Why does fdisk insist on starting the first partition at sector 65535 (MiB 31.9995…)

fdiskpartitionssd

So, I have one of these newfangled external Samsung T3 USB SSD drives here, sized at 250 GB.

Let's see what fdisk has to say about that:

Disk /dev/sdb: 232.9 GiB, 250059350016 bytes, 488397168 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 33553920 bytes
Disklabel type: dos
Disk identifier: 0x7df0da81

Device     Boot Start       End   Sectors   Size Id Type
/dev/sdb1  *       64 488392128 488392065 232.9G  7 HPFS/NTFS/exFAT

Sectors are still 512 byte large (I suspect underneath the software layer, things are a bit different), the minimum I/O size is 1 sector, the optimal I/O size is 3553920 byte = 65535 sectors ≈ 31.995 MiB

We see that the factory-performed formatting consists in a single partition from block 64 to block 488392128, which is ~232.88 GiB or ~244.19 GB.

So I just want to repartition using fdisk. fdisk insists that the first partition start at sector 65535. Maxing it out, it ends at sector 488397167, a bit further than the factory-issued partition:

Device     Boot Start       End   Sectors   Size Id Type
/dev/sdb1       65535 488397167 488331633 232.9G 83 Linux

I know about the 1 MiB alignment boundary which manifests itself by fdisk putting the first partition at sector 2048. The reason for that seems to be a historical accident due to a wart present in the Logical Disk Manager of Microsoft Vista.

But what about this new 65535 sector "alignment"?

Best Answer

If that 65535 refers to 512-byte sectors, it would not be aligned at all. Unfortunately fdisk with dos partitions can get funny ideas about drive geometry.

Ignore whatever fdisk is trying to do, just use 1MiB alignment anyway. If you don't need dos partitions for any particular reasons, switch to gpt. If fdisk doesn't work, switch to parted, gdisk, or whatever works.

parted /dev/sdb
) unit mib    # parted uses stupid unit by default
) mklabel gpt # beats msdos by a long shot
) mkpart boot 1 512
) mkpart swap 512 2048
) mkpart root 2048 20480
) mkpart home 20480 -1
) print free  # use this whenever you want to see what's available

Related Solutions

Debian – How to Move Root (/) to a New Partition

I'd suggest using LVM. To switch over, something like this:

Create a physical volume (pvcreate) on your existing /dev/sda2 (Windows partition)
Create a volume group consisting of just /dev/sda2 (vgcreate)
Create a logical volume to hold your / and /home (any reason you're splitting them? You don't need to.) (lvcreate)
mkfs the two new logical volumes (device: /dev/mapper/vgname-lvname)
If you're not already booted from a rescue disc, boot from one (you could use the Debian install disc, just switch to a terminal at the appropriate point, or alternatively a Ubuntu disc)
mount the new logical volumes (or just one, if you decide not to split). If you mount both, make sure to mount home inside the new root (e.g., as /new-root/ and /new-root/home) Mount your old rootfs somewhere, too.
use tar to copy the files over. Something like ( cd /old-root && tar --one-file-system -c . ) | ( cd /new-root && tar vx )
chroot into your new root filesystem, and:
1. change /etc/fstab to give the proper root filesystem
2. mount /proc and /sys and /boot (inside the chroot)
3. update-initramfs -u
4. update-grub
5. if you're paranoid, look at the generated /boot/grub/grub.cfg to check that it picked up the new rootfs.
6. exit the chroot
Finally, unmount the filesystems you mounted (starting with /new-root/{proc,sys,boot}), and reboot.
Confirm you've booted to the new root filesystem. E.g., check /proc/mounts. Confirm your data is present. (You do have backups too, right?) VERY IMPORTANT.
This is the point of no return. The next steps destroy your old partitions.
You can now pvcreate your old partitions, then add them to your volume group (vgextend)
You can now resize your / and /home logical volumes using lvextend followed by resize2fs (or whichever tool for your filesystem). This can be done online, with your system running.

Once you're on LVM, any future disk changes are much easier. You can use the LVM commands to do pretty much any disk change, with the system running even.

(Note: I haven't done this for a bit, I may have missed a step or two, but I'm pretty sure I haven't missed anything catastrophic. And of course, your original install & data is still around until after you've confirmed you're booted onto the new rootfs)

DD Command – How to Clone a Partition from Disk Image

# losetup --find --show --partscan --read-only 2013-02-09-wheezy-raspbian.img
/dev/loop7
# dd if=/dev/loop7p2 of=/dev/narnia bs=1M

If --partscan doesn't work, you can also use one of:

# partx -a /dev/loop7
# kpartx /dev/loop7

or similar partition mapping solutions.

You should probably mount it first just to see if it's the right thing or what.

Of course you can also read the fdisk output and give dd the skip=131072 or whatever directly, i.e. make it skip that many blocks of input so it starts reading at where the partition is located; but it's nicer to see actual partitions with a loop device.

Best Answer

Related Solutions

Debian – How to Move Root (/) to a New Partition

DD Command – How to Clone a Partition from Disk Image

Related Question