ObjectCacher: fix bh_read_finish offset logic

If we have an incoming read split across multiple BufferHeads, we want to
line up the BufferHead's bl with the incoming OSDOp's bl at the right offset. We
were erroneously using this nonsense calculation (always equal to zero!) when
a much simpler comparison of the BufferHead's logical object offset to the
incoming OSDOp's logical offset will do the trick nicely.

Fixes: http://tracker.ceph.com/issues/16002
Signed-off-by: Greg Farnum <gfarnum@redhat.com>
(cherry picked from commit 9ec6e7f608608088d51e449c9d375844631dcdde)

diff --git a/src/osdc/ObjectCacher.cc b/src/osdc/ObjectCacher.cc
index 7e76b3cdd6d06118f774beb3d1a1fe216d591349..8732325e41ff6029dd964c44a25dfd26b32e3dbf 100644 (file)
--- a/src/osdc/ObjectCacher.cc
+++ b/src/osdc/ObjectCacher.cc
@@ -833,7 +833,6 @@ void ObjectCacher::bh_read_finish(int64_t poolid, sobject_t oid,
       if (bh->error < 0)
        err = bh->error;
 
-      loff_t oldpos = opos;
       opos = bh->end();
 
       if (r == -ENOENT) {
@@ -853,7 +852,7 @@ void ObjectCacher::bh_read_finish(int64_t poolid, sobject_t oid,
        mark_error(bh);
       } else {
        bh->bl.substr_of(bl,
-                        oldpos-bh->start(),
+                        bh->start() - start,
                         bh->length());
        mark_clean(bh);
       }