On Wed, Apr 8, 2009 at 06:20, Alasdair G Kergon <agk@xxxxxxxxxx> wrote: > On Wed, Apr 08, 2009 at 05:49:11AM -0700, Kay Sievers wrote: >> I always thought we would return the driver core generated seqnum with >> the ioctl, > > When Peter tried to make that work it needed more and more hacks > piled onto it and instead he came up with this cookie method which seems > to satisfy all our requirements with a minimum amount of change, fingers > crossed. > > E.g. A problem with the other approach was: lvm2 userspace doesn't know the > seqnum till after the uevent has fired and udev could complete processing > before the lvm2 code has set itself up to wait for that seqnum so you have to > build extra interlocking into the process making it much more complex. If you > wait for all events up to a given number you could get stuck behind some > unrelated event on an unresponsive device, with is also unsatisfactory. Yeah, sounds sensible, and like a nice solution to the problem if you are able to watch the events udev sends out. Thanks, Kay -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
