Remove buffering from network layer and rework loopback interface.
loopback now has a separate recieve thread to allow concurrent sends and
prevent deadlocks
We now only send enough data to fill other ends window, not past that.
Previous logic had a but that allowed sending too much data leading to
retransmissions.
When the target sends zero window and later updates window size,
immediately retransmit non-acknowledged bytes.
Don't validate packets through listeing socket twice. The actual socket
will already verify the checksum so the listening socket does not have
to.
We now report actually available window size when sending packets. If
the available window size grows significantly we send an ACK to reflect
this to the remote.
This also removes the now old recvfrom and sendto syscalls. These are
now implemented as wrappers around recvmsg and sendmsg.
Also replace unnecessary spinlocks from unix socket with mutexes
Kernel can just use raw threads, pretty muchs the only thing that
process provides is syscalls which kernel threads of course don't
need.
Also this makes init process have pid 1 :D
All block functions now take an optional mutex parameter that is
atomically unlocked instead of having the user unlock it before hand.
This prevents a ton of race conditions everywhere in the code!
If connection on unix socket was closed and other end tries to recvfrom,
the thread would enter a fucked up state where it held the socket's
spinlock when returning to userspace.
- Fix race condition when adding packet to send buffer before other end
has acknowledged it
- Allow sending multiple packets before receiving ACK for previous ones
This implementation is on top of inodes instead of fds as linux does it.
If I start finding ports/software that relies on epoll allowing
duplicate inodes, I will do what linux does.
I'm probably missing multiple epoll_notify's which may cause hangs but
the system seems to work fine :dd:
