search

ublk-org / ublksrv

ublk: userspace block device driver
MIT License
147 stars 50 forks source link

wip: ublk zoned #28

Open metaspace opened 2 years ago

metaspace commented 2 years ago

Hi!

This is a work in progress effort to add zoned storage support to ublk. Kernel patches are here.

Any feedback is appreciated :)

ming1 commented 1 year ago

Hi metaspace,

I think we have reached agreement that zoned implementation should be done in tgt_null.c which is for test purpose, and should be kept as it is.

So please do not send PR for messing tgt_null.c with zoned.

Thanks,

ming1 commented 1 year ago

Hi metaspace,

I think we have reached agreement that zoned implementation should be done in tgt_null.c which is for test purpose, and should be kept as it is.

shouldn't be done.

metaspace commented 1 year ago

Hi Ming,

I think we have reached agreement that zoned implementation should be done in tgt_null.c which is for test purpose, and should be kept as it is.

So please do not send PR for messing tgt_null.c with zoned.

Thank you for your response. Please note that this PR is just a draft. I understand that you would prefer the zoned implementation as a separate target. To comply with that, I plan to move the zoned code out of loop and null targets, into a separate zoned target. I also need to rebase on master I think.

However, you suggested handling the kernel<-->user interface before putting too much effort into patching ubdsrv. So currently this PR is just demonstrating one possible user space implementation, not necessarily the final one.

If you would prefer, we can close this PR until the kernel patches are accepted and then I can submit a new PR with a separate zoned target later? But I think it is nice to have this PR to track the work.

In case you missed them, the kernel patches are here. I tried to CC you :)

Best regards Andreas

ming1 commented 1 year ago

Hi Ming,

I think we have reached agreement that zoned implementation should be done in tgt_null.c which is for test purpose, and should be kept as it is. So please do not send PR for messing tgt_null.c with zoned.

Thank you for your response. Please note that this PR is just a draft. I understand that you would prefer the zoned implementation as a separate target. To comply with that, I plan to move the zoned code out of loop and null targets, into a separate zoned target. I also need to rebase on master I think.

However, you suggested handling the kernel<-->user interface before putting too much effort into patching ubdsrv. So currently this PR is just demonstrating one possible user space implementation, not necessarily the final one.

If you would prefer, we can close this PR until the kernel patches are accepted and then I can submit a new PR with a separate zoned target later? But I think it is nice to have this PR to track the work.

In case you missed them, the kernel patches are here. I tried to CC you :)

I know this kernel patch, and I meant even for test purpose only, it is still better to write one clean & intact(at least for interface verification) target for verification & reviewing kabi interface, which won't be modified any more after kernel patch is merged(can only be extended). So here we need to pay more attention to.

I will review the kernel patch in this week.

Thanks, Ming