz exists in container and outside of the container,
413 * /proc/bc only outside of the container. */
414 if (access("/proc/vz", F_OK) >= 0 &&
415 access("/proc/bc", F_OK) < 0) {
416 r = VIRTUALIZATION_OPENVZ;
417 goto finish;
418 }
419
420 if (getpid() == 1) {
421 /* If we are PID 1 we can just check our own
422 * environment variable */
423
424 e = getenv("container");
425 if (isempty(e)) {
426 r = VIRTUALIZATION_NONE;
427 goto finish;
428 }
429 } else {
430
431 /* Otherwise, PID 1 dropped this information into a
432 * file in /run. This is better than accessing
433 * /proc/1/environ, since we don't need CAP_SYS_PTRACE
434 * for that. */
435
436 r = read_one_line_file("/run/systemd/container", &m);
437 if (r == -ENOENT) {
438
439 /* Fallback for cases where PID 1 was not
440 * systemd (for example, cases where
441 * init=/bin/sh is used. */
442
443 r = getenv_for_pid(1, "container", &m);
444 if (r <= 0) {
445
446 /* If that didn't work, give up,
447 * assume no container manager.
448 *
449 * Note: This means we still cannot
450 * detect containers if init=/bin/sh
451 * is passed but privileges dropped,
452 * as /proc/1/environ is only readable
453 * with privileges. */
454
455 r = VIRTUALIZATION_NONE;
456 goto finish;
457 }
458 }
459 if (r < 0)
460 return r;
461
462 e = m;
463 }
464
465 for (j = 0; j < ELEMENTSOF(value_table); j++)
466 if (streq(e, value_table[j].value)) {
467 r = value_table[j].id;
468 goto finish;
469 }
470
471 r = VIRTUALIZATION_CONTAINER_OTHER;
472
473 finish:
474 log_debug("Found container virtualization %s", virtualization_to_string(r));
475 cached_found = r;
476 return r;
477 }
478
479 int detect_virtualization(void) {
480 int r;
481
482 r = detect_container();
483 if (r == 0)
484 r = detect_vm();
485
486 return r;
487 }
488
489 static int userns_has_mapping(const char *name) {
490 _cleanup_fclose_ FILE *f = NULL;
491 _cleanup_free_ char *buf = NULL;
492 size_t n_allocated = 0;
493 ssize_t n;
494 uint32_t a, b, c;
495 int r;
496
497 f = fopen(name, "re");
498 if (!f) {
499 log_debug_errno(errno, "Failed to open %s: %m", name);
500 return errno == ENOENT ? false : -errno;
501 }
502
503 n = getline(&buf, &n_allocated, f);
504 if (n < 0) {
505 if (feof(f)) {
506 log_debug("%s is empty, we're in an uninitialized user namespace", name);
507 return true;
508 }
509
510 return log_debug_errno(errno, "Failed to read %s: %m", name);
511 }
512
513 r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
514 if (r < 3)
515 return log_debug_errno(errno, "Failed to parse %s: %m", name);
516
517 if (a == 0 && b == 0 && c == UINT32_MAX) {
518 /* The kernel calls mappings_overlap() and does not allow overlaps */
519 log_debug("%s has a full 1:1 mapping", name);
520 return false;
521 }
522
523 /* Anything else implies that we are in a user namespace */
524 log_debug("Mapping found in %s, we're in a user namespace", name);
525 return true;
526 }
527
528 int running_in_userns(void) {
529 _cleanup_free_ char *line = NULL;
530 int r;
531
532 r = userns_has_mapping("/proc/self/uid_map");
533 if (r != 0)
534 return r;
535
536 r = userns_has_mapping("/proc/self/gid_map");
537 if (r != 0)
538 return r;
539
540 /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
541 * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
542 * also does not exist. We cannot distinguish those two cases, so assume that
543 * we're running on a stripped-down recent kernel, rather than on an old one,
544 * and if the file is not found, return false.
545 */
546 r = read_one_line_file("/proc/self/setgroups", &line);
547 if (r < 0) {
548 log_debug_errno(r, "/proc/self/setgroups: %m");
549 return r == -ENOENT ? false : r;
550 }
551
552 truncate_nl(line);
553 r = streq(line, "deny");
554 /* See user_namespaces(7) for a description of this "setgroups" contents. */
555 log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
556 return r;
557 }
558
559 int running_in_chroot(void) {
560 _cleanup_free_ char *self_mnt = NULL, *pid1_mnt = NULL;
561 int r;
562
563 /* Try to detect whether we are running in a chroot() environment. Specifically, check whether we have a
564 * different root directory than PID 1, even though we live in the same mount namespace as it. */
565
566 if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
567 return 0;
568
569 r = files_same("/proc/1/root", "/");
570 if (r < 0)
571 return r;
572 if (r > 0)
573 return 0;
574
575 r = readlink_malloc("/proc/self/ns/mnt", &self_mnt);
576 if (r < 0)
577 return r;
578
579 r = readlink_malloc("/proc/1/ns/mnt", &pid1_mnt);
580 if (r < 0)
581 return r;
582
583 return streq(self_mnt, pid1_mnt); /* Only if we live in the same namespace! */
584 }
585
586 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
587 [VIRTUALIZATION_NONE] = "none",
588 [VIRTUALIZATION_KVM] = "kvm",
589 [VIRTUALIZATION_QEMU] = "qemu",
590 [VIRTUALIZATION_BOCHS] = "bochs",
591 [VIRTUALIZATION_XEN] = "xen",
592 [VIRTUALIZATION_UML] = "uml",
593 [VIRTUALIZATION_VMWARE] = "vmware",
594 [VIRTUALIZATION_ORACLE] = "oracle",
595 [VIRTUALIZATION_MICROSOFT] = "microsoft",
596 [VIRTUALIZATION_ZVM] = "zvm",
597 [VIRTUALIZATION_PARALLELS] = "parallels",
598 [VIRTUALIZATION_BHYVE] = "bhyve",
599 [VIRTUALIZATION_VM_OTHER] = "vm-other",
600
601 [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
602 [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
603 [VIRTUALIZATION_LXC] = "lxc",
604 [VIRTUALIZATION_OPENVZ] = "openvz",
605 [VIRTUALIZATION_DOCKER] = "docker",
606 [VIRTUALIZATION_RKT] = "rkt",
607 [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
608 };
609
610 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);