Allow reformatting of nvmes to different blocksizes

[fionera]

I run servers with NVMes that are formatted to a block size of 512b by default. It would be cool to have the ability to reformat these to a 4k block size. Is this in the scope of this project?

For now I will implement this by using a hook but a native way would be very nice

[Mic92]

Should this not already be possible by having additional mkfs arguments passed when formatting a partition?

disko/types/filesystem.nix

Line 9 in eb5def3

extraArgs = lib.mkOption {

It takes an extraArgs which could be set to -b 4096.

[fionera]

Afaik thats only for the filesystem and not the nvme itself

[Mic92]

Are you talking about aligning partitions to a 4k boundary? Can you give an example of what arguments to what command would needs to be changed?

[fionera]

I am talking about the lbaf parameter at eg nvme-cli https://www.mankier.com/1/nvme-format

[Mic92]

Ah. I see nvme namespaces could be a thing in disko. The api would probably looks similar to partitions?

[Lassulus]

I guess we would need a different table implementation. But I want to split the current one into a gpt and dos table type anyway. The namespaces could be an attrsOf in the new nvme_table type. An nvme_table would only work with nvme drives though. So we would need an assertion for that. For just blocksizes we could also add a option to the current table type. Not sure what reformatting means in this context though

[fionera]

when you call nvme format it is possible to change the blocksize. Its just a call to the internal nvme formatting function. there is an annoying side effect of having to rebind the devices, as the addressing on pcie is different afterwards

nix-shell -p pciutils nvme-cli jq
# the nvmes we use have 1 as 4k format. thats disk dependend tho
nvme list -o json | jq -r .Devices[].DevicePath | xargs -I{} nvme format -l 1 --force {}

for i in $(lspci | grep NVMe | awk '{print $1}'); do \
 echo 0000:$i > /sys/bus/pci/drivers/nvme/unbind; \
 echo 0000:$i > /sys/bus/pci/drivers/nvme/bind; \
done

[iFreilicht]

From my understanding, this would either require an extension to the disk type, or to clone the disk type into a new nvme type that then has additional options for running and passing arguments to nvme-format, right?