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 "uid-range.h"
  25 #include "virt.h"
  26
  27 enum {
  28       SMBIOS_VM_BIT_SET,
  29       SMBIOS_VM_BIT_UNSET,
  30       SMBIOS_VM_BIT_UNKNOWN,
  31 };
  32
  33 static Virtualization detect_vm_cpuid(void) {
  34
  35         /* CPUID is an x86 specific interface. */
  36 #if defined(__i386__) || defined(__x86_64__)
  37
  38         static const struct {
  39                 const char sig[13];
  40                 Virtualization id;
  41         } vm_table[] = {
  42                 { "XenVMMXenVMM", VIRTUALIZATION_XEN       },
  43                 { "KVMKVMKVM",    VIRTUALIZATION_KVM       }, /* qemu with KVM */
  44                 { "Linux KVM Hv", VIRTUALIZATION_KVM       }, /* qemu with KVM + HyperV Enlightenments */
  45                 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU      }, /* qemu without KVM */
  46                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
  47                 { "VMwareVMware", VIRTUALIZATION_VMWARE    },
  48                 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
  49                 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
  50                 /* https://wiki.freebsd.org/bhyve */
  51                 { "bhyve bhyve ", VIRTUALIZATION_BHYVE     },
  52                 { "QNXQVMBSQG",   VIRTUALIZATION_QNX       },
  53                 /* https://projectacrn.org */
  54                 { "ACRNACRNACRN", VIRTUALIZATION_ACRN      },
  55                 /* https://www.lockheedmartin.com/en-us/products/Hardened-Security-for-Intel-Processors.html */
  56                 { "SRESRESRESRE", VIRTUALIZATION_SRE       },
  57                 { "Apple VZ",     VIRTUALIZATION_APPLE     },
  58         };
  59
  60         uint32_t eax, ebx, ecx, edx;
  61         bool hypervisor;
  62
  63         /* http://lwn.net/Articles/301888/ */
  64
  65         /* First detect whether there is a hypervisor */
  66         if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) == 0)
  67                 return VIRTUALIZATION_NONE;
  68
  69         hypervisor = ecx & 0x80000000U;
  70
  71         if (hypervisor) {
  72                 union {
  73                         uint32_t sig32[3];
  74                         char text[13];
  75                 } sig = {};
  76
  77                 /* There is a hypervisor, see what it is */
  78                 __cpuid(0x40000000U, eax, ebx, ecx, edx);
  79
  80                 sig.sig32[0] = ebx;
  81                 sig.sig32[1] = ecx;
  82                 sig.sig32[2] = edx;
  83
  84                 log_debug("Virtualization found, CPUID=%s", sig.text);
  85
  86                 for (size_t i = 0; i < ELEMENTSOF(vm_table); i++)
  87                         if (memcmp_nn(sig.text, sizeof(sig.text),
  88                                       vm_table[i].sig, sizeof(vm_table[i].sig)) == 0)
  89                                 return vm_table[i].id;
  90
  91                 log_debug("Unknown virtualization with CPUID=%s. Add to vm_table[]?", sig.text);
  92                 return VIRTUALIZATION_VM_OTHER;
  93         }
  94 #endif
  95         log_debug("No virtualization found in CPUID");
  96
  97         return VIRTUALIZATION_NONE;
  98 }
  99
 100 static Virtualization detect_vm_device_tree(void) {
 101 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
 102         _cleanup_free_ char *hvtype = NULL;
 103         int r;
 104
 105         r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
 106         if (r == -ENOENT) {
 107                 _cleanup_closedir_ DIR *dir = NULL;
 108                 _cleanup_free_ char *compat = NULL;
 109
 110                 if (access("/proc/device-tree/ibm,partition-name", F_OK) == 0 &&
 111                     access("/proc/device-tree/hmc-managed?", F_OK) == 0 &&
 112                     access("/proc/device-tree/chosen/qemu,graphic-width", F_OK) != 0)
 113                         return VIRTUALIZATION_POWERVM;
 114
 115                 dir = opendir("/proc/device-tree");
 116                 if (!dir) {
 117                         if (errno == ENOENT) {
 118                                 log_debug_errno(errno, "/proc/device-tree: %m");
 119                                 return VIRTUALIZATION_NONE;
 120                         }
 121                         return -errno;
 122                 }
 123
 124                 FOREACH_DIRENT(de, dir, return -errno)
 125                         if (strstr(de->d_name, "fw-cfg")) {
 126                                 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", de->d_name);
