staging/lustre/lmv: fix duplicate directory entries

Previously, I accidentally introduced a new way for duplicate
directory entries to be returned from readdir().  That patch fails
to properly decrement the nlupgs counter when breaking out of the
inner-for loop.  This accounting error causes an extra iteration
of the inner-for loop when processing the next cfs page and a bad
ldp_hash_end value is then saved in the lu_dirpage.  To fix this,
always decrement the nlupgs counter on entry into the inner loop.

Note: this bug only affects architectures with > 4k-sized pages, e.g.
PowerPC.

Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3182
Lustre-change: http://review.whamcloud.com/6405Signed-off-by: Ned Bass <bass6@llnl.gov>
Reviewed-by: Fan Yong <fan.yong@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Bobi Jam <bobijam@gmail.com>
Signed-off-by: Peng Tao <tao.peng@emc.com>
Signed-off-by: Andreas Dilger <andreas.dilger@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/staging/lustre/lustre/lmv/lmv_obd.c

@@ -1959,7 +1959,7 @@ static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
 		__u64			hash_end = dp->ldp_hash_end;
 		__u32			flags = dp->ldp_flags;
 
-		for (; nlupgs > 1; nlupgs--) {
+		while (--nlupgs > 0) {
 			ent = lu_dirent_start(dp);
 			for (end_dirent = ent; ent != NULL;
 			     end_dirent = ent, ent = lu_dirent_next(ent));
@@ -1993,6 +1993,7 @@ static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
 
 		kunmap(pages[i]);
 	}
+	LASSERTF(nlupgs == 0, "left = %d", nlupgs);
 }
 #else
 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)