Running QEMU with persistent disks

It‘s useful to run QEMU with persistent disks, like you’d have on actual hardware.

Specifically you‘d want a /data minfs partition and a /blobstore blobfs partition, but /system and /boot from your local build. Here’s how to do that.

NOTE: Lines that begin with a $ should be typed. Other lines are example output.

Create the disk image

blk.bin is the name of the image that frun / mrun looks for when you pass -d. Make a 1g one in your $FUCHSIA_DIR.

Linux:

$ cd $FUCHSIA_DIR
$ truncate -s 1g blk.bin

macOS:

$ cd $FUCHSIA_DIR
$ mkfile -n 1g blk.bin

Start zircon

You don't need a full Fuchsia UI to set up the disk image so start just zircon but tell it to mount ``` $ mrun -d [00000.000] 00000.00000> multiboot: info @ 0xffffff8000009500 [00000.000] 00000.00000> multiboot: cmdline @ 0xffffff8000253059 [00000.000] 00000.00000> multiboot: ramdisk @ 00254000..0ec607e0 [00000.000] 00000.00000> bootdata: @ 0xffffff8000254000 (245417952 bytes) [00000.000] 00000.00000> [00000.000] 00000.00000> welcome to lk/MP (etc)


## Initialize the GPT You blank `blk.bin` image needs a partition table.

$ gpt init /dev/class/block/000 blocksize=0x200 blocks=2097152 WARNING: You are about to permanently alter /dev/class/block/000

Type ‘y’ to continue, any other key to cancel invalid header magic! [00031.068] 02004.02044> device: 0x4e3af554b000(sata0): ref=0, busy, not releasing [00031.070] 01043.01046> devcoord: drv=‘block’ bindable to dev=‘sata0’ [00031.072] 01043.01166> devmgr: new block device: /dev/class/block/001 GPT changes complete. [00031.077] 01043.01166> devmgr: /dev/class/block/001: GPT? [00031.078] 01043.01046> devcoord: dc_bind_device() ‘/boot/driver/gpt.so’ [00031.078] 01043.01046> devcoord: drv=‘gpt’ bindable to dev=‘block’


## Create the partitions Now that there's a blank partition table, create a 500m data and 500m blob partition: ``` $ gpt repartition /dev/class/block/001 data data 500m blob blobfs 500m blocksize=0x200 blocks=2097152 data: 524288000 bytes, 1024000 blocks, 48-1024063 blob: 524288000 bytes, 1024000 blocks, 1024064-2048079 [00242.582] 02004.02044> device: 0x4e3af554b000(sata0): ref=0, busy, not releasing [00242.584] 01043.01046> devcoord: drv='block' bindable to dev='sata0' [00242.587] 01043.01166> devmgr: new block device: /dev/class/block/002 GPT changes complete. [00242.594] 01043.01166> devmgr: /dev/class/block/002: GPT? [00242.596] 01043.01046> devcoord: dc_bind_device() '/boot/driver/gpt.so' [00242.596] 01043.01046> devcoord: drv='gpt' bindable to dev='block' [00242.619] 01043.01046> devcoord: drv='block' bindable to dev='part-000' [00242.622] 01043.01166> devmgr: new block device: /dev/class/block/003 [00242.624] 01043.01046> devcoord: drv='block' bindable to dev='part-001' [00242.628] 01043.01166> devmgr: new block device: /dev/class/block/004

Create the filesystems

And now create the minfs and blobstore filesystems on the new partitions:

$ mkfs /dev/class/block/003 minfs
$ mkfs /dev/class/block/004 blobstore

You're done. Now reboot and pass -d to frun to run with peristent disks.

Keeping a spare

Sometimes you will want to make a fresh clean disk image. After you have created a new, empty blk.bin you can gzip it as a backup:

$ gzip blk.bin

Then to get a decompressed blk.bin without getting rid of your gzipped backup:

$ gunzip -k blk.bin.gz

Multiple instances

If you want to run multiple instances of QEMU they must not be trying to use the same disk image. Each much have their own disk image. You can follow the instructions above but pass -D other-disk-name.bin to mrun and frun to specify the alternative location.