chiark / gitweb /
core: provide /dev/ptmx as symlink in PrivateDevices= execution environments
[elogind.git] / src / core / namespace.c
index 85147be13017c0fab13f97c0d58291b7e17dab34..1f987a4b9d03ceb225434d2e7d2d2f2a5a3a63c1 100644 (file)
@@ -39,6 +39,9 @@
 #include "missing.h"
 #include "execute.h"
 #include "loopback-setup.h"
+#include "mkdir.h"
+#include "dev-setup.h"
+#include "def.h"
 
 typedef enum MountMode {
         /* This is ordered by priority! */
@@ -46,6 +49,7 @@ typedef enum MountMode {
         READONLY,
         PRIVATE_TMP,
         PRIVATE_VAR_TMP,
+        PRIVATE_DEV,
         READWRITE
 } MountMode;
 
@@ -65,7 +69,7 @@ static int append_mounts(BindMount **p, char **strv, MountMode mode) {
 
                 (*p)->ignore = false;
 
-                if ((mode == INACCESSIBLE || mode == READONLY) && (*i)[0] == '-') {
+                if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
                         (*p)->ignore = true;
                         (*i)++;
                 }
@@ -129,6 +133,141 @@ static void drop_duplicates(BindMount *m, unsigned *n) {
         *n = t - m;
 }
 
+static int mount_dev(BindMount *m) {
+        static const char devnodes[] =
+                "/dev/null\0"
+                "/dev/zero\0"
+                "/dev/full\0"
+                "/dev/random\0"
+                "/dev/urandom\0"
+                "/dev/tty\0";
+
+        char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
+        const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devkdbus = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
+        _cleanup_umask_ mode_t u;
+        int r;
+
+        assert(m);
+
+        u = umask(0000);
+
+        if (!mkdtemp(temporary_mount))
+                return -errno;
+
+        dev = strappenda(temporary_mount, "/dev");
+        mkdir(dev, 0755);
+        if (mount("tmpfs", dev, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=755") < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        devpts = strappenda(temporary_mount, "/dev/pts");
+        mkdir(devpts, 0755);
+        if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        devptmx = strappenda(temporary_mount, "/dev/ptmx");
+        symlink("pts/ptmx", devptmx);
+
+        devshm = strappenda(temporary_mount, "/dev/shm");
+        mkdir(devshm, 01777);
+        r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
+        if (r < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        devmqueue = strappenda(temporary_mount, "/dev/mqueue");
+        mkdir(devmqueue, 0755);
+        mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
+
+        devkdbus = strappenda(temporary_mount, "/dev/kdbus");
+        mkdir(devkdbus, 0755);
+        mount("/dev/kdbus", devkdbus, NULL, MS_BIND, NULL);
+
+        devhugepages = strappenda(temporary_mount, "/dev/hugepages");
+        mkdir(devhugepages, 0755);
+        mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
+
+        devlog = strappenda(temporary_mount, "/dev/log");
+        symlink("/run/systemd/journal/dev-log", devlog);
+
+        NULSTR_FOREACH(d, devnodes) {
+                _cleanup_free_ char *dn = NULL;
+                struct stat st;
+
+                r = stat(d, &st);
+                if (r < 0) {
+
+                        if (errno == ENOENT)
+                                continue;
+
+                        r = -errno;
+                        goto fail;
+                }
+
+                if (!S_ISBLK(st.st_mode) &&
+                    !S_ISCHR(st.st_mode)) {
+                        r = -EINVAL;
+                        goto fail;
+                }
+
+                if (st.st_rdev == 0)
+                        continue;
+
+                dn = strappend(temporary_mount, d);
+                if (!dn) {
+                        r = -ENOMEM;
+                        goto fail;
+                }
+
+                r = mknod(dn, st.st_mode, st.st_rdev);
+                if (r < 0) {
+                        r = -errno;
+                        goto fail;
+                }
+        }
+
+        dev_setup(temporary_mount);
+
+        if (mount(dev, "/dev/", NULL, MS_MOVE, NULL) < 0) {
+                r = -errno;
+                goto fail;
+        }
+
+        rmdir(dev);
+        rmdir(temporary_mount);
+
+        return 0;
+
+fail:
+        if (devpts)
+                umount(devpts);
+
+        if (devshm)
+                umount(devshm);
+
+        if (devkdbus)
+                umount(devkdbus);
+
+        if (devhugepages)
+                umount(devhugepages);
+
+        if (devmqueue)
+                umount(devmqueue);
+
+        if (dev) {
+                umount(dev);
+                rmdir(dev);
+        }
+
+        rmdir(temporary_mount);
+
+        return r;
+}
+
 static int apply_mount(
                 BindMount *m,
                 const char *tmp_dir,
@@ -141,6 +280,9 @@ static int apply_mount(
 