 127                                 return VIRTUALIZATION_QEMU;
 128                         }
 129
 130                 r = read_one_line_file("/proc/device-tree/compatible", &compat);
 131                 if (r < 0 && r != -ENOENT)
 132                         return r;
 133                 if (r >= 0 && streq(compat, "qemu,pseries")) {
 134                         log_debug("Virtualization %s found in /proc/device-tree/compatible", compat);
 135                         return VIRTUALIZATION_QEMU;
 136                 }
 137
 138                 log_debug("No virtualization found in /proc/device-tree/*");
 139                 return VIRTUALIZATION_NONE;
 140         } else if (r < 0)
 141                 return r;
 142
 143         log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
 144         if (streq(hvtype, "linux,kvm"))
 145                 return VIRTUALIZATION_KVM;
 146         else if (strstr(hvtype, "xen"))
 147                 return VIRTUALIZATION_XEN;
 148         else if (strstr(hvtype, "vmware"))
 149                 return VIRTUALIZATION_VMWARE;
 150         else
 151                 return VIRTUALIZATION_VM_OTHER;
 152 #else
 153         log_debug("This platform does not support /proc/device-tree");
 154         return VIRTUALIZATION_NONE;
 155 #endif
 156 }
 157
 158 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || defined(__loongarch_lp64)
 159 static Virtualization detect_vm_dmi_vendor(void) {
 160         static const char* const dmi_vendors[] = {
 161                 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
 162                 "/sys/class/dmi/id/sys_vendor",
 163                 "/sys/class/dmi/id/board_vendor",
 164                 "/sys/class/dmi/id/bios_vendor",
 165                 "/sys/class/dmi/id/product_version", /* For Hyper-V VMs test */
 166                 NULL
 167         };
 168
 169         static const struct {
 170                 const char *vendor;
 171                 Virtualization id;
 172         } dmi_vendor_table[] = {
 173                 { "KVM",                   VIRTUALIZATION_KVM       },
 174                 { "OpenStack",             VIRTUALIZATION_KVM       }, /* Detect OpenStack instance as KVM in non x86 architecture */
 175                 { "KubeVirt",              VIRTUALIZATION_KVM       }, /* Detect KubeVirt instance as KVM in non x86 architecture */
 176                 { "Amazon EC2",            VIRTUALIZATION_AMAZON    },
 177                 { "QEMU",                  VIRTUALIZATION_QEMU      },
 178                 { "VMware",                VIRTUALIZATION_VMWARE    }, /* https://kb.vmware.com/s/article/1009458 */
 179                 { "VMW",                   VIRTUALIZATION_VMWARE    },
 180                 { "innotek GmbH",          VIRTUALIZATION_ORACLE    },
 181                 { "VirtualBox",            VIRTUALIZATION_ORACLE    },
 182                 { "Xen",                   VIRTUALIZATION_XEN       },
 183                 { "Bochs",                 VIRTUALIZATION_BOCHS     },
 184                 { "Parallels",             VIRTUALIZATION_PARALLELS },
 185                 /* https://wiki.freebsd.org/bhyve */
 186                 { "BHYVE",                 VIRTUALIZATION_BHYVE     },
 187                 { "Hyper-V",               VIRTUALIZATION_MICROSOFT },
 188                 { "Apple Virtualization",  VIRTUALIZATION_APPLE     },
 189                 { "Google Compute Engine", VIRTUALIZATION_GOOGLE    }, /* https://cloud.google.com/run/docs/container-contract#sandbox */
 190         };
 191         int r;
 192
 193         STRV_FOREACH(vendor, dmi_vendors) {
 194                 _cleanup_free_ char *s = NULL;
 195
 196                 r = read_one_line_file(*vendor, &s);
 197                 if (r < 0) {
 198                         if (r == -ENOENT)
 199                                 continue;
 200
 201                         return r;
 202                 }
 203
 204                 for (size_t i = 0; i < ELEMENTSOF(dmi_vendor_table); i++)
 205                         if (startswith(s, dmi_vendor_table[i].vendor)) {
 206                                 log_debug("Virtualization %s found in DMI (%s)", s, *vendor);
 207                                 return dmi_vendor_table[i].id;
 208                         }
 209         }
 210         log_debug("No virtualization found in DMI vendor table.");
 211         return VIRTUALIZATION_NONE;
 212 }
 213
 214 static int detect_vm_smbios(void) {
 215         /* The SMBIOS BIOS Characteristics Extension Byte 2 (Section 2.1.2.2 of
