X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fbasic%2Fvirt.c;h=35beebe7fa4116f5dedfd2b2a0e50fd0cc97486c;hp=1299a75ed54d242ef9b93964b7a24dfe302fff9c;hb=2a4a438b7c7306da5f698d2247e646263f3c45c0;hpb=1cfc78c91965df340cdde100ad6cb3ed50b28927
diff --git a/src/basic/virt.c b/src/basic/virt.c
index 1299a75ed..35beebe7f 100644
--- a/src/basic/virt.c
+++ b/src/basic/virt.c
@@ -1,5 +1,3 @@
-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
/***
This file is part of systemd.
@@ -19,34 +17,44 @@
along with systemd; If not, see .
***/
-#include
#include
+#include
+#include
+#include
#include
-#include "util.h"
+#include "alloc-util.h"
+#include "dirent-util.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "macro.h"
#include "process-util.h"
+#include "stat-util.h"
+#include "string-table.h"
+#include "string-util.h"
#include "virt.h"
-#include "fileio.h"
-static int detect_vm_cpuid(const char **_id) {
+#if 0 /// UNNEEDED by elogind
+static int detect_vm_cpuid(void) {
- /* Both CPUID and DMI are x86 specific interfaces... */
+ /* CPUID is an x86 specific interface. */
#if defined(__i386__) || defined(__x86_64__)
- static const char cpuid_vendor_table[] =
- "XenVMMXenVMM\0" "xen\0"
- "KVMKVMKVM\0" "kvm\0"
+ static const struct {
+ const char *cpuid;
+ int id;
+ } cpuid_vendor_table[] = {
+ { "XenVMMXenVMM", VIRTUALIZATION_XEN },
+ { "KVMKVMKVM", VIRTUALIZATION_KVM },
/* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
- "VMwareVMware\0" "vmware\0"
+ { "VMwareVMware", VIRTUALIZATION_VMWARE },
/* http://msdn.microsoft.com/en-us/library/ff542428.aspx */
- "Microsoft Hv\0" "microsoft\0";
+ { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
+ /* https://wiki.freebsd.org/bhyve */
+ { "bhyve bhyve ", VIRTUALIZATION_BHYVE },
+ };
uint32_t eax, ecx;
- union {
- uint32_t sig32[3];
- char text[13];
- } sig = {};
- const char *j, *k;
bool hypervisor;
/* http://lwn.net/Articles/301888/ */
@@ -74,6 +82,11 @@ static int detect_vm_cpuid(const char **_id) {
hypervisor = !!(ecx & 0x80000000U);
if (hypervisor) {
+ union {
+ uint32_t sig32[3];
+ char text[13];
+ } sig = {};
+ unsigned j;
/* There is a hypervisor, see what it is */
eax = 0x40000000U;
@@ -88,231 +101,319 @@ static int detect_vm_cpuid(const char **_id) {
: "0" (eax)
);
- NULSTR_FOREACH_PAIR(j, k, cpuid_vendor_table)
- if (streq(sig.text, j)) {
- *_id = k;
- return 1;
- }
+ log_debug("Virtualization found, CPUID=%s", sig.text);
- *_id = "other";
- return 0;
+ for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
+ if (streq(sig.text, cpuid_vendor_table[j].cpuid))
+ return cpuid_vendor_table[j].id;
+
+ return VIRTUALIZATION_VM_OTHER;
}
#endif
+ log_debug("No virtualization found in CPUID");
- return 0;
+ return VIRTUALIZATION_NONE;
}
-static int detect_vm_devicetree(const char **_id) {
+static int detect_vm_device_tree(void) {
#if defined(__arm__) || defined(__aarch64__) || defined(__powerpc__) || defined(__powerpc64__)
_cleanup_free_ char *hvtype = NULL;
int r;
r = read_one_line_file("/proc/device-tree/hypervisor/compatible", &hvtype);
- if (r >= 0) {
- if (streq(hvtype, "linux,kvm")) {
- *_id = "kvm";
- return 1;
- } else if (strstr(hvtype, "xen")) {
- *_id = "xen";
- return 1;
- }
- } else if (r == -ENOENT) {
+ if (r == -ENOENT) {
_cleanup_closedir_ DIR *dir = NULL;
struct dirent *dent;
dir = opendir("/proc/device-tree");
if (!dir) {
- if (errno == ENOENT)
- return 0;
+ if (errno == ENOENT) {
+ log_debug_errno(errno, "/proc/device-tree: %m");
+ return VIRTUALIZATION_NONE;
+ }
return -errno;
}
- FOREACH_DIRENT(dent, dir, return -errno) {
