this allows you to use exit nodes without forcing routes over the interface, useful for using lokinet with an exit and selectively routing over the lokinet interface using an external socks proxy or binding to device explicitly.
* make route poker configurable, defaults to enabled but allows disabling it on runtime if desired
* add config option [network]:auto-routing to enable/disable route poker
on win32/apple reading packets from the interface does not count as an io operation.
manually trigger pump on win32/apple to pretend that it is an io event.
add platform quark function MaybeWakeUpperLayers on vpn::Interface to manaully wake up the other components on platforms that need that (ones on which packet io is not done via io events).
on non linux platforms, use uv_prepare_t instead of uv_check_t as the former triggers before blocking for io, instead of after. this better matches linux's order of operations in libuv.
We are calling time_now() a huge amount, and it is a major consumer of
CPU cycles, but we don't need it: most of the time the current event
loop time is enough.
We have a few cases where we're making an extra shared_ptr which we copy
into a lambda, which then results in an extra unnecessary refcount
decrement in the parent; this changes them to give an rvalue reference
to the lambda to avoid the extra incr/decr instead.
The one in Session::Pump is particularly noticeable and shows up in
profiling.
- Replace m_FlushWakeup with a call to the router's god mode pump
method. m_FlushWakeup apparently isn't enough to get things out, and
we can end up with incoming packets that don't get properly handled
right away without it.
- The shared_ptr around the ihophandler queues isn't needed and is just
adding a layer of obfuscation; instead just exchange the list directly
into the lambda.
- Use std::exchange rather than swap
- A couple other small code cleanups.
The TriggerPump just below this is *already* going to trigger a flush,
so the extra flush call here can't do anything useful (and in
particular, it won't clear up the queue *immediately*, which is what
this code looks like it was aimed at doing).
- Make the main PumpLL also pump hidden services, rather than using
separate wakers in each TunEndpoint. It seems there is some
interactions that just one or the other is not enough.
- Eliminate TunEndpoint send queue -- it isn't needed as we can just
send directly.
If something needs to wake up the event loop it should be using an
async, as we are now with PumpLL(); but we had various code triggering a
wakeup, expecting that PumpLL gets called on every wakeup, which isn't
true anymore.
We trigger a pump immediately, but this is racey because we add to our
plaintext data in a worker thread; if the worker thread runs after the
pump then it ends up leaving plaintext to be handled, but there's no
wakeup until the next one.
This was the cause of seeing a random +1s and bunching added to ping
responses sometimes: it wasn't until the *next* ping goes through the
network that the plaintext queue gets processed, at which point it
flushes the old one and often the new one together.
The fix here gets rid of the map of sessions needing wakeups and instead
adds an atomic flag to all of them to let us figure out which ones
need to be flushed.
call_soon wakes up the mainloop and is often reentrant, we dont want to
ever way up the event loop but we also want to always defer the call so
we always use the workers
* attempt path timeout bullshittery fix
* make sure ServiceInfo always has its address set up
* do not copy intros in constuctor, ammend logging and
add assert
* use std::source_location instead of godawful macros in logging
* remove unused/absolutely haram af json logstream
* fix bug in android logger where it doesn't respect eLogNone
before when we get the list of router ids for gossip it was highly bias towards outbound sessions.
instead now we get a full list of link session router ids in random order, truncate them to be at most MaxGossipPeers number of keys, and then put them into an unordered set
Removes stuff we didn't end up needing/using:
- Lokinet.modulemap
- apple bits from lokinet.cpp (we don't use lokinet.cpp at all on macos
anymore).
- dnsproxy/extension C++ headers
- apple-specific network extension config in llarp::config::Config
When we enable/disable exit mode on this restarts the unbound DNS
responder with the DNS trampoline (or restores upstream, when disabling)
to properly route DNS requests through the tunnel (because libunbound's
direct requests don't get tunneled because unbound is inside the network
extension).
