src/basic/virt.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   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 <unistd.h>
  10
  11 #include "alloc-util.h"
  12 #include "cgroup-util.h"
  13 #include "dirent-util.h"
  14 #include "env-util.h"
  15 #include "errno-util.h"
  16 #include "fd-util.h"
  17 #include "fileio.h"
  18 #include "macro.h"
  19 #include "missing_threads.h"
  20 #include "process-util.h"
  21 #include "stat-util.h"
  22 #include "string-table.h"
  23 #include "string-util.h"
  24 #include "virt.h"
  25
  26 enum {
  27       SMBIOS_VM_BIT_SET,
  28       SMBIOS_VM_BIT_UNSET,
  29       SMBIOS_VM_BIT_UNKNOWN,
  30 };
  31
  32 static Virtualization detect_vm_cpuid(void) {
  33
  34         /* CPUID is an x86 specific interface. */
  35 #if defined(__i386__) || defined(__x86_64__)
  36
  37         static const struct {
  38                 const char sig[13];
  39                 Virtualization id;
  40         } vm_table[] = {
  41                 { "XenVMMXenVMM", VIRTUALIZATION_XEN       },
  42                 { "KVMKVMKVM",    VIRTUALIZATION_KVM       }, /* qemu with KVM */
  43                 { "Linux KVM Hv", VIRTUALIZATION_KVM       }, /* qemu with KVM + HyperV Enlightenments */
  44                 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU      }, /* qemu without KVM */
  45                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
  46                 { "VMwareVMware", VIRTUALIZATION_VMWARE    },
  47                 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
  48                 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
  49                 /* https://wiki.freebsd.org/bhyve */
  50                 { "bhyve bhyve ", VIRTUALIZATION_BHYVE     },
  51                 { "QNXQVMBSQG",   VIRTUALIZATION_QNX       },
  52                 /* https://projectacrn.org */
  53                 { "ACRNACRNACRN", VIRTUALIZATION_ACRN      },
  54                 /* https://www.lockheedmartin.com/en-us/products/Hardened-Security-for-Intel-Processors.html */
  55                 { "SRESRESRESRE", VIRTUALIZATION_SRE       },
  56         };
  57
  58         uint32_t eax, ebx, ecx, edx;
  59         bool hypervisor;
  60
  61         /* http://lwn.net/Articles/301888/ */
  62
  63         /* First detect whether there is a hypervisor */
  64         if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) == 0)
  65                 return VIRTUALIZATION_NONE;
  66
  67         hypervisor = ecx & 0x80000000U;
  68
  69         if (hypervisor) {
  70                 union {
  71                         uint32_t sig32[3];
  72                         char text[13];
  73                 } sig = {};
  74
  75                 /* There is a hypervisor, see what it is */
  76                 __cpuid(0x40000000U, eax, ebx, ecx, edx);
  77
  78                 sig.sig32[0] = ebx;
  79                 sig.sig32[1] = ecx;
  80                 sig.sig32[2] = edx;
  81
  82                 log_debug("Virtualization found, CPUID=%s", sig.text);
  83
  84                 for (size_t i = 0; i < ELEMENTSOF(vm_table); i++)
  85                         if (memcmp_nn(sig.text, sizeof(sig.text),
  86                                       vm_table[i].sig, sizeof(vm_table[i].sig)) == 0)
  87                                 return vm_table[i].id;
  88
  89                 log_debug("Unknown virtualization with CPUID=%s. Add to vm_table[]?", sig.text);
  90                 return VIRTUALIZATION_VM_OTHER;
  91         }
  92 #endif
  93         log_debug("No virtualization found in CPUID");
  94
  95         return VIRTUALIZATION_NONE;
  96 }
  97
  98 static Virtualization detect_vm_device_tree(void) {
  99 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
 100         _cleanup_free_ char *hvtype = NULL;
 101         int r;
 102
 103         r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
 104         if (r == -ENOENT) {
 105                 _cleanup_closedir_ DIR *dir = NULL;
 106                 _cleanup_free_ char *compat = NULL;
 107
 108                 if (access("/proc/device-tree/ibm,partition-name", F_OK) == 0 &&
 109                     access("/proc/device-tree/hmc-managed?", F_OK) == 0 &&
 110                     access("/proc/device-tree/chosen/qemu,graphic-width", F_OK) != 0)
 111                         return VIRTUALIZATION_POWERVM;
 112
 113                 dir = opendir("/proc/device-tree");
 114                 if (!dir) {
 115                         if (errno == ENOENT) {
 116                                 log_debug_errno(errno, "/proc/device-tree: %m");
 117                                 return VIRTUALIZATION_NONE;
 118                         }
 119                         return -errno;
 120                 }
 121
 122                 FOREACH_DIRENT(de, dir, return -errno)
 123                         if (strstr(de->d_name, "fw-cfg")) {
 124                                 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", de->d_name);
 125                                 return VIRTUALIZATION_QEMU;
 126                         }
 127
 128                 r = read_one_line_file("/proc/device-tree/compatible", &compat);
 129                 if (r < 0 && r != -ENOENT)
 130                         return r;
 131                 if (r >= 0 && streq(compat, "qemu,pseries")) {
 132                         log_debug("Virtualization %s found in /proc/device-tree/compatible", compat);
 133                         return VIRTUALIZATION_QEMU;
 134                 }
 135
 136                 log_debug("No virtualization found in /proc/device-tree/*");
 137                 return VIRTUALIZATION_NONE;
 138         } else if (r < 0)
 139                 return r;
 140
 141         log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
 142         if (streq(hvtype, "linux,kvm"))
 143                 return VIRTUALIZATION_KVM;
 144         else if (strstr(hvtype, "xen"))
 145                 return VIRTUALIZATION_XEN;
 146         else if (strstr(hvtype, "vmware"))
 147                 return VIRTUALIZATION_VMWARE;
