hey all, after further testing i discovered a serious bug in my code around snapshot renaming. specifically, you can't rename snapshots that have clones which are zoned or in use. (and all clones created by beadm or zone cloning have the zoned attribute set.) oops.
i've added a new block of logic to the main processing loop to deal with this. the unplesant bit is that we need to clear the zoned attr on clones before renaming snapshots and then restore it afterwards. i've also added logic to detect configurations where renames are going to fail so that we can present the user with an informative error message and abort (instead of just hitting a rename error after we've start modifying filesystems.) i'm planning to file another zfs rfe for a mechanism to allow the global zone to override the "zoned" attribute when doing operations. (since that would help simplify some of this logic.) ed On Tue, Apr 28, 2009 at 05:20:03PM -0700, Edward Pilatowicz wrote: > hey all, > > i've got the following webrev up for review: > > http://cr.opensolaris.org/~edp/pkg-bugs1/ > 7491 problems destroying zones with cloned dependents > 7640 zone datasets might not always promote correctly? > > this is basically a re-write of the ipkg brand uninstall script so that > it can property deal with uninstalls of zones that have cloned datasets. > > <optional background> > > before anyone asks, yes, this code would have been better suited to > python rather than ksh. but i don't know python yet and i'm hoping to > get this into the next os.o release. > > also, this work largely duplicates the zfs management code introduced > for the native nevada brands by 6787557 and 6825063. because of this > duplication (and the general complexity of automating the management of > zfs datasets which have clones), i've filed the some zfs RFEs so that > zfs can manage these issues itself. (see 6833599 and 6833600.) once > these zfs RFEs are implemented, my fixes above (and the fixes for > 6787557 and 6825063) can be removed. > > </optional background> > > thanks > ed _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