 216          * https://www.dmtf.org/sites/default/files/standards/documents/DSP0134_3.4.0.pdf), specifies that
 217          * the 4th bit being set indicates a VM. The BIOS Characteristics table is exposed via the kernel in
 218          * /sys/firmware/dmi/entries/0-0. Note that in the general case, this bit being unset should not
 219          * imply that the system is running on bare-metal.  For example, QEMU 3.1.0 (with or without KVM)
 220          * with SeaBIOS does not set this bit. */
 221         _cleanup_free_ char *s = NULL;
 222         size_t readsize;
 223         int r;
 224
 225         r = read_full_virtual_file("/sys/firmware/dmi/entries/0-0/raw", &s, &readsize);
 226         if (r < 0) {
 227                 log_debug_errno(r, "Unable to read /sys/firmware/dmi/entries/0-0/raw, "
 228                                 "using the virtualization information found in DMI vendor table, ignoring: %m");
 229                 return SMBIOS_VM_BIT_UNKNOWN;
 230         }
 231         if (readsize < 20 || s[1] < 20) {
 232                 /* The spec indicates that byte 1 contains the size of the table, 0x12 + the number of
 233                  * extension bytes. The data we're interested in is in extension byte 2, which would be at
 234                  * 0x13. If we didn't read that much data, or if the BIOS indicates that we don't have that
 235                  * much data, we don't infer anything from the SMBIOS. */
 236                 log_debug("Only read %zu bytes from /sys/firmware/dmi/entries/0-0/raw (expected 20). "
 237                           "Using the virtualization information found in DMI vendor table.", readsize);
 238                 return SMBIOS_VM_BIT_UNKNOWN;
 239         }
 240
 241         uint8_t byte = (uint8_t) s[19];
 242         if (byte & (1U<<4)) {
 243                 log_debug("DMI BIOS Extension table indicates virtualization.");
 244                 return SMBIOS_VM_BIT_SET;
 245         }
 246         log_debug("DMI BIOS Extension table does not indicate virtualization.");
 247         return SMBIOS_VM_BIT_UNSET;
 248 }
 249 #endif /* defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || defined(__loongarch_lp64) */
 250
 251 static Virtualization detect_vm_dmi(void) {
 252 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || defined(__loongarch_lp64)
 253
 254         int r;
 255         r = detect_vm_dmi_vendor();
 256
 257         /* The DMI vendor tables in /sys/class/dmi/id don't help us distinguish between Amazon EC2
 258          * virtual machines and bare-metal instances, so we need to look at SMBIOS. */
 259         if (r == VIRTUALIZATION_AMAZON) {
 260                 switch (detect_vm_smbios()) {
 261                 case SMBIOS_VM_BIT_SET:
 262                         return VIRTUALIZATION_AMAZON;
 263                 case SMBIOS_VM_BIT_UNSET:
 264                         return VIRTUALIZATION_NONE;
 265                 case SMBIOS_VM_BIT_UNKNOWN: {
 266                         /* The DMI information we are after is only accessible to the root user,
 267                          * so we fallback to using the product name which is less restricted
 268                          * to distinguish metal systems from virtualized instances */
 269                         _cleanup_free_ char *s = NULL;
 270                         const char *e;
 271
 272                         r = read_full_virtual_file("/sys/class/dmi/id/product_name", &s, NULL);
 273                         /* In EC2, virtualized is much more common than metal, so if for some reason
 274                          * we fail to read the DMI data, assume we are virtualized. */
 275                         if (r < 0) {
 276                                 log_debug_errno(r, "Can't read /sys/class/dmi/id/product_name,"
 277                                                 " assuming virtualized: %m");
 278                                 return VIRTUALIZATION_AMAZON;
 279                         }
 280                         e = strstrafter(truncate_nl(s), ".metal");
 281                         if (e && IN_SET(*e, 0, '-')) {
 282                                 log_debug("DMI product name has '.metal', assuming no virtualization");
 283                                 return VIRTUALIZATION_NONE;
 284                         } else
 285                                 return VIRTUALIZATION_AMAZON;
 286                 }
 287                 default:
 288                         assert_not_reached();
 289               }
 290         }
 291