 148         else
 149                 return VIRTUALIZATION_VM_OTHER;
 150 #else
 151         log_debug("This platform does not support /proc/device-tree");
 152         return VIRTUALIZATION_NONE;
 153 #endif
 154 }
 155
 156 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || defined(__loongarch_lp64)
 157 static Virtualization detect_vm_dmi_vendor(void) {
 158         static const char* const dmi_vendors[] = {
 159                 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
 160                 "/sys/class/dmi/id/sys_vendor",
 161                 "/sys/class/dmi/id/board_vendor",
 162                 "/sys/class/dmi/id/bios_vendor",
 163                 "/sys/class/dmi/id/product_version", /* For Hyper-V VMs test */
 164                 NULL
 165         };
 166
 167         static const struct {
 168                 const char *vendor;
 169                 Virtualization id;
 170         } dmi_vendor_table[] = {
 171                 { "KVM",                  VIRTUALIZATION_KVM       },
 172                 { "OpenStack",            VIRTUALIZATION_KVM       }, /* Detect OpenStack instance as KVM in non x86 architecture */
 173                 { "KubeVirt",             VIRTUALIZATION_KVM       }, /* Detect KubeVirt instance as KVM in non x86 architecture */
 174                 { "Amazon EC2",           VIRTUALIZATION_AMAZON    },
 175                 { "QEMU",                 VIRTUALIZATION_QEMU      },
 176                 { "VMware",               VIRTUALIZATION_VMWARE    }, /* https://kb.vmware.com/s/article/1009458 */
 177                 { "VMW",                  VIRTUALIZATION_VMWARE    },
 178                 { "innotek GmbH",         VIRTUALIZATION_ORACLE    },
 179                 { "VirtualBox",           VIRTUALIZATION_ORACLE    },
 180                 { "Xen",                  VIRTUALIZATION_XEN       },
 181                 { "Bochs",                VIRTUALIZATION_BOCHS     },
 182                 { "Parallels",            VIRTUALIZATION_PARALLELS },
 183                 /* https://wiki.freebsd.org/bhyve */
 184                 { "BHYVE",                VIRTUALIZATION_BHYVE     },
 185                 { "Hyper-V",              VIRTUALIZATION_MICROSOFT },
 186                 { "Apple Virtualization", VIRTUALIZATION_APPLE     },
 187         };
 188         int r;
 189
 190         STRV_FOREACH(vendor, dmi_vendors) {
 191                 _cleanup_free_ char *s = NULL;
 192
 193                 r = read_one_line_file(*vendor, &s);
 194                 if (r < 0) {
 195                         if (r == -ENOENT)
 196                                 continue;
 197
 198                         return r;
 199                 }
 200
 201                 for (size_t i = 0; i < ELEMENTSOF(dmi_vendor_table); i++)
 202                         if (startswith(s, dmi_vendor_table[i].vendor)) {
 203                                 log_debug("Virtualization %s found in DMI (%s)", s, *vendor);
 204                                 return dmi_vendor_table[i].id;
 205                         }
 206         }
 207         log_debug("No virtualization found in DMI vendor table.");
 208         return VIRTUALIZATION_NONE;
 209 }
 210
 211 static int detect_vm_smbios(void) {
 212         /* The SMBIOS BIOS Characteristics Extension Byte 2 (Section 2.1.2.2 of
 213          * https://www.dmtf.org/sites/default/files/standards/documents/DSP0134_3.4.0.pdf), specifies that
 214          * the 4th bit being set indicates a VM. The BIOS Characteristics table is exposed via the kernel in
 215          * /sys/firmware/dmi/entries/0-0. Note that in the general case, this bit being unset should not
 216          * imply that the system is running on bare-metal.  For example, QEMU 3.1.0 (with or without KVM)
 217          * with SeaBIOS does not set this bit. */
 218         _cleanup_free_ char *s = NULL;
 219         size_t readsize;
 220         int r;
 221
 222         r = read_full_virtual_file("/sys/firmware/dmi/entries/0-0/raw", &s, &readsize);
 223         if (r < 0) {
 224                 log_debug_errno(r, "Unable to read /sys/firmware/dmi/entries/0-0/raw, "
 225                                 "using the virtualization information found in DMI vendor table, ignoring: %m");
 226                 return SMBIOS_VM_BIT_UNKNOWN;
 227         }
 228         if (readsize < 20 || s[1] < 20) {
 229                 /* The spec indicates that byte 1 contains the size of the table, 0x12 + the number of
 230                  * extension bytes. The data we're interested in is in extension byte 2, which would be at
 231                  * 0x13. If we didn't read that much data, or if the BIOS indicates that we don't have that
 232                  * much data, we don't infer anything from the SMBIOS. */
 233                 log_debug("Only read %zu bytes from /sys/firmware/dmi/entries/0-0/raw (expected 20). "
 234                           "Using the virtualization information found in DMI vendor table.", readsize);
 235                 return SMBIOS_VM_BIT_UNKNOWN;
 236         }
 237
 238         uint8_t byte = (uint8_t) s[19];
 239         if (byte & (1U<<4)) {
 240                 log_debug("DMI BIOS Extension table indicates virtualization.");
 241                 return SMBIOS_VM_BIT_SET;
 242         }
 243         log_debug("DMI BIOS Extension table does not indicate virtualization.");
 244         return SMBIOS_VM_BIT_UNSET;
 245 }
 246 #endif /* defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || defined(__loongarch_lp64) */
 247
 248 static Virtualization detect_vm_dmi(void) {
 249 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || defined(__loongarch_lp64)
 250
 251         int r;
 252         r = detect_vm_dmi_vendor();
 253
 254         /* The DMI vendor tables in /sys/class/dmi/id don't help us distinguish between Amazon EC2
 255          * virtual machines and bare-metal instances, so we need to look at SMBIOS. */
 256         if (r == VIRTUALIZATION_AMAZON) {
 257                 switch (detect_vm_smbios()) {
 258                 case SMBIOS_VM_BIT_SET:
