- Settings updates were not thread-safe, as they were executed in
three separate steps:
1) Obtain settings value while acquiring the settings lock.
2) Modify settings value.
3) Overwrite settings value while acquiring the settings lock.
This approach allowed concurrent threads to modify the same base value
simultaneously, leading to data loss. When this occurred, the final
settings state would only reflect the changes from the last thread
that completed the operation, overwriting updates from other threads.
Fix this by making the settings update operation atomic.
- Add test coverage for this behavior.
Co-authored-by: furszy <matiasfurszyfer@protonmail.com>
d7707d9843 rpc: avoid copying into UniValue (Cory Fields)
Pull request description:
These are the simple (and hopefully obviously correct) copies that can be moves instead.
This is a follow-up from https://github.com/bitcoin/bitcoin/pull/30094#issuecomment-2108751842
As it turns out, there are hundreds of places where we copy UniValues needlessly. It should be the case that moves are always preferred over copies, so there should be no downside to these changes.
willcl-ark, however, noticed that memory usage may increase in some cases. Logically this makes no sense to me. The only plausible explanation imo is that because the moves are faster, more ops/second occur in some cases.
This list of moves was obtained by changing the function signatures of the UniValue functions to accept only rvalues, then compiling and fixing them up one by one. There still exist many places where copies are being made. These can/should be fixed up, but weren't done here for the sake of doing the easy ones first.
I ran these changes through clang-tidy with `performance-move-const-arg` and `bugprone-use-after-move` and no bugs were detected (though that's obviously not to say it can be trusted 100%).
As stated above, there are still lots of other less trivial fixups to do after these including:
- Using non-const UniValues where possible so that moves can happen
- Refactoring code in order to be able to move a UniValue without introducing a use-after-move
- Refactoring functions to accept UniValues by value rather than by const reference
ACKs for top commit:
achow101:
ACK d7707d9843
ryanofsky:
Code review ACK d7707d9843. No changes since last review other than rebase. I agree benchmarks showing increased peak memory usage and RSS are surprising, but number of allocations is down as expected, and runtime is also decreased.
willcl-ark:
ACK d7707d9843
Tree-SHA512: 7f511be73984553c278186286a7d161a34b2574c7f5f1a0edc87c2913b4c025a0af5241ef9af2df17547f2e4ef79710aa5bbb762fc9472435781c0488dba3435
d5228efb53 kernel: Remove dependency on CScheduler (TheCharlatan)
06069b3913 scripted-diff: Rename MainSignals to ValidationSignals (TheCharlatan)
0d6d2b650d scripted-diff: Rename SingleThreadedSchedulerClient to SerialTaskRunner (TheCharlatan)
4abde2c4e3 [refactor] Make MainSignals RAII styled (TheCharlatan)
84f5c135b8 refactor: De-globalize g_signals (TheCharlatan)
473dd4b97a [refactor] Prepare for g_signals de-globalization (TheCharlatan)
3fba3d5dee [refactor] Make signals optional in mempool and chainman (TheCharlatan)
Pull request description:
By defining a virtual interface class for the scheduler client, users of the kernel can now define their own event consuming infrastructure, without having to spawn threads or rely on the scheduler design.
Removing `CScheduler` also allows removing the thread and exception modules from the kernel library.
To make the `CMainSignals` class easier to use from a kernel library perspective, remove its global instantiation and adopt RAII practices.
Renames `CMainSignals` to `ValidationSignals`, which more accurately describes its purpose and scope.
Also make the `ValidationSignals` in the `ChainstateManager` and CTxMemPool` optional. This could be useful in the future for using or testing these classes without having to instantiate any form of signal handling.
---
This PR is part of the [libbitcoinkernel project](https://github.com/bitcoin/bitcoin/issues/27587). It improves the kernel API and removes two modules from the kernel library.
ACKs for top commit:
maflcko:
re-ACK d5228efb53🌄
ryanofsky:
Code review ACK d5228efb53. Just comment change since last review.
vasild:
ACK d5228efb53
furszy:
diff ACK d5228ef
Tree-SHA512: e93a5f10eb6182effb84bb981859a7ce750e466efd8171045d8d9e7fe46e4065631d9f6f533c5967c4d34c9bb7d7a67e9f4593bd4c5b30cd7b3bbad7be7b331b
Transactions are intended to be started on upper layers rather than
internally by the bdb batch object. This enables us to consolidate
different write operations within a procedure in the same db txn,
improving consistency due to the atomic property of the transaction,
as well as its performance due to the reduction of disk write
operations.
