92617b7a75 Make AddrMan support multiple ports per IP (Pieter Wuille)
Pull request description:
For a long part of Bitcoin's history, this codebase has aggressively avoided making automatic connections to anything but nodes running on port 8333. I'd like to propose changing that, and this is a first PR necessary for that.
The folklore justification (eventually actually added as a comment to the codebase in #20668) is that this is to prevent the Bitcoin P2P network from being leveraged to perform a DoS attack on other services, if their IP/port would get rumoured. It appears, at least the current network scale - and probably significantly larger - that the impact is very low at best (see calculations by vasild in https://github.com/bitcoin/bitcoin/issues/5150#issuecomment-853888909 e.g.). Another possible justification would be a risk that treating different IP:port combinations separately would help perform Eclipse attacks (by an attacker rumouring their own IP with many ports). This concern is (a) no different than what is possible with IPv6 (where large ranges of IP addresses are very cheaply available), and (b) already hopefully sufficiently addressed by addrman's design (which limits access through based selected based on network groups).
And this policy has downsides too; in particular, a fixed port is easy to detect, and a very obvious sign a Bitcoin node is running there.
One obstacle in moving away from a default port that is the fact that addrman is currently restricted to a single entry per IP address. If ports are no longer expected to be generally always the default one, we need to deal with the case where conflicting information is relayed. It turns out there is a very natural solution to this: treat (IP,port) combination exactly as we're treating IPs now; this automatically means that the same IP may appear with multiple ports, simply because those would be distinct entries. Given that indexing into addrman's bucket _already_ uses the port number, the only change required is making all addrman lookup be (IP,port) (aka `CService`) based, rather than IP (aka `CNetAddr`) based.
This PR doesn't include any change to the actual outbound connection preference logic, as perhaps that's something that we want to phase in more gradually.
ACKs for top commit:
jnewbery:
Code review ACK 92617b7a75
naumenkogs:
ACK 92617b7a75
ajtowns:
ACK 92617b7a75
vasild:
ACK 92617b7a75
Tree-SHA512: 9eef06ce97a8b54a3f05fb8acf6941f253a9a5e0be8ce383dd05c44bb567cea243b74ee5667178e7497f6df2db93adab97ac66edbc37c883fd8ec840ee69a33f
632aad9e6d Make CAddrman::Select_ select buckets, not positions, first (Pieter Wuille)
Pull request description:
The original CAddrMan behaviour (before #5941) was to pick a uniformly random non-empty bucket, and then pick a random element from that bucket. That commit, which introduced deterministic placement of entries in buckets, changed this to picking a uniformly random non-empty bucket position instead.
I believe that was a mistake. Buckets are our best metric for spreading out requests across independently-controlled nodes. That
does mean that if a bucket has fewer entries, its entries are supposed to be picked more frequently.
This PR reverts to the original high-level behavior, but on top of the deterministic placement logic.
ACKs for top commit:
jnewbery:
utACK 632aad9e6d
naumenkogs:
ACK 632aad9e6d
mzumsande:
ACK 632aad9e6d
Tree-SHA512: 60768afba2b6f0abd0dddff04381cab5acf374df48fc0e883ee16dde7cf7fd33056a04b573cff24a1b4d8e2a645bf0f0b3689eec84da4ff330e7b59ef142eca1
The original CAddrMan behaviour (before commit
e6b343d880) was to pick a uniformly
random non-empty bucket, and then pick a random element from that
bucket. That commit, which introduced deterministic placement
of entries in buckets, changed this to picking a uniformly random
non-empty bucket position instead.
This commit reverts the original high-level behavior, but in the
deterministic placement model.
CAddrInfo objects are an implementation detail of how AddrMan manages and adds
metadata to different records. Encapsulate this logic by updating Select &
SelectTriedCollision to return the additional info that the callers need.
Introduce the pimpl pattern for CAddrMan to separate the implementation details
from the externally used object representation. This reduces compile-time
dependencies and conceptually clarifies AddrMan's interface from the
implementation specifics.
Since the unit & fuzz tests currently rely on accessing CAddrMan internals, this
commit introduces addrman_impl.h, which is exclusively imported by addrman.cpp
and test files.
Review hint: git diff --color-moved=dimmed-zebra
--color-moved-ws=ignore-all-space
Also move `Check` and `ForceCheckAddrman` to be after the `FunctionName_` functions.
Review hint: use git diff --color-moved=dimmed-zebra
--color-moved-ws=ignore-all-space
In preparation for introducing the pimpl pattern to addrman, move all function
bodies out of the header file.
Review hint: use git diff --color-moved=dimmed-zebra
--color-moved-ws=ignore-all-space
This speeds up compilation of the whole program because the included
header file is smaller.
Can be reviewed with --color-moved=dimmed-zebra --color-moved-ws=ignore-all-space
This is done by removing an unnecessary loop in Good_() and looping
through the new tables in MakeTried() more efficiently, choosing a
starting value that allow us to stop early in typical cases.
