PROC_READ_ONLY("/proc/irq"),
PROC_READ_ONLY("/proc/scsi"),
- { "mqueue", "/dev/mqueue", "mqueue", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "mqueue", "/dev/mqueue", "mqueue", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
MOUNT_IN_USERNS|MOUNT_MKDIR },
/* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
- { "tmpfs", "/tmp", "tmpfs", "mode=1777" TMPFS_LIMITS_TMP, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/tmp", "tmpfs", "mode=1777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR },
- { "tmpfs", "/sys", "tmpfs", "mode=555" TMPFS_LIMITS_SYS, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "tmpfs", "/sys", "tmpfs", "mode=555" TMPFS_LIMITS_SYS, MS_NOSUID|MS_NOEXEC|MS_NODEV,
MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR },
- { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR }, /* skipped if above was mounted */
- { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
MOUNT_FATAL|MOUNT_MKDIR }, /* skipped if above was mounted */
- { "tmpfs", "/dev", "tmpfs", "mode=755" TMPFS_LIMITS_DEV, MS_NOSUID|MS_STRICTATIME,
+ { "tmpfs", "/dev", "tmpfs", "mode=755" TMPFS_LIMITS_DEV, MS_NOSUID|MS_STRICTATIME,
MOUNT_FATAL|MOUNT_MKDIR },
- { "tmpfs", "/dev/shm", "tmpfs", "mode=1777" TMPFS_LIMITS_DEV_SHM, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/dev/shm", "tmpfs", "mode=1777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
MOUNT_FATAL|MOUNT_MKDIR },
- { "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
MOUNT_FATAL|MOUNT_MKDIR },
-
+ { "/run/host", "/run/host", NULL, NULL, MS_BIND,
+ MOUNT_FATAL|MOUNT_MKDIR|MOUNT_PREFIX_ROOT }, /* Prepare this so that we can make it read-only when we are done */
+ { "/etc/os-release", "/run/host/os-release", NULL, NULL, MS_BIND,
+ MOUNT_TOUCH }, /* As per kernel interface requirements, bind mount first (creating mount points) and make read-only later */
+ { "/usr/lib/os-release", "/run/host/os-release", NULL, NULL, MS_BIND,
+ MOUNT_FATAL }, /* If /etc/os-release doesn't exist use the version in /usr/lib as fallback */
+ { NULL, "/run/host/os-release", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ MOUNT_FATAL },
+ { NULL, "/run/host", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ MOUNT_FATAL|MOUNT_IN_USERNS },
#if HAVE_SELINUX
- { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND,
+ { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND,
MOUNT_MKDIR }, /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */
- { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
0 }, /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */
#endif
};
int r;
for (k = 0; k < ELEMENTSOF(mount_table); k++) {
- _cleanup_free_ char *where = NULL, *options = NULL;
- const char *o;
+ _cleanup_free_ char *where = NULL, *options = NULL, *prefixed = NULL;
bool fatal = FLAGS_SET(mount_table[k].mount_settings, MOUNT_FATAL);
+ const char *o;
if (in_userns != FLAGS_SET(mount_table[k].mount_settings, MOUNT_IN_USERNS))
continue;
continue;
}
- if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_MKDIR)) {
+ if ((mount_table[k].mount_settings & (MOUNT_MKDIR|MOUNT_TOUCH)) != 0) {
uid_t u = (use_userns && !in_userns) ? uid_shift : UID_INVALID;
- r = mkdir_p_safe(dest, where, 0755, u, u, 0);
+
+ if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH))
+ r = mkdir_parents_safe(dest, where, 0755, u, u, 0);
+ else
+ r = mkdir_p_safe(dest, where, 0755, u, u, 0);
if (r < 0 && r != -EEXIST) {
if (fatal && r != -EROFS)
return log_error_errno(r, "Failed to create directory %s: %m", where);
}
}
+ if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH)) {
+ r = touch(where);
+ if (r < 0 && r != -EEXIST) {
+ if (fatal && r != -EROFS)
+ return log_error_errno(r, "Failed to create file %s: %m", where);
+
+ log_debug_errno(r, "Failed to create file %s: %m", where);
+ if (r != -EROFS)
+ continue;
+ }
+ }
+
o = mount_table[k].options;
if (streq_ptr(mount_table[k].type, "tmpfs")) {
r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
o = options;
}
+ if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_PREFIX_ROOT)) {
+ /* Optionally prefix the mount source with the root dir. This is useful in bind
+ * mounts to be created within the container image before we transition into it. Note
+ * that MOUNT_IN_USERNS is run after we transitioned hence prefixing is not ncessary
+ * for those. */
+ r = chase_symlinks(mount_table[k].what, dest, CHASE_PREFIX_ROOT, &prefixed, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].what);
+ }
+
r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG,
- mount_table[k].what,
+ prefixed ?: mount_table[k].what,
where,
mount_table[k].type,
mount_table[k].flags,
projects / thirdparty / systemd.git / blobdiff