chiark / gitweb /
virt: remove triple spurious newline
[elogind.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 "macro.h"
32 #include "process-util.h"
33 #include "stat-util.h"
34 #include "string-table.h"
35 #include "string-util.h"
36 #include "virt.h"
37
38 #if 0 /// UNNEEDED by elogind
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                 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU      },
51                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
52                 { "VMwareVMware", VIRTUALIZATION_VMWARE    },
53                 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
54                 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
55                 /* https://wiki.freebsd.org/bhyve */
56                 { "bhyve bhyve ", VIRTUALIZATION_BHYVE     },
57         };
58
59         uint32_t eax, ecx;
60         bool hypervisor;
61
62         /* http://lwn.net/Articles/301888/ */
63
64 #if defined (__i386__)
65 #define REG_a "eax"
66 #define REG_b "ebx"
67 #elif defined (__amd64__)
68 #define REG_a "rax"
69 #define REG_b "rbx"
70 #endif
71
72         /* First detect whether there is a hypervisor */
73         eax = 1;
74         __asm__ __volatile__ (
75                 /* ebx/rbx is being used for PIC! */
76                 "  push %%"REG_b"         \n\t"
77                 "  cpuid                  \n\t"
78                 "  pop %%"REG_b"          \n\t"
79
80                 : "=a" (eax), "=c" (ecx)
81                 : "0" (eax)
82         );
83
84         hypervisor = !!(ecx & 0x80000000U);
85
86         if (hypervisor) {
87                 union {
88                         uint32_t sig32[3];
89                         char text[13];
90                 } sig = {};
91                 unsigned j;
92
93                 /* There is a hypervisor, see what it is */
94                 eax = 0x40000000U;
95                 __asm__ __volatile__ (
96                         /* ebx/rbx is being used for PIC! */
97                         "  push %%"REG_b"         \n\t"
98                         "  cpuid                  \n\t"
99                         "  mov %%ebx, %1          \n\t"
100                         "  pop %%"REG_b"          \n\t"
101
102                         : "=a" (eax), "=r" (sig.sig32[0]), "=c" (sig.sig32[1]), "=d" (sig.sig32[2])
103                         : "0" (eax)
104                 );
105
106                 log_debug("Virtualization found, CPUID=%s", sig.text);
107
108                 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
109                         if (streq(sig.text, cpuid_vendor_table[j].cpuid))
110                                 return cpuid_vendor_table[j].id;
111
112                 return VIRTUALIZATION_VM_OTHER;
113         }
114 #endif
115         log_debug("No virtualization found in CPUID");
116
117         return VIRTUALIZATION_NONE;
118 }
119
120 static int detect_vm_device_tree(void) {
121 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
122         _cleanup_free_ char *hvtype = NULL;
123         int r;
124
125         r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
126         if (r == -ENOENT) {
127                 _cleanup_closedir_ DIR *dir = NULL;
128                 struct dirent *dent;
129
130                 dir = opendir("/proc/device-tree");
131                 if (!dir) {
132                         if (errno == ENOENT) {
133                                 log_debug_errno(errno, "/proc/device-tree: %m");
134                                 return VIRTUALIZATION_NONE;
135                         }
136                         return -errno;
137                 }
138
139                 FOREACH_DIRENT(dent, dir, return -errno)
140                         if (strstr(dent->d_name, "fw-cfg")) {
141                                 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
142                                 return VIRTUALIZATION_QEMU;
143                         }
144
145                 log_debug("No virtualization found in /proc/device-tree/*");
146                 return VIRTUALIZATION_NONE;
147         } else if (r < 0)
148                 return r;
149
150         log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
151         if (streq(hvtype, "linux,kvm"))
152                 return VIRTUALIZATION_KVM;
153         else if (strstr(hvtype, "xen"))
154                 return VIRTUALIZATION_XEN;
155         else
156                 return VIRTUALIZATION_VM_OTHER;
157 #else
158         log_debug("This platform does not support /proc/device-tree");
159         return VIRTUALIZATION_NONE;
160 #endif
161 }
162
163 static int detect_vm_dmi(void) {
164 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
165
166         static const char *const dmi_vendors[] = {
167                 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
168                 "/sys/class/dmi/id/sys_vendor",
169                 "/sys/class/dmi/id/board_vendor",
170                 "/sys/class/dmi/id/bios_vendor"
171         };
172
173         static const struct {
174                 const char *vendor;
175                 int id;
176         } dmi_vendor_table[] = {
177                 { "KVM",           VIRTUALIZATION_KVM       },
178                 { "QEMU",          VIRTUALIZATION_QEMU      },
179                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