This runs a DNS listener on localhost:1053 that bounces requests to the
upstream DNS through the tunnel. The idea here is that, when we turn on
exit mode, we start libunbound bouncing the requests through the
trampoline (since if it makes direct requests they won't go through the
tunnel).
(The actual libunbound configuration is still to follow).
Thus when a user goes looking for it they'll find the (commented out)
default in the right place and can edit it.
(That right place is: ~/Library/Containers/com.loki-project.lokinet.network-extension/Data/lokinet.ini)
Don't squash this commit so that the swift version stays around in
history in case we need to resurrect it again some day (i.e. when Apple
decides to kill off Objective-C support).
- Add a C callback interface (context_wrapper.h) between lokinet and the
objective-C code so that:
- we can use objective-C (rather than objective-C++), which seems more
likely to be supported by Apple into the future;
- we minimize the amount of code that needs to be aware of the Apple
APIs.
- this replaces apple logger objective c++ implementation with a plain
c++ implementation that takes a very simple C callback (provided
from the obj-c code) to actually make the call to NSLog.
- Add various documentation to the code of what is going on.
- Send all DNS traffic to the primary IP on the tun interface. The
match prefixes simply don't work as advertised, and have weird shit
(like even if you get it working for some domains, "instagram.com"
still doesn't because of god-knows-what Apple internal politics).
- Drop the dns proxy code as we don't need it anymore.
- Don't use 9.9.9.9 for default DNS. (We might consider the unfiltered
9.9.9.10 as an alternative default, but if we do it should be a global
lokinet change rather than a Mac-specific change).
- Parse a lokinet.ini in the data directory, if it exists. (Since we
are sandboxed, it is an app-specific "home" directory so is probably
buried god knows where, but at least the GUI ought to be able to get
it to let users add things to it).
- This commit also adds a swift version of the PacketTunnelProvider
glue, which ought to work in theory, but the *tooling* for cmake is so
underdeveloped that I couldn't find any way to actually get the damn
thing working. So I'm committing it here anyway (and will revert it
away in the next commit) in case we someday want to switch to it.
-
- Added contrib/macos/README.txt with description of the cancer
happening here.
- Add provisioningprofiles that Apple wants to make things work properly
- Made the entitlements files match the provisioningprofiles
- Remove configured entitlements files; we *can't* change any of the
things here because they are closedly tied to the provisioningprofiles
-- which means if someone wants to build their own Lokinet, they have
to replace a bunch of crap and change application IDs throughout.
This is the hostile-to-open-source Apple way.
- Remove unused old lokinet binary, as we're no longer using it on macos
- Use a POST_BUILD rather than install to copy things around into the
right places
- Convert all the configure_file's to consistently use @ONLY
- Misc cleanups
PR #1725 reversed argument orders but UnboundResolver was still using
(from,to) ordering in its callbacks, which leaked through to make a
wrong order in our reply function (which simply forwards arguments).
This fixes that bug by making UnboundResolver callback argument order
consistent (i.e. using to, from) with the PacketHandler argument order.
The reason the dns fix on android didn't work is that the DnsInterceptor
had a reversed to/from argument order for its
`SendServerMessageBufferTo` overload, and so android/mac needed the
to/from to be reversed so that the second reverse cancelled out the
first one.
Upon review, the DnsInterceptor order (to, from) is more intuitive than
the base order (from, to), so this reapplies the dns fix and swaps
everything *except* DnsInterceptor to match the (to, from) argument
order.
This reverts commit dace0224ec.
This reportedly didn't fix things on Android, and most definitely breaks
macOS (with this we get a bunch of errors about expecting inbound when
we have outbound).
- Reduce buffer size to INET6_ADDRSTRLEN, and use a single buf rather
than two identical ones in each branch.
- Don't pre-reserve because doing so is usually going to over-allocate,
but also because it prevents SSO, especially for the IPv4 case which
should fit in SSO for all IPv4 addresses.
The default upstream DNS was being set to 1.1.1.1:0, which doesn't work.