 292         /* If we haven't identified a VM, but the firmware indicates that there is one, indicate as much. We
 293          * have no further information about what it is. */
 294         if (r == VIRTUALIZATION_NONE && detect_vm_smbios() == SMBIOS_VM_BIT_SET)
 295                 return VIRTUALIZATION_VM_OTHER;
 296         return r;
 297 #else
 298         return VIRTUALIZATION_NONE;
 299 #endif
 300 }
 301
 302 #define XENFEAT_dom0 11 /* xen/include/public/features.h */
 303 #define PATH_FEATURES "/sys/hypervisor/properties/features"
 304 /* Returns -errno, or 0 for domU, or 1 for dom0 */
 305 static int detect_vm_xen_dom0(void) {
 306         _cleanup_free_ char *domcap = NULL;
 307         int r;
 308
 309         r = read_one_line_file(PATH_FEATURES, &domcap);
 310         if (r < 0 && r != -ENOENT)
 311                 return r;
 312         if (r >= 0) {
 313                 unsigned long features;
 314
 315                 /* Here, we need to use sscanf() instead of safe_atoul()
 316                  * as the string lacks the leading "0x". */
 317                 r = sscanf(domcap, "%lx", &features);
 318                 if (r == 1) {
 319                         r = !!(features & (1U << XENFEAT_dom0));
 320                         log_debug("Virtualization XEN, found %s with value %08lx, "
 321                                   "XENFEAT_dom0 (indicating the 'hardware domain') is%s set.",
 322                                   PATH_FEATURES, features, r ? "" : " not");
 323                         return r;
 324                 }
 325                 log_debug("Virtualization XEN, found %s, unhandled content '%s'",
 326                           PATH_FEATURES, domcap);
 327         }
 328
 329         r = read_one_line_file("/proc/xen/capabilities", &domcap);
 330         if (r == -ENOENT) {
 331                 log_debug("Virtualization XEN because /proc/xen/capabilities does not exist");
 332                 return 0;
 333         }
 334         if (r < 0)
 335                 return r;
 336
 337         for (const char *i = domcap;;) {
 338                 _cleanup_free_ char *cap = NULL;
 339
 340                 r = extract_first_word(&i, &cap, ",", 0);
 341                 if (r < 0)
 342                         return r;
 343                 if (r == 0) {
 344                         log_debug("Virtualization XEN DomU found (/proc/xen/capabilities)");
 345                         return 0;
 346                 }
 347
 348                 if (streq(cap, "control_d")) {
 349                         log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
 350                         return 1;
 351                 }
 352         }
 353 }
 354
 355 static Virtualization detect_vm_xen(void) {
 356         /* The presence of /proc/xen indicates some form of a Xen domain
 357            The check for Dom0 is handled outside this function */
 358         if (access("/proc/xen", F_OK) < 0) {
 359                 log_debug("Virtualization XEN not found, /proc/xen does not exist");
 360                 return VIRTUALIZATION_NONE;
 361         }
 362         log_debug("Virtualization XEN found (/proc/xen exists)");
 363         return VIRTUALIZATION_XEN;
 364 }
 365
 366 static Virtualization detect_vm_hypervisor(void) {
 367         _cleanup_free_ char *hvtype = NULL;
 368         int r;
 369
 370         r = read_one_line_file("/sys/hypervisor/type", &hvtype);
 371         if (r == -ENOENT)
 372                 return VIRTUALIZATION_NONE;
 373         if (r < 0)
 374                 return r;
 375
 376         log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
 377
 378         if (streq(hvtype, "xen"))
 379                 return VIRTUALIZATION_XEN;
 380         else
 381                 return VIRTUALIZATION_VM_OTHER;
 382 }
 383
 384 static Virtualization detect_vm_uml(void) {
 385         _cleanup_fclose_ FILE *f = NULL;
 386         int r;
 387
 388         /* Detect User-Mode Linux by reading /proc/cpuinfo */
 389         f = fopen("/proc/cpuinfo", "re");
 390         if (!f) {
 391                 if (errno == ENOENT) {
 392                         log_debug("/proc/cpuinfo not found, assuming no UML virtualization.");
 393                         return VIRTUALIZATION_NONE;
 394                 }
 395                 return -errno;
 396         }
 397
 398         for (;;) {
 399                 _cleanup_free_ char *line = NULL;
 400                 const char *t;
 401
 402                 r = read_line(f, LONG_LINE_MAX, &line);
 403                 if (r < 0)
 404                         return r;