 259                         return VIRTUALIZATION_AMAZON;
 260                 case SMBIOS_VM_BIT_UNSET:
 261                         return VIRTUALIZATION_NONE;
 262                 case SMBIOS_VM_BIT_UNKNOWN: {
 263                         /* The DMI information we are after is only accessible to the root user,
 264                          * so we fallback to using the product name which is less restricted
 265                          * to distinguish metal systems from virtualized instances */
 266                         _cleanup_free_ char *s = NULL;
 267                         const char *e;
 268
 269                         r = read_full_virtual_file("/sys/class/dmi/id/product_name", &s, NULL);
 270                         /* In EC2, virtualized is much more common than metal, so if for some reason
 271                          * we fail to read the DMI data, assume we are virtualized. */
 272                         if (r < 0) {
 273                                 log_debug_errno(r, "Can't read /sys/class/dmi/id/product_name,"
 274                                                 " assuming virtualized: %m");
 275                                 return VIRTUALIZATION_AMAZON;
 276                         }
 277                         e = strstrafter(truncate_nl(s), ".metal");
 278                         if (e && IN_SET(*e, 0, '-')) {
 279                                 log_debug("DMI product name has '.metal', assuming no virtualization");
 280                                 return VIRTUALIZATION_NONE;
 281                         } else
 282                                 return VIRTUALIZATION_AMAZON;
 283                 }
 284                 default:
 285                         assert_not_reached();
 286               }
 287         }
 288
 289         /* If we haven't identified a VM, but the firmware indicates that there is one, indicate as much. We
 290          * have no further information about what it is. */
 291         if (r == VIRTUALIZATION_NONE && detect_vm_smbios() == SMBIOS_VM_BIT_SET)
 292                 return VIRTUALIZATION_VM_OTHER;
 293         return r;
 294 #else
 295         return VIRTUALIZATION_NONE;
 296 #endif
 297 }
 298
 299 #define XENFEAT_dom0 11 /* xen/include/public/features.h */
 300 #define PATH_FEATURES "/sys/hypervisor/properties/features"
 301 /* Returns -errno, or 0 for domU, or 1 for dom0 */
 302 static int detect_vm_xen_dom0(void) {
 303         _cleanup_free_ char *domcap = NULL;
 304         int r;
 305
 306         r = read_one_line_file(PATH_FEATURES, &domcap);
 307         if (r < 0 && r != -ENOENT)
 308                 return r;
 309         if (r >= 0) {
 310                 unsigned long features;
 311
 312                 /* Here, we need to use sscanf() instead of safe_atoul()
 313                  * as the string lacks the leading "0x". */
 314                 r = sscanf(domcap, "%lx", &features);
 315                 if (r == 1) {
 316                         r = !!(features & (1U << XENFEAT_dom0));
 317                         log_debug("Virtualization XEN, found %s with value %08lx, "
 318                                   "XENFEAT_dom0 (indicating the 'hardware domain') is%s set.",
 319                                   PATH_FEATURES, features, r ? "" : " not");
 320                         return r;
 321                 }
 322                 log_debug("Virtualization XEN, found %s, unhandled content '%s'",
 323                           PATH_FEATURES, domcap);
 324         }
 325
 326         r = read_one_line_file("/proc/xen/capabilities", &domcap);
 327         if (r == -ENOENT) {
 328                 log_debug("Virtualization XEN because /proc/xen/capabilities does not exist");
 329                 return 0;
 330         }
 331         if (r < 0)
 332                 return r;
 333
 334         for (const char *i = domcap;;) {
 335                 _cleanup_free_ char *cap = NULL;
 336
 337                 r = extract_first_word(&i, &cap, ",", 0);
 338                 if (r < 0)
 339                         return r;
 340                 if (r == 0) {
 341                         log_debug("Virtualization XEN DomU found (/proc/xen/capabilities)");
 342                         return 0;
 343                 }
 344
 345                 if (streq(cap, "control_d")) {
 346                         log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
 347                         return 1;
 348                 }
 349         }
 350 }
 351
 352 static Virtualization detect_vm_xen(void) {
 353         /* The presence of /proc/xen indicates some form of a Xen domain
 354            The check for Dom0 is handled outside this function */
 355         if (access("/proc/xen", F_OK) < 0) {
 356                 log_debug("Virtualization XEN not found, /proc/xen does not exist");
 357                 return VIRTUALIZATION_NONE;
 358         }
 359         log_debug("Virtualization XEN found (/proc/xen exists)");
 360         return VIRTUALIZATION_XEN;
 361 }
 362
 363 static Virtualization detect_vm_hypervisor(void) {
 364         _cleanup_free_ char *hvtype = NULL;
 365         int r;
 366
 367         r = read_one_line_file("/sys/hypervisor/type", &hvtype);
 368         if (r == -ENOENT)
 369                 return VIRTUALIZATION_NONE;
 370         if (r < 0)
 371                 return r;
 372
 373         log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
 374
 375         if (streq(hvtype, "xen"))
 376                 return VIRTUALIZATION_XEN;
 377         else
 378                 return VIRTUALIZATION_VM_OTHER;
 379 }
 380
 381 static Virtualization detect_vm_uml(void) {
 382         _cleanup_fclose_ FILE *f = NULL;
 383         int r;
 384
 385         /* Detect User-Mode Linux by reading /proc/cpuinfo */
 386         f = fopen("/proc/cpuinfo", "re");
 387         if (!f) {
 388                 if (errno == ENOENT) {
 389                         log_debug("/proc/cpuinfo not found, assuming no UML virtualization.");
 390                         return VIRTUALIZATION_NONE;
