Re: BLKGETSIZE64 (bytes or sectors?)

[Andries.Brouwer@cwi.nl]

Matt_Domsch@dell.com wrote, and he is right:

    > Is the BLKGETSIZE64 ioctl supposed to return the size of the device in 
    > bytes (as the comment says, and is implemented in all places *except* 
    > blkpg.c), or in sectors (as is implemented in blkpg.c since 2.4.15)?

Yes, in bytes. blkpg.c has to be fixed.
Several people submitted patches. Sooner or later I suppose
this will be fixed.

Then Tim Pepper answered, and he is wrong:

    Wouldn't it be better to do the following (against 2.4.17).

    +            else {
    +                if (hardsect_size[MAJOR(dev)][MINOR(dev)]) {
    +                    ullval *= hardsect_size[MAJOR(dev)][MINOR(dev)];
    +                } else {
    +                    ullval *= 512;
    +                }
                     return put_user(ullval, (u64 *)arg);
    +            }

You see, the 512 here is 512, and has no relation to hardware
sector size. Multiplying with hardsect_size[][] is a bad bug.

Indeed, you can check this in fs/partitions/msdos.c, where
one reads
	int sector_size = get_hardsect_size(to_kdev_t(bdev->bd_dev)) / 512;
	...
	offs = START_SECT(p)*sector_size;
	size = NR_SECTS(p)*sector_size;
	...
	add_gd_partition(...);

So, indeed, we have already multiplied by hardsect_size, struct gendisk
uses sectors of size 512, independent of the hardware, and we must not
again multiply by hardsect_size.

Unfortunately Matt Domsch replied:

    > Yes, I agree.

but he meant: No!

Andries