 405                 if (r == 0)
 406                         break;
 407
 408                 t = startswith(line, "vendor_id\t: ");
 409                 if (t) {
 410                         if (startswith(t, "User Mode Linux")) {
 411                                 log_debug("UML virtualization found in /proc/cpuinfo");
 412                                 return VIRTUALIZATION_UML;
 413                         }
 414
 415                         break;
 416                 }
 417         }
 418
 419         log_debug("UML virtualization not found in /proc/cpuinfo.");
 420         return VIRTUALIZATION_NONE;
 421 }
 422
 423 static Virtualization detect_vm_zvm(void) {
 424
 425 #if defined(__s390__)
 426         _cleanup_free_ char *t = NULL;
 427         int r;
 428
 429         r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
 430         if (r == -ENOENT)
 431                 return VIRTUALIZATION_NONE;
 432         if (r < 0)
 433                 return r;
 434
 435         log_debug("Virtualization %s found in /proc/sysinfo", t);
 436         if (streq(t, "z/VM"))
 437                 return VIRTUALIZATION_ZVM;
 438         else
 439                 return VIRTUALIZATION_KVM;
 440 #else
 441         log_debug("This platform does not support /proc/sysinfo");
 442         return VIRTUALIZATION_NONE;
 443 #endif
 444 }
 445
 446 /* Returns a short identifier for the various VM implementations */
 447 Virtualization detect_vm(void) {
 448         static thread_local Virtualization cached_found = _VIRTUALIZATION_INVALID;
 449         bool other = false;
 450         int xen_dom0 = 0;
 451         Virtualization v, dmi;
 452
 453         if (cached_found >= 0)
 454                 return cached_found;
 455
 456         /* We have to use the correct order here:
 457          *
 458          * → First, try to detect Oracle Virtualbox, Amazon EC2 Nitro, Parallels, and Google Compute Engine, even if they use KVM,
 459          *   as well as Xen even if it cloaks as Microsoft Hyper-V. Attempt to detect uml at this stage also
 460          *   since it runs as a user-process nested inside other VMs. Also check for Xen now, because Xen PV
 461          *   mode does not override CPUID when nested inside another hypervisor.
 462          *
 463          * → Second, try to detect from CPUID, this will report KVM for whatever software is used even if
 464          *   info in DMI is overwritten.
 465          *
 466          * → Third, try to detect from DMI. */
 467
 468         dmi = detect_vm_dmi();
 469         if (IN_SET(dmi,
 470                    VIRTUALIZATION_ORACLE,
 471                    VIRTUALIZATION_XEN,
 472                    VIRTUALIZATION_AMAZON,
 473                    VIRTUALIZATION_PARALLELS,
 474                    VIRTUALIZATION_GOOGLE)) {
 475                 v = dmi;
 476                 goto finish;
 477         }
 478
 479         /* Detect UML */
 480         v = detect_vm_uml();
 481         if (v < 0)
 482                 return v;
 483         if (v != VIRTUALIZATION_NONE)
 484                 goto finish;
 485
 486         /* Detect Xen */
 487         v = detect_vm_xen();
 488         if (v < 0)
 489                 return v;
 490         if (v == VIRTUALIZATION_XEN) {
 491                  /* If we are Dom0, then we expect to not report as a VM. However, as we might be nested
 492                   * inside another hypervisor which can be detected via the CPUID check, wait to report this
 493                   * until after the CPUID check. */
 494                 xen_dom0 = detect_vm_xen_dom0();
 495                 if (xen_dom0 < 0)
 496                         return xen_dom0;
 497                 if (xen_dom0 == 0)
 498                         goto finish;
 499         } else if (v != VIRTUALIZATION_NONE)
 500                 assert_not_reached();
 501
 502         /* Detect from CPUID */
 503         v = detect_vm_cpuid();
 504         if (v < 0)
 505                 return v;
 506         if (v == VIRTUALIZATION_VM_OTHER)
 507                 other = true;
 508         else if (v != VIRTUALIZATION_NONE)
 509                 goto finish;
 510
 511         /* If we are in Dom0 and have not yet finished, finish with the result of detect_vm_cpuid */
 512         if (xen_dom0 > 0)
 513                 goto finish;
 514
 515         /* Now, let's get back to DMI */
 516         if (dmi < 0)
 517                 return dmi;
 518         if (dmi == VIRTUALIZATION_VM_OTHER)
 519                 other = true;
 520         else if (dmi != VIRTUALIZATION_NONE) {
