src/basic/virt.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2011 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <stdint.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25
  26 #include "alloc-util.h"
  27 #include "dirent-util.h"
  28 #include "env-util.h"
  29 #include "fd-util.h"
  30 #include "fileio.h"
  31 #include "macro.h"
  32 #include "process-util.h"
  33 #include "stat-util.h"
  34 #include "string-table.h"
  35 #include "string-util.h"
  36 #include "virt.h"
  37
  38 #if 0 /// UNNEEDED by elogind
  39 static int detect_vm_cpuid(void) {
  40
  41         /* CPUID is an x86 specific interface. */
  42 #if defined(__i386__) || defined(__x86_64__)
  43
  44         static const struct {
  45                 const char *cpuid;
  46                 int id;
  47         } cpuid_vendor_table[] = {
  48                 { "XenVMMXenVMM", VIRTUALIZATION_XEN       },
  49                 { "KVMKVMKVM",    VIRTUALIZATION_KVM       },
  50                 { "TCGTCGTCGTCG", VIRTUALIZATION_QEMU      },
  51                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
  52                 { "VMwareVMware", VIRTUALIZATION_VMWARE    },
  53                 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
  54                 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
  55                 /* https://wiki.freebsd.org/bhyve */
  56                 { "bhyve bhyve ", VIRTUALIZATION_BHYVE     },
  57         };
  58
  59         uint32_t eax, ecx;
  60         bool hypervisor;
  61
  62         /* http://lwn.net/Articles/301888/ */
  63
  64 #if defined (__i386__)
  65 #define REG_a "eax"
  66 #define REG_b "ebx"
  67 #elif defined (__amd64__)
  68 #define REG_a "rax"
  69 #define REG_b "rbx"
  70 #endif
  71
  72         /* First detect whether there is a hypervisor */
  73         eax = 1;
  74         __asm__ __volatile__ (
  75                 /* ebx/rbx is being used for PIC! */
  76                 "  push %%"REG_b"         \n\t"
  77                 "  cpuid                  \n\t"
  78                 "  pop %%"REG_b"          \n\t"
  79
  80                 : "=a" (eax), "=c" (ecx)
  81                 : "0" (eax)
  82         );
  83
  84         hypervisor = !!(ecx & 0x80000000U);
  85
  86         if (hypervisor) {
  87                 union {
  88                         uint32_t sig32[3];
  89                         char text[13];
  90                 } sig = {};
  91                 unsigned j;
  92
  93                 /* There is a hypervisor, see what it is */
  94                 eax = 0x40000000U;
  95                 __asm__ __volatile__ (
  96                         /* ebx/rbx is being used for PIC! */
  97                         "  push %%"REG_b"         \n\t"
  98                         "  cpuid                  \n\t"
  99                         "  mov %%ebx, %1          \n\t"
 100                         "  pop %%"REG_b"          \n\t"
 101
 102                         : "=a" (eax), "=r" (sig.sig32[0]), "=c" (sig.sig32[1]), "=d" (sig.sig32[2])
 103                         : "0" (eax)
 104                 );
 105
 106                 log_debug("Virtualization found, CPUID=%s", sig.text);
 107
 108                 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
 109                         if (streq(sig.text, cpuid_vendor_table[j].cpuid))
 110                                 return cpuid_vendor_table[j].id;
 111
 112                 return VIRTUALIZATION_VM_OTHER;
 113         }
 114 #endif
 115         log_debug("No virtualization found in CPUID");
 116
 117         return VIRTUALIZATION_NONE;
 118 }
 119
 120 static int detect_vm_device_tree(void) {
 121 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
 122         _cleanup_free_ char *hvtype = NULL;
 123         int r;
 124
 125         r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
 126         if (r == -ENOENT) {
 127                 _cleanup_closedir_ DIR *dir = NULL;
 128                 struct dirent *dent;
 129
 130                 dir = opendir("/proc/device-tree");
 131                 if (!dir) {
 132                         if (errno == ENOENT) {
 133                                 log_debug_errno(errno, "/proc/device-tree: %m");
