chiark / gitweb /
smack: bugfix the smack label of symlink when '--with-smack-run-label' is set
[elogind.git] / src / core / mount-setup.c
index b841f478c3fbc4bc30bb0071a776c4ed39963b7e..4713187e6c3c1b787000b4c42489f4f97b0ee149 100644 (file)
 
 #include <sys/mount.h>
 #include <errno.h>
-#include <sys/stat.h>
 #include <stdlib.h>
-#include <string.h>
-#include <libgen.h>
-#include <assert.h>
 #include <unistd.h>
 #include <ftw.h>
 
 #include "mount-setup.h"
-#include "dev-setup.h"
-#include "log.h"
+//#include "dev-setup.h"
+//#include "bus-util.h"
+//#include "log.h"
 #include "macro.h"
-#include "util.h"
+//#include "util.h"
 #include "label.h"
-#include "set.h"
-#include "strv.h"
+//#include "set.h"
+//#include "strv.h"
 #include "mkdir.h"
 #include "path-util.h"
-#include "missing.h"
+//#include "missing.h"
 #include "virt.h"
-#include "efivars.h"
-
-#ifndef TTY_GID
-#define TTY_GID 5
-#endif
+//#include "efivars.h"
+//#include "smack-util.h"
+#include "cgroup-util.h"
 
 typedef enum MountMode {
         MNT_NONE  =        0,
@@ -65,54 +60,79 @@ typedef struct MountPoint {
 
 /* The first three entries we might need before SELinux is up. The
  * fourth (securityfs) is needed by IMA to load a custom policy. The
- * other ones we can delay until SELinux and IMA are loaded. */
+ * other ones we can delay until SELinux and IMA are loaded. When
+ * SMACK is enabled we need smackfs, too, so it's a fifth one. */
+#ifdef HAVE_SMACK
 #define N_EARLY_MOUNT 5
+#else
+#define N_EARLY_MOUNT 4
+#endif
 
