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