 391                 }
 392                 return -errno;
 393         }
 394
 395         for (;;) {
 396                 _cleanup_free_ char *line = NULL;
 397                 const char *t;
 398
 399                 r = read_line(f, LONG_LINE_MAX, &line);
 400                 if (r < 0)
 401                         return r;
 402                 if (r == 0)
 403                         break;
 404
 405                 t = startswith(line, "vendor_id\t: ");
 406                 if (t) {
 407                         if (startswith(t, "User Mode Linux")) {
 408                                 log_debug("UML virtualization found in /proc/cpuinfo");
 409                                 return VIRTUALIZATION_UML;
 410                         }
 411
 412                         break;
 413                 }
 414         }
 415
 416         log_debug("UML virtualization not found in /proc/cpuinfo.");
 417         return VIRTUALIZATION_NONE;
 418 }
 419
 420 static Virtualization detect_vm_zvm(void) {
 421
 422 #if defined(__s390__)
 423         _cleanup_free_ char *t = NULL;
 424         int r;
 425
 426         r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
 427         if (r == -ENOENT)
 428                 return VIRTUALIZATION_NONE;
 429         if (r < 0)
 430                 return r;
 431
 432         log_debug("Virtualization %s found in /proc/sysinfo", t);
 433         if (streq(t, "z/VM"))
 434                 return VIRTUALIZATION_ZVM;
 435         else
 436                 return VIRTUALIZATION_KVM;
 437 #else
 438         log_debug("This platform does not support /proc/sysinfo");
 439         return VIRTUALIZATION_NONE;
 440 #endif
 441 }
 442
 443 /* Returns a short identifier for the various VM implementations */
 444 Virtualization detect_vm(void) {
 445         static thread_local Virtualization cached_found = _VIRTUALIZATION_INVALID;
 446         bool other = false;
 447         int xen_dom0 = 0;
 448         Virtualization v, dmi;
 449
 450         if (cached_found >= 0)
 451                 return cached_found;
 452
 453         /* We have to use the correct order here:
 454          *
 455          * → First, try to detect Oracle Virtualbox, Amazon EC2 Nitro, and Parallels, even if they use KVM,
 456          *   as well as Xen even if it cloaks as Microsoft Hyper-V. Attempt to detect uml at this stage also
 457          *   since it runs as a user-process nested inside other VMs. Also check for Xen now, because Xen PV
 458          *   mode does not override CPUID when nested inside another hypervisor.
 459          *
 460          * → Second, try to detect from CPUID, this will report KVM for whatever software is used even if
 461          *   info in DMI is overwritten.
 462          *
 463          * → Third, try to detect from DMI. */
 464
 465         dmi = detect_vm_dmi();
 466         if (IN_SET(dmi,
 467                    VIRTUALIZATION_ORACLE,
 468                    VIRTUALIZATION_XEN,
 469                    VIRTUALIZATION_AMAZON,
 470                    VIRTUALIZATION_PARALLELS)) {
 471                 v = dmi;
 472                 goto finish;
 473         }
 474
 475         /* Detect UML */
 476         v = detect_vm_uml();
 477         if (v < 0)
 478                 return v;
 479         if (v != VIRTUALIZATION_NONE)
 480                 goto finish;
 481
 482         /* Detect Xen */
 483         v = detect_vm_xen();
 484         if (v < 0)
 485                 return v;
 486         if (v == VIRTUALIZATION_XEN) {
 487                  /* If we are Dom0, then we expect to not report as a VM. However, as we might be nested
 488                   * inside another hypervisor which can be detected via the CPUID check, wait to report this
 489                   * until after the CPUID check. */
 490                 xen_dom0 = detect_vm_xen_dom0();
 491                 if (xen_dom0 < 0)
 492                         return xen_dom0;
 493                 if (xen_dom0 == 0)
 494                         goto finish;
 495         } else if (v != VIRTUALIZATION_NONE)
 496                 assert_not_reached();
 497
 498         /* Detect from CPUID */
 499         v = detect_vm_cpuid();
 500         if (v < 0)
 501                 return v;
 502         if (v == VIRTUALIZATION_VM_OTHER)
 503                 other = true;
 504         else if (v != VIRTUALIZATION_NONE)
 505                 goto finish;
 506
 507         /* If we are in Dom0 and have not yet finished, finish with the result of detect_vm_cpuid */
 508         if (xen_dom0 > 0)
 509                 goto finish;
 510
 511         /* Now, let's get back to DMI */
 512         if (dmi < 0)
 513                 return dmi;
 514         if (dmi == VIRTUALIZATION_VM_OTHER)
 515                 other = true;
 516         else if (dmi != VIRTUALIZATION_NONE) {
 517                 v = dmi;
 518                 goto finish;
 519         }
 520
 521         /* Check high-level hypervisor sysfs file */
 522         v = detect_vm_hypervisor();
 523         if (v < 0)
 524                 return v;
 525         if (v == VIRTUALIZATION_VM_OTHER)
 526                 other = true;
 527         else if (v != VIRTUALIZATION_NONE)
 528                 goto finish;
 529
 530         v = detect_vm_device_tree();
 531         if (v < 0)
 532                 return v;
 533         if (v == VIRTUALIZATION_VM_OTHER)
 534                 other = true;
 535         else if (v != VIRTUALIZATION_NONE)
 536                 goto finish;
 537
 538         v = detect_vm_zvm();
 539         if (v < 0)
 540                 return v;
 541
 542 finish:
 543         if (v == VIRTUALIZATION_NONE && other)
 544                 v = VIRTUALIZATION_VM_OTHER;
 545
 546         cached_found = v;
 547         log_debug("Found VM virtualization %s", virtualization_to_string(v));
 548         return v;
 549 }
 550
 551 static const char *const container_table[_VIRTUALIZATION_MAX] = {
 552         [VIRTUALIZATION_LXC]            = "lxc",
 553         [VIRTUALIZATION_LXC_LIBVIRT]    = "lxc-libvirt",
