On 25 April 2011 20:37, James Cameron <[email protected]> wrote: > No, because that would fail to detect blocks that have been read and > then written unchanged. While I don't know of any examples of that > being done, it is still a possibility, either now or in the future. > > Using FIEMAP will place a kernel version dependency on zhashfs > unnecessarily, since the task can be done with memcmp. > > It would also be more complex code to process the FIEMAP extents; more > complex than memcmp. > > It would also require the output directory to be on a filesystem that > supports sparse files. (Attempting to create a sparse file using dd on > a FAT filesystem results in a non-sparse file of the specified size ... > a FIEMAP ioctl on this would show no holes).
I think all of those are safe assumptions to make, and that FIEMAP is the more correct and efficient way of doing this. Daniel _______________________________________________ Devel mailing list [email protected] http://lists.laptop.org/listinfo/devel
