JeffleXu <jefflexu@xxxxxxxxxxxxxxxxx> wrote: > > @@ -466,7 +466,7 @@ static void netfs_rreq_short_read(struct netfs_read_request *rreq, > > netfs_get_read_subrequest(subreq); > > atomic_inc(&rreq->nr_rd_ops); > > if (subreq->source == NETFS_READ_FROM_CACHE) > > - netfs_read_from_cache(rreq, subreq, true); > > + netfs_read_from_cache(rreq, subreq, NETFS_READ_HOLE_CLEAR); > > Hi I'm not sure why NETFS_READ_HOLE_CLEAR style should be used in 'short > read' case. The cache backing filesystem (eg. ext4) might have excised a chunk of zeros from the cache in order to optimise its extent list. This instructs the cache to zero over the cracks. Actually, I need to think a bit further on this. This was written assuming that the cache tracks its content independently - but those patches are not in with this set. > I'm not sure why 'subreq->start' is not incremented with > 'subreq->transferred' when calling cres->ops->read() in 'short read' case. subreq->start shouldn't get changed. subreq->transferred is sufficient. David -- Linux-cachefs mailing list Linux-cachefs@xxxxxxxxxx https://listman.redhat.com/mailman/listinfo/linux-cachefs