 554         [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
 555         [VIRTUALIZATION_DOCKER]         = "docker",
 556         [VIRTUALIZATION_PODMAN]         = "podman",
 557         [VIRTUALIZATION_RKT]            = "rkt",
 558         [VIRTUALIZATION_WSL]            = "wsl",
 559         [VIRTUALIZATION_PROOT]          = "proot",
 560         [VIRTUALIZATION_POUCH]          = "pouch",
 561 };
 562
 563 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_FROM_STRING(container, int);
 564
 565 static int running_in_cgroupns(void) {
 566         int r;
 567
 568         if (!cg_ns_supported())
 569                 return false;
 570
 571         r = cg_all_unified();
 572         if (r < 0)
 573                 return r;
 574
 575         if (r) {
 576                 /* cgroup v2 */
 577
 578                 r = access("/sys/fs/cgroup/cgroup.events", F_OK);
 579                 if (r < 0) {
 580                         if (errno != ENOENT)
 581                                 return -errno;
 582                         /* All kernel versions have cgroup.events in nested cgroups. */
 583                         return false;
 584                 }
 585
 586                 /* There's no cgroup.type in the root cgroup, and future kernel versions
 587                  * are unlikely to add it since cgroup.type is something that makes no sense
 588                  * whatsoever in the root cgroup. */
 589                 r = access("/sys/fs/cgroup/cgroup.type", F_OK);
 590                 if (r == 0)
 591                         return true;
 592                 if (r < 0 && errno != ENOENT)
 593                         return -errno;
 594
 595                 /* On older kernel versions, there's no cgroup.type */
 596                 r = access("/sys/kernel/cgroup/features", F_OK);
 597                 if (r < 0) {
 598                         if (errno != ENOENT)
 599                                 return -errno;
 600                         /* This is an old kernel that we know for sure has cgroup.events
 601                          * only in nested cgroups. */
 602                         return true;
 603                 }
 604
 605                 /* This is a recent kernel, and cgroup.type doesn't exist, so we must be
 606                  * in the root cgroup. */
 607                 return false;
 608         } else {
 609                 /* cgroup v1 */
 610
 611                 /* If systemd controller is not mounted, do not even bother. */
 612                 r = access("/sys/fs/cgroup/systemd", F_OK);
 613                 if (r < 0) {
 614                         if (errno != ENOENT)
 615                                 return -errno;
 616                         return false;
 617                 }
 618
 619                 /* release_agent only exists in the root cgroup. */
 620                 r = access("/sys/fs/cgroup/systemd/release_agent", F_OK);
 621                 if (r < 0) {
 622                         if (errno != ENOENT)
 623                                 return -errno;
 624                         return true;
 625                 }
 626
 627                 return false;
 628         }
 629 }
 630
 631 static Virtualization detect_container_files(void) {
 632         static const struct {
 633                 const char *file_path;
 634                 Virtualization id;
 635         } container_file_table[] = {
 636                 /* https://github.com/containers/podman/issues/6192 */
 637                 /* https://github.com/containers/podman/issues/3586#issuecomment-661918679 */
 638                 { "/run/.containerenv", VIRTUALIZATION_PODMAN },
 639                 /* https://github.com/moby/moby/issues/18355 */
 640                 /* Docker must be the last in this table, see below. */
 641                 { "/.dockerenv",        VIRTUALIZATION_DOCKER },
 642         };
 643
 644         for (size_t i = 0; i < ELEMENTSOF(container_file_table); i++) {
 645                 if (access(container_file_table[i].file_path, F_OK) >= 0)
 646                         return container_file_table[i].id;
 647
 648                 if (errno != ENOENT)
 649                         log_debug_errno(errno,
 650                                         "Checking if %s exists failed, ignoring: %m",
 651                                         container_file_table[i].file_path);
 652         }
 653
 654         return VIRTUALIZATION_NONE;
 655 }
 656
 657 Virtualization detect_container(void) {
 658         static thread_local Virtualization cached_found = _VIRTUALIZATION_INVALID;
 659         _cleanup_free_ char *m = NULL, *o = NULL, *p = NULL;
 660         const char *e = NULL;
 661         Virtualization v;
 662         int r;
 663
 664         if (cached_found >= 0)
 665                 return cached_found;
 666
 667         /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
 668         if (access("/proc/vz", F_OK) < 0) {
 669                 if (errno != ENOENT)
 670                         log_debug_errno(errno, "Failed to check if /proc/vz exists, ignoring: %m");
 671         } else if (access("/proc/bc", F_OK) < 0) {
 672                 if (errno == ENOENT) {
 673                         v = VIRTUALIZATION_OPENVZ;
 674                         goto finish;
 675                 }
 676
 677                 log_debug_errno(errno, "Failed to check if /proc/bc exists, ignoring: %m");
 678         }
 679
 680         /* "Official" way of detecting WSL https://github.com/Microsoft/WSL/issues/423#issuecomment-221627364 */
 681         r = read_one_line_file("/proc/sys/kernel/osrelease", &o);
 682         if (r < 0)
 683                 log_debug_errno(r, "Failed to read /proc/sys/kernel/osrelease, ignoring: %m");
 684         else if (strstr(o, "Microsoft") || strstr(o, "WSL")) {
 685                 v = VIRTUALIZATION_WSL;
 686                 goto finish;
 687         }
 688
 689         /* proot doesn't use PID namespacing, so we can just check if we have a matching tracer for this
 690          * invocation without worrying about it being elsewhere.