 134                                 return VIRTUALIZATION_NONE;
 135                         }
 136                         return -errno;
 137                 }
 138
 139                 FOREACH_DIRENT(dent, dir, return -errno)
 140                         if (strstr(dent->d_name, "fw-cfg")) {
 141                                 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
 142                                 return VIRTUALIZATION_QEMU;
 143                         }
 144
 145                 log_debug("No virtualization found in /proc/device-tree/*");
 146                 return VIRTUALIZATION_NONE;
 147         } else if (r < 0)
 148                 return r;
 149
 150         log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
 151         if (streq(hvtype, "linux,kvm"))
 152                 return VIRTUALIZATION_KVM;
 153         else if (strstr(hvtype, "xen"))
 154                 return VIRTUALIZATION_XEN;
 155         else
 156                 return VIRTUALIZATION_VM_OTHER;
 157 #else
 158         log_debug("This platform does not support /proc/device-tree");
 159         return VIRTUALIZATION_NONE;
 160 #endif
 161 }
 162
 163 static int detect_vm_dmi(void) {
 164 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
 165
 166         static const char *const dmi_vendors[] = {
 167                 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
 168                 "/sys/class/dmi/id/sys_vendor",
 169                 "/sys/class/dmi/id/board_vendor",
 170                 "/sys/class/dmi/id/bios_vendor"
 171         };
 172
 173         static const struct {
 174                 const char *vendor;
 175                 int id;
 176         } dmi_vendor_table[] = {
 177                 { "KVM",           VIRTUALIZATION_KVM       },
 178                 { "QEMU",          VIRTUALIZATION_QEMU      },
 179                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
 180                 { "VMware",        VIRTUALIZATION_VMWARE    },
 181                 { "VMW",           VIRTUALIZATION_VMWARE    },
 182                 { "innotek GmbH",  VIRTUALIZATION_ORACLE    },
 183                 { "Xen",           VIRTUALIZATION_XEN       },
 184                 { "Bochs",         VIRTUALIZATION_BOCHS     },
 185                 { "Parallels",     VIRTUALIZATION_PARALLELS },
 186                 /* https://wiki.freebsd.org/bhyve */
 187                 { "BHYVE",         VIRTUALIZATION_BHYVE     },
 188         };
 189         unsigned i;
 190         int r;
 191
 192         for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
 193                 _cleanup_free_ char *s = NULL;
 194                 unsigned j;
 195
 196                 r = read_one_line_file(dmi_vendors[i], &s);
 197                 if (r < 0) {
 198                         if (r == -ENOENT)
 199                                 continue;
 200
 201                         return r;
 202                 }
 203
 204
 205
 206                 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
 207                         if (startswith(s, dmi_vendor_table[j].vendor)) {
 208                                 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
 209                                 return dmi_vendor_table[j].id;
 210                         }
 211         }
 212 #endif
 213
 214         log_debug("No virtualization found in DMI");
 215
 216         return VIRTUALIZATION_NONE;
 217 }
 218
 219 static int detect_vm_xen(void) {
 220
 221         /* Check for Dom0 will be executed later in detect_vm_xen_dom0
 222            The presence of /proc/xen indicates some form of a Xen domain */
 223         if (access("/proc/xen", F_OK) < 0) {
 224                 log_debug("Virtualization XEN not found, /proc/xen does not exist");
 225                 return VIRTUALIZATION_NONE;
 226         }
 227
 228         log_debug("Virtualization XEN found (/proc/xen exists)");
 229         return VIRTUALIZATION_XEN;
 230 }
 231
 232 static bool detect_vm_xen_dom0(void) {
 233         _cleanup_free_ char *domcap = NULL;
 234         char *cap, *i;
 235         int r;
 236
 237         r = read_one_line_file("/proc/xen/capabilities", &domcap);
 238         if (r == -ENOENT) {
 239                 log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
