On Sun, Jul 09, 2017 at 01:01:00PM -0700, Tahsin Erdogan wrote: > > What we could do is have ext4_new_inode check to see if there are > > enough credits to do add the xattr's (if necessary) in a single > > commit. If not, what we could do is to add the inode to the orphan > > list, and then set an inode state flag indicating we have done this. > > At this point, we *can* break the ext4_new_inode() operation into > > multiple commits, because if we crash in the middle the inode will be > > cleaned up when we do the orphan list processing. > > This makes sense. Also, we currently add the worst case credit > estimates of individual set xattr ops and start a journal handle with > the sum of it. A slight optimization is to do this lazily. > We can start with enough credits that can get us to a point where it > is safe to start a new transaction (safe because of orphan addition). I still am very concerned about the code complexity that this approach requires. I am also very concerned about the CPU scalability bottleneck that adding and removing the inode from the orphan list would entail. And if we have to wait for the new commit to start so that we can start a new handle, that's also a CPU scalability bottleneck and is guaranteed to add significant latency. One of the nice things about the xattr priority proposal is that it would guarantee that the security xattrs would never be in an ea_inode. (Since in the inode creation case, the only thing they would be competing with is the acl's, which are lower priority). So this reduces the chances of needing to do a lazy extend/restart in the first place. > > The downsides of this approach is that it causes the orphan list to be > > a bottleneck. So we would definitely not want to do this all time. > > Yes and I think lazy extend/restart should mitigate this. It mitigates it so long as we the lazy extent/restart is never/rarely *used*, since that's when we would incur the orphan list overhead. One other bit about the lazy extend/restart idea is that we need to make sure that there are enough credits left for the callers of ext4_new_inode() before it returns. Otherwise the complexity of this approach would infect all of the users of this interface (since they would have to potentially do the extend/restart of the transaction). - Ted
