On Thu, Apr 22, 2010 at 02:57:15AM +0900, [email protected] wrote: > Hi Oleg, > > Oleg: > > This dir is very old :-). And we don't plan to remove or rename it in the > > near future. > > Why I asked how old is that it contains .wh..wh.opq or something. > If it is so old and you have used it with UnionFS or aufs1 in > unionfs-compat mode, the name may be .wh.diropq or something.
We used only aufs2 on this server. > Anyway your "video/Films" is marked as "opaque" in the topmost branch > fs. ie. all the same named directories on the lower branches are > hidden. Since you are specifying "create=mfs" and all branches as > writable, this situation MAY not be what you expect. You may want to > distribute the newly created files among all branches. If so, you may > need to remove "video/Films/.wh..wh.opq" on the first branch. But be > careful, you should check "video/Films" dir on all branches. If there > are some files in the lower branch, they will appear after you remove > .wh..wh.opq. > > If "video/Films" has ever been marked as opaque by your intended > operation, then there is nothing to worry about. > > The root cause of the messages you saw is a bug in aufs. > It happens with the combination of create=<anything other than default> > and the opaque-ness in ancestor dir. Current aufs supports the opaqueness > of the parent dir (the nearest ancestor), but the upper ancestors. > Aufs should check all ancestors about the opaque-ness. > > In your case, you (or someone else) tried creating something > "video/Films/fileA" (fileA was unprintable characters actually). Since > create=mfs is specified, aufs tried finding the appropriate > branch. According to the log you sent, it was the second branch. > Before creating fileA, aufs tried acquiring a lock for video/Films on > the second branch. But the video/Films is marked as opaque in the first > branch. If aufs had succeeded creating fileA, you would never be able to > see it, because fileA exists in the second branch but the parent dir in > the second branch is hidden. So aufs detected this inconsistency and the > messages was produced. > Sorry about the long story (and my poor English), but do I make myself > clear? > > Here is the fix of this problem and I am testing currently, which may > take a day or two. > Would you test it? Yes. So, I will apply this patch. And what my next instructions? Do you need the kerndmesg or I simply can watch that everything ok? > If it succeed, it will be included in next Monday release. > > > J. R. Okajima > ------------------------------------------------------------------------------
