I realize that its considered pushy to reply to one's own post, but here is
the patch I'm suggesting I'll test it myself some time when my server is
less busy or the next time a drive goes down:
*** md.c.19990421-2.0.36 Tue Jun 22 16:56:52 1999
--- md.c Thu Jun 24 13:50:40 1999
***************
*** 682,687 ****
--- 682,688 ----
static void unlock_rdev (mdk_rdev_t *rdev)
{
blkdev_release(rdev->inode);
+ clear_inode(rdev->inode);
}
static void export_rdev (mdk_rdev_t * rdev)
----- Original Message -----
From: Richard Bollinger <[EMAIL PROTECTED]>
To: Linux-Raid <[EMAIL PROTECTED]>
Sent: Wednesday, June 23, 1999 3:47 PM
Subject: unlock_rdev() in md.c in raid0145-19990421-2.0.36 not clearing
inode
> With linux 2.0.36 and patch raid0145-19990421-2.0.36 applied, if a disk is
> looked at for autostart but rejected because it's superblock is out of
> date, you can't raidhotadd or raidhotremove it because the dummy inode is
> still out there. This would be no problem if I had really replaced the
disk
> with a fresh one that would have been rejected from the autostart process.
> In my case, it was just put back on line giving me the message "md: can
not
> import /dev/sda1, has active inodes!"
>
> At some time between patches 19990108 and 19990421 the line:
>
> clear_inode(rdev->inode);
>
> was replaced with
>
> blkdev_release(rdev->inode);
>
> I think _both_ lines are necessary to release the inode.
>
> Thoughts?
>
> Rich Bollinger, Elliott Company
>
>
>
>