 521                 v = dmi;
 522                 goto finish;
 523         }
 524
 525         /* Check high-level hypervisor sysfs file */
 526         v = detect_vm_hypervisor();
 527         if (v < 0)
 528                 return v;
 529         if (v == VIRTUALIZATION_VM_OTHER)
 530                 other = true;
 531         else if (v != VIRTUALIZATION_NONE)
 532                 goto finish;
 533
 534         v = detect_vm_device_tree();
 535         if (v < 0)
 536                 return v;
 537         if (v == VIRTUALIZATION_VM_OTHER)
 538                 other = true;
 539         else if (v != VIRTUALIZATION_NONE)
 540                 goto finish;
 541
 542         v = detect_vm_zvm();
 543         if (v < 0)
 544                 return v;
 545
 546 finish:
 547         if (v == VIRTUALIZATION_NONE && other)
 548                 v = VIRTUALIZATION_VM_OTHER;
 549
 550         cached_found = v;
 551         log_debug("Found VM virtualization %s", virtualization_to_string(v));
 552         return v;
 553 }
 554
 555 static const char *const container_table[_VIRTUALIZATION_MAX] = {
 556         [VIRTUALIZATION_LXC]            = "lxc",
 557         [VIRTUALIZATION_LXC_LIBVIRT]    = "lxc-libvirt",
 558         [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
 559         [VIRTUALIZATION_DOCKER]         = "docker",
 560         [VIRTUALIZATION_PODMAN]         = "podman",
 561         [VIRTUALIZATION_RKT]            = "rkt",
 562         [VIRTUALIZATION_WSL]            = "wsl",
 563         [VIRTUALIZATION_PROOT]          = "proot",
 564         [VIRTUALIZATION_POUCH]          = "pouch",
 565 };
 566
 567 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_FROM_STRING(container, int);
 568
 569 static int running_in_cgroupns(void) {
 570         int r;
 571
 572         if (!cg_ns_supported())
 573                 return false;
 574
 575         r = cg_all_unified();
 576         if (r < 0)
 577                 return r;
 578
 579         if (r) {
 580                 /* cgroup v2 */
 581
 582                 r = access("/sys/fs/cgroup/cgroup.events", F_OK);
 583                 if (r < 0) {
 584                         if (errno != ENOENT)
 585                                 return -errno;
 586                         /* All kernel versions have cgroup.events in nested cgroups. */
 587                         return false;
 588                 }
 589
 590                 /* There's no cgroup.type in the root cgroup, and future kernel versions
 591                  * are unlikely to add it since cgroup.type is something that makes no sense
 592                  * whatsoever in the root cgroup. */
 593                 r = access("/sys/fs/cgroup/cgroup.type", F_OK);
 594                 if (r == 0)
 595                         return true;
 596                 if (r < 0 && errno != ENOENT)
 597                         return -errno;
 598
 599                 /* On older kernel versions, there's no cgroup.type */
 600                 r = access("/sys/kernel/cgroup/features", F_OK);
 601                 if (r < 0) {
 602                         if (errno != ENOENT)
 603                                 return -errno;
 604                         /* This is an old kernel that we know for sure has cgroup.events
 605                          * only in nested cgroups. */
 606                         return true;
 607                 }
 608
 609                 /* This is a recent kernel, and cgroup.type doesn't exist, so we must be
 610                  * in the root cgroup. */
 611                 return false;
 612         } else {
 613                 /* cgroup v1 */
 614
 615                 /* If systemd controller is not mounted, do not even bother. */
 616                 r = access("/sys/fs/cgroup/systemd", F_OK);
 617                 if (r < 0) {
 618                         if (errno != ENOENT)
 619                                 return -errno;
 620                         return false;
 621                 }
 622
 623                 /* release_agent only exists in the root cgroup. */
 624                 r = access("/sys/fs/cgroup/systemd/release_agent", F_OK);
 625                 if (r < 0) {
 626                         if (errno != ENOENT)
 627                                 return -errno;
 628                         return true;
 629                 }
 630
 631                 return false;
 632         }
 633 }
 634
 635 static Virtualization detect_container_files(void) {
 636         static const struct {
 637                 const char *file_path;
 638                 Virtualization id;
