Matthew Dillon wrote: > :> Try using null mounts. The warning is in there because making the > :> null mount code work is a real hack and the authors aren't entirely > :> sure that everything's gotten covered. That said, use of a null mount > :> is certainly a lot safer if the stuff behind the mount is mostly > :> static. > : > :The problem is in the VM object alias code. Specifically, the > :getpages/putpages have to be implemented in terms of read/write, > :so that there are not two vm_object_t's that refer to the same > :data, since there is no "upcall" to notify of changes in a lower > :layer, and therefore guarantee coherency. > > I'm fairly sure the VM issues were fixed when VOP_GETVOBJECT was > added. A file accessed via a null mount will have the same VM object > as the file in the original filesystem. I'm not 100% sure about > that, I wasn't the one who did it, but I seem to recall it being > discussed.
VOP_GETVOBJECT is a different name, but the VOP was my suggestion, to allow an upper layer to obtain a backing object, and to collapse intermediate layers. The issue is that the NULLFS getpages falls through the the vfs_default.c vop_stdgetpages(), which calls the function vnode_pager_generic_getpages(), which in turn, calls VOP_BMAP(), which in null_vnops.c is vop_eopnotsupp(), so it falls back to vnode_pager_input_smlfs(), which VOP_BMAP()'s *again*, but off the device. At which point, you've lost coherency. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
