-----Original Message-----
From: linux-raid-owner@vger.kernel.org [mailto:linux-raid-
owner@vger.kernel.org] On Behalf Of NeilBrown
Sent: Wednesday, December 14, 2011 9:09 AM
To: Kwolek, Adam
Cc: linux-raid@vger.kernel.org; Ciechanowski, Ed; Labun, Marcin; Williams,
Dan J
Subject: Re: [PATCH 1/3] FIX: Do not allow for multiple reshape_array()
execution during reshape_container() call

On Tue, 13 Dec 2011 11:12:10 +0100 Adam Kwolek
[off-list ref] wrote:

quoted

It can happen during reshape restart that reshape_array() can exit
without error (e.g. Grow.c:1915) and reshape is not moved to next array.
reshape_array() is called again for the same device.
Do not allow for such execution and check if last reshaped array is
not the current one.
This patch can be treat not as solution, but it allows for such errors
detection.

Signed-off-by: Adam Kwolek <redacted>

Hi Adam.
 I'm afraid I understand the problem that requires this patch.
Could you please outline how it can happen that "reshape_array() can exit
without error (e.g. Grow.c:1915) and reshape is not moved to next array."

It happens (Lukasz Dorau saw it), when mdadm assembles array in initramfs and even mdadm is directed to assemble array only (--freeze-reshape is used),
filesystetem pivot occurs too early and array is in not stable state. During reshape continuation analyse_change() returns 0 in backup_blocks and reshape_container calls reshape_array() again.
This time restart variable is cleaned and mdadm finishes with error. For now I wanted to put better error description to help Lukasz in problem investigation.
At this moment I do not know more details, but probably when I finish checking raid0 md problem (using patch you sent) I'll go in to tihis.

Also the comment in the code seems to be incomplete.  Could you complete
it please?
Thanks.

The other 2 patches in the series are fine.

Thanks,
NeilBrown

quoted

---

 Grow.c |   12 ++++++++++++
 1 files changed, 12 insertions(+), 0 deletions(-)

diff --git a/Grow.c b/Grow.c
index 184a973..1828f83 100644
--- a/Grow.c
+++ b/Grow.c

@@ -2462,6 +2462,7 @@ int reshape_container(char *container, char

*devname,  {
 	struct mdinfo *cc = NULL;
 	int rv = restart;
+	int last_devnum = -1;

 	/* component_size is not meaningful for a container,
 	 * so pass '-1' meaning 'no change'

@@ -2546,6 +2547,17 @@ int reshape_container(char *container, char

*devname,

quoted

 		if (!adev)
 			adev = content->text_version;

+		if (last_devnum == mdstat->devnum) {
+			/* do not allow for reentry reshape_array()
+			 * for the same device. It can happen when

resahpe_array
This one.  Should there be more to this sentence?

You are right, I've missed it. I'll correct comment and resend this patch.

BR
Adam

NB

quoted

+			 */
+			printf(Name ": Multiple reshape execution detected

for "

quoted

+			       "device  %s.", adev);
+			close(fd);
+			break;
+		}
+		last_devnum = mdstat->devnum;
+
 		sysfs_init(content, fd, mdstat->devnum);

 		rv = reshape_array(container, fd, adev, st,