This resolves a bug introduced in
66aa1d58a1 where, if when responding
to a series of transaction requests in a getdata we hit the send
buffer limit and set fPauseSend, we will skip one transaction per
call to ProcessGetData.
Bug found by Cory Fields (@theuni).
Github-Pull: #12392
Rebased-From: c4af7387634765d254d1432746385cf35917d367
Tree-SHA512: d2f7707eb9f925a655f66e5e77ce406c5266f7b2feccd5bcdabf6d5bc27a3f6578e753fac83d9c8c3fd7cf7de6fee086eee2f95f77af99ea2c4e5ae77c322c58
Static analyzer (and humans!) will see ...
```
else if (state.m_chain_sync.m_timeout == 0 || (state.m_chain_sync.m_work_header != nullptr && ...
```
... and infer that state.m_chain_sync.m_work_header might be set to nullptr,
and thus flag `state.m_chain_sync.m_work_header->GetBlockHash().ToString()`
as a potential null pointer dereference.
This commit makes the tacit assumption (m_work_header != nullptr) explicit.
Code introduced in 5a6d00 ("Permit disconnection of outbound peers on
bad/slow chains") which was merged into master four days ago.
We should generally avoid writing to debug.log unconditionally for
inbound peers which misbehave (the peer being about to be banned
being an exception, since they cannot do this twice).
To avoid removing logs for outbound peers, a new log is added to
notify users when a new outbound peer is connected which mimics
the version print.
If our tip hasn't updated in a while, that may be because our peers are
not relaying blocks to us that we would consider valid. Allow connection
to an additional outbound peer in that circumstance.
Also, periodically check to see if we are exceeding our target number of
outbound peers, and disconnect the one which has least recently
announced a new block to us (choosing the newest such peer in the case
of tie).
There is no reason to wish to store blocks on disk always just
because a peer is whitelisted. This appears to be a historical
quirk to avoid breaking things when the accept limits were added.
Nowhere else in the protocol do we send headers which are for
blocks we have not fully validated except in response to getheaders
messages with a null locator. On my public node I have not seen any
such request (whether for an invalid block or not) in at least two
years of debug.log output, indicating that this should have minimal
impact.
Reading the variable mapBlockIndex requires holding the mutex cs_main.
The new "Disconnect outbound peers relaying invalid headers" code
added in commit 37886d5e2f and merged
as part of #11568 two days ago did not lock cs_main prior to accessing
mapBlockIndex.
Currently we have no rotation of outbound peers. If an outbound peer
stops serving us blocks, or is on a consensus-incompatible chain with
less work than our tip (but otherwise valid headers), then we will never
disconnect that peer, even though that peer is using one of our 8
outbound connection slots. Because we rely on our outbound peers to
find an honest node in order to reach consensus, allowing an
incompatible peer to occupy one of those slots is undesirable,
particularly if it is possible for all such slots to be occupied by such
peers.
Protect against this by always checking to see if a peer's best known
block has less work than our tip, and if so, set a 20 minute timeout --
if the peer is still not known to have caught up to a chain with as much
work as ours after 20 minutes, then send a single getheaders message,
wait 2 more minutes, and if a better header hasn't been received by then,
disconnect that peer.
Note:
- we do not require that our peer sync to the same tip as ours, just an
equal or greater work tip. (Doing otherwise would risk partitioning the
network in the event of a chain split, and is also unnecessary.)
- we pick 4 of our outbound peers and do not subject them to this logic,
to be more conservative. We don't wish to permit temporary network
issues (or an attacker) to excessively disrupt network topology.
Adds HasAllRelevantServices and GetRelevantServices, which check
for NETWORK|WITNESS.
This changes the following:
* Removes nRelevantServices from CConnman, disconnecting it a bit
more from protocol-level logic.
* Replaces our sometimes-connect-to-!WITNESS-nodes logic with
simply always requiring WITNESS|NETWORK for outbound non-feeler
connections (feelers still only require NETWORK).
* This has the added benefit of removing nServicesExpected from
CNode - instead letting net_processing's VERSION message
handling simply check HasAllRelevantServices.
* This implies we believe WITNESS nodes to continue to be a
significant majority of nodes on the network, but also because
we cannot sync properly from !WITNESS nodes, it is strange to
continue using our valuable outbound slots on them.
* In order to prevent this change from preventing connection to
-connect= nodes which have !WITNESS, -connect nodes are now
given the "addnode" flag. This also allows outbound connections
to !NODE_NETWORK nodes for -connect nodes (which was already true
of addnodes).
* Has the (somewhat unintended) consequence of changing one of the
eviction metrics from the same
sometimes-connect-to-!WITNESS-nodes metric to requiring
HasRelevantServices.
This should make NODE_NETWORK_LIMITED much simpler to implement.
Sending a getheaders message with an empty locator and a stop hash
is a request for a single header by hash. The node will respond with
headers for blocks not in the main chain as well as those in the main
chain. To avoid fingerprinting, the node should, however, ignore
requests for headers on side branches that are too old.
Combine fLimitFree and fOverrideMempoolLimit into a single boolean:
bypass_limits. This is used to indicate that mempool limiting based on feerate
should be bypassed. It is used when readding transactions from a reorg and then
the mempool is trimmed to size after all transactions are added and they can be
evaluated in the context of their descendants. No changes to behavior.
There are a few too many edge-cases here to make this a scripted diff.
The following commits will move a few functions into PeerLogicValidation, where
the local connman instance can be used. This change prepares for that usage.
At startup, we choose one peer to serve us the headers chain, until
our best header is close to caught up. Disconnect this peer if more
than 15 minutes + 1ms/expected_header passes and our best header
is still more than 1 day away from current time.
This patch makes several related changes:
* Changes the CCoinsView virtual methods (GetCoins, HaveCoins, ...)
to be COutPoint/Coin-based rather than txid/CCoins-based.
* Changes the chainstate db to a new incompatible format that is also
COutPoint/Coin based.
* Implements reconstruction code for hash_serialized_2.
* Adapts the coins_tests unit tests (thanks to Russell Yanofsky).
A side effect of the new CCoinsView model is that we can no longer
use the (unreliable) test for transaction outputs in the UTXO set
to determine whether we already have a particular transaction.
Matt Corallo