 691          */
 692         r = get_proc_field("/proc/self/status", "TracerPid", WHITESPACE, &p);
 693         if (r < 0)
 694                 log_debug_errno(r, "Failed to read our own trace PID, ignoring: %m");
 695         else if (!streq(p, "0")) {
 696                 pid_t ptrace_pid;
 697
 698                 r = parse_pid(p, &ptrace_pid);
 699                 if (r < 0)
 700                         log_debug_errno(r, "Failed to parse our own tracer PID, ignoring: %m");
 701                 else {
 702                         _cleanup_free_ char *ptrace_comm = NULL;
 703                         const char *pf;
 704
 705                         pf = procfs_file_alloca(ptrace_pid, "comm");
 706                         r = read_one_line_file(pf, &ptrace_comm);
 707                         if (r < 0)
 708                                 log_debug_errno(r, "Failed to read %s, ignoring: %m", pf);
 709                         else if (startswith(ptrace_comm, "proot")) {
 710                                 v = VIRTUALIZATION_PROOT;
 711                                 goto finish;
 712                         }
 713                 }
 714         }
 715
 716         /* The container manager might have placed this in the /run/host/ hierarchy for us, which is best
 717          * because we can be consumed just like that, without special privileges. */
 718         r = read_one_line_file("/run/host/container-manager", &m);
 719         if (r > 0) {
 720                 e = m;
 721                 goto translate_name;
 722         }
 723         if (!IN_SET(r, -ENOENT, 0))
 724                 return log_debug_errno(r, "Failed to read /run/host/container-manager: %m");
 725
 726         if (getpid_cached() == 1) {
 727                 /* If we are PID 1 we can just check our own environment variable, and that's authoritative.
 728                  * We distinguish three cases:
 729                  * - the variable is not defined → we jump to other checks
 730                  * - the variable is defined to an empty value → we are not in a container
 731                  * - anything else → some container, either one of the known ones or "container-other"
 732                  */
 733                 e = getenv("container");
 734                 if (!e)
 735                         goto check_files;
 736                 if (isempty(e)) {
 737                         v = VIRTUALIZATION_NONE;
 738                         goto finish;
 739                 }
 740
 741                 goto translate_name;
 742         }
 743
 744         /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
 745          * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
 746         r = read_one_line_file("/run/systemd/container", &m);
 747         if (r > 0) {
 748                 e = m;
 749                 goto translate_name;
 750         }
 751         if (!IN_SET(r, -ENOENT, 0))
 752                 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
 753
 754         /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
 755         r = getenv_for_pid(1, "container", &m);
 756         if (r > 0) {
 757                 e = m;
 758                 goto translate_name;
 759         }
 760         if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
 761                 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
 762
 763 check_files:
 764         /* Check for existence of some well-known files. We only do this after checking
 765          * for other specific container managers, otherwise we risk mistaking another
 766          * container manager for Docker: the /.dockerenv file could inadvertently end up
 767          * in a file system image. */
 768         v = detect_container_files();
 769         if (v < 0)
 770                 return v;
 771         if (v != VIRTUALIZATION_NONE)
 772                 goto finish;
 773
 774         r = running_in_cgroupns();
 775         if (r > 0) {
 776                 v = VIRTUALIZATION_CONTAINER_OTHER;
 777                 goto finish;
 778         }
 779         if (r < 0)
 780                 log_debug_errno(r, "Failed to detect cgroup namespace: %m");
 781
 782         /* If none of that worked, give up, assume no container manager. */
 783         v = VIRTUALIZATION_NONE;
 784         goto finish;
 785
 786 translate_name:
 787         if (streq(e, "oci")) {
 788                 /* Some images hardcode container=oci, but OCI is not a specific container manager.