         switch (m->mode) {
 
+        case PRIVATE_DEV:
+                return mount_dev(m);
+
         case INACCESSIBLE:
                 what = "/run/systemd/inaccessible";
                 break;
@@ -194,6 +336,9 @@ int setup_namespace(
                 char** inaccessible_dirs,
                 char* tmp_dir,
                 char* var_tmp_dir,
+                bool private_dev,
+                ProtectedHome protected_home,
+                bool read_only_system,
                 unsigned mount_flags) {
 
         BindMount *m, *mounts = NULL;
@@ -209,7 +354,10 @@ int setup_namespace(
         n = !!tmp_dir + !!var_tmp_dir +
                 strv_length(read_write_dirs) +
                 strv_length(read_only_dirs) +
-                strv_length(inaccessible_dirs);
+                strv_length(inaccessible_dirs) +
+                private_dev +
+                (protected_home != PROTECTED_HOME_NO ? 2 : 0) +
+                (read_only_system ? 2 : 0);
 
         if (n > 0) {
                 m = mounts = (BindMount *) alloca(n * sizeof(BindMount));
@@ -237,30 +385,52 @@ int setup_namespace(
                         m++;
                 }
 
+                if (private_dev) {
+                        m->path = "/dev";
+                        m->mode = PRIVATE_DEV;
+                        m++;
+                }
+
+                if (protected_home != PROTECTED_HOME_NO) {
+                        r = append_mounts(&m, STRV_MAKE("-/home", "-/run/user"), protected_home == PROTECTED_HOME_READ_ONLY ? READONLY : INACCESSIBLE);
+                        if (r < 0)
+                                return r;
+                }
+
+                if (read_only_system) {
+                        r = append_mounts(&m, STRV_MAKE("/usr", "-/boot"), READONLY);
+                        if (r < 0)
+                                return r;
+                }
+
                 assert(mounts + n == m);
 
                 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
                 drop_duplicates(mounts, &n);
         }
 
-        /* Remount / as SLAVE so that nothing now mounted in the namespace
-           shows up in the parent */
-        if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
-                return -errno;
+        if (n > 0) {
+                /* Remount / as SLAVE so that nothing now mounted in the namespace
+                   shows up in the parent */
+                if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
+                        return -errno;
 
-        for (m = mounts; m < mounts + n; ++m) {
-                r = apply_mount(m, tmp_dir, var_tmp_dir);
-                if (r < 0)
-                        goto fail;
-        }
+                for (m = mounts; m < mounts + n; ++m) {
+                        r = apply_mount(m, tmp_dir, var_tmp_dir);
+                        if (r < 0)
+                                goto fail;
+                }
 
-        for (m = mounts; m < mounts + n; ++m) {
-                r = make_read_only(m);
-                if (r < 0)
-                        goto fail;
+                for (m = mounts; m < mounts + n; ++m) {
+                        r = make_read_only(m);
+                        if (r < 0)
+                                goto fail;
+                }
         }
 
-        /* Remount / as the desired mode */
+        /* Remount / as the desired mode. Not that this will not
+         * reestablish propagation from our side to the host, since
+         * what's disconnected is disconnected. */
         if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
                 r = -errno;
                 goto fail;
@@ -269,9 +439,11 @@ int setup_namespace(
         return 0;
 
 fail:
-        for (m = mounts; m < mounts + n; ++m)
-                if (m->done)
-                        umount2(m->path, MNT_DETACH);
+        if (n > 0) {
+                for (m = mounts; m < mounts + n; ++m)
+                        if (m->done)
+                                umount2(m->path, MNT_DETACH);
+        }
 
         return r;
 }
@@ -431,3 +603,11 @@ fail:
 
         return r;
 }
+
+static const char *const protected_home_table[_PROTECTED_HOME_MAX] = {
+        [PROTECTED_HOME_NO] = "no",
+        [PROTECTED_HOME_YES] = "yes",
+        [PROTECTED_HOME_READ_ONLY] = "read-only",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(protected_home, ProtectedHome);