chiark / gitweb /
tests: clarify test_path_startswith return value (#4508)
[elogind.git] / src / basic / virt.c
index 4b6286567899fc2376ba8a30d30a91f742ea7480..7a2b9e9bd607b81f584818e8f53ffbe30e220404 100644 (file)
@@ -1,5 +1,3 @@
-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
 /***
   This file is part of systemd.
 
@@ -20,6 +18,8 @@
 ***/
 
 #include <errno.h>
+#include <stdint.h>
+#include <stdlib.h>
 #include <string.h>
 #include <unistd.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 "util.h"
 #include "virt.h"
 
 #if 0 /// UNNEEDED by elogind
@@ -50,6 +50,8 @@ static int detect_vm_cpuid(void) {
                 { "VMwareVMware", VIRTUALIZATION_VMWARE    },
                 /* http://msdn.microsoft.com/en-us/library/ff542428.aspx */
                 { "Microsoft Hv", VIRTUALIZATION_MICROSOFT },
+                /* https://wiki.freebsd.org/bhyve */
+                { "bhyve bhyve ", VIRTUALIZATION_BHYVE     },
         };
 
         uint32_t eax, ecx;
@@ -99,6 +101,8 @@ static int detect_vm_cpuid(void) {
                         : "0" (eax)
                 );
 
+                log_debug("Virtualization found, CPUID=%s", sig.text);
+
                 for (j = 0; j < ELEMENTSOF(cpuid_vendor_table); j ++)
                         if (streq(sig.text, cpuid_vendor_table[j].cpuid))
                                 return cpuid_vendor_table[j].id;
@@ -106,6 +110,7 @@ static int detect_vm_cpuid(void) {
                 return VIRTUALIZATION_VM_OTHER;
         }
 #endif
+        log_debug("No virtualization found in CPUID");
 
         return VIRTUALIZATION_NONE;
 }
@@ -122,19 +127,25 @@ static int detect_vm_device_tree(void) {
 
                 dir = opendir("/proc/device-tree");
                 if (!dir) {
-                        if (errno == ENOENT)
+                        if (errno == ENOENT) {
+                                log_debug_errno(errno, "/proc/device-tree: %m");
                                 return VIRTUALIZATION_NONE;
+                        }
                         return -errno;
                 }
 
                 FOREACH_DIRENT(dent, dir, return -errno)
-                        if (strstr(dent->d_name, "fw-cfg"))
+                        if (strstr(dent->d_name, "fw-cfg")) {
+                                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"))
@@ -142,6 +153,7 @@ static int detect_vm_device_tree(void) {
         else
                 return VIRTUALIZATION_VM_OTHER;
 #else
+        log_debug("This platform does not support /proc/device-tree");
         return VIRTUALIZATION_NONE;
 #endif
 }
@@ -169,6 +181,8 @@ static int detect_vm_dmi(void) {
                 { "Xen",           VIRTUALIZATION_XEN       },
                 { "Bochs",         VIRTUALIZATION_BOCHS     },
                 { "Parallels",     VIRTUALIZATION_PARALLELS },
+                /* https://wiki.freebsd.org/bhyve */
+                { "BHYVE",         VIRTUALIZATION_BHYVE     },
         };
         unsigned i;
         int r;
@@ -180,47 +194,77 @@ static int detect_vm_dmi(void) {
                 r = read_one_line_file(dmi_vendors[i], &s);
                 if (r < 0) {
                         if (r == -ENOENT)
-                        continue;
+                                continue;
 
                         return r;
-                        }
+                }
+
+
 
                 for (j = 0; j < ELEMENTSOF(dmi_vendor_table); j++)
-                        if (startswith(s, dmi_vendor_table[j].vendor))
+                        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
 
+        log_debug("No virtualization found in DMI");
+
         return VIRTUALIZATION_NONE;
 }
 
 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;
+        }
+
+        log_debug("Virtualization XEN found (/proc/xen/capabilities exists)");
+        return  VIRTUALIZATION_XEN;
+
+}
+
+static bool detect_vm_xen_dom0(void) {
         _cleanup_free_ char *domcap = NULL;
         char *cap, *i;
         int r;
 
         r = read_one_line_file("/proc/xen/capabilities", &domcap);
-        if (r == -ENOENT)
-                return VIRTUALIZATION_NONE;
+        if (r == -ENOENT) {
+                log_debug("Virtualization XEN not found, /proc/xen/capabilities does not exist");
+                return false;
+        }
+        if (r < 0)
+                return r;
 
         i = domcap;
-                while ((cap = strsep(&i, ",")))
-                        if (streq(cap, "control_d"))
-                                break;
+        while ((cap = strsep(&i, ",")))
+                if (streq(cap, "control_d"))
+                        break;
+        if (!cap) {
+                log_debug("Virtualization XEN DomU found (/proc/xen/capabilites)");
+                return false;
+        }
 
