On Sun 03 Mar 2013 at 18:07:26 -0300, Joao Luis Meloni Assirati wrote: > Brian wrote: > > > Had the USB hard drive been used instead of the 8-GB USB and *exactly* > > the same install attempted it too would have failed. The nature or size > > of the device being installed to is immaterial, as is whether it is a > > text mode or GUI install. The only thing that matters is that the device > > has previously had an isohybrid ISO written to it. > > [...] > > > Write a netinst image to two USB devices. Boot from one device and > > install to the second one. Try text and GUI modes. It's about an hour's > > work at most. > > But will this happen even if one formats the partition holding the iso? > What if the installation proccess is done normally, the target device that > happens to hold the iso is partitioned and all the partitions are > formatted? When grub gets to be installed, in the last installation step, > all the information that an iso existed before is gone, or no?
When the isohybrid ISO is written to to the drive the information about the iso9660 filesystem is put within the first 65 sectors of the drive. If fdisk is now used to partition the drive the first partition starts at sector 2048. Everything beyond this sector is destroyed but sectors below number 2048 are left intact. The drive is now useless as a device to boot Debian but information about the iso9660 file system is left intact. fdisk leaves space at the beginning of the drive because GRUB requires it to embed part of itself there. But GRUB will not go there because it thinks it is overwriting data on the disk when it detects the iso9660 signature. This is by design. D-I uses partman for partitioning. It too leaves an embedding area which contains the iso9660 data sector. The solution is to remember to do dd if=/dev/zero of=/dev/sdX count=65 before partitioning. > Anyway, it must be said that you did an impressive investigation with very > scarce resources to say the least, Brian! Thanks. I cheated though! I had already encountered the bug some time ago and the reported behaviour in this thread is very, very similar to it. -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20130304010743.GW14686@desktop
