Re: [PATCH v2] virtiofsd/passthrough_ll: don't remove O_DIRECT when cache=none

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1452 bytes --]

On Sat, Apr 11, 2020 at 04:35:44AM -0400, Catherine Ho wrote:
> @@ -1702,10 +1703,11 @@ static void update_open_flags(int writeback, struct fuse_file_info *fi)
>  
>      /*
>       * O_DIRECT in guest should not necessarily mean bypassing page
> -     * cache on host as well. If somebody needs that behavior, it
> -     * probably should be a configuration knob in daemon.
> +     * cache on host as well. If cache=none, keep the flag unchanged
>       */
> -    fi->flags &= ~O_DIRECT;
> +    if (cache_mode != CACHE_NONE) {
> +        fi->flags &= ~O_DIRECT;
> +    }
>  }
>  

Thanks for the patch!  I have CCed the virtio-fs mailing list so more
people see it.

Please add a new command-line option to control O_DIRECT behavior.

There are two cases:

1. O_DIRECT bypasses the guest page cache but not the host page cache.
   This makes sense when the DAX feature is enabled.

2. O_DIRECT bypasses both the guest and host page cache.  This make
   sense for non-DAX and for I/O performance benchmarking.

Today only #1 is supported.

Your patch makes the behavior dependent on the cache mode option, but
the cache mode doesn't necessarily determine how O_DIRECT should be
handled.  For example, in the DAX case the guest page cache is bypassed
and cache=none can be used, but we do want to use the host page cache.

You can add a new option so that O_DIRECT handling is configurable for
all cases.

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]