On 22/02/2010 20:29, Renee Danson Sommerfeld wrote: > On Mon, Feb 22, 2010 at 07:41:33PM +0000, Alan Maguire wrote: > >> I think I have a fix for this. The problem was we >> were unconditionally enabling network/location >> which confuses manifest-import. The fix is >> to check if network/location exists before enabling, >> see >> >> http://zhadum.east.sun.com/export/ws/amaguire/nwam1-fix/webrev/ >> > These changes look reasonable. > > One thing I noticed: could you fix the typo on line 732 in net_include.sh > (just after the bit you added)? s/temporarirly/temporarily/ > > Will do. > One concern, though. We're using /usr/bin/svcs, which is not guaranteed > to be present the first time net-nwam runs (the nwam service does not-- > and cannot--depend on filesystem/usr). I'm pretty sure this is not the > only place where this assumption is made (it was made in the check we > just removed, in fact); and the net-nwam script is likely not the only > one that does this. And it is--at least as far as I know--primarily an > academic concern, as folks don't generally separate root and usr. > > But this seems like an issue for the smf utilities in particular, which > are in /usr/bin and are really handy when one is starting up a service. > Do you know if there's any general solution for this? Work in progress, > by any chance? > Not that I know of, but as you say the assumption is made in a lot of places. Seems like a straightforward RFE would be to move svccfg and svcadm to /sbin and symlink from /usr/sbin. Both currently reside in /usr/sbin (while svcs is in /usr/bin), and the libraries they rely on all live in /lib already - libscf, libcontract, libuutil and libumem. Given that svcs resides in /usr/bin, I'm thinking the service_exists() check I've added should be changed to use svccfg in anticipation of the above change being made - what do you think?
Alan
