On 11/12/26 03:49, Niklas Cassel wrote:
> A block driver that emulates zones (e.g. qcow2 with the zoned format
> feature) rejects requests that violate the zone protocol, and the device
> model has to report each of these to the guest with the matching zone
> error (ZAC/ZBC sense data, virtio-blk zone status codes, ...). It can
> only do so if each condition comes with an errno of its own, rather than
> a generic one such as EIO.
> 
> Define one for each condition:
> 
>   BLK_ZE_UNALIGNED_WRITE  ESPIPE        write not at the write pointer
>   BLK_ZE_BOUNDARY         ERANGE        write past the zone capacity
>   BLK_ZE_OPEN_RESOURCE    ETOOMANYREFS  open zone limit reached
>   BLK_ZE_ACTIVE_RESOURCE  EOVERFLOW     active zone limit reached
>   BLK_ZE_FULL             EXFULL        write to a full zone
> 
> ETOOMANYREFS and EOVERFLOW are what Linux returns for
> BLK_STS_ZONE_OPEN_RESOURCE and BLK_STS_ZONE_ACTIVE_RESOURCE, so a host
> zoned device used through file-posix already reports the same values.
> For the boundary case, ERANGE is used rather than EFBIG, which the host
> file system can legitimately return for an image file.
> 
> A write that is not at the write pointer gets ESPIPE, which file and
> block device I/O never return, rather than EINVAL, which a zoned driver
> also uses for invalid requests such as an invalid zone state
> transition. So a device model can map each of these errnos to its zone
> error without having to know which request failed.
> 
> A full zone has no write pointer, and ZAC, ZBC, NVMe ZNS and virtio-blk
> all report a write to a full zone differently from a write that is not
> at the write pointer, so it gets its own errno, EXFULL.
> 
> ETOOMANYREFS is not POSIX and EXFULL is Linux-specific, so provide
> fallbacks for hosts that lack them.
> 
> Signed-off-by: Niklas Cassel <[email protected]>

Looks OK to me.

Reviewed-by: Damien Le Moal <[email protected]>

-- 
Damien Le Moal
Western Digital Research

Reply via email to