On May 6, 2009, at 6:54 PM, Richard Todd wrote: > Hi. I updated my main -current box to the current sources as of > yesterday, > and I've managed to twice get the "wrong vnode panic" out of the > cache_enter > code. As you can see from the backtrace down at cache_enter at > frame 11, > the code is apparently expecting vp to point to a VDIR vnode, but > instead > it's pointing to a VLNK. (I've got another, similar, dump where vp > is > pointing to a VREG node.) In both cases, I was running tinderbox to > build > packages, with resulting heavy use of both zfs (the underlying > filesystem) > and nullfs, if that's relevant. > (kgdb) l > 697 if (dvp->v_cache_dd != NULL) { > 698 CACHE_WUNLOCK(); > 699 cache_free(ncp); > 700 return; > 701 } > 702 KASSERT(vp == NULL || vp->v_type == VDIR, > 703 ("wrong vnode type %p", vp)); > 704 dvp->v_cache_dd = ncp; > 705 } > 706 It looks like this KASSERT was added relatively recently as part of the dotdot changes to the namecache: http://svn.freebsd.org/viewvc/base?view=revision&revision=190945 Perhaps there is further fallout from the changes that only occur with ZFS? - BenReceived on Thu May 07 2009 - 10:59:49 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:47 UTC