Edward Pilatowicz wrote: > On Fri, Mar 03, 2006 at 09:58:20AM +0000, Darren J Moffat wrote: >> Edward Pilatowicz wrote: >>> because it's much cleaner than doing pathmapping. ;) >>> >>> this is for brandz and the application is a solaris daemon running in >>> a linux branded zone. >> So this is a specific application rather than a general problem. >> >> Which Solaris daemon is this ? The reason I ask is that I had > > lockd and statd.
Hmn okay so it isn't ncsd which is the one I talked to the old Janus team about. >> some very long discussions with the then Janus team around how >> to deal with some of the cases where they wanted a Solaris daemon >> in an otherwise pure Linux zone. >> > > i'm confused. the janus team didn't use zones for running linux > processes. the brandz project is the follow on to janus and it > introduces the concept of a linux branded zone. No they didn't but I was talking to them about how it could be done. There were prototypes. It didn't use a branded zone though. >> Building chroot's are evil they destroy all the benefits of patching >> and effectively bring back so many of the bad things about static >> linking. Now if you chroot to the "real" patchable bits thats okay >> but chroot'ing to something you have build isn't nice. >> > > i'm not sure i understand your concern here. > > in a linux branded zone we have parts of the global zone filesystem > read only loopback mounted to /native. (this is necessary for just > running normal linux applications since we need access to solaris > libraries, soalris /proc, etc.) when i want to run lockd and > statd, i use and LD_PRELOADed library to cause these applications > to chroot to /native so i don't have to worry about remapping > paths that the daemons try to access. Thats the bit of missing info I didn't have /native being an lofs mount from the global zone. That means the patching problem goes away because it isn't a "copy" of the libraries but a pointer to the patchable versions. -- Darren J Moffat