 240                 return false;
 241         }
 242         if (r < 0)
 243                 return r;
 244
 245         i = domcap;
 246         while ((cap = strsep(&i, ",")))
 247                 if (streq(cap, "control_d"))
 248                         break;
 249         if (!cap) {
 250                 log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
 251                 return false;
 252         }
 253
 254         log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
 255         return true;
 256 }
 257
 258 static int detect_vm_hypervisor(void) {
 259         _cleanup_free_ char *hvtype = NULL;
 260         int r;
 261
 262         r = read_one_line_file("/sys/hypervisor/type", &hvtype);
 263         if (r == -ENOENT)
 264                 return VIRTUALIZATION_NONE;
 265         if (r < 0)
 266                 return r;
 267
 268         log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
 269
 270         if (streq(hvtype, "xen"))
 271                 return VIRTUALIZATION_XEN;
 272         else
 273                 return VIRTUALIZATION_VM_OTHER;
 274 }
 275
 276 static int detect_vm_uml(void) {
 277         _cleanup_free_ char *cpuinfo_contents = NULL;
 278         int r;
 279
 280         /* Detect User-Mode Linux by reading /proc/cpuinfo */
 281         r = read_full_file("/proc/cpuinfo", &cpuinfo_contents, NULL);
 282         if (r < 0)
 283                 return r;
 284
 285         if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
 286                 log_debug("UML virtualization found in /proc/cpuinfo");
 287                 return VIRTUALIZATION_UML;
 288         }
 289
 290         log_debug("No virtualization found in /proc/cpuinfo.");
 291         return VIRTUALIZATION_NONE;
 292 }
 293
 294 static int detect_vm_zvm(void) {
 295
 296 #if defined(__s390__)
 297         _cleanup_free_ char *t = NULL;
 298         int r;
 299
 300         r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
 301         if (r == -ENOENT)
 302                 return VIRTUALIZATION_NONE;
 303         if (r < 0)
 304                 return r;
 305
 306         log_debug("Virtualization %s found in /proc/sysinfo", t);
 307         if (streq(t, "z/VM"))
 308                 return VIRTUALIZATION_ZVM;
 309         else
 310                 return VIRTUALIZATION_KVM;
 311 #else
 312         log_debug("This platform does not support /proc/sysinfo");
 313         return VIRTUALIZATION_NONE;
 314 #endif
 315 }
 316
 317 /* Returns a short identifier for the various VM implementations */
 318 int detect_vm(void) {
 319         static thread_local int cached_found = _VIRTUALIZATION_INVALID;
 320         int r, dmi;
 321         bool other = false;
 322
 323         if (cached_found >= 0)
 324                 return cached_found;
 325
 326         /* We have to use the correct order here:
 327          *
 328          * -> First try to detect Oracle Virtualbox, even if it uses KVM.
 329          * -> Second try to detect from cpuid, this will report KVM for
 330          *    whatever software is used even if info in dmi is overwritten.
 331          * -> Third try to detect from dmi. */
 332
 333         dmi = detect_vm_dmi();
 334         if (dmi == VIRTUALIZATION_ORACLE) {
 335                 r = dmi;
 336                 goto finish;
 337         }
 338
 339         r = detect_vm_cpuid();
 340         if (r < 0)
 341                 return r;
 342         if (r != VIRTUALIZATION_NONE) {
 343                 if (r == VIRTUALIZATION_VM_OTHER)
 344                         other = true;
 345                 else
 346                         goto finish;
 347         }
 348
 349         r = dmi;
 350         if (r < 0)
 351                 return r;
 352         if (r != VIRTUALIZATION_NONE) {
 353                 if (r == VIRTUALIZATION_VM_OTHER)
 354                         other = true;
 355                 else
 356                         goto finish;
 357         }
 358
 359         /* x86 xen will most likely be detected by cpuid. If not (most likely
 360          * because we're not an x86 guest), then we should try the /proc/xen
 361          * directory next. If that's not found, then we check for the high-level
 362          * hypervisor sysfs file.
