criu-ns: Add tests for criu-ns script

.cirrus.yml

@@ -134,11 +134,11 @@ task:
     memory: 8G
 
   setup_script: |
-    # EPEL is needed for python2-future, python2-junit_xml, python-flake8 and libbsd-devel.
+    # EPEL is needed for python2-future, python2-junit_xml, python-pathlib, python-flake8 and libbsd-devel.
     # Do not fail if latest epel repository definition is already installed
     yum install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm || :
     ln -sf /usr/include/google/protobuf/descriptor.proto images/google/protobuf/descriptor.proto
-    yum install -y findutils gcc git gnutls-devel iproute iptables libaio-devel libasan libcap-devel libnet-devel libnl3-devel libbsd-devel make procps-ng protobuf-c-devel protobuf-devel protobuf-python python python-flake8 python-ipaddress python2-future python2-junit_xml python-yaml python-six sudo tar which e2fsprogs python2-pip rubygem-asciidoctor libselinux-devel
+    yum install -y findutils gcc git gnutls-devel iproute iptables libaio-devel libasan libcap-devel libnet-devel libnl3-devel libbsd-devel make procps-ng protobuf-c-devel protobuf-devel protobuf-python python python-flake8 python-ipaddress python2-future python2-junit_xml python-yaml python-six python-pathlib sudo tar which e2fsprogs python2-pip rubygem-asciidoctor libselinux-devel
     # Even with selinux in permissive mode the selinux tests will be executed
     # The Cirrus CI user runs as a service from selinux point of view and is
     # much more restricted than a normal shell (system_u:system_r:unconfined_service_t:s0)

Makefile

@@ -438,6 +438,7 @@ lint:
 	flake8 --config=scripts/flake8.cfg lib/py/images/pb2dict.py
 	flake8 --config=scripts/flake8.cfg lib/py/images/images.py
 	flake8 --config=scripts/flake8.cfg scripts/criu-ns
+	flake8 --config=scripts/flake8.cfg test/others/criu-ns/run.py
 	flake8 --config=scripts/flake8.cfg crit/setup.py
 	flake8 --config=scripts/flake8.cfg scripts/uninstall_module.py
 	flake8 --config=scripts/flake8.cfg coredump/

scripts/ci/run-ci-tests.sh

@@ -260,6 +260,7 @@ if [ -n "$TRAVIS" ] || [ -n "$CIRCLECI" ]; then
        # Error (criu/tty.c:1014): tty: Don't have tty to inherit session from, aborting
        make -C test/others/shell-job/ run
 fi
+make -C test/others/criu-ns/ run
 make -C test/others/skip-file-rwx-check/ run
 make -C test/others/rpc/ run
 

scripts/criu-ns

@@ -1,4 +1,4 @@
-#!/usr/bin/env python3
+#!/usr/bin/env python
 import ctypes
 import ctypes.util
 import errno
@@ -71,7 +71,19 @@ def _wait_for_process_status(criu_pid):
         try:
             (pid, status) = os.wait()
             if pid == criu_pid:
-                return os.waitstatus_to_exitcode(status)
+                # The following code block is based on
+                # os.waitstatus_to_exitcode() introduced in Python 3.9
+                # and we implement this for comparability with older
+                # versions of Python.
+                if os.WIFSIGNALED(status):
+                    return os.WTERMSIG(status)
+                elif os.WIFEXITED(status):
+                    return os.WEXITSTATUS(status)
+                elif os.WIFSTOPPED(status):
+                    return os.WSTOPSIG(status)
+                else:
+                    raise Exception("CRIU was terminated by an "
+                                    "unidentified reason")
         except OSError:
             return -251
 
@@ -81,8 +93,21 @@ def run_criu(args):
     Spawn CRIU binary
     """
     print(sys.argv)
-    os.execlp('criu', *['criu'] + args)
-    raise OSError(errno.ENOENT, "No such command")
+
+    if "--criu-binary" in args:
+        try:
+            opt_index = args.index("--criu-binary")
+            path = args[opt_index + 1]
+            del args[opt_index:opt_index + 2]
+            args.insert(0, "criu")
+            os.execv(path, args)
+            raise OSError(errno.ENOENT, "No such command")
+        except (ValueError, IndexError, FileNotFoundError):
+            raise OSError(errno.ENOENT, "--criu-binary missing argument")
+    else:
+        args.insert(0, "criu")
+        os.execvp("criu", args)
+        raise OSError(errno.ENOENT, "No such command")
 