 static const MountPoint mount_table[] = {
-        { "proc",       "/proc",                     "proc",       NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          NULL,       MNT_FATAL|MNT_IN_CONTAINER },
-        { "sysfs",      "/sys",                      "sysfs",      NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          NULL,       MNT_FATAL|MNT_IN_CONTAINER },
-        { "devtmpfs",   "/dev",                      "devtmpfs",   "mode=755", MS_NOSUID|MS_STRICTATIME,
-          NULL,       MNT_FATAL|MNT_IN_CONTAINER },
-        { "securityfs", "/sys/kernel/security",      "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          NULL,       MNT_NONE },
-        { "smackfs",    "/sys/fs/smackfs",           "smackfs",    "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
-          NULL,       MNT_NONE },
-        { "tmpfs",      "/dev/shm",                  "tmpfs",      "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
-          NULL,       MNT_FATAL|MNT_IN_CONTAINER },
-        { "devpts",     "/dev/pts",                  "devpts",     "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
-          NULL,       MNT_IN_CONTAINER },
-        { "tmpfs",      "/run",                      "tmpfs",      "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
-          NULL,       MNT_FATAL|MNT_IN_CONTAINER },
-        { "tmpfs",      "/sys/fs/cgroup",            "tmpfs",      "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
-          NULL,       MNT_IN_CONTAINER },
-#ifdef HAVE_XATTR
-        { "cgroup",     "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          NULL,       MNT_IN_CONTAINER },
+/// UNNEEDED by elogind
+#if 0
+        { "sysfs",       "/sys",                      "sysfs",      NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          NULL,          MNT_FATAL|MNT_IN_CONTAINER },
+        { "proc",        "/proc",                     "proc",       NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          NULL,          MNT_FATAL|MNT_IN_CONTAINER },
+        { "devtmpfs",    "/dev",                      "devtmpfs",   "mode=755",                MS_NOSUID|MS_STRICTATIME,
+          NULL,          MNT_FATAL|MNT_IN_CONTAINER },
+        { "securityfs",  "/sys/kernel/security",      "securityfs", NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          NULL,          MNT_NONE                   },
+#ifdef HAVE_SMACK
+        { "smackfs",     "/sys/fs/smackfs",           "smackfs",    "smackfsdef=*",            MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          mac_smack_use, MNT_FATAL                  },
+        { "tmpfs",       "/dev/shm",                  "tmpfs",      "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+          mac_smack_use, MNT_FATAL                  },
+#endif
+        { "tmpfs",       "/dev/shm",                  "tmpfs",      "mode=1777",               MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+          NULL,          MNT_FATAL|MNT_IN_CONTAINER },
+        { "devpts",      "/dev/pts",                  "devpts",     "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
+          NULL,          MNT_IN_CONTAINER           },
+#ifdef HAVE_SMACK
+        { "tmpfs",       "/run",                      "tmpfs",      "mode=755,smackfsroot=*",  MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+          mac_smack_use, MNT_FATAL                  },
 #endif
-        { "cgroup",     "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          NULL,       MNT_IN_CONTAINER },
-        { "pstore",     "/sys/fs/pstore",            "pstore",     NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          NULL,       MNT_NONE },
+        { "tmpfs",       "/run",                      "tmpfs",      "mode=755",                MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+          NULL,          MNT_FATAL|MNT_IN_CONTAINER },
+        { "cgroup",      "/sys/fs/cgroup",            "cgroup",     "__DEVEL__sane_behavior",  MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          cg_is_unified_wanted, MNT_FATAL|MNT_IN_CONTAINER },
+#endif // 0
+        { "tmpfs",       "/sys/fs/cgroup",            "tmpfs",      "mode=755",                MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
+          cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
+/// UNNEEDED by elogind
+#if 0
+        { "cgroup",      "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          cg_is_legacy_wanted, MNT_IN_CONTAINER           },
+        { "cgroup",      "/sys/fs/cgroup/systemd",    "cgroup",     "none,name=systemd",       MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
+#else
+        { "cgroup",      "/sys/fs/cgroup/elogind",    "cgroup",     "none,name=elogind,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          cg_is_legacy_wanted, MNT_IN_CONTAINER           },
+        { "cgroup",      "/sys/fs/cgroup/elogind",    "cgroup",     "none,name=elogind",       MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
+#endif // 0
+/// UNNEEDED by elogind
+#if 0
+        { "pstore",      "/sys/fs/pstore",            "pstore",     NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          NULL,          MNT_NONE                   },
 #ifdef ENABLE_EFI
-        { "efivarfs",   "/sys/firmware/efi/efivars", "efivarfs",   NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
-          is_efi_boot, MNT_NONE },
+        { "efivarfs",    "/sys/firmware/efi/efivars", "efivarfs",   NULL,                      MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          is_efi_boot,   MNT_NONE                   },
 #endif
+        { "kdbusfs",    "/sys/fs/kdbus",             "kdbusfs",    NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+          is_kdbus_wanted,       MNT_IN_CONTAINER },
+#endif // 0
 };
 
+/// UNNEEDED by elogind
+#if 0
 /* These are API file systems that might be mounted by other software,
  * we just list them here so that we know that we should ignore them */
 
 static const char ignore_paths[] =
         /* SELinux file systems */
         "/sys/fs/selinux\0"
-        "/selinux\0"
-        /* Legacy cgroup mount points */
-        "/dev/cgroup\0"
-        "/cgroup\0"
-        /* Legacy kernel file system */
-        "/proc/bus/usb\0"
         /* Container bind mounts */
         "/proc/sys\0"
         "/dev/console\0"
@@ -140,6 +160,7 @@ bool mount_point_ignore(const char *path) {
 
         return false;
 }
+#endif // 0
 
 static int mount_one(const MountPoint *p, bool relabel) {
         int r;
@@ -153,10 +174,9 @@ static int mount_one(const MountPoint *p, bool relabel) {
         if (relabel)
                 label_fix(p->where, true, true);
 
-        r = path_is_mount_point(p->where, true);
-        if (r < 0)
+        r = path_is_mount_point(p->where, AT_SYMLINK_FOLLOW);
+        if (r < 0 && r != -ENOENT)
                 return r;
-
         if (r > 0)
                 return 0;
 
