mdadm: Do not start reshape before switchroot

There are numerous issues for --grow --continue in switchroot phrase,
they include:
* Events being missed for restarting grow-continue service. This is
  apparent mostly on OS on RAID scenarios. When a checkpoint (next step)
  is committed, we have no reliable way to gracefully stop reshape until
  it reaches that checkpoint. During boot, there's heavy I/O utilisation,
  which causes sync speed drop, and naturally checkpoint takes longer to
  reach. This further causes systemd to forcefully kill grow-continue
  service due to timeouts, which results in udev event being missed for
  grow-continue service restart.
* Grow-continue (seemingly) was not designed to be restarted without
  reassembly, some things like stopping chunksize (to lower) migration
  were straight up not working until recently.
This patch makes grow-continue (actual reshape) start after switchroot
phrase. This way we should not encounter issues related to restarting
the service.

Add checks not start a reshape if in initrd, let it initialise only.
Change grow-continue udev rule to be triggered whenever there's a
reshape happening in metadata, rely on udev event to kick reshape after
switchroot. Add handle_forking helper function for reshapes to avoid
duplicating code.

Signed-off-by: Mateusz Kusiak <mateusz.kusiak@intel.com>

1 files changed, 1 insertions, 0 deletions

diff --git a/util.c b/util.c
index 6aa44a80..8099852f 100644
--- a/util.c
+++ b/util.c
@@ -2307,6 +2307,7 @@ int continue_via_systemd(char *devnm, char *service_name, char *prefix)
 	int pid, status;
 	char pathbuf[1024];
 
+	dprintf("Start %s service\n", service_name);
 	/* Simply return that service cannot be started */
 	if (check_env("MDADM_NO_SYSTEMCTL"))
 		return 0;