Dave Hansen <hansendc at us.ibm.com> writes: > Here's a set of patches which I believe allows multiple > instances of /proc to exist. We want these because it > allows each pid_namespace to have a completely diffferent > view of /proc. > > --- > > /proc currently has some special code to make sure > that the root directory gets set up correctly. It > uses the global proc_mnt variable in order to find > its way to the root inode. > > In order to try to get multiple /proc mounts, I'm > attempting to kill this variable. But, I noticed > that we have all of the data to fill in the inode's > pid in proc_fill_super(), where it takes a wee bit > more work in proc_get_sb(). > > Any reason not to just do it in proc_fill_super()? Because we are using get_sb_single. So proc_fill_super is only called once when proc_mnt is established. Since proc_mnt is created before pid == 1 exists, ei->pid on the root_inode is never filled in. Once proc_mnt disappears we can make this change, but it should definitely not come first in the patch set or horrible things will happen to git-bisect users. Eric