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 "fd-util.h"
  29 #include "fileio.h"
  30 #include "macro.h"
  31 #include "process-util.h"
  32 #include "stat-util.h"
  33 #include "string-table.h"
  34 #include "string-util.h"
  35 #include "virt.h"
  36
  37 #if 0 /// UNNEEDED by elogind
  38 static int detect_vm_cpuid(void) {
  39
  40         /* CPUID is an x86 specific interface. */
  41 #if defined(__i386__) || defined(__x86_64__)
  42
  43         static const struct {
  44                 const char *cpuid;
  45                 int id;
  46         } cpuid_vendor_table[] = {
  47                 { "XenVMMXenVMM", VIRTUALIZATION_XEN       },
  48                 { "KVMKVMKVM",    VIRTUALIZATION_KVM       },
  49                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
  50                 { "VMwareVMware", VIRTUALIZATION_VMWARE    },
  51                 /* https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs */
  52                 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
  53                 /* https://wiki.freebsd.org/bhyve */
  54                 { "bhyve bhyve ", VIRTUALIZATION_BHYVE     },
  55         };
  56
  57         uint32_t eax, ecx;
  58         bool hypervisor;
  59
  60         /* http://lwn.net/Articles/301888/ */
  61
  62 #if defined (__i386__)
  63 #define REG_a "eax"
  64 #define REG_b "ebx"
  65 #elif defined (__amd64__)
  66 #define REG_a "rax"
  67 #define REG_b "rbx"
  68 #endif
  69
  70         /* First detect whether there is a hypervisor */
  71         eax = 1;
  72         __asm__ __volatile__ (
  73                 /* ebx/rbx is being used for PIC! */
  74                 "  push %%"REG_b"         \n\t"
  75                 "  cpuid                  \n\t"
  76                 "  pop %%"REG_b"          \n\t"
  77
  78                 : "=a" (eax), "=c" (ecx)
  79                 : "0" (eax)
  80         );
  81
  82         hypervisor = !!(ecx & 0x80000000U);
  83
  84         if (hypervisor) {
  85                 union {
  86                         uint32_t sig32[3];
  87                         char text[13];
  88                 } sig = {};
  89                 unsigned j;
  90
  91                 /* There is a hypervisor, see what it is */
  92                 eax = 0x40000000U;
  93                 __asm__ __volatile__ (
  94                         /* ebx/rbx is being used for PIC! */
  95                         "  push %%"REG_b"         \n\t"
  96                         "  cpuid                  \n\t"
  97                         "  mov %%ebx, %1          \n\t"
  98                         "  pop %%"REG_b"          \n\t"
  99
 100                         : "=a" (eax), "=r" (sig.sig32[0]), "=c" (sig.sig32[1]), "=d" (sig.sig32[2])
 101                         : "0" (eax)
 102                 );
 103
 104                 log_debug("Virtualization found, CPUID=%s", sig.text);
 105
 106                 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
 107                         if (streq(sig.text, cpuid_vendor_table[j].cpuid))
 108                                 return cpuid_vendor_table[j].id;
 109
 110                 return VIRTUALIZATION_VM_OTHER;
 111         }
 112 #endif
 113         log_debug("No virtualization found in CPUID");
 114
 115         return VIRTUALIZATION_NONE;
 116 }
 117
 118 static int detect_vm_device_tree(void) {
 119 #if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
 120         _cleanup_free_ char *hvtype = NULL;
 121         int r;
 122
 123         r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
 124         if (r == -ENOENT) {
 125                 _cleanup_closedir_ DIR *dir = NULL;
 126                 struct dirent *dent;
 127
 128                 dir = opendir("/proc/device-tree");
 129                 if (!dir) {
 130                         if (errno == ENOENT) {
 131                                 log_debug_errno(errno, "/proc/device-tree: %m");
 132                                 return VIRTUALIZATION_NONE;
 133                         }
 134                         return -errno;
 135                 }
 136
 137                 FOREACH_DIRENT(dent, dir, return -errno)
 138                         if (strstr(dent->d_name, "fw-cfg")) {
 139                                 log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
 140                                 return VIRTUALIZATION_QEMU;
 141                         }
 142
