Is NFS/Samba the only way to share files between host and guest?

[jgonera]

I was wondering if I am missing something. Can I somehow mount the disk image both on guest and host? I suspect that this would cause concurrency issues even if it is possible.

[xez]

Network based solutions are probably the only options right now. I suggested virtio-9p as means of guest <-> host file sharing. But thats quite a bit of work.

[geoff-nixon]

@jgonera osxfuse + fuse-ext2 should allow you to mount R/W concurrently, though I can't vouch for it personally. Myself I use Paragon's commercial ExtFS for Mac, which is quite fast and the write locks are good enough I've never seen a race condition or corruption.

[geoff-nixon]

@johanneswuerbach Good point... osxfuse requires root and a codesigned kext. So I'll venture a different "no" answer — there's also afp (netatalk/avahi), http/webdav, ftp, scp, netcat... 😉 .

[galaxy001]

It should be possible to mimic the Parallel Desktop's solution that installs a driver to the guest and export filesystem to host.

If we export it to host under fuse, no other kext needed except for osxfuse itself.

Well, I just read about VirtFS. Seems enough.

---

@geoff-codes mentioned mount R/W for a disk image with write locks, I wonder whether the "write locks" means it can make both the virtual machine and host to write on same disk image ?

[jceel]

Hi guys,

I'm working on virtio-9p (so-called VirtFS) support for bhyve/xhyve. Right now I have almost complete 9p2000.u protocol support (except atime/mtime modification in Twstat).

You can check it out here: https://github.com/jceel/xhyve

Here's how to use it:

1. add -s <n>,virtio-9p,sharename=/host/path to your xhyve arguments
2. boot Linux
3. mount -t 9p -o trans=virtio -o version=9p2000.u sharename /mnt/guest/path

Using 9p as root filesystem should work too.

This code is of course totally experimental and untested - I'm looking forward to your feeback!

[xez]

@jceel Awesome, great work!!!

[ailispaw]

Hi @jceel great work!!
I tried to use virtfs with your xhyve and it works, but I have one problem here.
When the owner of a file and a user in a VM have different uid, the user (except root) can't modify the file from the VM.
Is there any configuration for ACL mapping like -mapall option for NFS?

[jceel]

@ailispaw: You can try the following mount options: access=any (and optionally dfltuid=<n> if going to attach as non-root). That should make Linux do single attach as root/dfltuid.

Here's mount -t 9p options reference: https://www.kernel.org/doc/Documentation/filesystems/9p.txt

[ailispaw]

@jceel Thank you for the reply.
Yes. I have added access=any to it, but I can read files but write.
I will try dfltuid=<n>. Should <n> be a uid on host, shouldn't it?

[jceel]

@ailispaw: yes, it should refer to the host UID.

[ailispaw]

Kernel version: 4.4.1 with

CONFIG_NET_9P=y
CONFIG_NET_9P_VIRTIO=y
CONFIG_9P_FS=y
CONFIG_9P_FS_POSIX_ACL=y

501 is my uid in OSX 10.10.5 host.

$ sudo mkdir -p /mnt/host
$ sudo mount -t 9p -o trans=virtio -o version=9p2000.u -o access=any -o dfltuid=501 host /mnt/host
$ touch /mnt/host/aaa
touch: /mnt/host/aaa: Permission denied

Something missing?

-s 5,virtio-9p,host=/Users/ailispaw/Desktop/test in my xhyverun.sh

[jceel]

Ah, I see what you're trying to do. There's no UID/GID translation mechanism (yet) and it turns out dfltuid works a bit differently than I thought:

• it's effective only when using legacy 9P2000 protocol (because original 9P2000 doesn't have concept of numeric uids/gids)
• it's a client uid

So, a very poor man's solution would be to use 9P2000 and map everything to uid specified by dfltuid. For example:

mount -t 9p -o version=9p2000 -o trans=virtio -o access=any -o uname=root -o dftlduid=1000

but with that approach, you'll lose ability to create symlinks/pipes/devices and ability see/modify file ownerships - basically everything in shared filesystem will be shown as owned by UID 1000. on the host side, newly created files and directories will be owned by root.

