gfs2-utils: master - gfs2_convert: Use proper header size when reordering meta pointers

Bob Peterson rpeterso at fedoraproject.org
Fri Dec 21 16:39:52 UTC 2012


Gitweb:        http://git.fedorahosted.org/git/?p=gfs2-utils.git;a=commitdiff;h=e7820f91c1b5bf5dbe5966dcf0b982df0d3e4ba7
Commit:        e7820f91c1b5bf5dbe5966dcf0b982df0d3e4ba7
Parent:        b180bc7302d47319674be977d1d621405eca406f
Author:        Bob Peterson <rpeterso at redhat.com>
AuthorDate:    Mon Dec 17 15:01:02 2012 -0600
Committer:     Bob Peterson <rpeterso at redhat.com>
CommitterDate: Thu Dec 20 14:30:58 2012 -0600

gfs2_convert: Use proper header size when reordering meta pointers

This patch changes function fix_metatree to use a proper metadata
header size. Before, it was using sizeof(struct gfs2_meta_header).
That's correct in almost all cases. But if you make a big file,
such that it goes into height==2 (two levels of indirection), then
truncate the file back to where it only would normally only need
height==1, then run gfs2_convert, it gets into trouble. That's
because you have a small file size, which calculates a much smaller
number of GFS2 pointers needed, due to the truncation. Function
fix_metatree will ensure at least one level of indirection by
unstuffing the converted dinode, but we need to start pushing out
the pointers onto the dinode's buffer, and for that, we need to
calculate the right header size.

rhbz#887374
---
 gfs2/convert/gfs2_convert.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/gfs2/convert/gfs2_convert.c b/gfs2/convert/gfs2_convert.c
index 61d1504..36ca1a3 100644
--- a/gfs2/convert/gfs2_convert.c
+++ b/gfs2/convert/gfs2_convert.c
@@ -356,7 +356,8 @@ static void fix_metatree(struct gfs2_sbd *sbp, struct gfs2_inode *ip,
 			gfs2_meta_header_out(&mh, bh);
 		}
 
-		hdrsize = sizeof(struct gfs2_meta_header);
+		hdrsize = blk->height ? sizeof(struct gfs2_meta_header) :
+			sizeof(struct gfs2_dinode);
 
 		if (amount > sbp->bsize - hdrsize - ptramt)
 			amount = sbp->bsize - hdrsize - ptramt;


More information about the cluster-commits mailing list