 639         } container_file_table[] = {
 640                 /* https://github.com/containers/podman/issues/6192 */
 641                 /* https://github.com/containers/podman/issues/3586#issuecomment-661918679 */
 642                 { "/run/.containerenv", VIRTUALIZATION_PODMAN },
 643                 /* https://github.com/moby/moby/issues/18355 */
 644                 /* Docker must be the last in this table, see below. */
 645                 { "/.dockerenv",        VIRTUALIZATION_DOCKER },
 646         };
 647
 648         for (size_t i = 0; i < ELEMENTSOF(container_file_table); i++) {
 649                 if (access(container_file_table[i].file_path, F_OK) >= 0)
 650                         return container_file_table[i].id;
 651
 652                 if (errno != ENOENT)
 653                         log_debug_errno(errno,
 654                                         "Checking if %s exists failed, ignoring: %m",
 655                                         container_file_table[i].file_path);
 656         }
 657
 658         return VIRTUALIZATION_NONE;
 659 }
 660
 661 Virtualization detect_container(void) {
 662         static thread_local Virtualization cached_found = _VIRTUALIZATION_INVALID;
 663         _cleanup_free_ char *m = NULL, *o = NULL, *p = NULL;
 664         const char *e = NULL;
 665         Virtualization v;
 666         int r;
 667
 668         if (cached_found >= 0)
 669                 return cached_found;
 670
 671         /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
 672         if (access("/proc/vz", F_OK) < 0) {
 673                 if (errno != ENOENT)
 674                         log_debug_errno(errno, "Failed to check if /proc/vz exists, ignoring: %m");
 675         } else if (access("/proc/bc", F_OK) < 0) {
 676                 if (errno == ENOENT) {
 677                         v = VIRTUALIZATION_OPENVZ;
 678                         goto finish;
 679                 }
 680
 681                 log_debug_errno(errno, "Failed to check if /proc/bc exists, ignoring: %m");
 682         }
 683
 684         /* "Official" way of detecting WSL https://github.com/Microsoft/WSL/issues/423#issuecomment-221627364 */
 685         r = read_one_line_file("/proc/sys/kernel/osrelease", &o);
 686         if (r < 0)
 687                 log_debug_errno(r, "Failed to read /proc/sys/kernel/osrelease, ignoring: %m");
 688         else if (strstr(o, "Microsoft") || strstr(o, "WSL")) {
 689                 v = VIRTUALIZATION_WSL;
 690                 goto finish;
 691         }
 692
 693         /* proot doesn't use PID namespacing, so we can just check if we have a matching tracer for this
 694          * invocation without worrying about it being elsewhere.
 695          */
 696         r = get_proc_field("/proc/self/status", "TracerPid", WHITESPACE, &p);
 697         if (r < 0)
 698                 log_debug_errno(r, "Failed to read our own trace PID, ignoring: %m");
 699         else if (!streq(p, "0")) {
 700                 pid_t ptrace_pid;
 701
 702                 r = parse_pid(p, &ptrace_pid);
 703                 if (r < 0)
 704                         log_debug_errno(r, "Failed to parse our own tracer PID, ignoring: %m");
 705                 else {
 706                         _cleanup_free_ char *ptrace_comm = NULL;
 707                         const char *pf;
 708
 709                         pf = procfs_file_alloca(ptrace_pid, "comm");
 710                         r = read_one_line_file(pf, &ptrace_comm);
 711                         if (r < 0)
 712                                 log_debug_errno(r, "Failed to read %s, ignoring: %m", pf);
 713                         else if (startswith(ptrace_comm, "proot")) {
 714                                 v = VIRTUALIZATION_PROOT;
 715                                 goto finish;
 716                         }
 717                 }
 718         }
 719
 720         /* The container manager might have placed this in the /run/host/ hierarchy for us, which is best
 721          * because we can be consumed just like that, without special privileges. */
 722         r = read_one_line_file("/run/host/container-manager", &m);
 723         if (r > 0) {
 724                 e = m;
 725                 goto translate_name;
 726         }
 727         if (!IN_SET(r, -ENOENT, 0))
 728                 return log_debug_errno(r, "Failed to read /run/host/container-manager: %m");
 729
 730         if (getpid_cached() == 1) {
 731                 /* If we are PID 1 we can just check our own environment variable, and that's authoritative.