This fixes it to also set the port so that default upstream resolution
(i.e. with an empty config) works again.
This function had a bug in stable (fixed in dev) when `last` returns
npos, but the function also appears to basically be duplicating what the
next split version can do, so this just removes it and uses the single
more generic split(strview, strview) method.
* make it so that we don't set up unbound resolver when we have no resolvers provided by config
* clean up dns codepath and make it use llarp::SockAddr instead of llarp::IpAddress
* add option to persist address mappings between restarts using [network]:persist-addrmap-file
* make it work
* only persist address map for inbound convos
* turn persisting address map on by default
* dont load addrmap file if it has been modified last over a minute ago to prevent foot cannons fired from loading a really old version of it
* when a path build times out, shitlist every router in the path except the first hop, this way eventually we get the nodedb pruned to only the routers that are currently actually alive, any ones we nuke that we need later we can always do lookups for.
* introset path haodver tweaks
* improve warn/error messages to convey more information
* dont block on queue insertion
* reset convotag on decrypt/verify fail
* add multiple ready hooks on outbound context
* lookup introsets from close routers on dht
* continue to tick dead sessions so they expire their paths
* introset spacing
* reduce lns lookup diversity requirement for speed
* add a function to send reset convotag message
* only have 1 outbound context at a time
* include stricter router profiling checks in path::Builder hop slection algorithm
* make intro selection function nicer by returning a std::optional instead of a bool with an "out" variable
* do FEC for latency tests so if we fail one test it doesn't kill the entire path
* ignore FEC'd responses on latency tests
* track latency history and report the mean latency instead of just the last sample
* add remove_node_from_failing to remove a node by pubkey from the failing set
* if a router is deregistered we remove it from the failing set so we don't retest it
* remove a router from the failing set if we get a test success
currently creating an outbound session will cancel if we have any session
at all with the relay. instead, only cancel if we have an outbound session
to that relay. this is useful for reachability testing.
It's there for polling, which we aren't doing anymore; we just got the
hash from oxend's push notification, so if it pass it in then we will
always get an "unchanged" result because we're telling oxend that we
already have the data for that hash updated.
This just drops the hash completely because we don't need it anymore.
* immediately poke routes when we are told to use an exit so that packets get pushed which makes an exit path happen
* fix up cmake oddity in nsis section
* not gossip our rc
* not explore the network to prevent outbound session attempts
* not establish sessions to other service nodes
* close all open sessions we have to tell clients we don't want them
* catch exceptions flushing peerdb in disk thread
* don't connect out to non allowed routers
* simplify logic in RCLookupHandler::RemoteIsAllowed()
* add HaveReceivedWhitelist to I_RCLookupHandler base type
* add LooksDeregistered to Router type that tells us if we think we are deregistered
* don't allow building paths over us if we are deregistered
so it doesn't time out and get into a state that's totally screwed.
add virtual function service::Endpont::DefaultPathAlignmentTimeout() to get the timeout for path alignment
and use it for resetablishing outbound sessions
* increase publish introset timeout so that it does not time out on the network
* remove pedantic log warn
* make sure the path we are using for replying on inbound sessions is alive
* include convotag in log message so we know wtf is going on
* appease tom's autism, improve log message text
adds [network] section parameter called path-alignment-timeout that allows configring the timeout
for optional name lookup + introset lookup + aligned path build, used by tun endpoint dns, provided
as milliseconds.
Fixes a subtle memory leak that was a result of outbound messages which
were in the shared queue (not yet sorted into a per-path queue) when a
path was removed, resulting in a ghost path queue (and thus round-robin
order entry as well).
Adds much needed documentation to the outbound message handler class.
Wires up systemd support to configure DNS on startup and when
enabling/disabling exit mode.
On startup (and when turning off an exit) we tell systemd-resolved to
direct .loki and .snode lookups to lokinet (leaving other DNS traffic
alone).
On exit enabling, we reconfigure it to resolve "." (i.e. the root DNS
domain) so that all lookups come into it.