 789                  * Try to detect one based on well-known files. */
 790                 v = detect_container_files();
 791                 if (v == VIRTUALIZATION_NONE)
 792                         v = VIRTUALIZATION_CONTAINER_OTHER;
 793                 goto finish;
 794         }
 795         v = container_from_string(e);
 796         if (v < 0)
 797                 v = VIRTUALIZATION_CONTAINER_OTHER;
 798
 799 finish:
 800         log_debug("Found container virtualization %s.", virtualization_to_string(v));
 801         cached_found = v;
 802         return v;
 803 }
 804
 805 Virtualization detect_virtualization(void) {
 806         int v;
 807
 808         v = detect_container();
 809         if (v != VIRTUALIZATION_NONE)
 810                 return v;
 811
 812         return detect_vm();
 813 }
 814
 815 static int userns_has_mapping(const char *name) {
 816         _cleanup_fclose_ FILE *f = NULL;
 817         uid_t a, b, c;
 818         int r;
 819
 820         f = fopen(name, "re");
 821         if (!f) {
 822                 log_debug_errno(errno, "Failed to open %s: %m", name);
 823                 return errno == ENOENT ? false : -errno;
 824         }
 825
 826         errno = 0;
 827         r = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &a, &b, &c);
 828         if (r == EOF) {
 829                 if (ferror(f))
 830                         return log_debug_errno(errno_or_else(EIO), "Failed to read %s: %m", name);
 831
 832                 log_debug("%s is empty, we're in an uninitialized user namespace", name);
 833                 return true;
 834         }
 835         if (r != 3)
 836                 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), "Failed to parse %s: %m", name);
 837
 838         if (a == 0 && b == 0 && c == UINT32_MAX) {
 839                 /* The kernel calls mappings_overlap() and does not allow overlaps */
 840                 log_debug("%s has a full 1:1 mapping", name);
 841                 return false;
 842         }
 843
 844         /* Anything else implies that we are in a user namespace */
 845         log_debug("Mapping found in %s, we're in a user namespace", name);
 846         return true;
 847 }
 848
 849 int running_in_userns(void) {
 850         _cleanup_free_ char *line = NULL;
 851         int r;
 852
 853         r = userns_has_mapping("/proc/self/uid_map");
 854         if (r != 0)
 855                 return r;
 856
 857         r = userns_has_mapping("/proc/self/gid_map");
 858         if (r != 0)
 859                 return r;
 860
 861         /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also possible to compile a
 862          * kernel without CONFIG_USER_NS, in which case "setgroups" also does not exist. We cannot
 863          * distinguish those two cases, so assume that we're running on a stripped-down recent kernel, rather
 864          * than on an old one, and if the file is not found, return false. */
 865         r = read_virtual_file("/proc/self/setgroups", SIZE_MAX, &line, NULL);
 866         if (r < 0) {
 867                 log_debug_errno(r, "/proc/self/setgroups: %m");
 868                 return r == -ENOENT ? false : r;
 869         }
 870
 871         strstrip(line); /* remove trailing newline */
 872
 873         r = streq(line, "deny");
 874         /* See user_namespaces(7) for a description of this "setgroups" contents. */
 875         log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
 876         return r;
 877 }
 878
 879 int running_in_chroot(void) {
 880         int r;
 881
 882         /* If we're PID1, /proc may not be mounted (and most likely we're not in a chroot). But PID1 will
 883          * mount /proc, so all other programs can assume that if /proc is *not* available, we're in some
 884          * chroot. */
 885
 886         if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
 887                 return 0;
 888
 889         r = inode_same("/proc/1/root", "/", 0);
 890         if (r == -ENOENT) {
 891                 r = proc_mounted();
 892                 if (r == 0) {
 893                         if (getpid_cached() == 1)
 894                                 return false; /* We will mount /proc, assuming we're not in a chroot. */
 895
 896                         log_debug("/proc is not mounted, assuming we're in a chroot.");
 897                         return true;
 898                 }
 899                 if (r > 0)  /* If we have fake /proc/, we can't do the check properly. */
 900                         return -ENOSYS;
 901         }
 902         if (r < 0)
 903                 return r;
 904
 905         return r == 0;
 906 }
 907
 908 #if defined(__i386__) || defined(__x86_64__)
 909 struct cpuid_table_entry {
 910         uint32_t flag_bit;
 911         const char *name;
 912 };
 913
 914 static const struct cpuid_table_entry leaf1_edx[] = {
 915         {  0, "fpu"     },
 916         {  1, "vme"     },
 917         {  2, "de"      },
 918         {  3, "pse"     },
 919         {  4, "tsc"     },
 920         {  5, "msr"     },
 921         {  6, "pae"     },
 922         {  7, "mce"     },
 923         {  8, "cx8"     },
 924         {  9, "apic"    },
 925         { 11, "sep"     },
 926         { 12, "mtrr"    },
 927         { 13, "pge"     },
 928         { 14, "mca"     },
 929         { 15, "cmov"    },
