9p: use fscache by default for RO mounts

[AkihiroSuda]

Fix #786

[matejsp]

Shouldn't default be based on not having any writeable mounts?
As in: any writeable -> default mmap else fscache?

[jandubois]

Shouldn't default be based on not having any writeable mounts?
As in: any writeable -> default mmap else fscache?

I don't think the default for one mount should depend on the writable property of a different mount.

This is not a global setting, so it would make more sense (to me) to say: defaults to fscache for readonly mounts and mmap for writable ones.

[AkihiroSuda]

Updated PR to keep mmap as the default for RW mounts

[jandubois]

This is not a global setting, so it would make more sense (to me) to say: defaults to fscache for readonly mounts and mmap for writable ones.

Already implemented before I can press the "Comment" button! 😂

[afbjorklund]

Sounds good to me, it is more complex than just keeping the upstream default (of "none") but performance could be worth it...

Never exposed the "cache" parameter on Podman Machine or for Minikube (in 2017), so there it uses cache=none always

---

I guess the same already applied to the msize parameter, where the upstream default (8K on Mac) caused runtime warnings.

qemu-system-aarch64: warning: 9p: degraded performance - fixed by increasing the default msize to 128K on all platforms

[jandubois]

LGTM

[LionsAd]

@AkihiroSuda one thing to note:

fscache failed miserably for me when doing the dd test.

eG as it’s read only,

dd if=largefile of=/dev/null bs=1M count=1000

that said I tested writing and it failed, so maybe reading is fine but wanted to note it anyway.