Re: Re: [PATCH v11 02/10] block/raw: add persistent reservation in/out driver

[zhenwei pi]

On 9/10/24 04:18, Keith Busch wrote:
> On Mon, Sep 09, 2024 at 07:34:45PM +0800, Changqi Lu wrote:
> > +static int coroutine_fn GRAPH_RDLOCK
> > +raw_co_pr_register(BlockDriverState *bs, uint64_t old_key,
> > +                   uint64_t new_key, BlockPrType type,
> > +                   bool ptpl, bool ignore_key)
> > +{
> > +    return bdrv_co_pr_register(bs->file->bs, old_key, new_key,
> > +                               type, ptpl, ignore_key);
> > +}
>
> The nvme parts look fine, but could you help me understand how this all
> works? I was looking for something utilizing ioctl's, like
> IOC_PR_REGISTER for this one, chained through the file-posix block
> driver. Is this only supposed to work with iscsi?

Hi Keith,

IOC_PR_REGISTER family supports PR OUT direction only in Linux kernel, 
so the command `blkpr` command (from util-linux since v2.39) supports PR 
OUT direction only too.

- In a guest:
* `blkpr` command could test PR OUT commands.
* `sg_persist` command (from sg3-utils) works fine on a SCSI device
* `nvme` command (from nvme-cli) works fine on a NVMe device

- On a host:
* libiscsi supports PR, and LIO/SPDK supports PR from the target side, 
tgt supports uncompleted PR(lack of PTPL), so QEMU libiscsi driver works 
fine.
* `iscsi-pr` command (from libiscsi-bin since v1.20.0) supports the full 
PR family command.
* because of the lack of PR IN commands from linux block layer, so QEMU 
posix block driver can't support PR currently. Once this series is 
merged into QEMU, I think we have a scenario on posix block PR IN 
family, it's a hint to promote it for linux block layer.
* I wrote a user space nvme-of library `libnvmf` 
(https://github.com/bytedance/libnvmf), it does not support PR family 
command, but I don't think it's difficult to support if necessary.
* As far as I know, several private vendor block driver support PR 
family, this QEMU block framework will make the private drivers easy to 
integrate.