>>On Wed, Jun 10, 2015 at 5:48 PM, Greg Kroah-Hartman ><gregkh@xxxxxxxxxxxxxxxxxxx> wrote: >> >> Are you sure all of these are correct? The kernel/user api for lustre >> is a complex beast, and just casting away the pointer types isn't >> usually the proper thing to do in order to resolve the issues here. >> >> thanks, >> >> greg k-h > >I'm not 100% sure, but the pointers that I added the annotation to end >up being used as user memory. (eg. passed to copy_to_user, etc.) >Sometimes these pointers are passed to functions that already have >__user annotation in their signatures (eg. ll_getname, copy_and_ioctl, >ll_fid2path, etc.). Using these simple cast are not the proper fix. We had a lot of issues with user land tools breaking due to leakage of kernel space stuff and other problems. Some work went into cleaning that up in the OpenSFS branch but it is not totally complete yet. Evans you wanted something challenging to work on well this is up your alley. I would recommend looking at JIRA ticket LU-6401 and all its sub tickets. You could start the port of those to the upstream client. At the same time we can finish the cleanup in the OpenSFS branch as well. _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
