ocfs2: skip existing hole when removing the last extent_rec in punching-hole codes.
Tristan Ye [Thu, 12 May 2011 12:47:07 +0000 (20:47 +0800)]
In the case of removing a partial extent record which covers a hole, current
punching-hole logic will try to remove more than the length of whole extent
record, which leads to the failure of following assert(fs/ocfs2/alloc.c):

5507         BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);

This patch tries to skip existing hole at the last attempt of removing a partial
extent record, what's more, it also adds some necessary comments for better
understanding of punching-hole codes.

Signed-off-by: Tristan Ye <tristan.ye@oracle.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>

fs/ocfs2/file.c

index 41565ae..89659d6 100644 (file)
@@ -1607,6 +1607,9 @@ static void ocfs2_calc_trunc_pos(struct inode *inode,
        range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
 
        if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
+               /*
+                * remove an entire extent record.
+                */
                *trunc_cpos = le32_to_cpu(rec->e_cpos);
                /*
                 * Skip holes if any.
@@ -1617,7 +1620,16 @@ static void ocfs2_calc_trunc_pos(struct inode *inode,
                *blkno = le64_to_cpu(rec->e_blkno);
                *trunc_end = le32_to_cpu(rec->e_cpos);
        } else if (range > trunc_start) {
+               /*
+                * remove a partial extent record, which means we're
+                * removing the last extent record.
+                */
                *trunc_cpos = trunc_start;
+               /*
+                * skip hole if any.
+                */
+               if (range < *trunc_end)
+                       *trunc_end = range;
                *trunc_len = *trunc_end - trunc_start;
                coff = trunc_start - le32_to_cpu(rec->e_cpos);
                *blkno = le64_to_cpu(rec->e_blkno) +