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