Important Note:
This approach also ensures that the BerkeleyBatch::ErasePrefix
function behaves exactly as the SQLiteBatch::ErasePrefix function,
which does not create a db txn internally.
Furthermore, since the `BerkeleyBatch::ErasePrefix' implementation
erases records one by one (by traversing the db), this change
ensures that the function is always called within an active txn
context. Without this measure, there's a potential risk to consistency;
certain records may be removed while others could persist due to an
internal failure during the procedure.
9a3c5c8697 scripted-diff: rename ZapSelectTx to RemoveTxs (furszy)
83b762845f wallet: batch and simplify ZapSelectTx process (furszy)
595d50a103 wallet: migration, remove extra NotifyTransactionChanged call (furszy)
a2b071f992 wallet: ZapSelectTx, remove db rewrite code (furszy)
Pull request description:
Work decoupled from #28574. Brother of #28894.
Includes two different, yet interconnected, performance and code improvements to the zap wallet transactions process.
1) As the goal of the `ZapSelectTx` function is to erase tx records that match any of the inputted hashes. There is no need to traverse the whole database record by record. We could just check if the tx exist, and remove it directly by calling `EraseTx()`.
2) Instead of performing single write operations per removed tx record, this PR batches them all within a single atomic db txn.
Moreover, these changes will enable us to consolidate all individual write operations that take place during the wallet migration process into a single db txn in the future.
ACKs for top commit:
achow101:
ACK 9a3c5c8697
josibake:
ACK 9a3c5c8697
Tree-SHA512: fb2ecc48224c400ab3b1fbb32e174b5b13bf03794717727f80f01f55fb183883b067a68c0a127b2de8885564da15425d021a96541953bf38a72becc2e9929ccf
The goal of the function is to erase the wallet transactions that
match the inputted hashes. There is no need to traverse the database,
reading record by record, to then perform single entry removals for
each of them.
To ensure consistency and improve performance, this change-set removes
all tx records within a single atomic db batch operation, as well as
it cleans up code, improves error handling and simplifies the
transactions removal process entirely.
This optimizes the removal of watch-only transactions during the wallet
migration process and the 'removeprunedfunds' RPC command.
Making the `GenerateRandomKey` helper available to other modules via
key.{h.cpp} allows us to create random private keys directly at
instantiation of CKey, in contrast to the two-step process of creating
the instance and then having to call `MakeNewKey(...)`.
Since #25273, the behavior of 'inserting change at a random
position' is instructed by passing std::nullopt instead of -1.
Also, added missing documentation about the meaning of
'change_pos=std::nullopt' inside 'CWallet::CreateTransaction()'
fa79a881ce refactor: P2P transport without serialize version and type (MarcoFalke)
fa9b5f4fe3 refactor: NetMsg::Make() without nVersion (MarcoFalke)
66669da4a5 Remove unused Make() overload in netmessagemaker.h (MarcoFalke)
fa0ed07941 refactor: VectorWriter without nVersion (MarcoFalke)
Pull request description:
Now that the serialize framework ignores the serialize version and serialize type, everything related to it can be removed from the code.
This is the first step, removing dead code from the P2P stack. A different pull will remove it from the wallet and other parts.
ACKs for top commit:
ajtowns:
reACK fa79a881ce
Tree-SHA512: 785b413580d980f51f0d4f70ea5e0a99ce14cd12cb065393de2f5254891be94a14f4266110c8b87bd2dbc37467676655bce13bdb295ab139749fcd8b61bd5110
This should fix the bug reported in
https://github.com/bitcoin/bitcoin/pull/28246#discussion_r1371640502,
which caused the GUI to not detect the destination type of recipients,
thus picking the wrong change destination type.
Also, add missing lifetimebound attribute to a getter method.
fac29a0ab1 Remove SER_GETHASH, hard-code client version in CKeyPool serialize (MarcoFalke)
fa72f09d6f Remove CHashWriter type (MarcoFalke)
fa4a9c0f43 Remove unused GetType() from OverrideStream, CVectorWriter, SpanReader (MarcoFalke)
Pull request description:
Removes a bunch of redundant, dead or duplicate code.
