Author: Jason A. Donenfeld
Author Date: 2019-12-19 00:12:35 UTC

version: bump snapshot

Author: Matt Dunwoodie
Author Date: 2019-10-27 10:13:20 UTC

Rework ipc.c to new OpenBSD IOCTL interface

Author: Jason A. Donenfeld
Author Date: 2019-10-28 16:55:33 UTC

peerlookup: fall back to get_random_bytes for Ryzen 3000 bug

In case get_random_u32() fails after 32 tries -- something that should
only happen on the Ryzen 3000 which returns -1 everytime if you have the
wrong CPU microcode -- we fall back to get_random_bytes(), which is
slower, but at least works.

Author: Nicolas Douma
Author Date: 2019-10-02 19:10:51 UTC

wg-quick: android: use Binder for setting DNS on Android 10

Signed-off-by: Nicolas Douma <nicolas@serveur.io>

Author: Jason A. Donenfeld
Author Date: 2019-06-11 17:22:52 UTC

tools: add syncconf command

Author: Shawn Landden
Author Date: 2019-05-11 17:19:51 UTC

[zinc] Add PowerPC accelerated poly1305 from openssl/cryptograms

Unfortunately I am not seeing a speed up with this patch,
but it does decrease CPU usage.

Only (currently) runs on the outbound path, as the in-bound path is in
an interrupt, but that can be fixed in Linux.

v2: - Do not include the FPU version, as +10% performance on POWER8
    (admittedly better on really old CPUs, like old world macs) is not
    worth it, especially when there is a fast VSX version available.
    - Honor CONFIG_VSX.
Signed-off-by: Shawn Landden <shawn@git.icu>

Author: Joe Holden
Author Date: 2019-05-12 17:41:50 UTC

compat: on older kernels, ignore_df might be local_df

Author: Luis Ressel
Author Date: 2019-04-11 13:13:11 UTC

wg.8: Rewrite AllowedIPs description

* The current text doesn't describe how overlapping values are handled.
* "[addrs] to which outgoing traffic for this peer is directed" is vague
  and misleading.
* 0.0.0.0/0 and ::/0 don't need to be mentioned, since they aren't
  special cases. (Should they be mentioned in the example section,
  though?)

Thanks-to: jrb0001, MacGyver, zanijwa
Signed-off-by: Luis Ressel <aranea@aixah.de>

Author: Luis Ressel
Author Date: 2019-03-25 13:53:09 UTC

[WIP] wg: Support human-readable peer names

TODO:
* Refactor the inflatable_buffer code from ipc.c and use it for both
  the wgpeer_name array and the names.
* ugly_print, dumb_print
* Make wgpeer_names part of wgdevice?
* Check for dupes?
* Test!

Signed-off-by: Luis Ressel <aranea@aixah.de>

Author: Florent Daigniere
Author Date: 2019-02-26 17:21:55 UTC

receive: fix the ECN-related behaviour

Match what the kernel would do; drop packets when it would

Signed-off-by: Florent Daigniere <nextgens@freenetproject.org>

Author: Florent Daigniere
Author Date: 2019-02-24 18:27:21 UTC

net: implement ECN handling, rfc6040 style

To decide whether we should use the compatibility mode or the normal
mode with a peer, we use the handshake messages as a signaling channel.

If we receive the expected ECN bits, it most likely means they're
running a compatible version.

Signed-off-by: Florent Daigniere <nextgens@freenetproject.org>

Author: Jason A. Donenfeld
Author Date: 2019-02-04 03:38:28 UTC

noise: compute timestamps from a given base time

Author: Julian Orth
Author Date: 2018-09-08 11:45:57 UTC

tests: add test for transit-net

Author: Jason A. Donenfeld
Author Date: 2018-11-17 00:41:46 UTC

qemu: use arnd's toolchains

Author: Jason A. Donenfeld
Author Date: 2018-11-13 03:41:56 UTC

wg-quick: experiment with ifgroup suppression

Author: Jason A. Donenfeld
Author Date: 2018-11-11 21:06:56 UTC

chacha20: enable sse2 path

Author: Jason A. Donenfeld
Author Date: 2018-10-08 16:06:37 UTC

curve25519-arm: simply call setend

Probably not good if this runs on a v8 CPU.

Author: Jason A. Donenfeld
Author Date: 2018-09-23 16:12:43 UTC

Real life sort of benchmark

Author: Jonathan Neuschäfer
Author Date: 2018-07-21 01:34:46 UTC

hashtables: switch to rhashtable

NOTE: Due to a limitation in the rhashtable API, the siphash key (or
"seed") is reduced to 32 bits of random, rather than 128 bits that were
used before.

Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>

Author: Jason A. Donenfeld
Author Date: 2018-07-18 15:26:03 UTC

queueing: remove per-peer queues

Previously, having many peers would result in many napi structs, which
could make lookups in the napi_hash in net/core/dev.c slow. So, we move
to using a single napi struct per device.

The best solution would be to replace napi_hash with an idr or just get
rid of it all together and use straight pointers. However, that isn't the
case currently, so we work with what is and begrudgingly remove per-peer
queues. On the upside, it means we reduce the per-peer memory usage by
about 8k/16k, but on the downside it means that napi_gro_receive is
called on a unified list, which might result in less GRO speedups on
systems with many peers active at once.

However, if napi_hash does ever go away, we should consider reverting
this commit.

Since this means moving to unified packet queues, flushing at peer
removal is something of a problem. So we make the slightly dubious
modification of just not flushing, and letting our reference counters do
the work. This in turn required some small changes to ensure that the
reference counter will, at some point in the future, still reach zero,
and not be kept alive by non-stop packet ingress.

