Date: Mon, 20 Oct 1997 05:55:35 -0500 (EST) From: "John S. Dyson" <toor@dyson.iquest.net> To: phk@critter.freebsd.dk (Poul-Henning Kamp) Cc: toor@dyson.iquest.net, roberto@keltia.freenix.fr, current@FreeBSD.ORG Subject: Re: nullfs & current Message-ID: <199710201055.FAA01099@dyson.iquest.net> In-Reply-To: <406.877333677@critter.freebsd.dk> from Poul-Henning Kamp at "Oct 20, 97 09:47:57 am"
next in thread | previous in thread | raw e-mail | index | archive | help
Poul-Henning Kamp said: > In message <199710200143.UAA05747@dyson.iquest.net>, "John S. Dyson" writes: > >Ollivier Robert said: > >> According to John S. Dyson: > >> > The VM system holds a reference. You have to do a vnode_pager_uncache whe > >n > >> > deleting a file. > >> > >> Wait... Isn't ufs_remove() supposed to do that ? ufs_remove() is called by > >> null_bypass() by an indirect call (VCALL()) and it seemed safe to consider > >> that the lower level was supposed to DTR. > >> > >> Do I have to create a null_remove() function in order to call > >> vnode_pager_uncache inside it ? > >> > >nullfs should really share the VM object with the lower filesystem, but genera > >lly > >what you say is true, even if you don't share the object. The complexity is > >one reason that I haven't fixed it. It IS fixable reasonably though. > > No, this is wrong. Nullfs can only share the VM object if the > permissions are the same, consider a read-only mounted nullfs for > instance. > The VM object should be shared to maintain coherency. Permissions are maintained by the VM_MAP layer. Objects are mosly globs of data. For some apps, an object can be RO, and others RW. -- John dyson@freebsd.org jdyson@nc.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199710201055.FAA01099>