[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [virtio-comment] Block Device VIRTIO_BLK_F_RO clarification needed.
On Wed, Jan 22, 2020 at 12:53:37AM -0500, Michael S. Tsirkin wrote: > On Tue, Jan 21, 2020 at 04:08:18PM +0000, Stefan Hajnoczi wrote: > > On Mon, Jan 20, 2020 at 11:21:36AM -0500, Michael S. Tsirkin wrote: > > > On Mon, Jan 20, 2020 at 11:45:05AM +0000, Damir Shaikhutdinov wrote: > > > > Hello! > > > > > > > > While reading section 5.2 of the spec, I found that VIRTIO_BLK_F_RO feature could use > > > > some additional clarification regarding handling requests when it was offered. > > > > > > > > This RO feature is described as "Device is read only" (Section 5.2.3). There are some explicit requirements for this feature, but they only describe a "write request". > > > > > > > > 5.2.5 p3 says "If the VIRTIO_BLK_F_RO feature is set by the device, any write requests will fail." > > > > > > > > 5.2.6 says "The type of the request is either a read (VIRTIO_BLK_T_IN), a write (VIRTIO_BLK_T_OUT) ...", so "write request" mentioned in 5.2.5 p3 is VIRTIO_BLK_T_OUT. > > > > > > > > 5.2.6.2 says "A device MUST set the status byte to VIRTIO_BLK_S_IOERR for a write request if the VIRTIO_BLK_F_RO > > > > feature if offered, and MUST NOT write any data." > > > > > > > > What is missing here is a requirement for features and other types of requests in case of VIRTIO_BLK_F_RO: > > > > > > > > * config.writeback field (guarded by VIRTIO_BLK_F_CONFIG_WCE feature) > > > > * VIRTIO_BLK_T_ FLUSH (guarded by VIRTIO_BLK_F_FLUSH feature) > > > > * VIRTIO_BLK_T_DISCARD (guarded by VIRTIO_BLK_F_DISCARD feature) > > > > * VIRTIO_BLK_T_WRITE_ZEROES (guarded by VIRTIO_BLK_F_WRITE_ZEROES feature) > > > > > > > > Configuration field, requests and features described above make no sense in presence of VIRTIO_BLK_F_RO. > > > > > > > > This can be clarified in several ways, for example: > > > > > > > > 1. Device should not offer CONFIG_WCE, FLUSH, DISCARD and WRITE_ZEROES features if it offers RO feature (partially contradicts 5.2.5.2 "Device SHOULD always offer VIRTIO_BLK_F_FLUSH") > > > > 2. Device must set the status byte to IOERR for FLUSH, DISCARD and WRITE_ZEROES requests if it offers RO feature. > > > > > > > > > > > > What do you think about that? > > > > > > I guess flush can even be allowed, it's harmless if nothing is written, > > > right? > > > > I agree. There seems to be no benefit in failing it and there's a small > > chance that some drivers could get upset if it fails (typically when a > > common code path submits a flush request and expects it to succeed). > > > > > I agree DISCARD and WRITE_ZEROES must fail. And set VIRTIO_BLK_S_IOERR > > > I guess? > > > > Yes. > > > > > It's too late to disallow configurations such as a combination of VIRTIO_BLK_F_CONFIG_WCE > > > and VIRTIO_BLK_F_RO, devices out there might be setting this combination. > > > > Yes. > > > > Stefan > > Thanks! Stefan do you have the time to work on a spec patch to fix this? > Damir do you? Damir, you can find instructions for contributing to the virtio specification here: https://github.com/oasis-tcs/virtio-spec/ If you don't have time to write a patch please let me know. Stefan
Attachment:
signature.asc
Description: PGP signature
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]