-        return cap ? VIRTUALIZATION_NONE : VIRTUALIZATION_XEN;
-                }
+        log_debug("Virtualization XEN Dom0 ignored (/proc/xen/capabilities)");
+        return true;
+}
 
 static int detect_vm_hypervisor(void) {
-                _cleanup_free_ char *hvtype = NULL;
+        _cleanup_free_ char *hvtype = NULL;
         int r;
 
-                r = read_one_line_file("/sys/hypervisor/type", &hvtype);
+        r = read_one_line_file("/sys/hypervisor/type", &hvtype);
         if (r == -ENOENT)
                 return VIRTUALIZATION_NONE;
         if (r < 0)
                 return r;
 
+        log_debug("Virtualization %s found in /sys/hypervisor/type", hvtype);
+
         if (streq(hvtype, "xen"))
                 return VIRTUALIZATION_XEN;
         else
@@ -235,9 +279,13 @@ static int detect_vm_uml(void) {
         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"))
+
+        if (strstr(cpuinfo_contents, "\nvendor_id\t: User Mode Linux\n")) {
+                log_debug("UML virtualization found in /proc/cpuinfo");
                 return VIRTUALIZATION_UML;
+        }
 
+        log_debug("No virtualization found in /proc/cpuinfo.");
         return VIRTUALIZATION_NONE;
 }
 
@@ -253,14 +301,16 @@ static int detect_vm_zvm(void) {
         if (r < 0)
                 return r;
 
+        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
-        }
+}
 
 /* Returns a short identifier for the various VM implementations */
 int detect_vm(void) {
@@ -270,13 +320,20 @@ int detect_vm(void) {
         if (cached_found >= 0)
                 return cached_found;
 
-        r = detect_vm_cpuid();
+        /* 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;
 
-        r = detect_vm_dmi();
+        r = detect_vm_cpuid();
         if (r < 0)
                 return r;
         if (r != VIRTUALIZATION_NONE)
@@ -311,14 +368,21 @@ int detect_vm(void) {
         if (r < 0)
                 return r;
         if (r != VIRTUALIZATION_NONE)
-                        goto finish;
+                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
@@ -407,6 +471,7 @@ int detect_container(void) {
         r = VIRTUALIZATION_CONTAINER_OTHER;
 
 finish:
+        log_debug("Found container virtualization %s", virtualization_to_string(r));
         cached_found = r;
         return r;
 }
@@ -416,16 +481,89 @@ int detect_virtualization(void) {
         int r;
 
         r = detect_container();
+        if (r == 0)
+                r = detect_vm();
+
+        return r;
+}
+#endif // 0
+
+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;
+
+        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;
 
-        return detect_vm();
+        r = userns_has_mapping("/proc/self/gid_map");
+        if (r != 0)
+                return r;
+
+        /* "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 (getenv_bool("SYSTEMD_IGNORE_CHROOT") > 0)
+                return 0;
+
         ret = files_same("/proc/1/root", "/");
         if (ret < 0)
                 return ret;
@@ -433,7 +571,6 @@ int running_in_chroot(void) {
         return ret == 0;
 }
 
-#if 0 /// UNNEEDED by elogind
 static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
         [VIRTUALIZATION_NONE] = "none",
         [VIRTUALIZATION_KVM] = "kvm",
@@ -446,6 +583,7 @@ static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
         [VIRTUALIZATION_MICROSOFT] = "microsoft",
         [VIRTUALIZATION_ZVM] = "zvm",
         [VIRTUALIZATION_PARALLELS] = "parallels",
+        [VIRTUALIZATION_BHYVE] = "bhyve",
         [VIRTUALIZATION_VM_OTHER] = "vm-other",
 
         [VIRTUALIZATION_SYSTEMD_NSPAWN] = "systemd-nspawn",
@@ -458,4 +596,3 @@ static const char *const virtualization_table[_VIRTUALIZATION_MAX] = {
 };
 
 DEFINE_STRING_TABLE_LOOKUP(virtualization, int);
-#endif // 0