On Thu, May 27, 2010 at 08:47:48AM -0400, Mike Snitzer wrote: > Introduce 'type_lock' in mapped_device structure and use it to protect > md->type access. table_load() sets md->type without concern for: > - another table_load() racing to set conflicting md->type. which leads to non-deterministic behaviour that is of no practical use and so there is no need to support it. > - do_resume() making a conflicting table live. Any table being made live must already have passed through table_load so I don't see how there can be conflicting types. The mutex here seems to be overkill: instead of protecting one field to cope with a useless race, we should be making the output of the race deterministic (e.g. perhaps giving an error on the parallel table_load attempt). Anyway, I can take this patch for now and we can deal with the mutex/race differently later. Alasdair -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