 
 # pidns_holder creates a process that is reparented to the init.

test/Makefile

@@ -12,7 +12,7 @@ all:
 	$(MAKE) zdtm-freezer
 .PHONY: all
 
-TESTS = unix-callback mem-snap rpc libcriu mounts/ext security pipes crit socketpairs overlayfs mnt-ext-dev shell-job skip-file-rwx-check
+TESTS = unix-callback mem-snap rpc libcriu mounts/ext security pipes crit socketpairs overlayfs mnt-ext-dev shell-job criu-ns skip-file-rwx-check
 
 other:
 	for t in $(TESTS); do				\

test/others/criu-ns/Makefile

@@ -0,0 +1,3 @@
+run:
+	@make -C ../.. zdtm_ct
+	../../zdtm_ct run.py

test/others/criu-ns/run.py

@@ -0,0 +1,258 @@
+#!/usr/bin/env python
+
+import fcntl
+import os
+import pathlib
+import pty
+import shutil
+import subprocess
+import sys
+import termios
+import time
+
+
+CRIU_BIN = "../../../criu/criu"
+CRIU_NS = "../../../scripts/criu-ns"
+IMG_DIR = "dumpdir"
+DUMP_LOG = "dump.log"
+RESTORE_LOG = "restore.log"
+PIDFILE = "pidfile"
+
+
+def check_dumpdir(path=IMG_DIR):
+    if os.path.isdir(path):
+        shutil.rmtree(path)
+    os.mkdir(path, 0o755)
+
+
+def set_blocking(fd, blocking):
+    """Implement os.set_blocking() for compatibility with Python
+    versions earlier than 3.5"""
+    flags = fcntl.fcntl(fd, fcntl.F_GETFL)
+
+    if blocking:
+        flags &= ~os.O_NONBLOCK
+    else:
+        flags |= os.O_NONBLOCK
+
+    fcntl.fcntl(fd, fcntl.F_SETFL, flags)
+
+
+def run_task_with_own_pty(task):
+    fd_m, fd_s = pty.openpty()
+
+    pid = os.fork()
+    if pid == 0:
+        os.close(fd_m)
+        os.setsid()
+        os.dup2(fd_s, 0)
+        os.dup2(fd_s, 1)
+        os.dup2(fd_s, 2)
+        fcntl.ioctl(fd_s, termios.TIOCSCTTY, 1)
+        os.close(fd_s)
+        task()
+        exit(0)
+
+    os.close(fd_s)
+    fd_m = os.fdopen(fd_m, "rb")
+    set_blocking(fd_m.fileno(), False)
+
+    while True:
+        try:
+            data = fd_m.read()
+        except IOError:
+            break
+        if data is not None:
+            print(data.decode("utf-8"))
+
+    _, status = os.waitpid(pid, 0)
+
+    try:
+        data = fd_m.read()
+    except IOError as err:
+        print(err)
+
+    if data is not None:
+        print(data.decode("utf-8"))
+    fd_m.close()
+
+    if status != 0:
+        print("task %s exited badly: %d" % (task.__name__, status))
+        exit(1)
+
+    return 0
+
+
+def create_pty():
+    fd_m, fd_s = pty.openpty()
+    return (os.fdopen(fd_m, "wb"), os.fdopen(fd_s, "wb"))
+
+
+def create_isolated_dumpee():
+    pathlib.Path("running").touch()
+    fd_m, fd_s = create_pty()
+    pid = os.fork()
+    if pid == 0:
+        os.setsid()
+        os.dup2(fd_s.fileno(), 0)
+        os.dup2(fd_s.fileno(), 1)
+        os.dup2(fd_s.fileno(), 2)
+        fcntl.ioctl(fd_s.fileno(), termios.TIOCSCTTY, 1)
+        while True:
+            if not os.access("running", os.F_OK):
+                sys.exit(0)
+            time.sleep(1)
+    fd_m.close()
+    fd_s.close()
+    return pid
+
+
+def criu_ns_dump(pid, shell_job=False):
+    cmd = [CRIU_NS, "dump", "-D", IMG_DIR, "-v4", "-t", str(pid),
+           "--log-file", DUMP_LOG, "--criu-binary", CRIU_BIN]
+    if shell_job:
+        cmd.append("--shell-job")
+    ret = subprocess.Popen(cmd).wait()
+    return ret
+
+
+def criu_ns_restore(shell_job=False, restore_detached=False):
+    cmd = [CRIU_NS, "restore", "-D", IMG_DIR, "-v4", "--log-file",
+           RESTORE_LOG, "--criu-binary", CRIU_BIN]
+    if shell_job:
+        cmd.append("--shell-job")
+    if restore_detached:
+        cmd += ["--restore-detached", "--pidfile", PIDFILE]