Co-developed-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>

Author: Thomas Gschwantner
Author Date: 2018-07-04 21:11:26 UTC

uint64_t and need_resched()

Author: Thomas Gschwantner
Author Date: 2018-07-01 03:28:38 UTC

mpmc_ptr_ring: add {,un}likely() annotations

Author: Gauvain "GovanifY" Roussel-Tarbouriech
Author Date: 2018-06-28 13:38:57 UTC

GRO does indeed concatenate packets, should we store them somewhere?

Author: Jason A. Donenfeld
Author Date: 2018-06-25 04:05:14 UTC

send: wait for suspend to complete before sending handshake

Otherwise the transmission might be dropped on Android devices.

Author: Jonathan Neuschäfer
Author Date: 2018-06-20 22:56:05 UTC

[BROKEN] mpmc_ptr_ring: Rely on null pointers to avoid p_tail

This should fix a performance problem when two (or more) producers run
on the same CPU.

[ TODO: more text ]

previously:

        Thread A Thread B
        producer_head=1
        write item

                <preemptive context switch>

                                        producer_head=2
                                        write item
                                        <stalled because producer_tail
                                         is still 0>

                <preemptive context switch>

        producer_tail=1

        producer_head=3
        write item
        <stalled because producer_tail is still 1>

                                ...

with this patch:

        Thread A Thread B
        producer_head=1
        write item

                <preemptive context switch>

                                        producer_head=2
                                        write item
                                        producer_head=3
                                        write item
                                        producer_head=4
                                        write item
                                        producer_head=5
                                        write item

                <preemptive context switch>

        producer_head=6
        write item
        producer_head=7
        write item

                                ...

Author: Jason A. Donenfeld
Author Date: 2018-06-18 13:10:14 UTC

selftest/mpmc_ring: lower test count for perf testing

Author: Jason A. Donenfeld
Author Date: 2018-06-18 01:14:20 UTC

mpmc_ptr_ring: nits

Author: Jason A. Donenfeld
Author Date: 2018-06-13 02:30:44 UTC

compat: backport ipvs_reset

Author: Thomas Gschwantner
Author Date: 2018-06-04 17:44:17 UTC

WIP9

Author: René van Dorst
Author Date: 2018-05-30 15:29:19 UTC

poly1305: generic: Reduce opaque size.

Signed-off-by: René van Dorst <opensource@vdorst.com>

Author: René van Dorst
Author Date: 2018-05-30 10:05:20 UTC

poly1305: Add a helper function which will padded the input to full poly1305 block when needed.
In the current code, every poly1305 input is a multiple of POLY1305_BLOCK_SIZE or will padded with zero's.
Put it in one function makes it more efficient.

Signed-off-by: René van Dorst <opensource@vdorst.com>

Author: Jason A. Donenfeld
Author Date: 2018-05-30 01:09:05 UTC

chacha20poly1305: combine stack variables into union

Author: Filippo Valsorda
Author Date: 2018-05-21 04:10:54 UTC

tools: fix OpenBSD build

License: MIT
Signed-off-by: Filippo Valsorda <valsorda@google.com>

Author: Jason A. Donenfeld
Author Date: 2018-05-21 17:31:40 UTC

wg-quick: linux: check for right route

Author: Jonathan Neuschäfer
Author Date: 2018-05-14 14:13:22 UTC

tests/qemu: Pin most of the tarball hashes

When src/tests/qemu/Makefile downloads tarballs, they may be corrupted
by accident or malice. Detect such errors by comparing the downloaded
files to their known-good hashes.

In the case of iperf there is actually a chance of a successful MitM
attack because the tarball is downloaded over plain HTTP if the mirror
(https://download.wireguard.com/...) isn't reachable for some reason.

Only the kernel tarball isn't checked, because the makefile needs to
work with many different kernel versions.

Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>

Author: Jason A. Donenfeld
Author Date: 2018-05-04 16:11:21 UTC

allowedips: single expression endian choice

This will upset millions of curious onlookers, and people will begin to
doubt my judgement and sanity. But this really is cleaner to express as
a single statement.

Author: Luis Ressel
Author Date: 2018-01-13 09:53:52 UTC

WIP: Automatically update PMTU for tunneled destinations

Author: Jason A. Donenfeld
Author Date: 2017-12-11 11:32:22 UTC

device: disable rp_filter for wireguard devices

Just like ipsec disables it with secpath, we can too, since allowedips
is really the right mechanism for this.

Author: Jason A. Donenfeld
Author Date: 2017-11-01 20:28:39 UTC

Allow disabling roaming

Author: Jason A. Donenfeld
Author Date: 2017-10-24 17:54:55 UTC

qemu: work around bogus QEMU virt ARM machines

The virt machine type in QEMU only supports one native serial port,
which means we're forced to add in a whole PCI bus if we want to get any
more information out. Bad situation.

This should be reverted as soon as upstream QEMU does something about
it.

Author: Jason A. Donenfeld
Author Date: 2017-09-15 21:58:38 UTC

data: move from spinlocks to lockless data structure

Reference: https://www.cs.rochester.edu/research/synchronization/pseudocode/queues.html
Original-code-from: @aegis

Author: Jason A. Donenfeld
Author Date: 2017-09-15 14:49:42 UTC

Try only advancing to next CPU if dql says so

Author: Jason A. Donenfeld
Author Date: 2017-09-15 13:25:22 UTC

DQL trial 1

Author: Jason A. Donenfeld
Author Date: 2017-08-03 21:45:26 UTC

ratelimiter: remove and use in-kernel functionality