mdrecoveryd oops

Hi, it's me again.
I also wanted to ask if it is supported to use md array during recovery? (with some journaling fs on it). When a system fails I have to wait for about 1 hour while recovery finishes before i mount fs on md, or else it crashes. I tried many kernel versions from 2.4.16 to 2.4.19-pre8, with the same result.

Thanks,

VanTo

It *should* work. Are you using this on a SMP machine? There are some race situations causing crashes, but it should work with an UP machine.

Here is a patch produced by [email protected] that should help in SMP machine. It not pretty and it's not '100% right', but it seems to work.

D.

diff -uNr linux.orig/drivers/md/md.c linux/drivers/md/md.c
@@ -468,6 +468,19 @@
 	}
 }
 
+static inline int mark_rdev_faulty (mdk_rdev_t * rdev)
+{
+       if (!rdev) {
+               MD_BUG();
+               return 0;
+       }
+       if (rdev->faulty)
+               return 0;
+       free_disk_sb(rdev);
+       rdev->faulty = 1;
+       return 1;
+}
+
 static int read_disk_sb(mdk_rdev_t * rdev)
 {
 	int ret = -EINVAL;

@@ -3048,10 +3061,14 @@
 	rrdev = find_rdev(mddev, rdev);
 	if (!rrdev || rrdev->faulty)
 		return 0;
+	if (!mark_rdev_faulty(rrdev))
+		return 0;
 	if (!mddev->pers->error_handler
 			|| mddev->pers->error_handler(mddev,rdev) <= 0) {
+/*
 		free_disk_sb(rrdev);
 		rrdev->faulty = 1;
+*/
 	} else
 		return 1;
 	/*

It's an UP machine. I was asking this question because we had raid0 and it was OK. Now with raid5, machine dies after about 5-6 days. Usually kswapd oopses in some raid5 routine while manipulating with stripe list.