@@ -166,7 +186,10 @@ static int mount_one(const MountPoint *p, bool relabel) {
 
         /* The access mode here doesn't really matter too much, since
          * the mounted file system will take precedence anyway. */
-        mkdir_p_label(p->where, 0755);
+        if (relabel)
+                mkdir_p_label(p->where, 0755);
+        else
+                mkdir_p(p->where, 0755);
 
         log_debug("Mounting %s to %s of type %s with options %s.",
                   p->what,
@@ -179,7 +202,7 @@ static int mount_one(const MountPoint *p, bool relabel) {
                   p->type,
                   p->flags,
                   p->options) < 0) {
-                log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s: %s", p->where, strerror(errno));
+                log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s at %s: %m", p->type, p->where);
                 return (p->mode & MNT_FATAL) ? -errno : 0;
         }
 
@@ -190,6 +213,8 @@ static int mount_one(const MountPoint *p, bool relabel) {
         return 1;
 }
 
+/// UNNEEDED by elogind
+#if 0
 int mount_setup_early(void) {
         unsigned i;
         int r = 0;
@@ -210,57 +235,30 @@ int mount_setup_early(void) {
 }
 
 int mount_cgroup_controllers(char ***join_controllers) {
+        _cleanup_set_free_free_ Set *controllers = NULL;
         int r;
-        FILE *f;
-        char buf[LINE_MAX];
-        Set *controllers;
 
-        /* Mount all available cgroup controllers that are built into the kernel. */
-
-        f = fopen("/proc/cgroups", "re");
-        if (!f) {
-                log_error("Failed to enumerate cgroup controllers: %m");
+        if (!cg_is_legacy_wanted())
                 return 0;
-        }
 
-        controllers = set_new(string_hash_func, string_compare_func);
-        if (!controllers) {
-                r = log_oom();
-                goto finish;
-        }
-
-        /* Ignore the header line */
-        (void) fgets(buf, sizeof(buf), f);
-
-        for (;;) {
-                char *controller;
-                int enabled = 0;
-
-                if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
-
-                        if (feof(f))
-                                break;
-
-                        log_error("Failed to parse /proc/cgroups.");
-                        r = -EIO;
-                        goto finish;
-                }
+        /* Mount all available cgroup controllers that are built into the kernel. */
 
-                if (!enabled) {
-                        free(controller);
-                        continue;
-                }
+        controllers = set_new(&string_hash_ops);
+        if (!controllers)
+                return log_oom();
 
-                r = set_consume(controllers, controller);
-                if (r < 0) {
-                        log_error("Failed to add controller to set.");
-                        goto finish;
-                }
-        }
+        r = cg_kernel_controllers(controllers);
+        if (r < 0)
+                return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
 
         for (;;) {
-                MountPoint p;
-                char *controller, *where, *options;
+                _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
+                MountPoint p = {
+                        .what = "cgroup",
+                        .type = "cgroup",
+                        .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
+                        .mode = MNT_IN_CONTAINER,
+                };
                 char ***k = NULL;
 
                 controller = set_steal_first(controllers);
@@ -278,14 +276,13 @@ int mount_cgroup_controllers(char ***join_controllers) {
                         for (i = *k, j = *k; *i; i++) {
 
                                 if (!streq(*i, controller)) {
-                                        char *t;
+                                        _cleanup_free_ char *t;
 
                                         t = set_remove(controllers, *i);
                                         if (!t) {
                                                 free(*i);
                                                 continue;
                                         }
-                                        free(t);
                                 }
 
                                 *(j++) = *i;
@@ -294,77 +291,54 @@ int mount_cgroup_controllers(char ***join_controllers) {
                         *j = NULL;
 
                         options = strv_join(*k, ",");
-                        if (!options) {
-                                free(controller);
-                                r = log_oom();
-                                goto finish;
-                        }
-
+                        if (!options)
+                                return log_oom();
                 } else {
                         options = controller;
                         controller = NULL;
                 }
 
                 where = strappend("/sys/fs/cgroup/", options);
-                if (!where) {
-                        free(options);
-                        r = log_oom();
-                        goto finish;
-                }
+                if (!where)
+                        return log_oom();
 
