*: rebase the tailscale branch onto upstream #16
Conversation
|
note: PR won't merge, I'll need to push, but I'd appreciate a quick eyeball. |
device/channels.go
Outdated
| @@ -19,13 +19,13 @@ import ( | |||
| // call wg.Done to remove the initial reference. | |||
| // When the refcount hits 0, the queue's channel is closed. | |||
| type outboundQueue struct { | |||
| c chan *QueueOutboundElement | |||
| c chan *[]*QueueOutboundElement | |||
There was a problem hiding this comment.
why pointer to slice out of curiosity? seems weird enough to be worth calling out somewhere to explain what's happening.
There was a problem hiding this comment.
We were pooling the slice headers as well as the content, but we since moved to a container in order to shard a mutex, which is also cleaner for this reason.
tun/tcp_offload_linux.go
Outdated
| coalesceInsufficientCap coalesceResult = iota | ||
| coalescePSHEnding | ||
| coalesceItemInvalidCSum | ||
| coalescePktInvalidCSum | ||
| coalesceSuccess |
There was a problem hiding this comment.
these and the tcpGROResult const values below never go across the network or to disk or to the kernel, do they? (iota should only be used when the values truly don't matter and renumbering them wouldn't break anything)
There was a problem hiding this comment.
These are all internal yeah, there are some different buffer &/ state handling semantics for specific cases.
device/channels.go
Outdated
| @@ -19,13 +19,13 @@ import ( | |||
| // call wg.Done to remove the initial reference. | |||
| // When the refcount hits 0, the queue's channel is closed. | |||
| type outboundQueue struct { | |||
| c chan *[]*QueueOutboundElement | |||
| c chan *QueueOutboundElementsContainer | |||
There was a problem hiding this comment.
oh yay the pointer to slice is gone
that's great, thank you. It's good to get some eyes on. |
This required integrating the changes from 25eb973 into the GRO patch that had included an earlier iteration.