+ FOREACH_DIRENT(dent, dir, return -errno)
if (strstr(dent->d_name, "fw-cfg")) {
- *_id = "qemu";
- return 1;
+ log_debug("Virtualization QEMU: \"fw-cfg\" present in /proc/device-tree/%s", dent->d_name);
+ return VIRTUALIZATION_QEMU;
}
- }
- }
+
+ log_debug("No virtualization found in /proc/device-tree/*");
+ return VIRTUALIZATION_NONE;
+ } else if (r < 0)
+ return r;
+
+ log_debug("Virtualization %s found in /proc/device-tree/hypervisor/compatible", hvtype);
+ if (streq(hvtype, "linux,kvm"))
+ return VIRTUALIZATION_KVM;
+ else if (strstr(hvtype, "xen"))
+ return VIRTUALIZATION_XEN;
+ else
+ return VIRTUALIZATION_VM_OTHER;
+#else
+ log_debug("This platform does not support /proc/device-tree");
+ return VIRTUALIZATION_NONE;
#endif
- return 0;
}
-static int detect_vm_dmi(const char **_id) {
-
- /* Both CPUID and DMI are x86 specific interfaces... */
-#if defined(__i386__) || defined(__x86_64__)
+static int detect_vm_dmi(void) {
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
static const char *const dmi_vendors[] = {
+ "/sys/class/dmi/id/product_name", /* Test this before sys_vendor to detect KVM over QEMU */
"/sys/class/dmi/id/sys_vendor",
"/sys/class/dmi/id/board_vendor",
"/sys/class/dmi/id/bios_vendor"
};
- static const char dmi_vendor_table[] =
- "QEMU\0" "qemu\0"
+ static const struct {
+ const char *vendor;
+ int id;
+ } dmi_vendor_table[] = {
+ { "KVM", VIRTUALIZATION_KVM },
+ { "QEMU", VIRTUALIZATION_QEMU },
/* http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1009458 */
- "VMware\0" "vmware\0"
- "VMW\0" "vmware\0"
- "innotek GmbH\0" "oracle\0"
- "Xen\0" "xen\0"
- "Bochs\0" "bochs\0";
+ { "VMware", VIRTUALIZATION_VMWARE },
+ { "VMW", VIRTUALIZATION_VMWARE },
+ { "innotek GmbH", VIRTUALIZATION_ORACLE },
+ { "Xen", VIRTUALIZATION_XEN },
+ { "Bochs", VIRTUALIZATION_BOCHS },
+ { "Parallels", VIRTUALIZATION_PARALLELS },
+ /* https://wiki.freebsd.org/bhyve */
+ { "BHYVE", VIRTUALIZATION_BHYVE },
+ };
unsigned i;
+ int r;
for (i = 0; i < ELEMENTSOF(dmi_vendors); i++) {
_cleanup_free_ char *s = NULL;
- const char *j, *k;
- int r;
+ unsigned j;
r = read_one_line_file(dmi_vendors[i], &s);
if (r < 0) {
- if (r != -ENOENT)
- return r;
+ if (r == -ENOENT)
+ continue;
- continue;
+ return r;
}
- NULSTR_FOREACH_PAIR(j, k, dmi_vendor_table)
- if (startswith(s, j)) {
- *_id = k;
- return 1;
+
+
+ for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
+ if (startswith(s, dmi_vendor_table[j].vendor)) {
+ log_debug("Virtualization %s found in DMI (%s)", s, dmi_vendors[i]);
+ return dmi_vendor_table[j].id;
}
}
#endif
- return 0;
+ log_debug("No virtualization found in DMI");
+
+ return VIRTUALIZATION_NONE;
}
-/* Returns a short identifier for the various VM implementations */
-int detect_vm(const char **id) {
- _cleanup_free_ char *domcap = NULL, *cpuinfo_contents = NULL;
- static thread_local int cached_found = -1;
- static thread_local const char *cached_id = NULL;
- const char *_id = NULL;
- int r;
+static int detect_vm_xen(void) {
+ /* Check for Dom0 will be executed later in detect_vm_xen_dom0
+ Thats why we dont check the content of /proc/xen/capabilities here. */
+ if (access("/proc/xen/capabilities", F_OK) < 0) {
+ log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
+ return VIRTUALIZATION_NONE;
+ }
- if (_likely_(cached_found >= 0)) {
+ log_debug("Virtualization XEN found (/proc/xen/capabilities exists)");
+ return VIRTUALIZATION_XEN;
- if (id)
- *id = cached_id;
+}
- return cached_found;
- }
+static bool detect_vm_xen_dom0(void) {
+ _cleanup_free_ char *domcap = NULL;