-                zero(p);
-                p.what = "cgroup";
                 p.where = where;
-                p.type = "cgroup";
                 p.options = options;
-                p.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV;
-                p.mode = MNT_IN_CONTAINER;
 
                 r = mount_one(&p, true);
-                free(controller);
-                free(where);
-
-                if (r < 0) {
-                        free(options);
-                        goto finish;
-                }
+                if (r < 0)
+                        return r;
 
                 if (r > 0 && k && *k) {
                         char **i;
 
                         for (i = *k; *i; i++) {
-                                _cleanup_free_ char *t;
+                                _cleanup_free_ char *t = NULL;
 
                                 t = strappend("/sys/fs/cgroup/", *i);
-                                if (!t) {
-                                        r = log_oom();
-                                        free(options);
-                                        goto finish;
-                                }
+                                if (!t)
+                                        return log_oom();
 
                                 r = symlink(options, t);
-                                if (r < 0 && errno != EEXIST) {
-                                        log_error("Failed to create symlink %s: %m", t);
-                                        r = -errno;
-                                        free(options);
-                                        goto finish;
-                                }
+                                if (r < 0 && errno != EEXIST)
+                                        return log_error_errno(errno, "Failed to create symlink %s: %m", t);
+#ifdef SMACK_RUN_LABEL
+                                r = mac_smack_copy(t, options);
+                                if (r < 0 && r != -EOPNOTSUPP)
+                                        return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", options, t);
+#endif
                         }
                 }
-
-                free(options);
         }
 
-        r = 0;
+        /* Now that we mounted everything, let's make the tmpfs the
+         * cgroup file systems are mounted into read-only. */
+        (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
 
-finish:
-        set_free_free(controllers);
-
-        fclose(f);
-
-        return r;
+        return 0;
 }
 
+#if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
 static int nftw_cb(
                 const char *fpath,
                 const struct stat *sb,
@@ -386,24 +360,27 @@ static int nftw_cb(
 
         return FTW_CONTINUE;
 };
+#endif
+#endif // 0
 
 int mount_setup(bool loaded_policy) {
-
-        static const char relabel[] =
-                "/run/initramfs/root-fsck\0"
-                "/run/initramfs/shutdown\0";
-
-        int r;
         unsigned i;
-        const char *j;
+        int r = 0;
 
         for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
-                r = mount_one(mount_table + i, true);
+                int j;
 
-                if (r < 0)
-                        return r;
+                j = mount_one(mount_table + i, loaded_policy);
+                if (r == 0)
+                        r = j;
         }
 
+        if (r < 0)
+                return r;
+
+/// elogind does not control /, /dev, /run and /run/systemd/* are setup elsewhere.
+#if 0
+#if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
         /* Nodes in devtmpfs and /run need to be manually updated for
          * the appropriate labels, after mounting. The other virtual
          * API file systems like /sys and /proc do not need that, they
@@ -417,20 +394,17 @@ int mount_setup(bool loaded_policy) {
                 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
                 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
 
-                /* Explicitly relabel these */
-                NULSTR_FOREACH(j, relabel)
-                        label_fix(j, true, false);
-
                 after_relabel = now(CLOCK_MONOTONIC);
 
                 log_info("Relabelled /dev and /run in %s.",
                          format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
         }
+#endif
 
         /* Create a few default symlinks, which are normally created
          * by udevd, but some scripts might need them before we start
          * udevd. */
-        dev_setup(NULL);
+        dev_setup(NULL, UID_INVALID, GID_INVALID);
 
         /* Mark the root directory as shared in regards to mount
          * propagation. The kernel defaults to "private", but we think
@@ -440,7 +414,7 @@ int mount_setup(bool loaded_policy) {
          * propagation mode to private if needed. */
         if (detect_container(NULL) <= 0)
                 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
-                        log_warning("Failed to set up the root directory for shared mount propagation: %m");
+                        log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
 
         /* Create a few directories we always want around, Note that
          * sd_booted() checks for /run/systemd/system, so this mkdir
@@ -450,6 +424,7 @@ int mount_setup(bool loaded_policy) {
         mkdir_label("/run/systemd", 0755);
         mkdir_label("/run/systemd/system", 0755);
         mkdir_label("/run/systemd/inaccessible", 0000);
+#endif // 0
 
         return 0;
 }