 363          */
 364
 365         r = detect_vm_xen();
 366         if (r < 0)
 367                 return r;
 368         if (r != VIRTUALIZATION_NONE) {
 369                 if (r == VIRTUALIZATION_VM_OTHER)
 370                         other = true;
 371                 else
 372                         goto finish;
 373         }
 374
 375         r = detect_vm_hypervisor();
 376         if (r < 0)
 377                 return r;
 378         if (r != VIRTUALIZATION_NONE) {
 379                 if (r == VIRTUALIZATION_VM_OTHER)
 380                         other = true;
 381                 else
 382                         goto finish;
 383         }
 384
 385         r = detect_vm_device_tree();
 386         if (r < 0)
 387                 return r;
 388         if (r != VIRTUALIZATION_NONE) {
 389                 if (r == VIRTUALIZATION_VM_OTHER)
 390                         other = true;
 391                 else
 392                         goto finish;
 393         }
 394
 395         r = detect_vm_uml();
 396         if (r < 0)
 397                 return r;
 398         if (r != VIRTUALIZATION_NONE) {
 399                 if (r == VIRTUALIZATION_VM_OTHER)
 400                         other = true;
 401                 else
 402                         goto finish;
 403         }
 404
 405         r = detect_vm_zvm();
 406         if (r < 0)
 407                 return r;
 408
 409 finish:
 410         /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
 411          * In order to detect the Dom0 as not virtualization we need to
 412          * double-check it */
 413         if (r == VIRTUALIZATION_XEN && detect_vm_xen_dom0())
 414                 r = VIRTUALIZATION_NONE;
 415         else if (r == VIRTUALIZATION_NONE && other)
 416                 r = VIRTUALIZATION_VM_OTHER;
 417
 418         cached_found = r;
 419         log_debug("Found VM virtualization %s", virtualization_to_string(r));
 420         return r;
 421 }
 422 #endif // 0
 423
 424 int detect_container(void) {
 425
 426         static const struct {
 427                 const char *value;
 428                 int id;
 429         } value_table[] = {
 430                 { "lxc",            VIRTUALIZATION_LXC            },
 431                 { "lxc-libvirt",    VIRTUALIZATION_LXC_LIBVIRT    },
 432                 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
 433                 { "docker",         VIRTUALIZATION_DOCKER         },
 434                 { "rkt",            VIRTUALIZATION_RKT            },
 435         };
 436
 437         static thread_local int cached_found = _VIRTUALIZATION_INVALID;
 438         _cleanup_free_ char *m = NULL;
 439         const char *e = NULL;
 440         unsigned j;
 441         int r;
 442
 443         if (cached_found >= 0)
 444                 return cached_found;
 445
 446         /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
 447         if (access("/proc/vz", F_OK) >= 0 &&
 448             access("/proc/bc", F_OK) < 0) {
 449                 r = VIRTUALIZATION_OPENVZ;
 450                 goto finish;
 451         }
 452
 453         if (getpid_cached() == 1) {
 454                 /* If we are PID 1 we can just check our own environment variable, and that's authoritative. */
 455
 456                 e = getenv("container");
 457                 if (isempty(e)) {
 458                         r = VIRTUALIZATION_NONE;
 459                         goto finish;
 460                 }
 461
 462                 goto translate_name;
 463         }
 464
 465         /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
 466          * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
 467         r = read_one_line_file("/run/systemd/container", &m);
 468         if (r >= 0) {
 469                 e = m;
 470                 goto translate_name;
 471         }
 472         if (r != -ENOENT)
 473                 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
 474
 475         /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
 476         r = getenv_for_pid(1, "container", &m);
 477         if (r > 0) {
 478                 e = m;
 479                 goto translate_name;
 480         }
 481         if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
 482                 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
 483
 484         /* Interestingly /proc/1/sched actually shows the host's PID for what we see as PID 1. Hence, if the PID shown
 485          * there is not 1, we know we are in a PID namespace. and hence a container. */
 486         r = read_one_line_file("/proc/1/sched", &m);
 487         if (r >= 0) {
 488                 const char *t;
 489
 490                 t = strrchr(m, '(');
 491                 if (!t)
 492                         return -EIO;
 493
 494                 if (!startswith(t, "(1,")) {
 495                         r = VIRTUALIZATION_CONTAINER_OTHER;
 496                         goto finish;
 497                 }
 498         } else if (r != -ENOENT)
 499                 return r;
 500
 501         /* If that didn't work, give up, assume no container manager. */
