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