RE: [PATCH 07/14] FIX: Extend size of raid0 array

On Tuesday, April 17, 2012 4:53 AM NeilBrown wrote:

On Fri, 13 Apr 2012 16:52:01 +0200 Adam Kwolek [off-list ref]
wrote:

quoted

For raid0, takeover operation is required for size change.
Add takeover to degraded raid4 before size change and back to raid0 after.
Array information has to be read again from md after takeover.

Signed-off-by: Adam Kwolek <redacted>
---
 Grow.c |   35 +++++++++++++++++++++++++++++++++++
 1 files changed, 35 insertions(+), 0 deletions(-)

diff --git a/Grow.c b/Grow.c
index 8c7bafc..e8f6554 100644
--- a/Grow.c
+++ b/Grow.c

@@ -1626,6 +1626,7 @@ int Grow_reshape(char *devname, int fd, int

quiet, char *backup_file,

quoted

 		long long orig_size = get_component_size(fd)/2;
 		long long min_csize;
 		struct mdinfo *mdi;
+		int raid0_takeover = 0;

 		if (orig_size == 0)
 			orig_size = array.size;

@@ -1674,6 +1675,28 @@ int Grow_reshape(char *devname, int fd, int

quiet, char *backup_file,

quoted

 				"2TB per device\n");
 			size = min_csize;
 		}
+		if (st->ss->external) {
+			if (sra->array.level == 0) {
+				rv = sysfs_set_str(sra, NULL, "level",
+						   "raid5");
+				if (!rv) {
+					raid0_takeover = 1;
+					/* get array parametes after takeover
+					 * to chane one parameter at time

only

quoted

+					 */
+					rv = ioctl(fd, GET_ARRAY_INFO,

&array);

quoted

+				}
+			}
+			/* make sure mdmon is
+			 * aware of the new level */
+			if (!mdmon_running(st->container_dev))
+				start_mdmon(st->container_dev);
+			ping_monitor(container);
+			if (mdmon_running(st->container_dev) &&
+					st->update_tail == NULL)
+				st->update_tail = &st->updates;
+		}
+
 		array.size = size;
 		if (array.size != size) {
 			/* got truncated to 32bit, write to

@@ -1686,12 +1709,24 @@ int Grow_reshape(char *devname, int fd, int

quiet, char *backup_file,

quoted

 				rv = -1;
 		} else {
 			rv = ioctl(fd, SET_ARRAY_INFO, &array);
+
 			/* manage array size when it is managed externally
 			 */
 			if ((rv == 0) && st->ss->external)
 				rv = set_array_size(st, sra, sra->text_version);
 		}

+		if (raid0_takeover) {
+			/* do not recync non-existing parity,
+			 * we will drop it anyway
+			 */
+			sysfs_set_str(sra, NULL, "sync_action", "idle");

This is probably ineffective.  If the array thinks it needs to resync, then
just setting it to 'idle' will only stop in temporarily.  It will immediately
start syncing again.

However given that it is a RAID5 array which is degraded, it shouldn't try to
resync at all.

So:  does it still work if you remove this line?  If so, could you send a
patch to remove it?
If not - we should work out why.

It doesn't work without this line.
The N-disks raid0 changes to degraded (N+1)-disks raid5 (N of N+1 disks) 
in /proc/mdstat after size extension, however the metadata on all disks is correct.

Regards,
Lukasz
 

quoted

+			/* go back to raid0, drop parity disk
+			 */
+			sysfs_set_str(sra, NULL, "level", "raid0");
+			ioctl(fd, GET_ARRAY_INFO, &array);
+		}
+
 		if (rv != 0) {
 			int err = errno;