 143                 log_debug("No virtualization found in /proc/device-tree/*");
 144                 return VIRTUALIZATION_NONE;
 145         } else if (r < 0)
 146                 return r;
 147
 148         log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
 149         if (streq(hvtype, "linux,kvm"))
 150                 return VIRTUALIZATION_KVM;
 151         else if (strstr(hvtype, "xen"))
 152                 return VIRTUALIZATION_XEN;
 153         else
 154                 return VIRTUALIZATION_VM_OTHER;
 155 #else
 156         log_debug("This platform does not support /proc/device-tree");
 157         return VIRTUALIZATION_NONE;
 158 #endif
 159 }
 160
 161 static int detect_vm_dmi(void) {
 162 #if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
 163
 164         static const char *const dmi_vendors[] = {
 165                 "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
 166                 "/sys/class/dmi/id/sys_vendor",
 167                 "/sys/class/dmi/id/board_vendor",
 168                 "/sys/class/dmi/id/bios_vendor"
 169         };
 170
 171         static const struct {
 172                 const char *vendor;
 173                 int id;
 174         } dmi_vendor_table[] = {
 175                 { "KVM",           VIRTUALIZATION_KVM       },
 176                 { "QEMU",          VIRTUALIZATION_QEMU      },
 177                 /* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
 178                 { "VMware",        VIRTUALIZATION_VMWARE    },
 179                 { "VMW",           VIRTUALIZATION_VMWARE    },
 180                 { "innotek GmbH",  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         };
 187         unsigned i;
 188         int r;
 189
 190         for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
 191                 _cleanup_free_ char *s = NULL;
 192                 unsigned j;
 193
 194                 r = read_one_line_file(dmi_vendors[i], &s);
 195                 if (r < 0) {
 196                         if (r == -ENOENT)
 197                                 continue;
 198
 199                         return r;
 200                 }
 201
 202
 203
 204                 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
 205                         if (startswith(s, dmi_vendor_table[j].vendor)) {
 206                                 log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
 207                                 return dmi_vendor_table[j].id;
 208                         }
 209         }
 210 #endif
 211
 212         log_debug("No virtualization found in DMI");
 213
 214         return VIRTUALIZATION_NONE;
 215 }
 216
 217 static int detect_vm_xen(void) {
 218         /* Check for Dom0 will be executed later in detect_vm_xen_dom0
 219            Thats why we dont check the content of /proc/xen/capabilities here. */
 220         if (access("/proc/xen/capabilities", F_OK) < 0) {
 221                 log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
 222                 return VIRTUALIZATION_NONE;
 223         }
 224
 225         log_debug("Virtualization XEN found (/proc/xen/capabilities exists)");
 226         return  VIRTUALIZATION_XEN;
 227
 228 }
 229
 230 static bool detect_vm_xen_dom0(void) {
 231         _cleanup_free_ char *domcap = NULL;
 232         char *cap, *i;
 233         int r;
 234
 235         r = read_one_line_file("/proc/xen/capabilities", &domcap);
 236         if (r == -ENOENT) {
 237                 log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
 238                 return false;
 239         }
 240         if (r < 0)
 241                 return r;
 242
 243         i = domcap;
 244         while ((cap = strsep(&i, ",")))
 245                 if (streq(cap, "control_d"))
 246                         break;
 247         if (!cap) {
 248                 log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
 249                 return false;
 250         }
 251
 252         log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
 253         return true;
 254 }
 255
 256 static int detect_vm_hypervisor(void) {
 257         _cleanup_free_ char *hvtype = NULL;
 258         int r;
 259
 260         r = read_one_line_file("/sys/hypervisor/type", &hvtype);
 261         if (r == -ENOENT)
 262                 return VIRTUALIZATION_NONE;
 263         if (r < 0)
 264                 return r;
 265
 266         log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
 267
 268         if (streq(hvtype, "xen"))
