wireguard-openbsd/sys/netinet, branch jd/queueboosts WireGuard implementation for the OpenBSD kernel https://git.zx2c4.com/wireguard-openbsd/atom/sys/netinet?h=jd%2Fqueueboosts 2020-06-21T05:19:27Z wrap a long line. no functional change. 2020-06-21T05:19:27Z dlg dlg@openbsd.org 2020-06-21T05:19:27Z urn:sha1:e7446642031c218e3629adf1bb790ad9b7043c2b if an inp_upcall is set, let it look at and maybe steal the udp packet. 2020-06-21T05:17:15Z dlg dlg@openbsd.org 2020-06-21T05:17:15Z urn:sha1:63a788dae37771312ea8081c21a6a4bf68bd1bbb i wrote the original version of this, but it was tweaked by Matt Dunwoodie and Jason A. Donenfeld for use with wireguard. knf: the inp_upcall line was too long. 2020-06-21T05:14:04Z dlg dlg@openbsd.org 2020-06-21T05:14:04Z urn:sha1:f078884af493dcfa8d4c1b98344467fc12ab9bbe add a inp_upcall function pointer and inp_upcall_arg to struct in_pcb. 2020-06-21T05:12:06Z dlg dlg@openbsd.org 2020-06-21T05:12:06Z urn:sha1:418165f529c2f1871c09378d773a3f9af2b669ba this is so protocols (eg, udp) can let things (eg, kernel support for wireguard or vxlan or geneve) look at and possibly steal packets before they get added to a socket buffer. i wrote the original version of this, but it was tweaked by Matt Dunwoodie and Jason A. Donenfeld for use with wireguard. Break a glass ceiling on cwnd due to integer division during congestion 2020-06-19T22:47:22Z procter procter@openbsd.org 2020-06-19T22:47:22Z urn:sha1:0c48a0c979b0f3897fe6b2c44dd76df5ebe75265 avoidance. The problem and fix is noted in RFC5681 section 3.1, page 7. Report, diff and testing from Brian Brombacher, thanks! Testing and a cosmetic tweak by myself. ok claudio Refuse to set 0 or a negative value for net.inet.tcp.synbucketlimit. 2020-06-18T14:52:51Z mpi mpi@openbsd.org 2020-06-18T14:52:51Z urn:sha1:2b10bfc1e665e7267e288a6f7558e12495aa488e Prevent a panic in syn_cache_insert() found by syzbot. Reported-by: syzbot+aee24ad9b7bf5665912d@syzkaller.appspotmail.com ok sashan@, anton@, millert@ Connectionless sockets like UDP can be re-connected to a different 2020-05-27T20:44:07Z bluhm bluhm@openbsd.org 2020-05-27T20:44:07Z urn:sha1:e425060fbaec9ae9ef36af44a3fa14b13d0f72e6 address. In that case, the linking to the pf state must be dissolved as the latter still contains the old address. If it is a divert state, also remove the state as any divert state must be associated with a matching socket. Call pf_remove_divert_state() and pf_inp_unlink() from in_pcbconnect(). reported by Tim Kuijsten; OK sashan@ claudio@ Document the various flavors of NET_LOCK() and rename the reader version. 2020-05-27T11:19:28Z mpi mpi@openbsd.org 2020-05-27T11:19:28Z urn:sha1:4d5cbc65b3f249a15784c3b64c2933d8466f7caa Since our last concurrency mistake only ioctl(2) ans sysctl(2) code path take the reader lock. This is mostly for documentation purpose as long as the softnet thread is converted back to use a read lock. dlg@ said that comments should be good enough. ok sashan@ don't count packets in the carp protocol handling against an interface. 2020-05-21T05:24:59Z dlg dlg@openbsd.org 2020-05-21T05:24:59Z urn:sha1:f503af1577e35c83a5c76a8bbe18623e8e2005d6 these packets have generally already been counted on the interface because that's where they were sent or received from. the protocol handling side of things already counts things like packets, which you see with netstat -sp carp. implement a carp_transmit that bypasses the ifq on output. 2020-05-21T03:33:44Z dlg dlg@openbsd.org 2020-05-21T03:33:44Z urn:sha1:18daf6c0a930925e176c39abedbdcc33edb0f7c5 this is modelled on vlan_transmit, and basically enqueues the packet directly on the parent interface. even though carp is generally not used to transmit packets, we run dhcp relays on it at work and hit a situation where we unecessarily dropped packets because it's ifq maxlen was 1. i've been running this for a month in production. ok jmatthew@