Goldwyn Rodrigues
2011-Jun-07 15:41 UTC
[Ocfs2-devel] Do filesystem need to turn RO on dx corruptions?
Hi, There have been index tree corruptions, where the directory entries are correct however the index directory is not. In this case, the filesystem turns read-only where as all the information to complete the operation is still available because the primary directory structure is sane. The ocfs2 module could just disable indexed directory for this directory and continue handling the operation like the normal directories without indexes. Of course, it still writes the warning message, to run fsck, to the kernel log but avoid the RO filesystem which halt all I/O writes. Th ocfs2-tools fix is posted at: http://oss.oracle.com/pipermail/ocfs2-tools-devel/2011-May/003812.html However, it would need changes for indexed directory rebuilds. Let me know what you think. -- Goldwyn
Sunil Mushran
2011-Jun-07 17:30 UTC
[Ocfs2-devel] Do filesystem need to turn RO on dx corruptions?
On 06/07/2011 08:41 AM, Goldwyn Rodrigues wrote:> There have been index tree corruptions, where the directory entries > are correct however the index directory is not. In this case, the > filesystem turns read-only where as all the information to complete > the operation is still available because the primary directory > structure is sane. The ocfs2 module could just disable indexed > directory for this directory and continue handling the operation like > the normal directories without indexes. Of course, it still writes the > warning message, to run fsck, to the kernel log but avoid the RO > filesystem which halt all I/O writes. > > Th ocfs2-tools fix is posted at: > http://oss.oracle.com/pipermail/ocfs2-tools-devel/2011-May/003812.html > However, it would need changes for indexed directory rebuilds. > > Let me know what you think.I think it looks good. Though I would prefer if Mark also acked this as he knows dxdir best.
Joel Becker
2011-Jun-08 06:41 UTC
[Ocfs2-devel] Do filesystem need to turn RO on dx corruptions?
On Tue, Jun 07, 2011 at 10:41:42AM -0500, Goldwyn Rodrigues wrote:> There have been index tree corruptions, where the directory entries > are correct however the index directory is not. In this case, the > filesystem turns read-only where as all the information to complete > the operation is still available because the primary directory > structure is sane. The ocfs2 module could just disable indexed > directory for this directory and continue handling the operation like > the normal directories without indexes. Of course, it still writes the > warning message, to run fsck, to the kernel log but avoid the RO > filesystem which halt all I/O writes.Goldwyn, You're correct, this is a more graceful solution. We've just never gotten around to it. Joel -- "Behind every successful man there's a lot of unsuccessful years." - Bob Brown http://www.jlbec.org/ jlbec at evilplan.org