 269                 return VIRTUALIZATION_XEN;
 270         else
 271                 return VIRTUALIZATION_VM_OTHER;
 272 }
 273
 274 static int detect_vm_uml(void) {
 275         _cleanup_free_ char *cpuinfo_contents = NULL;
 276         int r;
 277
 278         /* Detect User-Mode Linux by reading /proc/cpuinfo */
 279         r = read_full_file("/proc/cpuinfo", &cpuinfo_contents, NULL);
 280         if (r < 0)
 281                 return r;
 282
 283         if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
 284                 log_debug("UML virtualization found in /proc/cpuinfo");
 285                 return VIRTUALIZATION_UML;
 286         }
 287
 288         log_debug("No virtualization found in /proc/cpuinfo.");
 289         return VIRTUALIZATION_NONE;
 290 }
 291
 292 static int detect_vm_zvm(void) {
 293
 294 #if defined(__s390__)
 295         _cleanup_free_ char *t = NULL;
 296         int r;
 297
 298         r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
 299         if (r == -ENOENT)
 300                 return VIRTUALIZATION_NONE;
 301         if (r < 0)
 302                 return r;
 303
 304         log_debug("Virtualization %s found in /proc/sysinfo", t);
 305         if (streq(t, "z/VM"))
 306                 return VIRTUALIZATION_ZVM;
 307         else
 308                 return VIRTUALIZATION_KVM;
 309 #else
 310         log_debug("This platform does not support /proc/sysinfo");
 311         return VIRTUALIZATION_NONE;
 312 #endif
 313 }
 314
 315 /* Returns a short identifier for the various VM implementations */
 316 int detect_vm(void) {
 317         static thread_local int cached_found = _VIRTUALIZATION_INVALID;
 318         int r, dmi;
 319
 320         if (cached_found >= 0)
 321                 return cached_found;
 322
 323         /* We have to use the correct order here:
 324          *
 325          * -> First try to detect Oracle Virtualbox, even if it uses KVM.
 326          * -> Second try to detect from cpuid, this will report KVM for
 327          *    whatever software is used even if info in dmi is overwritten.
 328          * -> Third try to detect from dmi. */
 329
 330         dmi = detect_vm_dmi();
 331         if (dmi == VIRTUALIZATION_ORACLE)
 332                 return dmi;
 333
 334         r = detect_vm_cpuid();
 335         if (r < 0)
 336                 return r;
 337         if (r != VIRTUALIZATION_NONE)
 338                 goto finish;
 339
 340         r = dmi;
 341         if (r < 0)
 342                 return r;
 343         if (r != VIRTUALIZATION_NONE)
 344                 goto finish;
 345
 346         /* x86 xen will most likely be detected by cpuid. If not (most likely
 347          * because we're not an x86 guest), then we should try the xen capabilities
 348          * file next. If that's not found, then we check for the high-level
 349          * hypervisor sysfs file:
 350          *
 351          * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
 352
 353         r = detect_vm_xen();
 354         if (r < 0)
 355                 return r;
 356         if (r != VIRTUALIZATION_NONE)
 357                 goto finish;
 358
 359         r = detect_vm_hypervisor();
 360         if (r < 0)
 361                 return r;
 362         if (r != VIRTUALIZATION_NONE)
 363                 goto finish;
 364
 365         r = detect_vm_device_tree();
 366         if (r < 0)
 367                 return r;
 368         if (r != VIRTUALIZATION_NONE)
 369                 goto finish;
 370
 371         r = detect_vm_uml();
 372         if (r < 0)
 373                 return r;
 374         if (r != VIRTUALIZATION_NONE)
 375                 goto finish;
 376
 377         r = detect_vm_zvm();
 378         if (r < 0)
 379                 return r;
 380
 381 finish:
 382         /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
 383          * In order to detect the Dom0 as not virtualization we need to
 384          * double-check it */
 385         if (r == VIRTUALIZATION_XEN && detect_vm_xen_dom0())
 386                 r = VIRTUALIZATION_NONE;
 387
 388         cached_found = r;
 389         log_debug("Found VM virtualization %s", virtualization_to_string(r));
 390         return r;
 391 }
 392 #endif // 0
 393
 394 int detect_container(void) {
 395
 396         static const struct {
 397                 const char *value;
 398                 int id;
 399         } value_table[] = {
