[Cluster-devel] gfs2-utils: mkfs segfault with 4.16 kernel

Andrew Price anprice at redhat.com
Mon May 14 13:24:38 UTC 2018


On 13/05/18 10:13, Valentin Vidic wrote:
> On Wed, May 09, 2018 at 05:53:19PM +0100, Andrew Price wrote:
>> You should be able to fix this by using the configure.ac and
>> gfs2/libgfs2/meta.c files from master. The diff should be very small. It'll
>> need an ./autogen.sh to rebuild the configure script.
>>
>> The downside of that approach is that the new fields will be recognised but
>> not used in the utils. The next release of gfs2-utils will add that
>> functionality.
> 
> As I feared, the gfs2 kernel module does not want to mount the
> filesystem created with patched gfs2-utils. I guess that I will
> need to take the current master branch to get it working again?

We can still keep this minimal if that's better for the Debian package.

The issue is due to the log header checksum (lh_hash) being calculated 
using sizeof(struct gfs2_log_header) and that size has increased in the 
latest kernel gfs2_ondisk.h.

For the purposes of the package we can fix that by replacing those 
sizeof's with the correct value, as below.

Andy

diff --git a/gfs2/libgfs2/recovery.c b/gfs2/libgfs2/recovery.c
index cad723f0..7326574f 100644
--- a/gfs2/libgfs2/recovery.c
+++ b/gfs2/libgfs2/recovery.c
@@ -66,7 +66,7 @@ int get_log_header(struct gfs2_inode *ip, unsigned int 
blk,
         tmp = (struct gfs2_log_header *)bh->b_data;
         saved_hash = tmp->lh_hash;
         tmp->lh_hash = 0;
-       hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
+       hash = gfs2_disk_hash(bh->b_data, 48);
         tmp->lh_hash = saved_hash;
         gfs2_log_header_in(&lh, bh);
         brelse(bh);
@@ -234,7 +234,7 @@ int clean_journal(struct gfs2_inode *ip, struct 
gfs2_log_header *head)
         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
         lh->lh_blkno = cpu_to_be32(lblock);
-       hash = gfs2_disk_hash((const char *)lh, sizeof(struct 
gfs2_log_header));
+       hash = gfs2_disk_hash((const char *)lh, 48);
         lh->lh_hash = cpu_to_be32(hash);
         bmodified(bh);
         brelse(bh);
diff --git a/gfs2/libgfs2/structures.c b/gfs2/libgfs2/structures.c
index 1cc88b50..a8da4a0c 100644
--- a/gfs2/libgfs2/structures.c
+++ b/gfs2/libgfs2/structures.c
@@ -133,7 +133,7 @@ int lgfs2_write_journal_data(struct gfs2_inode *ip)
                 lh.lh_sequence = seq;
                 lh.lh_blkno = bh->b_blocknr - jext0;
                 gfs2_log_header_out(&lh, bh);
-               hash = gfs2_disk_hash(bh->b_data, sizeof(struct 
gfs2_log_header));
+               hash = gfs2_disk_hash(bh->b_data, 48);
                 ((struct gfs2_log_header *)bh->b_data)->lh_hash = 
cpu_to_be32(hash);

                 if (bwrite(bh)) {
@@ -185,7 +185,7 @@ int write_journal(struct gfs2_inode *jnl, unsigned 
bsize, unsigned int blocks)
                 lh.lh_sequence = seq;
                 lh.lh_blkno = x;
                 gfs2_log_header_out(&lh, bh);
-               hash = gfs2_disk_hash(bh->b_data, sizeof(struct 
gfs2_log_header));
+               hash = gfs2_disk_hash(bh->b_data, 48);
                 ((struct gfs2_log_header *)bh->b_data)->lh_hash = 
cpu_to_be32(hash);

                 bmodified(bh);




More information about the Cluster-devel mailing list