+ char *cap, *i;
+ int r;
- /* Try xen capabilities file first, if not found try high-level hypervisor sysfs file:
- *
- * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
r = read_one_line_file("/proc/xen/capabilities", &domcap);
- if (r >= 0) {
- char *cap, *i = domcap;
-
- while ((cap = strsep(&i, ",")))
- if (streq(cap, "control_d"))
- break;
+ if (r == -ENOENT) {
+ log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
+ return false;
+ }
+ if (r < 0)
+ return r;
- if (!cap) {
- _id = "xen";
- r = 1;
- }
+ i = domcap;
+ while ((cap = strsep(&i, ",")))
+ if (streq(cap, "control_d"))
+ break;
+ if (!cap) {
+ log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
+ return false;
+ }
- goto finish;
+ log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
+ return true;
+}
- } else if (r == -ENOENT) {
- _cleanup_free_ char *hvtype = NULL;
+static int detect_vm_hypervisor(void) {
+ _cleanup_free_ char *hvtype = NULL;
+ int r;
- r = read_one_line_file("/sys/hypervisor/type", &hvtype);
- if (r >= 0) {
- if (streq(hvtype, "xen")) {
- _id = "xen";
- r = 1;
- goto finish;
- }
- } else if (r != -ENOENT)
- return r;
- } else
+ r = read_one_line_file("/sys/hypervisor/type", &hvtype);
+ if (r == -ENOENT)
+ return VIRTUALIZATION_NONE;
+ if (r < 0)
return r;
- /* this will set _id to "other" and return 0 for unknown hypervisors */
- r = detect_vm_cpuid(&_id);
- if (r != 0)
- goto finish;
-
- r = detect_vm_dmi(&_id);
- if (r != 0)
- goto finish;
+ log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
- r = detect_vm_devicetree(&_id);
- if (r != 0)
- goto finish;
+ if (streq(hvtype, "xen"))
+ return VIRTUALIZATION_XEN;
+ else
+ return VIRTUALIZATION_VM_OTHER;
+}
- if (_id) {
- /* "other" */
- r = 1;
- goto finish;
- }
+static int detect_vm_uml(void) {
+ _cleanup_free_ char *cpuinfo_contents = NULL;
+ int r;
/* Detect User-Mode Linux by reading /proc/cpuinfo */
r = read_full_file("/proc/cpuinfo", &cpuinfo_contents, NULL);
if (r < 0)
return r;
+
if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
- _id = "uml";
- r = 1;
- goto finish;
+ log_debug("UML virtualization found in /proc/cpuinfo");
+ return VIRTUALIZATION_UML;
}
+ log_debug("No virtualization found in /proc/cpuinfo.");
+ return VIRTUALIZATION_NONE;
+}
+
+static int detect_vm_zvm(void) {
+
#if defined(__s390__)
- {
- _cleanup_free_ char *t = NULL;
+ _cleanup_free_ char *t = NULL;
+ int r;
- r = get_status_field("/proc/sysinfo", "VM00 Control Program:", &t);
- if (r >= 0) {
- if (streq(t, "z/VM"))
- _id = "zvm";
- else
- _id = "kvm";
- r = 1;
+ r = get_proc_field("/proc/sysinfo", "VM00 Control Program", WHITESPACE, &t);
+ if (r == -ENOENT)
+ return VIRTUALIZATION_NONE;
+ if (r < 0)
+ return r;
- goto finish;
- }
- }
+ log_debug("Virtualization %s found in /proc/sysinfo", t);
+ if (streq(t, "z/VM"))
+ return VIRTUALIZATION_ZVM;
+ else
+ return VIRTUALIZATION_KVM;
+#else
+ log_debug("This platform does not support /proc/sysinfo");
+ return VIRTUALIZATION_NONE;
#endif
+}
- r = 0;
+/* Returns a short identifier for the various VM implementations */
+int detect_vm(void) {
+ static thread_local int cached_found = _VIRTUALIZATION_INVALID;
+ int r;
-finish:
- cached_found = r;
+ if (cached_found >= 0)
+ return cached_found;
- cached_id = _id;
- if (id)
- *id = _id;
+ /* We have to use the correct order here:
+ * Some virtualization technologies do use KVM hypervisor but are
+ * expected to be detected as something else. So detect DMI first.
+ *
+ * An example is Virtualbox since version 5.0, which uses KVM backend.