(to use -o uname option you need to update your sources, I've checked in support for it few minutes ago)

It seems to me that there are two ultimate solutions:

• stick to "align your UIDs, sir" policy forever
• implement an UID translation mechanism on the xhyve side (eg. make it read uid1=uid2 pairs from a text file specified on the command line)

[ailispaw]

@jceel Thank you so much for the explanation.
I thought security_mode=mapped on the virtfs server side was for that purpose.

[ailispaw]

@jceel I would like to let you know the result of 9p2000 case.
I tried mount -t 9p -o version=9p2000 -o trans=virtio -o access=any -o uname=root -o dfltuid=1000 host /mnt/host, but I got a xhyve panic as below.

$ sudo mkdir -p /mnt/host
$ sudo mount -t 9p -o version=9p2000 -o trans=virtio -o access=any -o uname=root -o dfltuid=1000 host /mnt/host
xhyve(5451,0x14468d000) malloc: *** error for object 0x7ffbd07006b0: pointer being freed was not allocated
     *** set a breakpoint in malloc_error_break to debug
                                                        ./xhyverun.sh: line 68:  5451 Abort trap: 6

[ailispaw]

@jceel Oh, I'm sorry I didn't update xhyve. I will test it again.

[ailispaw]

@jceel
That solution works for me!!

$ sudo mount -t 9p -o version=9p2000 -o trans=virtio -o access=any -o uname=ailispaw -o dfltuid=1000 -o dfltgid=1000 host /mnt/host

Thanks. :)

[stv0g]

Do you plan to push those changes to @mist64's repo?

[galaxy001]

I tried jceel/xhyve with sysresccd which provides fs/9p/9p.ko. However, when I try tomount -t 9p -o trans=virtio,version=9p2000.L hostshare /mnt/ after modprobe 9p, it says:

mount: special device hostshare does not exist

And find /dev/ |grep hostshare is empty.

My lsmod is:

% lsmod|grep -P '9p|virtio'
9p                     41056  0
fscache                47757  1 9p
9pnet_virtio           12537  0
9pnet                  58902  2 9p,9pnet_virtio
virtio_pci             12487  0
virtio_ring            14405  2 virtio_pci,9pnet_virtio
virtio                 13313  2 virtio_pci,9pnet_virtio

[jceel]

9p2000.L is not supported yet - try 9p2000.u.

[galaxy001]

I keep get this in dmesg each time run mount:

9pnet_virtio: no channels available

I googled to https://bugs.launchpad.net/qemu/+bug/648128 and find 2 patch for trans_virtio.c and mount.c.

How can I check whether xhyve works ?

My lspci do shows a Virtio filesystem, but I do not know how to dump its tagname.

% lspci
00:00.0 Host bridge: Network Appliance Corporation Device 1275
00:02.0 Ethernet controller: Red Hat, Inc Virtio network device
00:03.0 SATA controller: Intel Corporation 82801HR/HO/HH (ICH8R/DO/DH) 6 port SATA Controller [AHCI mode]
00:05.0 SCSI storage controller: Red Hat, Inc Virtio filesystem
00:1f.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II]

[jceel]

What did you pass to xhyve command line?

[galaxy001]

build/xhyve -A -m 1G -s 0:0,hostbridge -s 31,lpc -l com1,stdio -s 2:0,virtio-net -s 3,ahci-cd,/Users/Galaxy/Downloads/systemrescuecd-x86-4.7.2.iso -s 5,virtio-9p,hostshare=/Users/Galaxy/git/xhyve/t,ro -f kexec,systemrescuecd/rescue64,systemrescuecd/initram.igz,earlyprintk=serial console=ttyS0

[galaxy001]

@jceel Would you update a kernel binary in your repo for testing ?

If my CLI was correct, it should due to the systemrescuecd kernel cannot support even after modprobe 9p.