Co-authored-by: John Newbery <john@johnnewbery.com>
853c4edb70 [net] Remove asmap argument from CNode::CopyStats() (John Newbery)
9fd5618610 [asmap] Make DecodeAsmap() a utility function (John Newbery)
bfdf4ef334 [asmap] Remove SanityCheckASMap() from netaddress (John Newbery)
07a9eccb60 [net] Remove CConnman::Options.m_asmap (John Newbery)
Pull request description:
These small cleanups to the asmap code are the first 4 commits from #22910. They're minor improvements that are independently useful whether or not 22910 is merged.
ACKs for top commit:
naumenkogs:
ACK 853c4edb70
theStack:
Concept and code-review ACK 853c4edb70🗺️
fanquake:
ACK 853c4edb70
Tree-SHA512: 64783743182592ac165df6ff8d18870b63861e9204ed722c207fca6938687aac43232a5ac4d8228cf8b92130ab0349de1b410a2467bb5a9d60dd9a7221b3b85b
fade9a1a4d Remove confusing CAddrDB (MarcoFalke)
fa7f77b7d1 Fix addrdb includes (MarcoFalke)
fa3f5d0dae Move addrman includes from .h to .cpp (MarcoFalke)
Pull request description:
Split out from #22762 to avoid having to carry it around in (an)other rebase(s)
ACKs for top commit:
ryanofsky:
Code review ACK fade9a1a4d
lsilva01:
Code Review ACK fade9a1a4d
Tree-SHA512: 7615fb0b6235d0c1e6f8cd6263dd18c4d95890567c2b797fe6fce6cb12cc85ce6eacbe07dbb6d81b05d179ef03b42edfd61c940e35a1044ce6d363b54c2dae5c
fab0b55cf0 addrman: Fix format string in deserialize error (MarcoFalke)
facce4ca44 test: Remove useless overwrite (MarcoFalke)
Pull request description:
The format string is evaluated differently on modern compilers (clang 10 and later, as well as gcc 10 and later).
Work around the behaviour change in compilers by pinning the underlying type of the format arguments.
Can be tested by observing a failing test when running against master compiled with clang 10 or gcc 10 (or later).
ACKs for top commit:
jonatack:
ACK fab0b55cf0 verified the test fails on master as expected only at line 61 (assertion fixed by the code change); the last two test additions pass as expected
mzumsande:
ACK fab0b55cf0
Tree-SHA512: 07462901435107f3bc79098fd7d06446bfe8fe065fffdd35adfcba8f1dd3c499575006557afe7bc74b79d690c5ef7b58e3e031e908161be5529cf237e3b30609
DecopeAsmap is a pure utility function and doesn't have any
dependencies on addrman, so move it to util/asmap.
Reviewer hint: use:
`git diff --color-moved=dimmed-zebra --color-moved-ws=ignore-all-space`
724c497562 [fuzz] Add ConsumeAsmap() function (John Newbery)
5840476714 [addrman] Make m_asmap private (John Newbery)
f9002cb5db [net] Rename the copyStats arg from m_asmap to asmap (John Newbery)
f572f2b204 [addrman] Set m_asmap in CAddrMan initializer list (John Newbery)
593247872d [net] Remove CConnMan::SetAsmap() (John Newbery)
50fd77045e [init] Read/decode asmap before constructing addrman (John Newbery)
Pull request description:
Commit 181a1207 introduced an initialization order bug: CAddrMan's m_asmap must be set before deserializing peers.dat.
The first commit restores the correct initialization order. The remaining commits make `CAddrMan::m_asmap` usage safer:
- don't reach into `CAddrMan`'s internal data from `CConnMan`
- set `m_asmap` in the initializer list and make it const
- make `m_asmap` private, and access it (as a reference to const) from a getter.
This ensures that peers.dat deserialization must happen after setting m_asmap, since m_asmap is set during CAddrMan construction.
ACKs for top commit:
mzumsande:
Tested ACK 724c497562
amitiuttarwar:
code review but utACK 724c497562
naumenkogs:
utACK 724c497562
vasild:
ACK 724c497562
MarcoFalke:
review ACK 724c497562👫
Tree-SHA512: 684a4cf9e3d4496c9997fb2bc4ec874809987055c157ec3fad1d2143b8223df52b5a0af787d028930b27388c8efeba0aeb2446cb35c337a5552ae76112ade726
Clear() is now only called from the ctor, so just inline the code into
that function.
The LOCK(cs) can be removed, since there can be no data races in the ctor.
Also move the function definition out of the header and into the cpp file.
Currently addrman consistency checks are a compile time option, and are not
enabled in our CI. It's unlikely anyone is running these consistency checks.
