netdev - Re: [External] Re: [PATCH 0/4] Add multi-cid support for vsock driver

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [day] [month] [year] [list]

Date:   Wed, 4 Aug 2021 17:23:49 +0200
From:   Stefano Garzarella <sgarzare@...hat.com>
To:     傅关丞 <fuguancheng@...edance.com>
Cc:     "Michael S. Tsirkin" <mst@...hat.com>, jasowang@...hat.com,
        stefanha@...hat.com, davem@...emloft.net, kuba@...nel.org,
        arseny.krasnov@...persky.com, andraprs@...zon.com,
        Colin King <colin.king@...onical.com>, kvm@...r.kernel.org,
        virtualization@...ts.linux-foundation.org, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [External] Re: [PATCH 0/4] Add multi-cid support for vsock driver

On Wed, Aug 04, 2021 at 03:09:41PM +0800, 傅关丞 wrote:
>Sorry I cannot figure out a good use case.
>
>It is normal for a host to have multiple ip addresses used for
>communication.
>So I thought it might be nice to have both  host and guest use multiple
>CIDs for communication.
>I know this is not a very strong argument.

Maybe there could be a use case for guests (which I don't see now), but 
for the host it seems pointless. The strength of vsock is that the guest 
knows that using CID=2 always reaches the host.

Moreover we have recently merged VMADDR_FLAG_TO_HOST that when set 
allows you to forward any packet to the host, regardless of the CID (not 
yet supported by vhost-vsock).

>
>The vsock driver does not work if one of the two peers doesn't support
>multiple CIDs.

This is absolutely to be avoided.

I think the virtio device feature negotiation can help here.

>
>I have a possible solution here, but there may be some problems with it
>that I haven't noticed.
>
>Hypervisors will use different ways to send CIDs setup to the kernel based
>on their vsock setup.
>
>------For host-------
>If host vsock driver supports multi-cid, the hypervisor will use the
>modified VHOST_VSOCK_SET_GUEST_CID call to set its CIDs.
>Otherwise, the original call is used.
>
>------For guest-------
>Now the virtio_vsock_config looks like this:
>u64 guest_cid
>u32 num_guest_cid;
>u32 num_host_cid;
>u32 index;
>u64 cid;
>
>If the guest vsock driver supports multi-cid, it will read num_guest_cid
>and num_host_cid from the device config space.
>Then it writes an index register, which is the cid it wants to read.  After
>hypervisors handle this issue, it can read the cid
>from the cid register.
>
>If it does not support multi-cid, it will just read the guest_cid from the
>config space, which should work just fine.
>

Why not add a new device feature to enable or disable multi-cid?


>
>-------Communication--------
>For communication issues, we might need to use a new feature bit.  Let's
>call it VHOST_VSOCK_SUPPORT_MULTI_CID.
>The basic idea is that this feature bit is set when both host and guest
>support using multiple CIDs.  After negotiation, if the feature bit
>is set, the host can use all the CIDs specified to communicate with the
>guest.  Otherwise, the first cid passed in will
>be used as the guest_cid to communicate with guests.

I think the same feature bit can be used for the virtio_vsock_config, 
no?

>
>Also, if the bit is set for guests, all the CIDs can be used to communicate
>with the host.  Otherwise, the first cid with index 0 will be
>used as the guest_cid while the VMADDR_HOST_CID will be used for host cid.

We already have VMADDR_FLAG_TO_HOST to forward all packets to the host, 
we only need to support in some way in vhost-vsock.

Thanks,
Stefano