We discovered two cases where edge connections can stall during testing:
1. Due to final data sitting in the edge inbuf when it was resumed
2. Due to flag synchronization between the token bucket and XON/XOFF
The first issue has always existed in C-Tor, but we were able to tickle it
in scp testing. If the last data from the protocol is able to fit in the
inbuf, but not large enough to send, if an XOFF or connection block comes in
at exactly that point, when the edge connection resumes, there will be no
data to read from the socket, but the inbuf can just sit there, never
draining.
We noticed the second issue along the way to finding the first. It seems
wrong, but it didn't seem to affect anything in practice.
These are extremely rare in normal operation, but with conflux, XON/XOFF
activity is more common, so we hit these.
Signed-off-by: David Goulet <dgoulet@torproject.org>
Streams can get blocked on a circuit in two ways:
1. When the circuit package window is full
2. When the channel's cell queue is too high
Conflux needs to decouple stream blocking from both of these conditions,
because streams can continue on another circuit, even if the primary circuit
is blocked for either of these cases.
However, both conflux and congestion control need to know if the channel's
cell queue hit the highwatermark and is still draining, because this condition
is used by those components, independent of stream state.
Therefore, this commit renames the 'streams_blocked_on_chan' variable to
signify that it refers to the cell queue state, and also refactors the actual
stream blocking bits out, so they can be handled separately if conflux is
present.
Using tor_free is wrong; event_free must be called for objects obtained from
event_new. Additionally, this slightly simplifies the code.
Also, add a static_assert to prevent further instances.
Regular relays are about to get their DirPort removed so that reachability
test is not useful anymore
Authorities will still use the DirPort but because network reentry towards
their DirPort is now denied network wide, this test is not useful anymore and
so it should simply be considered reachable at all time.
Part of #40282
Signed-off-by: David Goulet <dgoulet@torproject.org>
This is unfortunately massive but both functionalities were extremely
intertwined and it would have required us to actually change the HSv2 code in
order to be able to split this into multiple commits.
After this commit, there are still artefacts of v2 in the code but there is no
more support for service, intro point and HSDir.
The v2 support for rendezvous circuit is still available since that code is
the same for the v3 and we will leave it in so if a client is able to
rendezvous on v2 then it can still transfer traffic. Once the entire network
has moved away from v2, we can remove v2 rendezvous point support.
Related to #40266
Signed-off-by: David Goulet <dgoulet@torproject.org>
Some days before this commit, the network experienced a DDoS on the directory
authorities that prevented them to generate a consensus for more than 5 hours
straight.
That in turn entirely disabled onion service v3, client and service side, due
to the subsystem requiring a live consensus to function properly.
We know require a reasonably live consensus which means that the HSv3
subsystem will to its job for using the best consensus tor can find. If the
entire network is using an old consensus, than this should be alright.
If the service happens to use a live consensus while a client is not, it
should still work because the client will use the current SRV it sees which
might be the previous SRV for the service for which it still publish
descriptors for.
If the service is using an old one and somehow can't get a new one while
clients are on a new one, then reachability issues might arise. However, this
is a situation we already have at the moment since the service will simply not
work if it doesn't have a live consensus while a client has one.
Fixes#40237
Signed-off-by: David Goulet <dgoulet@torproject.org>
Typos found with codespell.
Please keep in mind that this should have impact on actual code
and must be carefully evaluated:
src/core/or/lttng_circuit.inc
- ctf_enum_value("CONTROLER", CIRCUIT_PURPOSE_CONTROLLER)
+ ctf_enum_value("CONTROLLER", CIRCUIT_PURPOSE_CONTROLLER)
Simplify ip_address_changed() by removing redundant checks
(all exits are relays, and need to rebuild their descriptor when their
IP addresses change).
Rewrite the function comment, and rename the argument, so the function
is easier to understand and use.
Closes 33091.
This is an automated commit, generated by this command:
./scripts/maint/rename_c_identifier.py \
directory_must_use_begindir dirclient_must_use_begindir \
directory_fetches_from_authorities dirclient_fetches_from_authorities \
directory_fetches_dir_info_early dirclient_fetches_dir_info_early \
directory_fetches_dir_info_later dirclient_fetches_dir_info_later \
directory_too_idle_to_fetch_descriptors dirclient_too_idle_to_fetch_descriptors
Doing so caused us to crash in some unusual circumstances, such as
using --verify-config to verify a configuration that failed during
the options_act() stage.
Fixes bug 32407; bugfix on 0.3.3.1-alpha.
Conflicts:
src/feature/dirparse/authcert_parse.c
src/feature/dirparse/ns_parse.c
src/feature/hs/hs_service.c
src/lib/conf/conftesting.h
src/lib/log/log.h
src/lib/thread/threads.h
src/test/test_options.c
These conflicts were mostly related to autostyle improvements, with
one or two due to doxygen fixes.
Relays now respect their AccountingMax bandwidth again. When relays
entered "soft" hibernation (which typically starts when we've hit
90% of our AccountingMax), we had stopped checking whether we should
enter hard hibernation. Soft hibernation refuses new connections and
new circuits, but the existing circuits can continue, meaning that
relays could have exceeded their configured AccountingMax.
This commit rolls back some of the cpu-saving fixes, where we tried
to avoid calling so many of our events while we're off the network.
That's because PERIODIC_EVENT_FLAG_NEED_NET checks net_is_disabled(),
which returns true even if we're only in soft hibernation.
Fixes bug 32108; bugfix on 0.4.0.1-alpha.
Code adapted from Rob's proposed patch in #30344.
Also add a comment in connection_mark_for_close_internal_() on why we should
not be adding extra code there without a very good reason.
David Goulet
Mike Perry
Gabriela Moldovan
Alex Xu (Hello71)
George Kadianakis
Nick Mathewson
Neel Chauhan
Samanta Navarro
teor
Roger Dingledine