On Wednesday May 13, goswin-v-b@xxxxxx wrote: > Neil Brown <neilb@xxxxxxx> writes: > > > Online changing of chunk size for raid5 is very close to being > > implemented. > > All the kernel side support will be in 2.6.30. > > I just need to spend some time getting mdadm to make it work. > > Because data is being rearranged in-place, and because you don't want > > to lose everything if your systems crashes, mdadm will have to work in > > tandem with the kernel making a backup copy of a section of the array, > > allowing the kernel to reshape that section, and then moving on to the > > next section. It will be slow, but it will be reliable. > > > > NeilBrown > > Shouldn't mdadm create a file and pass the FD to the kernel or a > loopback device and pass the major/minor and the kernel then moves > data there itself? Seems like a bad idea to pass all the data of the > raid to userspace just so it can be passed back to the kernel again. The data doesn't really have to go to user-space. If we use mmap instead of read, and pass the mmapped buffer to write, then the data just lives in kernel buffers (I think) - it never gets faulted in to the address space. I want to keep the functionality in the kernel to a minimum and do any fiddly bits in user-space. I think keeping that back - and restoring from it in the event of a crash - is definitely a fiddly bit that does not belong in the kernel. NeilBrown -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