+ * Detection via DMI works corretly, the CPU ID would find KVM
+ * only. */
+ r = detect_vm_dmi();
+ if (r < 0)
+ return r;
+ if (r != VIRTUALIZATION_NONE)
+ goto finish;
- return r;
-}
+ r = detect_vm_cpuid();
+ if (r < 0)
+ return r;
+ if (r != VIRTUALIZATION_NONE)
+ goto finish;
-int detect_container(const char **id) {
+ /* x86 xen will most likely be detected by cpuid. If not (most likely
+ * because we're not an x86 guest), then we should try the xen capabilities
+ * file next. If that's not found, then we check for the high-level
+ * hypervisor sysfs file:
+ *
+ * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
- static thread_local int cached_found = -1;
- static thread_local const char *cached_id = NULL;
+ r = detect_vm_xen();
+ if (r < 0)
+ return r;
+ if (r != VIRTUALIZATION_NONE)
+ goto finish;
- _cleanup_free_ char *m = NULL;
- const char *_id = NULL, *e = NULL;
- int r;
+ r = detect_vm_hypervisor();
+ if (r < 0)
+ return r;
+ if (r != VIRTUALIZATION_NONE)
+ goto finish;
+
+ r = detect_vm_device_tree();
+ if (r < 0)
+ return r;
+ if (r != VIRTUALIZATION_NONE)
+ goto finish;
- if (_likely_(cached_found >= 0)) {
+ r = detect_vm_uml();
+ if (r < 0)
+ return r;
+ if (r != VIRTUALIZATION_NONE)
+ goto finish;
+
+ r = detect_vm_zvm();
+ if (r < 0)
+ return r;
+
+finish:
+ /* x86 xen Dom0 is detected as XEN in hypervisor and maybe others.
+ * In order to detect the Dom0 as not virtualization we need to
+ * double-check it */
+ if (r == VIRTUALIZATION_XEN && detect_vm_xen_dom0())
+ r = VIRTUALIZATION_NONE;
+
+ cached_found = r;
+ log_debug("Found VM virtualization %s", virtualization_to_string(r));
+ return r;
+}
+#endif // 0
+
+int detect_container(void) {
+
+ static const struct {
+ const char *value;
+ int id;
+ } value_table[] = {
+ { "lxc", VIRTUALIZATION_LXC },
+ { "lxc-libvirt", VIRTUALIZATION_LXC_LIBVIRT },
+ { "systemd-nspawn", VIRTUALIZATION_SYSTEMD_NSPAWN },
+ { "docker", VIRTUALIZATION_DOCKER },
+ { "rkt", VIRTUALIZATION_RKT },
+ };
- if (id)
- *id = cached_id;
+ static thread_local int cached_found = _VIRTUALIZATION_INVALID;
+ _cleanup_free_ char *m = NULL;
+ const char *e = NULL;
+ unsigned j;
+ int r;
+ if (cached_found >= 0)
return cached_found;
- }
/* /proc/vz exists in container and outside of the container,
* /proc/bc only outside of the container. */
if (access("/proc/vz", F_OK) >= 0 &&
access("/proc/bc", F_OK) < 0) {
- _id = "openvz";
- r = 1;
+ r = VIRTUALIZATION_OPENVZ;
goto finish;
}
@@ -322,7 +423,7 @@ int detect_container(const char **id) {
e = getenv("container");
if (isempty(e)) {
- r = 0;
+ r = VIRTUALIZATION_NONE;
goto finish;
}
} else {
@@ -351,7 +452,7 @@ int detect_container(const char **id) {
* as /proc/1/environ is only readable
* with privileges. */
- r = 0;
+ r = VIRTUALIZATION_NONE;
goto finish;
}
}
@@ -361,46 +462,139 @@ int detect_container(const char **id) {
e = m;
}
- /* We only recognize a selected few here, since we want to
- * enforce a redacted namespace */
- if (streq(e, "lxc"))
- _id ="lxc";
- else if (streq(e, "lxc-libvirt"))
- _id = "lxc-libvirt";
- else if (streq(e, "systemd-nspawn"))
- _id = "systemd-nspawn";
- else if (streq(e, "docker"))
- _id = "docker";
- else
- _id = "other";
+ for (j = 0; j < ELEMENTSOF(value_table); j++)
+ if (streq(e, value_table[j].value)) {
+ r = value_table[j].id;
+ goto finish;
+ }
- r = 1;
+ r = VIRTUALIZATION_CONTAINER_OTHER;
finish:
+ log_debug("Found container virtualization %s", virtualization_to_string(r));
cached_found = r;
+ return r;
+}
+
+#if 0 /// UNNEEDED by elogind
+int detect_virtualization(void) {
+ int r;
- cached_id = _id;
- if (id)
- *id = _id;
+ r = detect_container();
