include, block: add krun_add_disk3 API

[jakecorrenti]

Add the krun_add_disk3 API that allows users to explicitly configure the sync and cache mode for a virtio-blk device.

This is a followup to #428

[tylerfanelli]

On non macOS targets, should we define the API to just return EINVAL?

[jakecorrenti]

On non macOS targets, should we define the API to just return EINVAL?

On a non-macOS target it will be a compile-time error. IMO this better than a runtime error so it's clear it's not even an option, but if you think it should be a runtime error I'd love to hear your reasoning

[slp]

Hm... I think I would favor a more generic approach, implementing a krun_add_disk3 instead. This new API should have two additional arguments:

• cache_mode: to choose between "CACHED" and "UNCACHED".
  • CACHED: default behavior, how libkrun works right now.
  • UNCACHED: imago configured with direct_io.
• sync_mode: to choose between "FULL", "RELAXED" and "NONE".
  • FULL: default behavior, how libkrun worked until the relaxed mode on macOS was introduced.
  • RELAXED: on macOS, enable imago relaxed mode. On Linux, we may consider doing a partial fsync.
  • NONE: ignore VIRTIO_BLK_F_FLUSH.

[jakecorrenti]

Hm... I think I would favor a more generic approach, implementing a krun_add_disk3 instead. This new API should have two additional arguments:

* cache_mode: to choose between "CACHED" and "UNCACHED".
  
  * CACHED: default behavior, how libkrun works right now.
  * UNCACHED: imago configured with direct_io.

* sync_mode: to choose between "FULL", "RELAXED" and "NONE".
  
  * FULL: default behavior, how libkrun worked until the relaxed mode on macOS was introduced.
  * RELAXED: on macOS, enable imago relaxed mode. On Linux, we may consider doing a partial fsync.
  * NONE: ignore VIRTIO_BLK_F_FLUSH.

Ah, ok I see now. I'll do that instead. Thanks