 732                  * We distinguish three cases:
 733                  * - the variable is not defined → we jump to other checks
 734                  * - the variable is defined to an empty value → we are not in a container
 735                  * - anything else → some container, either one of the known ones or "container-other"
 736                  */
 737                 e = getenv("container");
 738                 if (!e)
 739                         goto check_files;
 740                 if (isempty(e)) {
 741                         v = VIRTUALIZATION_NONE;
 742                         goto finish;
 743                 }
 744
 745                 goto translate_name;
 746         }
 747
 748         /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
 749          * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
 750         r = read_one_line_file("/run/systemd/container", &m);
 751         if (r > 0) {
 752                 e = m;
 753                 goto translate_name;
 754         }
 755         if (!IN_SET(r, -ENOENT, 0))
 756                 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
 757
 758         /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
 759         r = getenv_for_pid(1, "container", &m);
 760         if (r > 0) {
 761                 e = m;
 762                 goto translate_name;
 763         }
 764         if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
 765                 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
 766
 767 check_files:
 768         /* Check for existence of some well-known files. We only do this after checking
 769          * for other specific container managers, otherwise we risk mistaking another
 770          * container manager for Docker: the /.dockerenv file could inadvertently end up
 771          * in a file system image. */
 772         v = detect_container_files();
 773         if (v < 0)
 774                 return v;
 775         if (v != VIRTUALIZATION_NONE)
 776                 goto finish;
 777
 778         r = running_in_cgroupns();
 779         if (r > 0) {
 780                 v = VIRTUALIZATION_CONTAINER_OTHER;
 781                 goto finish;
 782         }
 783         if (r < 0)
 784                 log_debug_errno(r, "Failed to detect cgroup namespace: %m");
 785
 786         /* If none of that worked, give up, assume no container manager. */
 787         v = VIRTUALIZATION_NONE;
 788         goto finish;
 789
 790 translate_name:
 791         if (streq(e, "oci")) {
 792                 /* Some images hardcode container=oci, but OCI is not a specific container manager.
 793                  * Try to detect one based on well-known files. */
 794                 v = detect_container_files();
 795                 if (v == VIRTUALIZATION_NONE)
 796                         v = VIRTUALIZATION_CONTAINER_OTHER;
 797                 goto finish;
 798         }
 799         v = container_from_string(e);
 800         if (v < 0)
 801                 v = VIRTUALIZATION_CONTAINER_OTHER;
 802
 803 finish:
 804         log_debug("Found container virtualization %s.", virtualization_to_string(v));
 805         cached_found = v;
 806         return v;
 807 }
 808
 809 Virtualization detect_virtualization(void) {
 810         int v;
 811
 812         v = detect_container();
 813         if (v != VIRTUALIZATION_NONE)
 814                 return v;
 815
 816         return detect_vm();
 817 }
 818
 819 static int userns_has_mapping(const char *name) {
 820         _cleanup_fclose_ FILE *f = NULL;
 821         uid_t base, shift, range;
 822         int r;
 823
 824         f = fopen(name, "re");
 825         if (!f) {
 826                 log_debug_errno(errno, "Failed to open %s: %m", name);
 827                 return errno == ENOENT ? false : -errno;
 828         }
 829
 830         r = uid_map_read_one(f, &base, &shift, &range);
 831         if (r == -ENOMSG) {
 832                 log_debug("%s is empty, we're in an uninitialized user namespace.", name);
 833                 return true;
 834         }
 835         if (r < 0)
 836                 return log_debug_errno(r, "Failed to read %s: %m", name);
 837
 838         if (base == 0 && shift == 0 && range == 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_count(7, 0, &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_GOOGLE]          = "google",
1053         [VIRTUALIZATION_VM_OTHER]        = "vm-other",
1054
1055         [VIRTUALIZATION_SYSTEMD_NSPAWN]  = "systemd-nspawn",
1056         [VIRTUALIZATION_LXC_LIBVIRT]     = "lxc-libvirt",
1057         [VIRTUALIZATION_LXC]             = "lxc",
1058         [VIRTUALIZATION_OPENVZ]          = "openvz",
1059         [VIRTUALIZATION_DOCKER]          = "docker",
1060         [VIRTUALIZATION_PODMAN]          = "podman",
1061         [VIRTUALIZATION_RKT]             = "rkt",
1062         [VIRTUALIZATION_WSL]             = "wsl",
1063         [VIRTUALIZATION_PROOT]           = "proot",
1064         [VIRTUALIZATION_POUCH]           = "pouch",
1065         [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
1066 };
1067
1068 DEFINE_STRING_TABLE_LOOKUP(virtualization, Virtualization);