 400                 { "lxc",            VIRTUALIZATION_LXC            },
 401                 { "lxc-libvirt",    VIRTUALIZATION_LXC_LIBVIRT    },
 402                 { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
 403                 { "docker",         VIRTUALIZATION_DOCKER         },
 404                 { "rkt",            VIRTUALIZATION_RKT            },
 405         };
 406
 407         static thread_local int cached_found = _VIRTUALIZATION_INVALID;
 408         _cleanup_free_ char *m = NULL;
 409         const char *e = NULL;
 410         unsigned j;
 411         int r;
 412
 413         if (cached_found >= 0)
 414                 return cached_found;
 415
 416         /* /proc/vz exists in container and outside of the container, /proc/bc only outside of the container. */
 417         if (access("/proc/vz", F_OK) >= 0 &&
 418             access("/proc/bc", F_OK) < 0) {
 419                 r = VIRTUALIZATION_OPENVZ;
 420                 goto finish;
 421         }
 422
 423         if (getpid() == 1) {
 424                 /* If we are PID 1 we can just check our own environment variable, and that's authoritative. */
 425
 426                 e = getenv("container");
 427                 if (isempty(e)) {
 428                         r = VIRTUALIZATION_NONE;
 429                         goto finish;
 430                 }
 431
 432                 goto translate_name;
 433         }
 434
 435         /* Otherwise, PID 1 might have dropped this information into a file in /run. This is better than accessing
 436          * /proc/1/environ, since we don't need CAP_SYS_PTRACE for that. */
 437         r = read_one_line_file("/run/systemd/container", &m);
 438         if (r >= 0) {
 439                 e = m;
 440                 goto translate_name;
 441         }
 442         if (r != -ENOENT)
 443                 return log_debug_errno(r, "Failed to read /run/systemd/container: %m");
 444
 445         /* Fallback for cases where PID 1 was not systemd (for example, cases where init=/bin/sh is used. */
 446         r = getenv_for_pid(1, "container", &m);
 447         if (r > 0) {
 448                 e = m;
 449                 goto translate_name;
 450         }
 451         if (r < 0) /* This only works if we have CAP_SYS_PTRACE, hence let's better ignore failures here */
 452                 log_debug_errno(r, "Failed to read $container of PID 1, ignoring: %m");
 453
 454         /* Interestingly /proc/1/sched actually shows the host's PID for what we see as PID 1. Hence, if the PID shown
 455          * there is not 1, we know we are in a PID namespace. and hence a container. */
 456         r = read_one_line_file("/proc/1/sched", &m);
 457         if (r >= 0) {
 458                 const char *t;
 459
 460                 t = strrchr(m, '(');
 461                 if (!t)
 462                         return -EIO;
 463
 464                 if (!startswith(t, "(1,")) {
 465                         r = VIRTUALIZATION_CONTAINER_OTHER;
 466                         goto finish;
 467                 }
 468         } else if (r != -ENOENT)
 469                 return r;
 470
 471         /* If that didn't work, give up, assume no container manager. */
 472         r = VIRTUALIZATION_NONE;
 473         goto finish;
 474
 475 translate_name:
 476         for (j = 0; j < ELEMENTSOF(value_table); j++)
 477                 if (streq(e, value_table[j].value)) {
 478                         r = value_table[j].id;
 479                         goto finish;
 480                 }
 481
 482         r = VIRTUALIZATION_CONTAINER_OTHER;
 483
 484 finish:
 485         log_debug("Found container virtualization %s.", virtualization_to_string(r));
 486         cached_found = r;
 487         return r;
 488 }
 489
 490 #if 0 /// UNNEEDED by elogind
 491 int detect_virtualization(void) {
 492         int r;
 493
 494         r = detect_container();
 495         if (r == 0)
 496                 r = detect_vm();
 497
 498         return r;
 499 }
 500
 501 static int userns_has_mapping(const char *name) {
 502         _cleanup_fclose_ FILE *f = NULL;
 503         _cleanup_free_ char *buf = NULL;
 504         size_t n_allocated = 0;
 505         ssize_t n;
 506         uint32_t a, b, c;
 507         int r;
 508
 509         f = fopen(name, "re");
 510         if (!f) {
 511                 log_debug_errno(errno, "Failed to open %s: %m", name);
 512                 return errno == ENOENT ? false : -errno;
 513         }
 514
 515         n = getline(&buf, &n_allocated, f);
