Because we handle messages (which can take some time, persisting
things to disk or validating cryptographic signatures) with the
top-level read lock, but require the top-level write lock to
connect new peers or handle disconnection, we are particularly
sensitive to writer starvation issues.
Rust's libstd RwLock does not provide any fairness guarantees,
using whatever the OS provides as-is. On Linux, pthreads defaults
to starving writers, which Rust's RwLock exposes to us (without
any configurability).
Here we work around that issue by blocking readers if there are
pending writers, optimizing for readable code over
perfectly-optimized blocking.
This avoids any extra calls to `read_event` after a write fails to
flush the write buffer fully, as is required by the PeerManager
API (though it isn't critical).
Only one instance of PeerManager::process_events can run at a time,
and each run always finishes all available work before returning.
Thus, having several threads blocked on the process_events lock
doesn't accomplish anything but blocking more threads.
Here we limit the number of blocked calls on process_events to two
- one processing events and one blocked at the top which will
process all available events after the first completes.
Because the peers write lock "blocks the world", and happens after
each read event, always taking the write lock has pretty severe
impacts on parallelism. Instead, here, we only take the global
write lock if we have to disconnect a peer.
Unlike very ancient versions of lightning-net-tokio, this does not
rely on a single global process_events future, but instead has one
per connection. This could still cause significant contention, so
we'll ensure only two process_events calls can exist at once in
the next few commits.
Users are required to only ever call `read_event` serially
per-peer, thus we actually don't need any locks while we're
processing messages - we can only be processing messages in one
thread per-peer.
That said, we do need to ensure that another thread doesn't
disconnect the peer we're processing messages for, as that could
result in a peer_disconencted call while we're processing a
message for the same peer - somewhat nonsensical.
This significantly improves parallelism especially during gossip
processing as it avoids waiting on the entire set of individual
peer locks to forward a gossip message while several other threads
are validating gossip messages with their individual peer locks
held.
This adds the required locking to process messages from different
peers simultaneously in `PeerManager`. Note that channel messages
are still processed under a global lock in `ChannelManager`, and
most work is still processed under a global lock in gossip message
handling, but parallelizing message deserialization and message
decryption is somewhat helpful.
Use the `counterparty_max_htlc_value_in_flight_msat` value, and not the
`holder_max_htlc_value_in_flight_msat` value when creating the
`htlc_maximum_msat` value for `ChannelUpdate` messages.
BOLT 7 specifies that the field MUST be less than or equal to
`max_htlc_value_in_flight_msat` received from the peer, which we
currently are not guaranteed to adhere to by using the holder value.
Add a config field
`ChannelHandshakeConfig::max_inbound_htlc_value_in_flight_percent_of_channel`
which sets the percentage of the channel value we cap the total value of
outstanding inbound HTLCs to.
This field can be set to a value between 1-100, where the value
corresponds to the percent of the channel value in whole percentages.
Note that:
* If configured to another value than the default value 10, any new
channels created with the non default value will cause versions of LDK
prior to 0.0.104 to refuse to read the `ChannelManager`.
* This caps the total value for inbound HTLCs in-flight only, and
there's currently no way to configure the cap for the total value of
outbound HTLCs in-flight.
* The requirements for your node being online to ensure the safety of
HTLC-encumbered funds are different from the non-HTLC-encumbered funds.
This makes this an important knob to restrict exposure to loss due to
being offline for too long. See
`ChannelHandshakeConfig::our_to_self_delay` and
`ChannelConfig::cltv_expiry_delta` for more information.
Default value: 10.
Minimum value: 1, any values less than 1 will be treated as 1 instead.
Maximum value: 100, any values larger than 100 will be treated as 100
instead.
We only use it to check the amount when processing MPP parts, but
store the full object (including new payment metadata) in it.
Because we now store the amount in the parent structure, there is
no need for it at all in the `OnionPayload`. Sadly, for
serialization compatibility, we need it to continue to exist, at
least temporarily, but we can avoid populating the new fields in
that case.
...instead of accessing it via the `OnionPayload::Invoice` form.
This may be useful if we add MPP keysend support, but is directly
useful to allow us to drop `FinalOnionHopData` from `OnionPayload`.
The `full_stack_target` fuzzer managed to find a subtraction
underflow in the new `Channel::get_htlc_maximum` function where we
subtract both sides' reserve values from the channel funding. Such
a channel is obviously completely useless, so we should reject it
during opening instead of integer-underflowing later.
Thanks to Chaincode Labs for providing the fuzzing resources which
found this bug!
When we receive multiple HTLCs which claim to be a part of the same
MPP but which are inconsistent for some reason, we should fail the
inconsistent HTLCs but keep the first HTLCs up until the first
inconsistency.
This works, but it turns out there was no test coverage, so we add
some here.
The `chain::Listen` interface provides a block-connection-based
alternative to the `chain::Confirm` interface, which supports
providing transaction data at a time separate from the block
connection time.
For users who are downloading the full headers tree (e.g. from a
node over the Bitcoin P2P protocol) but who are not downloading
full blocks (e.g. because they're using BIP 157/158 filtering)
there is no API that matches exactly their event stream -
`chain::Listen` requries full blocks for each block,
`chain::Confirm` requires breaking each connection event into two
calls.
Given its incredibly trivial to take a `TransactionData` in
addition to a `Block` in `chain::Listen` we do so here, adding a
default-implementation `block_connected` which simply creates the
`TransactionData`, which ultimately all of the `chain::Listen`
implementations currently do anyway.
Closes#1128.
The main loop of the background processor has this line:
`peer_manager.process_events(); // Note that this may block on ChannelManager's locking`
which does, indeed, sometimes block waiting on the `ChannelManager`
to finish whatever its doing. Specifically, its the only place in
the background processor loop that we block waiting on the
`ChannelManager`, so if the `ChannelManager` is relatively busy, we
may end up being blocked there most of the time.
This should be fine, except today we had a user who's node was
particularly slow in processing some channel updates, resulting in
the background processor being blocked there (as expected). Then,
when the channel updates were completed (and persisted) the next
thing the background processor did was hand the user events to
process, creating yet more channel updates. Ultimately, the users'
node crashed before finishing the event processing. This left us
with an updated monitor on disk and an outdated manager, and they
lost the channel on startup.
Here we simply move the above quoted line to after the normal event
processing, ensuring the next thing we do after blocking on
`ChannelManager` locks is persist the manager, prior to event
handling.
MAX_FUNDING_SATOSHIS will no longer be accurately named once wumbo is merged.
Also, we'll want to check that wumbo channels don't exceed the total bitcoin supply
`ChannelDetails::outbound_capacity_msat` describes the total amount
available for sending across several HTLCs, basically just our
balance minus the reserve value maintained by our counterparty.
However, when routing we use it to guess the maximum amount we can
send in a single additional HTLC, which it is not.
There are numerous reasons why our balance may not match the amount
we can send in a single HTLC, whether the HTLC in-flight limit, the
channe's HTLC maximum, or our feerate buffer.
This commit splits the `outbound_capacity_msat` field into two -
`outbound_capacity_msat` and `outbound_htlc_limit_msat`, setting us
up for correctly handling our next-HTLC-limit in the future.
This also addresses the first of the reasons why the values may
not match - the max-in-flight limit. The inaccuracy is ultimately
tracked as #1126.