Uses the idea from and finishes the idea https://github.com/bitcoin/bitcoin/pull/28428 by theuni
ACKs for top commit:
ajtowns:
ACK fac29a0ab1
kevkevinpal:
added one nit but otherwise ACK [fac29a0](fac29a0ab1)
Tree-SHA512: cc805e2f38e73869a6691fdb5da09fa48524506b87fc93f05d32c336ad3033425a2d7608e317decd3141fde3f084403b8de280396c0c39132336fe0f7510af9e
This removes unused includes, primitives/block found manually, and the
others by iwyu:
blockfilter.h should remove these lines:
- #include <serialize.h> // lines 16-16
- #include <undo.h> // lines 18-18
CTxDestination is really our internal representation of an address and
doesn't really have anything to do with standard script types, so move
them to their own file.
and drop the util/random dependency on util/setup_common.
This improves code separation and avoids creating a circular dependency if
setup_common needs to call the util/random functions.
The wallet tests and benchmarks both had helper functions for loading
and unloading the wallet for the test that were almost identical.
These functions are consolidated and reused.
33e2b82a4f wallet, bench: Remove unused database options from WalletBenchLoading (Andrew Chow)
80ace042d8 tests: Modify records directly in wallet ckey loading test (Andrew Chow)
b3bb17d5d0 tests: Update DuplicateMockDatabase for MockableDatabase (Andrew Chow)
f0eecf5e40 scripted-diff: Replace CreateMockWalletDB with CreateMockableWalletDB (Andrew Chow)
075962bc25 wallet, tests: Include wallet/test/util.h (Andrew Chow)
14aa4cb1e4 wallet: Move DummyDatabase to salvage (Andrew Chow)
f67a385556 wallet, tests: Replace usage of dummy db with mockable db (Andrew Chow)
33c6245ac1 Introduce MockableDatabase for wallet unit tests (Andrew Chow)
Pull request description:
For the wallet's unit tests, we currently use either `DummyDatabase` or memory-only versions of either BDB or SQLite. The tests that use `DummyDatabase` just need a `WalletDatabase` so that the `CWallet` can be constructed, while the tests using the memory-only databases just need a backing data store. There is also a `FailDatabase` that is similar to `DummyDatabase` except it fails be default or can have a configured return value. Having all of these different database types can make it difficult to write tests, particularly tests that work when either BDB or SQLite is disabled.
This PR unifies all of these different unit test database classes into a single `MockableDatabase`. Like `DummyDatabase`, most functions do nothing and just return true. Like `FailDatabase`, the return value of some functions can be configured on the fly to test various failure cases. Like the memory-only databases, records can actually be written to the `MockableDatabase` and be retrieved later, but all of this is still held in memory. Using `MockableDatabase` completely removes the need for having BDB or SQLite backed wallets in the unit tests for the tests that are not actually testing specific database behaviors.
Because `MockableDatabase`s can be created by each unit test, we can also control what records are stored in the database. Records can be added and removed externally from the typical database modification functions. This will give us greater ability to test failure conditions, particularly those involving corrupted records.
Possible alternative to #26644
ACKs for top commit:
furszy:
ACK 33e2b82
TheCharlatan:
ACK 33e2b82a4f
Tree-SHA512: c2b09eff9728d063d2d4aea28a0f0e64e40b76483e75dc53f08667df23bd25834d52656cd4eafb02e552db0b9e619cfdb1b1c65b26b5436ee2c971d804768bcc
This is a commit in preparation for the next few commits. The functions
are moved to methods to avoid their re-declaration for the purpose of
passing in BlockManager options.
The functions that were now moved into the BlockManager should no longer
use the params as an argument, but instead use the member variable.
In the moved ReadBlockFromDisk and UndoReadFromDisk, change
the function signature to accept a reference to a CBlockIndex instead of
a raw pointer. The pointer is expected to be non-null, so reflect that
in the type.
To allow for the move of functions to BlockManager methods all call
sites require an instantiated BlockManager, or a callback to one.
Since we have a mockable wallet database, we don't really need to be
using BDB or SQLite's in-memory database capabilities. It doesn't really
help us to be using those as we aren't doing anything that requires one
type of db over the other, and will just prefer SQLite if it's
available.
MockableDatabase is suitable for these uses, so use
CreateMockableWalletDatabase to use that.