+    ret = subprocess.Popen(cmd).wait()
+    return ret
+
+
+def read_log_file(filename):
+    logfile_path = os.path.join(IMG_DIR, filename)
+    with open(logfile_path) as logfile:
+        print(logfile.read())
+
+
+def test_dump_and_restore_with_shell_job():
+    print("Test criu-ns dump and restore with --shell-job option")
+    check_dumpdir()
+    pathlib.Path("running").touch()
+    pid = os.fork()
+    if pid == 0:
+        while True:
+            if not os.access("running", os.F_OK):
+                sys.exit(0)
+            time.sleep(1)
+
+    ret = criu_ns_dump(pid, shell_job=True)
+    if ret != 0:
+        read_log_file(DUMP_LOG)
+        sys.exit(ret)
+
+    os.unlink("running")
+    fd_m, fd_s = create_pty()
+    pid = os.fork()
+    if pid == 0:
+        os.setsid()
+        fd_m.close()
+        # since criu-ns takes control of the tty stdin
+        os.dup2(fd_s.fileno(), 0)
+        ret = criu_ns_restore(shell_job=True)
+        if ret != 0:
+            read_log_file(RESTORE_LOG)
+            sys.exit(ret)
+        os._exit(0)
+
+    fd_s.close()
+    os.waitpid(pid, 0)
+
+
+def test_dump_and_restore_without_shell_job(restore_detached=False):
+    print("Test criu-ns dump and restore with an isolated process"
+          "(%d)" % restore_detached)
+    check_dumpdir()
+    pid = create_isolated_dumpee()
+    ret = criu_ns_dump(pid)
+    if ret != 0:
+        read_log_file(DUMP_LOG)
+        sys.exit(ret)
+
+    if not restore_detached:
+        os.unlink("running")
+
+    pid = os.fork()
+    if pid == 0:
+        os.setsid()
+        ret = criu_ns_restore(restore_detached=restore_detached)
+        if ret != 0:
+            read_log_file(RESTORE_LOG)
+            sys.exit(ret)
+        os._exit(0)
+
+    os.waitpid(pid, 0)
+
+
+def test_dump_and_restore_in_pidns():
+    if os.system("grep NSpid /proc/self/status"):
+        return
+
+    print("Test criu-ns dump and restore in namespaces")
+
+    def _dump():
+        pid = create_isolated_dumpee()
+        ret = criu_ns_dump(pid)
+        if ret != 0:
+            read_log_file(DUMP_LOG)
+            sys.exit(ret)
+
+    def _restore():
+        ret = criu_ns_restore(restore_detached=True)
+        if ret != 0:
+            read_log_file(RESTORE_LOG)
+            sys.exit(ret)
+
+    def _get_restored_pid():
+        restored_pid = 0
+        pidfile_path = os.path.join(IMG_DIR, PIDFILE)
+        if not os.path.exists(pidfile_path):
+            raise FileNotFoundError("pidfile not found")
+        with open(pidfile_path, "r") as pidfile:
+            restored_pid = pidfile.read().strip()
+        return int(restored_pid)
+
+    def _redump():
+        global IMG_DIR
+        try:
+            restored_pid = _get_restored_pid()
+        except FileNotFoundError:
+            sys.exit(1)
+        IMG_DIR = "dumpdir2"
+        check_dumpdir(IMG_DIR)
+        ret = criu_ns_dump(restored_pid)
+        if ret != 0:
+            read_log_file(DUMP_LOG)
+            sys.exit(ret)
+
+    def _re_restore():
+        os.unlink("running")
+        ret = criu_ns_restore()
+        if ret != 0:
+            read_log_file(RESTORE_LOG)
+            sys.exit(ret)
+
+    check_dumpdir()
+    _dump()
+    _restore()
+    _redump()
+    _re_restore()
+
+
+def main():
+    test_dump_and_restore_with_shell_job()
+    test_dump_and_restore_without_shell_job()
+    test_dump_and_restore_without_shell_job(restore_detached=True)
+    test_dump_and_restore_in_pidns()
+
+
+if __name__ == "__main__":
+    run_task_with_own_pty(main)