 502         r = VIRTUALIZATION_NONE;
 503         goto finish;
 504
 505 translate_name:
 506         for (j = 0; j < ELEMENTSOF(value_table); j++)
 507                 if (streq(e, value_table[j].value)) {
 508                         r = value_table[j].id;
 509                         goto finish;
 510                 }
 511
 512         r = VIRTUALIZATION_CONTAINER_OTHER;
 513
 514 finish:
 515         log_debug("Found container virtualization %s.", virtualization_to_string(r));
 516         cached_found = r;
 517         return r;
 518 }
 519
 520 #if 0 /// UNNEEDED by elogind
 521 int detect_virtualization(void) {
 522         int r;
 523
 524         r = detect_container();
 525         if (r == 0)
 526                 r = detect_vm();
 527
 528         return r;
 529 }
 530
 531 static int userns_has_mapping(const char *name) {
 532         _cleanup_fclose_ FILE *f = NULL;
 533         _cleanup_free_ char *buf = NULL;
 534         size_t n_allocated = 0;
 535         ssize_t n;
 536         uint32_t a, b, c;
 537         int r;
 538
 539         f = fopen(name, "re");
 540         if (!f) {
 541                 log_debug_errno(errno, "Failed to open %s: %m", name);
 542                 return errno == ENOENT ? false : -errno;
 543         }
 544
 545         n = getline(&buf, &n_allocated, f);
 546         if (n < 0) {
 547                 if (feof(f)) {
 548                         log_debug("%s is empty, we're in an uninitialized user namespace", name);
 549                         return true;
 550                 }
 551
 552                 return log_debug_errno(errno, "Failed to read %s: %m", name);
 553         }
 554
 555         r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
 556         if (r < 3)
 557                 return log_debug_errno(errno, "Failed to parse %s: %m", name);
 558
 559         if (a == 0 && b == 0 && c == UINT32_MAX) {
 560                 /* The kernel calls mappings_overlap() and does not allow overlaps */
 561                 log_debug("%s has a full 1:1 mapping", name);
 562                 return false;
 563         }
 564
 565         /* Anything else implies that we are in a user namespace */
 566         log_debug("Mapping found in %s, we're in a user namespace", name);
 567         return true;
 568 }
 569
 570 int running_in_userns(void) {
 571         _cleanup_free_ char *line = NULL;
 572         int r;
 573
 574         r = userns_has_mapping("/proc/self/uid_map");
 575         if (r != 0)
 576                 return r;
 577
 578         r = userns_has_mapping("/proc/self/gid_map");
 579         if (r != 0)
 580                 return r;
 581
 582         /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
 583          * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
 584          * also does not exist. We cannot distinguish those two cases, so assume that
 585          * we're running on a stripped-down recent kernel, rather than on an old one,
 586          * and if the file is not found, return false.
 587          */
 588         r = read_one_line_file("/proc/self/setgroups", &line);
 589         if (r < 0) {
 590                 log_debug_errno(r, "/proc/self/setgroups: %m");
 591                 return r == -ENOENT ? false : r;
 592         }
 593
 594         truncate_nl(line);
 595         r = streq(line, "deny");
 596         /* See user_namespaces(7) for a description of this "setgroups" contents. */
 597         log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
 598         return r;
 599 }
 600 #endif // 0
 601
 602 int running_in_chroot(void) {
 603         int ret;
 604
 605 #if 0 /// elogind does not allow to ignore chroots, we are never init!
 606         if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
 607                 return 0;
 608 #endif // 0
 609
 610         ret = files_same("/proc/1/root", "/", 0);
 611         if (ret < 0)
 612                 return ret;
 613
 614         return ret == 0;
 615 }
 616
 617 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
 618         [VIRTUALIZATION_NONE] = "none",
 619         [VIRTUALIZATION_KVM] = "kvm",
 620         [VIRTUALIZATION_QEMU] = "qemu",
 621         [VIRTUALIZATION_BOCHS] = "bochs",
 622         [VIRTUALIZATION_XEN] = "xen",
 623         [VIRTUALIZATION_UML] = "uml",
 624         [VIRTUALIZATION_VMWARE] = "vmware",
 625         [VIRTUALIZATION_ORACLE] = "oracle",
 626         [VIRTUALIZATION_MICROSOFT] = "microsoft",
 627         [VIRTUALIZATION_ZVM] = "zvm",
 628         [VIRTUALIZATION_PARALLELS] = "parallels",
 629         [VIRTUALIZATION_BHYVE] = "bhyve",
 630         [VIRTUALIZATION_VM_OTHER] = "vm-other",
 631
 632         [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
 633         [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
 634         [VIRTUALIZATION_LXC] = "lxc",
 635         [VIRTUALIZATION_OPENVZ] = "openvz",
 636         [VIRTUALIZATION_DOCKER] = "docker",
 637         [VIRTUALIZATION_RKT] = "rkt",
 638         [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
 639 };
 640
 641 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);