Make them a runtime option instead, where users can enable addrman
consistency checks every n operations (similar to mempool tests). Update
the addrman unit tests to do internal consistency checks every 100
operations (checking on every operations causes the test runtime to
increase by several seconds).
Also assert on a failed addrman consistency check to terminate program
execution.
65332b1178 [addrman] Remove RemoveInvalid() (John Newbery)
Pull request description:
PRs #22179 and #22112 (EDIT: later reverted in #22497) added hotfix code to addrman to remove invalid addresses and mutate the ports of I2P entries after entering into addrman. Those hotfixes included at least two addrman data corruption bugs:
- #22467 (Assertion `nRndPos1 < vRandom.size() && nRndPos2 < vRandom.size()' failed)
- #22470 (Changing I2P ports in addrman may wronly skip some entries from "new" buckets)
Hotfixing addrman is inherently dangerous. There are many members that have implicit assumptions on each others' state, and mutating those directly can lead to violating addrman's internal invariants.
Instead of trying to hotfix addrman, just don't insert any invalid addresses. For now, those are addresses which fail `CNetAddr::IsValid()`.
ACKs for top commit:
sipa:
utACK 65332b1178. I tried to reason through scenarios that could introduce inconsistencies with this code, but can't find any.
fanquake:
ACK 65332b1178 - Skipping the addition of invalid addresses (this code was initially added for Tor addrs) rather than adding all the invalids then removing them all when finishing unserializing seems like an improvement. Especially if it can be achieved with less code.
Tree-SHA512: 023113764cb475572f15da7bf9824b62b79e10a7e359af2eee59017df354348d2aeed88de0fd4ad7a9f89a0dad10827f99d70af6f1cb20abb0eca2714689c8d7
Instead of deserializing addresses, placing them in the buckets, and
then removing them if they're invalid, check first and don't place in
the buckets if they're invalid.
This is a temporary change to convert I2P addresses that have propagated
with port 8333 to ones with port 0.
It would cause a problem some day if indeed some bitcoin software is
listening on port 8333 only and rejects connections to port 0 and we are
still using SAM 3.1 which only supports port 0. In this case we would
replace 8333 with 0 and try to connect to such nodes.
This commit should be included in 22.0 and be reverted before 23.0 is
released.
00b875ba94 addrman: remove invalid addresses when unserializing (Vasil Dimov)
bdb62096f0 fuzz: reduce possible networks check (Vasil Dimov)
a164cd3ba6 net: simplify CNetAddr::IsRoutable() (Vasil Dimov)
Pull request description:
* Simplify some code, now that we know `CNetAddr::IsRFC4193()` and `CNetAddr::IsTor()` cannot be `true` at the same time.
* Drop Tor v2 addresses when loading addrman from `peers.dat` - they would have been loaded as dummy-all-zeros IPv6 addresses and linger in addrman, wasting space.
ACKs for top commit:
sipa:
ACK 00b875ba94. Reviewed the code, and tested with -DDEBUG_ADDRMAN (unit tests + mainnet run with peers.dat that contained v2 onions).
laanwj:
Code review and lightly tested ACK 00b875ba94
jonatack:
ACK 00b875ba94 reviewed, debug-built with -DEBUG_ADDRMAN rebased to current master, restarted node on mainnet/signet/testnet and verified that on each chain -addrinfo shows no change in address counts (as expected). Added some sanity check asserts, rebuilt/re-ran test. Checked that the new test fails on master with "test/addrman_tests.cpp(824): error: in "addrman_tests/remove_invalid": check addrman.size() == 2 has failed [4 != 2]"
jarolrod:
ACK 00b875ba94
Tree-SHA512: 6ed8e6745134b1b94fffaba28482de909ea39483b46b7f57bda61cdbae7a51251d15cb674de3631772fbeabe153d77a19269f96e62a89102a2d5c01e48f0ba06
a92485b2c2 addrman: use unordered_map instead of map (Vasil Dimov)
Pull request description:
`CAddrMan` uses `std::map` internally even though it does not require
that the map's elements are sorted. `std::map`'s access time is
`O(log(map size))`. `std::unordered_map` is more suitable as it has a
`O(1)` access time.
This patch lowers the execution times of `CAddrMan`'s methods as follows
(as per `src/bench/addrman.cpp`):
```
AddrMan::Add(): -3.5%
AddrMan::GetAddr(): -76%
AddrMan::Good(): -0.38%
AddrMan::Select(): -45%
```
ACKs for top commit:
jonatack:
ACK a92485b2c2
achow101:
ACK a92485b2c2
hebasto:
re-ACK a92485b2c2, only suggested changes and rebased since my [previous](https://github.com/bitcoin/bitcoin/pull/18722#pullrequestreview-666663681) review.
Tree-SHA512: d82959a00e6bd68a6c4c5a265dd08849e6602ac3231293b7a3a3b7bf82ab1d3ba77f8ca682919c15c5d601b13e468b8836fcf19595248116635f7a50d02ed603
