diff --git a/criu/seize.c b/criu/seize.c index 9bd1832d9b..c9d23365d4 100644 --- a/criu/seize.c +++ b/criu/seize.c @@ -542,7 +542,7 @@ static int freeze_processes(void) enum freezer_state state = THAWED; static const unsigned long step_ms = 100; - unsigned long nr_attempts = (opts.timeout * 1000000) / step_ms; + unsigned long nr_attempts = (opts.timeout * 1000) / step_ms; unsigned long i = 0; const struct timespec req = { @@ -555,7 +555,7 @@ static int freeze_processes(void) * If timeout is turned off, lets * wait for at least 10 seconds. */ - nr_attempts = (10 * 1000000) / step_ms; + nr_attempts = (10 * 1000) / step_ms; } pr_debug("freezing processes: %lu attempts with %lu ms steps\n", nr_attempts, step_ms); @@ -594,9 +594,39 @@ static int freeze_processes(void) if (state == FROZEN) break; - if (alarm_timeouted()) + if (alarm_timeouted()) { + pr_err("Unable to freeze cgroup %s (timed out)\n", opts.freeze_cgroup); goto err; + } nanosleep(&req, NULL); + + if (cgroup_v2) + continue; + + /* As per older kernel docs (freezer-subsystem.txt before + * the kernel commit ef9fe980c6fcc1821), if FREEZING is seen, + * userspace should either retry or thaw. While current + * kernel cgroup v1 docs no longer mention a need to retry, + * even recent kernels can't reliably freeze a cgroup v1. + * + * Let's keep asking the kernel to freeze from time to time. + * In addition, do occasional thaw/sleep/freeze. + * + * This is still a game of chances (the real fix belongs to the kernel) + * but these kludges might improve the probability of success. + * + * Cgroup v2 does not have this problem. + */ + switch (i%32) { + case 30: + freezer_write_state(fd, THAWED); + break; + case 9: + case 20: + case 31: + freezer_write_state(fd, FROZEN); + break; + } } if (i > nr_attempts) {