Brian C. Lane wrote: > On Tue, Sep 04, 2012 at 04:51:53PM -0700, Brian C. Lane wrote: >> >> I'm looking for comments on this quick test patch. The problem is this: >> >> In Fedora we have a hybrid iso that also contains a couple of EFI boot >> partitions in order to make it bootable on EFI systems and on Apple >> harware. Parted is sometimes throwing an error when they are written to >> a USB stick using dd and then examined with 'parted /dev/sdd p' or >> parted -l >> >> I think the right way to fix this is to reallocate the buf with the new >> size. The only other alternative I can see is to make >> __disk_analyse_block_size always return an error when there is a >> mismatch instead of prompting the user. >> >> I'm also not sure how we could write a good test case for this. > > On re-reading that I realize I wasn't too clear as to what actually > happens. When it examines the USB it hits this warning in > _disk_analyse_block_size: > > Warning: The driver descriptor says the physical block size is 2048 > bytes, but Linux says it is 512 bytes. > > If you tell it to ignore it will change the disk->dev->sector_size to > 2048, then when it tries to read the ghost partitions in mac_read it > will overrun the buffer and clobber the heap, which results in the next > error causing a SIGSEGV as it tries to malloc space for the error > message.
Thanks for the added detail. Would you please include something like that to the commit log for your patch? Re testing, would it be enough to use parted to create a MAC partition table with "mklabel mac", add at least one partition, and then use dd or perl to poke a new, larger number into its 16-bit big-endian raw_disk->block_size field ?
