Martin Lucina

Yes. You probably want to look at #325 first though.

That depends on what the backend implementation needs to do. As I've never needed to invoke a `discard`-type op from Linux, I don't know. The Solo5 interface I would prefer...

> With spt bindings on Linux, this will use fallocate with FALLOC_FL_PUNCH_HOLE and fallback to zeroing manually if the filesystem doesn't support it. That man page does not actually mention...

Oh. One more thing, just re-reading the manpage. It only talks about _filesystem-backed_ fds. How do we do the equivalent if the `fd` is pointing directly to a block device...

> At any rate, the logic for this will be the same regardless of what the backing store is; if ENOTSUP is returned, write zeroes manually. Right, just make sure...

@g2p Moving your questions from Slack here, as I'd like the design discussion to be in one place: > What would I use to allocate a buffer of zeroes from...

> I need it to actually ensure later reads return zeroes. Ok. It seems to me then that the simplest possible interface at the Solo5 layer is that mentioned in...

Sounds good to me.

/cc @avsm @yomimono

FWIW, I've done (3) for Solo5 using the standalone C unikernels only, no Mirage involved. Also, only supports QEMU/KVM/ukvm/bhyve for now (solo5/solo5#120).