180                 { "VMware",        VIRTUALIZATION_VMWARE    },
181                 { "VMW",           VIRTUALIZATION_VMWARE    },
182                 { "innotek GmbH",  VIRTUALIZATION_ORACLE    },
183                 { "Xen",           VIRTUALIZATION_XEN       },
184                 { "Bochs",         VIRTUALIZATION_BOCHS     },
185                 { "Parallels",     VIRTUALIZATION_PARALLELS },
186                 /* https://wiki.freebsd.org/bhyve */
187                 { "BHYVE",         VIRTUALIZATION_BHYVE     },
188         };
189         unsigned i;
190         int r;
191
192         for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
193                 _cleanup_free_ char *s = NULL;
194                 unsigned j;
195
196                 r = read_one_line_file(dmi_vendors[i], &s);
197                 if (r < 0) {
198                         if (r == -ENOENT)
199                                 continue;
200
201                         return r;
202                 }
203
204                 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
205                         if (startswith(s, dmi_vendor_table[j].vendor)) {
206                                 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
207                                 return dmi_vendor_table[j].id;
208                         }
209         }
210 #endif
211
212         log_debug("No virtualization found in DMI");
213
214         return VIRTUALIZATION_NONE;
215 }
216
217 static int detect_vm_xen(void) {
218
219         /* Check for Dom0 will be executed later in detect_vm_xen_dom0
220            The presence of /proc/xen indicates some form of a Xen domain */
221         if (access("/proc/xen", F_OK) < 0) {
222                 log_debug("Virtualization XEN not found, /proc/xen does not exist");
223                 return VIRTUALIZATION_NONE;
224         }
225
226         log_debug("Virtualization XEN found (/proc/xen exists)");
227         return VIRTUALIZATION_XEN;
228 }
229
230 static bool detect_vm_xen_dom0(void) {
231         _cleanup_free_ char *domcap = NULL;
232         char *cap, *i;
233         int r;
234
235         r = read_one_line_file("/proc/xen/capabilities", &domcap);
236         if (r == -ENOENT) {
237                 log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
238                 return false;
239         }
240         if (r < 0)
241                 return r;
242
243         i = domcap;
244         while ((cap = strsep(&i, ",")))
245                 if (streq(cap, "control_d"))
246                         break;
247         if (!cap) {
248                 log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
249                 return false;
250         }
251
252         log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
253         return true;
254 }
255
256 static int detect_vm_hypervisor(void) {
257         _cleanup_free_ char *hvtype = NULL;
258         int r;
259
260         r = read_one_line_file("/sys/hypervisor/type", &hvtype);
261         if (r == -ENOENT)
262                 return VIRTUALIZATION_NONE;
263         if (r < 0)
264                 return r;
265
266         log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
267
268         if (streq(hvtype, "xen"))
269                 return VIRTUALIZATION_XEN;
270         else
271                 return VIRTUALIZATION_VM_OTHER;
272 }
273
274 static int detect_vm_uml(void) {
275         _cleanup_free_ char *cpuinfo_contents = NULL;
276         int r;
277
278         /* Detect User-Mode Linux by reading /proc/cpuinfo */
279         r = read_full_file("/proc/cpuinfo", &cpuinfo_contents, NULL);
280         if (r < 0)
281                 return r;
282
283         if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
284                 log_debug("UML virtualization found in /proc/cpuinfo");
285                 return VIRTUALIZATION_UML;
286         }
287
288         log_debug("No virtualization found in /proc/cpuinfo.");
289         return VIRTUALIZATION_NONE;
290 }
291
292 static int detect_vm_zvm(void) {
293
294 #if defined(__s390__)
295         _cleanup_free_ char *t = NULL;
296         int r;
297
298         r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
299         if (r == -ENOENT)
300                 return VIRTUALIZATION_NONE;
301         if (r < 0)
302                 return r;
303
304         log_debug("Virtualization %s found in /proc/sysinfo", t);
305         if (streq(t, "z/VM"))
306                 return VIRTUALIZATION_ZVM;
307         else
308                 return VIRTUALIZATION_KVM;
309 #else
310         log_debug("This platform does not support /proc/sysinfo");
311         return VIRTUALIZATION_NONE;
312 #endif
313 }
314
315 /* Returns a short identifier for the various VM implementations */
316 int detect_vm(void) {
317         static thread_local int cached_found = _VIRTUALIZATION_INVALID;
318         int r, dmi;
319         bool other = false;
320
321         if (cached_found >= 0)
322                 return cached_found;
323
324         /* We have to use the correct order here:
325          *
326          * -> First try to detect Oracle Virtualbox, even if it uses KVM.
327          * -> Second try to detect from cpuid, this will report KVM for
328          *    whatever software is used even if info in dmi is overwritten.
