[Virtio-fs] [virtiofsd] Issue opened: Unexpected behavior observed when using mmap on virtiofs backed paths

[Virtio-fs] [virtiofsd] Issue opened: Unexpected behavior observed when using mmap on virtiofs backed paths

[virtiofs-bot]

When using python and writing to a mmap file, I am seeing with cache mode auto and always that the file is increasing in size each time I write to it.  This is seen with virtiofsd C version ( 6.1.0 ) as well as RS version (1.3.0), and can be reproduced using QEMU (6.2) as well as Cloud Hypervisor (v0.26). When switching to a different vhost-user-fs backend (nydusd), we can no longer reproduce the failure.  

The following python script will show the failure behavior if run on a virtiofs backed path:
```
import mmap
import struct
l = 1024
f = open("./test.db", "a+b")
f.truncate(l)
m = mmap.mmap(f.fileno(), 1024, access=mmap.ACCESS_WRITE)
m[0:4] = b"0000"
m.flush()
m.size()
m[0:4] = b"0000"
m.flush()
m.size()
```

See https://github.com/kata-containers/kata-containers/issues/5634 for more details on reproduction.
---
https://gitlab.com/virtio-fs/virtiofsd/-/issues/67