(Illustration by Gaich Muramatsu)
On Fri, Jan 03, 2003 at 10:04:13AM -0500, Greg Troxel wrote: > > Why can't we just refrain from doing the stat at mount time, and do it > on demand when it is actually needed, with a variable to record > whether the stat info is valid (or time of last stat, if we want to > avoid calling more than once a second and get fancier). > Currently, there are a couple of reasons, first is that sys_mount always stats the file system on mount at the moment - we would have to hack the code to avoid doing this for coda (I did consider doing this), seondly, it still means that we will have the artifact I described before - the Coda partition will not appear in a df output until the Coda mount has been stat'ed which maybe... disturbing. > I can't explain why, but it strikes me that having to fork for the > mount call is at least somewhat bletcherous. Other processes blocking > on calls to venus is one thing, but venus shouldn't be in that boat in > the first place. > I do agree but at the moment we have a chicken and egg situation - venus needs to be answering to provide the stat information but it is currently in the kernel doing a mount. One thing that does occur to me is what about venus feeding the stat information into the message queue prior to doing the mount call? Would that work? -- Brett LymnReceived on 2003-01-04 03:10:03