329          * -> Third try to detect from dmi. */
330
331         dmi = detect_vm_dmi();
332         if (dmi == VIRTUALIZATION_ORACLE) {
333                 r = dmi;
334                 goto finish;
335         }
336
337         r = detect_vm_cpuid();
338         if (r < 0)
339                 return r;
340         if (r != VIRTUALIZATION_NONE) {
341                 if (r == VIRTUALIZATION_VM_OTHER)
342                         other = true;
343                 else
344                         goto finish;
345         }
346
347         r = dmi;
348         if (r < 0)
349                 return r;
350         if (r != VIRTUALIZATION_NONE) {
351                 if (r == VIRTUALIZATION_VM_OTHER)
352                         other = true;
353                 else
354                         goto finish;
355         }
356
357         /* x86 xen will most likely be detected by cpuid. If not (most likely
358          * because we're not an x86 guest), then we should try the /proc/xen
359          * directory next. If that's not found, then we check for the high-level
360          * hypervisor sysfs file.
361          */
362
363         r = detect_vm_xen();
364         if (r < 0)
365                 return r;
366         if (r != VIRTUALIZATION_NONE) {
367                 if (r == VIRTUALIZATION_VM_OTHER)
368                         other = true;
369                 else
370                         goto finish;
371         }
372
373         r = detect_vm_hypervisor();
374         if (r < 0)
375                 return r;
376         if (r != VIRTUALIZATION_NONE) {
377                 if (r == VIRTUALIZATION_VM_OTHER)
378                         other = true;
379                 else
380                         goto finish;
381         }
382
383         r = detect_vm_device_tree();
384         if (r < 0)
385                 return r;
386         if (r != VIRTUALIZATION_NONE) {
387                 if (r == VIRTUALIZATION_VM_OTHER)
388                         other = true;
389                 else
390                         goto finish;
391         }
392
393         r = detect_vm_uml();
394         if (r < 0)
395                 return r;
396         if (r != VIRTUALIZATION_NONE) {
397                 if (r == VIRTUALIZATION_VM_OTHER)
398                         other = true;
399                 else
400                         goto finish;
401         }
402
403         r = detect_vm_zvm();
404         if (r < 0)
405                 return r;
406
407 finish:
408         /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
409          * In order to detect the Dom0 as not virtualization we need to
410          * double-check it */
411         if (r == VIRTUALIZATION_XEN && detect_vm_xen_dom0())
412                 r = VIRTUALIZATION_NONE;
413         else if (r == VIRTUALIZATION_NONE && other)
414                 r = VIRTUALIZATION_VM_OTHER;
415
416         cached_found = r;
417         log_debug("Found VM virtualization %s", virtualization_to_string(r));
418         return r;
419 }
420 #endif // 0
421
422 int detect_container(void) {
423
424         static const struct {
425                 const char *value;
426                 int id;
427         } value_table[] = {
428                 { "lxc",            VIRTUALIZATION_LXC            },
429                 { "lxc-libvirt",    VIRTUALIZATION_LXC_LIBVIRT    },
430                 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
431                 { "docker",         VIRTUALIZATION_DOCKER         },
432                 { "rkt",            VIRTUALIZATION_RKT            },
433         };
434
435         static thread_local int cached_found = _VIRTUALIZATION_INVALID;
436         _cleanup_free_ char *m = NULL;
437         const char *e = NULL;
438         unsigned j;
439         int r;
440
441         if (cached_found >= 0)
442                 return cached_found;
443
444         /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
445         if (access("/proc/vz", F_OK) >= 0 &&
446             access("/proc/bc", F_OK) < 0) {
447                 r = VIRTUALIZATION_OPENVZ;
448                 goto finish;
449         }
450
451         if (getpid_cached() == 1) {
452                 /* If we are PID 1 we can just check our own environment variable, and that's authoritative. */
453
454                 e = getenv("container");
455                 if (isempty(e)) {
456                         r = VIRTUALIZATION_NONE;
457                         goto finish;
458                 }
459
460                 goto translate_name;
461         }
462
463         /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
464          * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
465         r = read_one_line_file("/run/systemd/container", &m);
466         if (r >= 0) {
467                 e = m;
468                 goto translate_name;
469         }
470         if (r != -ENOENT)
471                 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
472
473         /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
474         r = getenv_for_pid(1, "container", &m);
475         if (r > 0) {
476                 e = m;
477                 goto translate_name;
478         }
479         if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
480                 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
481
482         /* Interestingly /proc/1/sched actually shows the host's PID for what we see as PID 1. Hence, if the PID shown
483          * there is not 1, we know we are in a PID namespace. and hence a container. */
