Am 01.12.2015 um 11:26 hat Fam Zheng geschrieben: > On Tue, 12/01 10:54, Kevin Wolf wrote: > > Am 01.12.2015 um 10:36 hat Fam Zheng geschrieben: > > > The assertion problem was noticed in 06c3916b35a, but it wasn't > > > completely fixed, because even though the req is not marked as > > > serialising, it still gets serialised by wait_serialising_requests > > > against other serialising requests, which could lead to the same > > > assertion failure. > > > > > > Fix it by even more explicitly skipping the serialising for this > > > specific case. > > > > > > Signed-off-by: Fam Zheng <f...@redhat.com> > > > > And this, my friends, is another example why read/write notifiers are > > wrong and should die sooner rather than later. </broken-record> > > > > Yes, I agree, except it's not clear to me what a better alternative solution > should be. A immediate question is, with whatever approach we will have, > wouldn't we still need to do this sort of "reentrant" COW before the data is > overwritten?
If the backup job could temporarily insert a filter driver, you wouldn't get reentrance, but another BDS in the stack. Now that much of the blockdev-add stuff seems to be falling into place at last, dynamic reconfiguration and filters might be the next big nut for us to crack. Kevin