+ if (r == 0)
+ r = detect_vm();
return r;
}
-/* Returns a short identifier for the various VM/container implementations */
-int detect_virtualization(const char **id) {
+static int userns_has_mapping(const char *name) {
+ _cleanup_fclose_ FILE *f = NULL;
+ _cleanup_free_ char *buf = NULL;
+ size_t n_allocated = 0;
+ ssize_t n;
+ uint32_t a, b, c;
int r;
- r = detect_container(id);
- if (r < 0)
+ f = fopen(name, "re");
+ if (!f) {
+ log_debug_errno(errno, "Failed to open %s: %m", name);
+ return errno == -ENOENT ? false : -errno;
+ }
+
+ n = getline(&buf, &n_allocated, f);
+ if (n < 0) {
+ if (feof(f)) {
+ log_debug("%s is empty, we're in an uninitialized user namespace", name);
+ return true;
+ }
+
+ return log_debug_errno(errno, "Failed to read %s: %m", name);
+ }
+
+ r = sscanf(buf, "%"PRIu32" %"PRIu32" %"PRIu32, &a, &b, &c);
+ if (r < 3)
+ return log_debug_errno(errno, "Failed to parse %s: %m", name);
+
+ if (a == 0 && b == 0 && c == UINT32_MAX) {
+ /* The kernel calls mappings_overlap() and does not allow overlaps */
+ log_debug("%s has a full 1:1 mapping", name);
+ return false;
+ }
+
+ /* Anything else implies that we are in a user namespace */
+ log_debug("Mapping found in %s, we're in a user namespace", name);
+ return true;
+}
+
+int running_in_userns(void) {
+ _cleanup_free_ char *line = NULL;
+ int r;
+
+ r = userns_has_mapping("/proc/self/uid_map");
+ if (r != 0)
return r;
- if (r > 0)
- return VIRTUALIZATION_CONTAINER;
- r = detect_vm(id);
- if (r < 0)
+ r = userns_has_mapping("/proc/self/gid_map");
+ if (r != 0)
return r;
- if (r > 0)
- return VIRTUALIZATION_VM;
- return VIRTUALIZATION_NONE;
+ /* "setgroups" file was added in kernel v3.18-rc6-15-g9cc46516dd. It is also
+ * possible to compile a kernel without CONFIG_USER_NS, in which case "setgroups"
+ * also does not exist. We cannot distinguish those two cases, so assume that
+ * we're running on a stripped-down recent kernel, rather than on an old one,
+ * and if the file is not found, return false.
+ */
+ r = read_one_line_file("/proc/self/setgroups", &line);
+ if (r < 0) {
+ log_debug_errno(r, "/proc/self/setgroups: %m");
+ return r == -ENOENT ? false : r;
+ }
+
+ truncate_nl(line);
+ r = streq(line, "deny");
+ /* See user_namespaces(7) for a description of this "setgroups" contents. */
+ log_debug("/proc/self/setgroups contains \"%s\", %s user namespace", line, r ? "in" : "not in");
+ return r;
}
+#endif // 0
+
+int running_in_chroot(void) {
+ int ret;
+
+#if 0 /// elogind does not allow to ignore chroots, we are never init!
+ if (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
+ return 0;
+#endif // 0
+
+ ret = files_same("/proc/1/root", "/");
+ if (ret < 0)
+ return ret;
+
+ return ret == 0;
+}
+
+static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
+ [VIRTUALIZATION_NONE] = "none",
+ [VIRTUALIZATION_KVM] = "kvm",
+ [VIRTUALIZATION_QEMU] = "qemu",
+ [VIRTUALIZATION_BOCHS] = "bochs",
+ [VIRTUALIZATION_XEN] = "xen",
+ [VIRTUALIZATION_UML] = "uml",
+ [VIRTUALIZATION_VMWARE] = "vmware",
+ [VIRTUALIZATION_ORACLE] = "oracle",
+ [VIRTUALIZATION_MICROSOFT] = "microsoft",
+ [VIRTUALIZATION_ZVM] = "zvm",
+ [VIRTUALIZATION_PARALLELS] = "parallels",
+ [VIRTUALIZATION_BHYVE] = "bhyve",
+ [VIRTUALIZATION_VM_OTHER] = "vm-other",
+
+ [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
+ [VIRTUALIZATION_LXC_LIBVIRT] = "lxc-libvirt",
+ [VIRTUALIZATION_LXC] = "lxc",
+ [VIRTUALIZATION_OPENVZ] = "openvz",
+ [VIRTUALIZATION_DOCKER] = "docker",
+ [VIRTUALIZATION_RKT] = "rkt",
+ [VIRTUALIZATION_CONTAINER_OTHER] = "container-other",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(virtualization, int);