484         r = read_one_line_file("/proc/1/sched", &m);
485         if (r >= 0) {
486                 const char *t;
487
488                 t = strrchr(m, '(');
489                 if (!t)
490                         return -EIO;
491
492                 if (!startswith(t, "(1,")) {
493                         r = VIRTUALIZATION_CONTAINER_OTHER;
494                         goto finish;
495                 }
496         } else if (r != -ENOENT)
497                 return r;
498
499         /* If that didn't work, give up, assume no container manager. */
500         r = VIRTUALIZATION_NONE;
501         goto finish;
502
503 translate_name:
504         for (j = 0; j < ELEMENTSOF(value_table); j++)
505                 if (streq(e, value_table[j].value)) {
506                         r = value_table[j].id;
507                         goto finish;
508                 }
509
510         r = VIRTUALIZATION_CONTAINER_OTHER;
511
512 finish:
513         log_debug("Found container virtualization %s.", virtualization_to_string(r));
514         cached_found = r;
515         return r;
516 }
517
518 #if 0 /// UNNEEDED by elogind
519 int detect_virtualization(void) {
520         int r;
521
522         r = detect_container();
523         if (r == 0)
524                 r = detect_vm();
525
526         return r;
527 }
528
529 static int userns_has_mapping(const char *name) {
530         _cleanup_fclose_ FILE *f = NULL;
531         _cleanup_free_ char *buf = NULL;
532         size_t n_allocated = 0;
533         ssize_t n;
534         uint32_t a, b, c;
535         int r;
536
537         f = fopen(name, "re");
538         if (!f) {
539                 log_debug_errno(errno, "Failed to open %s: %m", name);
540                 return errno == ENOENT ? false : -errno;
541         }
542
543         n = getline(&buf, &n_allocated, f);
544         if (n < 0) {
545                 if (feof(f)) {
546                         log_debug("%s is empty, we're in an uninitialized user namespace", name);
547                         return true;
548                 }
549
550                 return log_debug_errno(errno, "Failed to read %s: %m", name);
551         }
552
553         r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
554         if (r < 3)
555                 return log_debug_errno(errno, "Failed to parse %s: %m", name);
556
557         if (a == 0 && b == 0 && c == UINT32_MAX) {
558                 /* The kernel calls mappings_overlap() and does not allow overlaps */
559                 log_debug("%s has a full 1:1 mapping", name);
560                 return false;
561         }
562
563         /* Anything else implies that we are in a user namespace */
564         log_debug("Mapping found in %s, we're in a user namespace", name);
565         return true;
566 }
567
568 int running_in_userns(void) {
569         _cleanup_free_ char *line = NULL;
570         int r;
571
572         r = userns_has_mapping("/proc/self/uid_map");
573         if (r != 0)
574                 return r;
575
576         r = userns_has_mapping("/proc/self/gid_map");
577         if (r != 0)
578                 return r;
579
580         /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
581          * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
582          * also does not exist. We cannot distinguish those two cases, so assume that
583          * we're running on a stripped-down recent kernel, rather than on an old one,
584          * and if the file is not found, return false.
585          */
586         r = read_one_line_file("/proc/self/setgroups", &line);
587         if (r < 0) {
588                 log_debug_errno(r, "/proc/self/setgroups: %m");
589                 return r == -ENOENT ? false : r;
590         }
591
592         truncate_nl(line);
593         r = streq(line, "deny");
594         /* See user_namespaces(7) for a description of this "setgroups" contents. */
595         log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
596         return r;
597 }
598 #endif // 0
599
600 int running_in_chroot(void) {
601         int ret;
602
603 #if 0 /// elogind does not allow to ignore chroots, we are never init!
604         if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
605                 return 0;
606 #endif // 0
607
608         ret = files_same("/proc/1/root", "/", 0);
609         if (ret < 0)
610                 return ret;
611
612         return ret == 0;
613 }
614
615 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
616         [VIRTUALIZATION_NONE] = "none",
617         [VIRTUALIZATION_KVM] = "kvm",
618         [VIRTUALIZATION_QEMU] = "qemu",
619         [VIRTUALIZATION_BOCHS] = "bochs",
620         [VIRTUALIZATION_XEN] = "xen",
621         [VIRTUALIZATION_UML] = "uml",
622         [VIRTUALIZATION_VMWARE] = "vmware",
623         [VIRTUALIZATION_ORACLE] = "oracle",
624         [VIRTUALIZATION_MICROSOFT] = "microsoft",
625         [VIRTUALIZATION_ZVM] = "zvm",
626         [VIRTUALIZATION_PARALLELS] = "parallels",
627         [VIRTUALIZATION_BHYVE] = "bhyve",
628         [VIRTUALIZATION_VM_OTHER] = "vm-other",
629
630         [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
631         [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
632         [VIRTUALIZATION_LXC] = "lxc",
633         [VIRTUALIZATION_OPENVZ] = "openvz",
634         [VIRTUALIZATION_DOCKER] = "docker",
635         [VIRTUALIZATION_RKT] = "rkt",
636         [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
637 };
638
639 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);