On Fri, Feb 6, 2009 at 12:32 PM, Peter Memishian <peter.memishian at sun.com> wrote: > > > Note that ti's the same value as / . > > > > What is happening is a bit of a (rather neat IMO) optimization. > > During startup, the OS dynamically picks an optimized version of libc > > for the running system (based on the capabilities of the cpu) and > > overlays the default libc with that version (note it doesn't modify > > anything on disk, it just 'hides' the default one). In your case, it > > determined that /usr/lib/libc/libc_hwcap2.so.1 was the best fit, and > > so it loopback mounts that as /lib/libc.so.1. Because of that, the > > used/free/etc. stats for the 'filesystem' (which is just a single > > file) are the same as the underlying filesystem (i.e. /). > > This optimization causes a lot of problems too -- e.g., how to back up or > upgrade the hidden generic libc.so.1. It also leads to this meaningless > and confusing output in df.
Not to get too far off topic here, but I think that could be dealt with fairly easily (at least going forward). df could easily be modified to ignore well-known loopback mount points and other filesystems where such output isn't meaningful (i.e. the various pseudo filesystems). Seems like a simple RFE. For updates/backups, if the future is zfs / + beadm/lu for updates, that should take care of it -- backup your snapshot (TSM can already do this with a little scripting, not sure about other packages), update your clones. I know it doesn't help the UFS / systems, but it at least mean it should be less of a problem over time.