-BEGIN VERIFY SCRIPT-
sed -i "s/CreateMockWalletDatabase(options)/CreateMockableWalletDatabase()/" $(git grep -l "CreateMockWalletDatabase(options)" -- ":(exclude)src/wallet/walletdb.*")
sed -i "s/CreateMockWalletDatabase/CreateMockableWalletDatabase/" $(git grep -l "CreateMockWalletDatabase" -- ":(exclude)src/wallet/walletdb.*")
-END VERIFY SCRIPT-
MockableDatabase is a WalletDatabase that allows us to interact with the
records to change them independently from the wallet, as well as
changing the return values from within the tests. This will give us
greater flexibility in testing the wallet.
This is cleanup that doesn't change external behavior.
- Removes awkward `StringMap` intermediate representation
- Simplifies CWallet code, deals with used address and received request
serialization in walletdb.cpp
- Adds test coverage and documentation
- Reduces memory usage
This PR doesn't change externally observable behavior. Internally, only change
in behavior is that EraseDestData deletes directly from database because the
`StringMap` is gone. This is more direct and efficient because it uses a single
btree lookup and scan instead of multiple lookups
Motivation for this cleanup is making changes like #18550, #18192, #13756
easier to reason about and less likely to result in unintended behavior and
bugs
Co-authored-by: furszy <matiasfurszyfer@protonmail.com>
The previous behavior for getAvailableBalance when coin control
has selected coins was to return the sum of them. Instead, we
are currently returning the wallet's available total balance minus
the selected coins total amount.
This turns into a GUI-only issue for the "use available balance"
button when the user manually select coins in the send screen.
Reason:
We missed to update the GetAvailableBalance function to include
the coin control selected coins on #25685.
Context:
Since #25685 we skip the selected coins inside `AvailableCoins`,
the reason is that there is no need to traverse the wallet's
txes map just to get coins that can directly be fetched by
their id.
4aebd832a4 db: Change DatabaseCursor::Next to return status enum (Andrew Chow)
d79e8dcf29 wallet: Have cursor users use DatabaseCursor directly (Andrew Chow)
7a198bba0a wallet: Introduce DatabaseCursor RAII class for managing cursor (Andrew Chow)
69efbc011b Move SafeDbt out of BerkeleyBatch (Andrew Chow)
Pull request description:
Instead of having database cursors be tied to a particular `DatabaseBatch` object and requiring its setup and teardown be separate functions in that batch, we can have cursors be separate RAII classes. This makes it easier to create and destroy cursors as well as having cursors that have slightly different behaviors.
Additionally, since reading data from a cursor is a tri-state, this PR changes the return value of the `Next` function (formerly `ReadAtCursor`) to return an Enum rather than the current system of 2 booleans. This greatly simplifies and unifies the code that deals with cursors as now there is no confusion as to what the function returns when there are no records left to be read.
Extracted from #24914
ACKs for top commit:
furszy:
diff ACK 4aebd83
theStack:
Code-review ACK 4aebd832a4
Tree-SHA512: 5d0be56a18de5b08c777dd5a73ba5a6ef1e696fdb07d1dca952a88ded07887b7c5c04342f9a76feb2f6fe24a45dc31f094f1f5d9500e6bdf4a44f4edb66dcaa1
f2fc03ec85 refactor: use braced init for integer constants instead of c style casts (Pasta)
Pull request description:
See https://github.com/bitcoin/bitcoin/pull/23810 for more context. This is broken out from that PR, as it is less breaking, and should be trivial to review and merge.
EDIT: Long term, the intention is to remove all C-style casts, as they can dangerously introduce reinterpret_casts. This is one step which removes a number of trivially removable C-style casts
ACKs for top commit:
aureleoules:
ACK f2fc03ec85
Tree-SHA512: 2fd11b92c9147e3f970ec3e130e3b3dce70e707ff02950a8c697d4b111ddcbbfa16915393db20cfc8f384bc76f13241c9b994a187987fcecd16a61f8cc0af14c
55696a0ac3 wallet: remove `mempool_sequence` from `transactionRemovedFromMempool` (w0xlt)
bf19069c53 wallet: remove `mempool_sequence` from `transactionAddedToMempool` (w0xlt)
Pull request description:
This PR removes `mempool_sequence` from `transactionRemovedFromMempool` and `transactionAddedToMempool`.
`mempool_sequence` is not used in these methods, only in ZMQ notifications.
ACKs for top commit:
instagibbs:
ACK 55696a0ac3
Tree-SHA512: 621e89230bcb6edfed83e2758601a2b093822fc2dc4e9bfb00487e340f2bc4c5ac3bf6df3ca00b7fe55bb3df15858820f2bf698f403d2e48b915dd9eb47b63e0