 516         if (n < 0) {
 517                 if (feof(f)) {
 518                         log_debug("%s is empty, we're in an uninitialized user namespace", name);
 519                         return true;
 520                 }
 521
 522                 return log_debug_errno(errno, "Failed to read %s: %m", name);
 523         }
 524
 525         r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
 526         if (r < 3)
 527                 return log_debug_errno(errno, "Failed to parse %s: %m", name);
 528
 529         if (a == 0 && b == 0 && c == UINT32_MAX) {
 530                 /* The kernel calls mappings_overlap() and does not allow overlaps */
 531                 log_debug("%s has a full 1:1 mapping", name);
 532                 return false;
 533         }
 534
 535         /* Anything else implies that we are in a user namespace */
 536         log_debug("Mapping found in %s, we're in a user namespace", name);
 537         return true;
 538 }
 539
 540 int running_in_userns(void) {
 541         _cleanup_free_ char *line = NULL;
 542         int r;
 543
 544         r = userns_has_mapping("/proc/self/uid_map");
 545         if (r != 0)
 546                 return r;
 547
 548         r = userns_has_mapping("/proc/self/gid_map");
 549         if (r != 0)
 550                 return r;
 551
 552         /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
 553          * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
 554          * also does not exist. We cannot distinguish those two cases, so assume that
 555          * we're running on a stripped-down recent kernel, rather than on an old one,
 556          * and if the file is not found, return false.
 557          */
 558         r = read_one_line_file("/proc/self/setgroups", &line);
 559         if (r < 0) {
 560                 log_debug_errno(r, "/proc/self/setgroups: %m");
 561                 return r == -ENOENT ? false : r;
 562         }
 563
 564         truncate_nl(line);
 565         r = streq(line, "deny");
 566         /* See user_namespaces(7) for a description of this "setgroups" contents. */
 567         log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
 568         return r;
 569 }
 570 #endif // 0
 571
 572 int running_in_chroot(void) {
 573         _cleanup_free_ char *self_mnt = NULL, *pid1_mnt = NULL;
 574         int r;
 575
 576         /* Try to detect whether we are running in a chroot() environment. Specifically, check whether we have a
 577          * different root directory than PID 1, even though we live in the same mount namespace as it. */
 578
 579 #if 0 /// elogind does not allow to ignore chroots, we are never init!
 580         if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
 581                 return 0;
 582 #endif // 0
 583
 584         r = files_same("/proc/1/root", "/");
 585         if (r < 0)
 586                 return r;
 587         if (r > 0)
 588                 return 0;
 589
 590         r = readlink_malloc("/proc/self/ns/mnt", &self_mnt);
 591         if (r < 0)
 592                 return r;
 593
 594         r = readlink_malloc("/proc/1/ns/mnt", &pid1_mnt);
 595         if (r < 0)
 596                 return r;
 597
 598         return streq(self_mnt, pid1_mnt); /* Only if we live in the same namespace! */
 599 }
 600
 601 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
 602         [VIRTUALIZATION_NONE] = "none",
 603         [VIRTUALIZATION_KVM] = "kvm",
 604         [VIRTUALIZATION_QEMU] = "qemu",
 605         [VIRTUALIZATION_BOCHS] = "bochs",
 606         [VIRTUALIZATION_XEN] = "xen",
 607         [VIRTUALIZATION_UML] = "uml",
 608         [VIRTUALIZATION_VMWARE] = "vmware",
 609         [VIRTUALIZATION_ORACLE] = "oracle",
 610         [VIRTUALIZATION_MICROSOFT] = "microsoft",
 611         [VIRTUALIZATION_ZVM] = "zvm",
 612         [VIRTUALIZATION_PARALLELS] = "parallels",
 613         [VIRTUALIZATION_BHYVE] = "bhyve",
 614         [VIRTUALIZATION_VM_OTHER] = "vm-other",
 615
 616         [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
 617         [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
 618         [VIRTUALIZATION_LXC] = "lxc",
 619         [VIRTUALIZATION_OPENVZ] = "openvz",
 620         [VIRTUALIZATION_DOCKER] = "docker",
 621         [VIRTUALIZATION_RKT] = "rkt",
 622         [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
 623 };
 624
 625 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);