 930         { 16, "pat"     },
 931         { 17, "pse36"   },
 932         { 19, "clflush" },
 933         { 23, "mmx"     },
 934         { 24, "fxsr"    },
 935         { 25, "sse"     },
 936         { 26, "sse2"    },
 937         { 28, "ht"      },
 938 };
 939
 940 static const struct cpuid_table_entry leaf1_ecx[] = {
 941         {  0, "pni"     },
 942         {  1, "pclmul"  },
 943         {  3, "monitor" },
 944         {  9, "ssse3"   },
 945         { 12, "fma3"    },
 946         { 13, "cx16"    },
 947         { 19, "sse4_1"  },
 948         { 20, "sse4_2"  },
 949         { 22, "movbe"   },
 950         { 23, "popcnt"  },
 951         { 25, "aes"     },
 952         { 26, "xsave"   },
 953         { 27, "osxsave" },
 954         { 28, "avx"     },
 955         { 29, "f16c"    },
 956         { 30, "rdrand"  },
 957 };
 958
 959 static const struct cpuid_table_entry leaf7_ebx[] = {
 960         {  3, "bmi1"   },
 961         {  5, "avx2"   },
 962         {  8, "bmi2"   },
 963         { 18, "rdseed" },
 964         { 19, "adx"    },
 965         { 29, "sha_ni" },
 966 };
 967
 968 static const struct cpuid_table_entry leaf81_edx[] = {
 969         { 11, "syscall" },
 970         { 27, "rdtscp"  },
 971         { 29, "lm"      },
 972 };
 973
 974 static const struct cpuid_table_entry leaf81_ecx[] = {
 975         {  0, "lahf_lm" },
 976         {  5, "abm"     },
 977 };
 978
 979 static const struct cpuid_table_entry leaf87_edx[] = {
 980         {  8, "constant_tsc" },
 981 };
 982
 983 static bool given_flag_in_set(const char *flag, const struct cpuid_table_entry *set, size_t set_size, uint32_t val) {
 984         for (size_t i = 0; i < set_size; i++) {
 985                 if ((UINT32_C(1) << set[i].flag_bit) & val &&
 986                                 streq(flag, set[i].name))
 987                         return true;
 988         }
 989         return false;
 990 }
 991
 992 static bool real_has_cpu_with_flag(const char *flag) {
 993         uint32_t eax, ebx, ecx, edx;
 994
 995         if (__get_cpuid(1, &eax, &ebx, &ecx, &edx)) {
 996                 if (given_flag_in_set(flag, leaf1_ecx, ELEMENTSOF(leaf1_ecx), ecx))
 997                         return true;
 998
 999                 if (given_flag_in_set(flag, leaf1_edx, ELEMENTSOF(leaf1_edx), edx))
1000                         return true;
1001         }
1002
1003         if (__get_cpuid(7, &eax, &ebx, &ecx, &edx)) {
1004                 if (given_flag_in_set(flag, leaf7_ebx, ELEMENTSOF(leaf7_ebx), ebx))
1005                         return true;
1006         }
1007
1008         if (__get_cpuid(0x80000001U, &eax, &ebx, &ecx, &edx)) {
1009                 if (given_flag_in_set(flag, leaf81_ecx, ELEMENTSOF(leaf81_ecx), ecx))
1010                         return true;
1011
1012                 if (given_flag_in_set(flag, leaf81_edx, ELEMENTSOF(leaf81_edx), edx))
1013                         return true;
1014         }
1015
1016         if (__get_cpuid(0x80000007U, &eax, &ebx, &ecx, &edx))
1017                 if (given_flag_in_set(flag, leaf87_edx, ELEMENTSOF(leaf87_edx), edx))
1018                         return true;
1019
1020         return false;
1021 }
1022 #endif
1023
1024 bool has_cpu_with_flag(const char *flag) {
1025         /* CPUID is an x86 specific interface. Assume on all others that no CPUs have those flags. */
1026 #if defined(__i386__) || defined(__x86_64__)
1027         return real_has_cpu_with_flag(flag);
1028 #else
1029         return false;
1030 #endif
1031 }
1032
1033 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
1034         [VIRTUALIZATION_NONE]            = "none",
1035         [VIRTUALIZATION_KVM]             = "kvm",
1036         [VIRTUALIZATION_AMAZON]          = "amazon",
1037         [VIRTUALIZATION_QEMU]            = "qemu",
1038         [VIRTUALIZATION_BOCHS]           = "bochs",
1039         [VIRTUALIZATION_XEN]             = "xen",
1040         [VIRTUALIZATION_UML]             = "uml",
1041         [VIRTUALIZATION_VMWARE]          = "vmware",
1042         [VIRTUALIZATION_ORACLE]          = "oracle",
1043         [VIRTUALIZATION_MICROSOFT]       = "microsoft",
1044         [VIRTUALIZATION_ZVM]             = "zvm",
1045         [VIRTUALIZATION_PARALLELS]       = "parallels",
1046         [VIRTUALIZATION_BHYVE]           = "bhyve",
1047         [VIRTUALIZATION_QNX]             = "qnx",
1048         [VIRTUALIZATION_ACRN]            = "acrn",
1049         [VIRTUALIZATION_POWERVM]         = "powervm",
1050         [VIRTUALIZATION_APPLE]           = "apple",
1051         [VIRTUALIZATION_SRE]             = "sre",
1052         [VIRTUALIZATION_VM_OTHER]        = "vm-other",
1053
1054         [VIRTUALIZATION_SYSTEMD_NSPAWN]  = "systemd-nspawn",
1055         [VIRTUALIZATION_LXC_LIBVIRT]     = "lxc-libvirt",
1056         [VIRTUALIZATION_LXC]             = "lxc",
1057         [VIRTUALIZATION_OPENVZ]          = "openvz",
1058         [VIRTUALIZATION_DOCKER]          = "docker",
1059         [VIRTUALIZATION_PODMAN]          = "podman",
1060         [VIRTUALIZATION_RKT]             = "rkt",
1061         [VIRTUALIZATION_WSL]             = "wsl",
1062         [VIRTUALIZATION_PROOT]           = "proot",
1063         [VIRTUALIZATION_POUCH]           = "pouch",
1064         [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
1065 };
1066
1067 DEFINE_STRING_TABLE_LOOKUP(virtualization, Virtualization);