It will enable different error handling flows for different error types.
Also, `ReadRawBlockBench` performance has decreased due to no longer reusing a vector
with an unchanging capacity - mirroring our production code behavior.
Co-authored-by: Hodlinator <172445034+hodlinator@users.noreply.github.com>
Co-authored-by: Lőrinc <pap.lorinc@gmail.com>
The return value of BlockManager::GetFirstBlock must always be non-null. This
can be inferred by the implementation, which has an assertion that the return
value is not null. A raw pointer should only be returned if the result may be
null. In this case a reference is more appropriate.
fa6c0bedd3 refactor: Return uint64_t from GetSerializeSize (MarcoFalke)
fad0c8680e refactor: Use uint64_t over size_t for serialized-size values (MarcoFalke)
fa4f388fc9 refactor: Use fixed size ints over (un)signed ints for serialized values (MarcoFalke)
fa01f38e53 move-only: Move CBlockFileInfo to kernel namespace (MarcoFalke)
fa2bbc9e4c refactor: [rpc] Remove cast when reporting serialized size (MarcoFalke)
fa364af89b test: Remove outdated comment (MarcoFalke)
Pull request description:
Consensus code should arrive at the same conclusion, regardless of the architecture it runs on. Using architecture-specific types such as `size_t` can lead to issues, such as the low-severity [CVE-2025-46597](https://bitcoincore.org/en/2025/10/24/disclose-cve-2025-46597/).
The CVE was already worked around, but it may be good to still fix the underlying issue.
Fixes https://github.com/bitcoin/bitcoin/issues/33709 with a few refactors to use explicit fixed-sized integer types in serialization-size related code and concluding with a refactor to return `uint64_t` from `GetSerializeSize`. The refactors should not change any behavior, because the CVE was already worked around.
ACKs for top commit:
Crypt-iQ:
crACK fa6c0bedd3
l0rinc:
ACK fa6c0bedd3
laanwj:
Code review ACK fa6c0bedd3
Tree-SHA512: f45057bd86fb46011e4cb3edf0dc607057d72ed869fd6ad636562111ae80fea233b2fc45c34b02256331028359a9c3f4fa73e9b882b225bdc089d00becd0195e
743abbcbde refactor: inline constant return value of `BlockTreeDB::WriteBatchSync` and `BlockManager::WriteBlockIndexDB` and `BlockTreeDB::WriteFlag` (Lőrinc)
e030240e90 refactor: inline constant return value of `CDBWrapper::Erase` and `BlockTreeDB::WriteReindexing` (Lőrinc)
cdab9480e9 refactor: inline constant return value of `CDBWrapper::Write` (Lőrinc)
d1847cf5b5 refactor: inline constant return value of `TxIndex::DB::WriteTxs` (Lőrinc)
50b63a5698 refactor: inline constant return value of `CDBWrapper::WriteBatch` (Lőrinc)
Pull request description:
Related to https://github.com/bitcoin/bitcoin/pull/31144#discussion_r2223587480
### Summary
`WriteBatch` always returns `true` - the errors are handled by throwing `dbwrapper_error` instead.
### Context
This boolean return value of the `Write` methods is confusing because it's inconsistent with `CDBWrapper::Read`, which catches exceptions and returns a boolean to indicate success/failure. It's bad that `Read` returns and `Write` throws - but it's a lot worse that `Write` advertises a return value when it actually communicates errors through exceptions.
### Solution
This PR removes the constant return values from write methods and inlines `true` at their call sites. Many upstream methods had boolean return values only because they were propagating these constants - those have been cleaned up as well.
Methods that returned a constant `true` value that now return `void`:
- `CDBWrapper::WriteBatch`, `CDBWrapper::Write`, `CDBWrapper::Erase`
- `TxIndex::DB::WriteTxs`
- `BlockTreeDB::WriteReindexing`, `BlockTreeDB::WriteBatchSync`, `BlockTreeDB::WriteFlag`
- `BlockManager::WriteBlockIndexDB`
### Note
`CCoinsView::BatchWrite` (and transitively `CCoinsViewCache::Flush` & `CCoinsViewCache::Sync`) were intentionally not changed here. While all implementations return `true`, the base `CCoinsView::BatchWrite` returns `false`. Changing this would cause `coins_view` tests to fail with:
> terminating due to uncaught exception of type std::logic_error: Not all unspent flagged entries were cleared
We can fix that in a follow-up PR.
ACKs for top commit:
achow101:
ACK 743abbcbde
janb84:
ACK 743abbcbde
TheCharlatan:
ACK 743abbcbde
sipa:
ACK 743abbcbde
Tree-SHA512: b2a550bff066216f1958d2dd9a7ef6a9949de518cc636f8ab9c670e0b7a330c1eb8c838e458a8629acb8ac980cea6616955cd84436a7b8ab9096f6d648073b1e
Did both in this commit, since the return value of `WriteReindexing` was ignored anyway - which existed only because of the constant `Erase` being called
`WriteBatch` can only ever return `true` - its errors are handled by throwing a `throw dbwrapper_error` instead.
The boolean return value is quite confusing, especially since it's symmetric with `CDBWrapper::Read`, which catches the exceptions and returns a boolean instead.
We're removing the constant return value and inlining `true` for its usages.
Before this, if we had two (or more) same work tip candidates and restarted our node,
it could be the case that the block set as tip after bootstrap didn't match the one
before stopping. That's because the work and `nSequenceId` of both block will be the same
(the latter is only kept in memory), so the active chain after restart would have depended
on what tip candidate was loaded first.
This makes sure that we are consistent over reboots.
It is redundant with -logsourcelocations and the log messages are
clearer without it.
Also, remove a double-space.
Also, add braces around `if` touched in the next commit.
This tiny behavior change requires a test fixup.
Mechanical refactor of the low-level "xor" wording to signal the intent instead of the implementation used.
The renames are ordered by heaviest-hitting substitutions first, and were constructed such that after each replacement the code is still compilable.
-BEGIN VERIFY SCRIPT-
sed -i \
-e 's/\bGetObfuscateKey\b/GetObfuscation/g' \
-e 's/\bxor_key\b/obfuscation/g' \
-e 's/\bxor_pat\b/obfuscation/g' \
-e 's/\bm_xor_key\b/m_obfuscation/g' \
-e 's/\bm_xor\b/m_obfuscation/g' \
-e 's/\bobfuscate_key\b/m_obfuscation/g' \
-e 's/\bOBFUSCATE_KEY_KEY\b/OBFUSCATION_KEY_KEY/g' \
-e 's/\bSetXor(/SetObfuscation(/g' \
-e 's/\bdata_xor\b/obfuscation/g' \
-e 's/\bCreateObfuscateKey\b/CreateObfuscation/g' \
-e 's/\bobfuscate key\b/obfuscation key/g' \
$(git ls-files '*.cpp' '*.h')
-END VERIFY SCRIPT-
Since 31 byte xor-keys are not used in the codebase, using the common size (8 bytes) makes the benchmarks more realistic.
Co-authored-by: maflcko <6399679+maflcko@users.noreply.github.com>
c10e382d2a flatfile: check whether the file has been closed successfully (Vasil Dimov)
4bb5dd78ea util: check that a file has been closed before ~AutoFile() is called (Vasil Dimov)
8bb34f07df Explicitly close all AutoFiles that have been written (Vasil Dimov)
a69c4098b2 rpc: take ownership of the file by WriteUTXOSnapshot() (Hodlinator)
Pull request description:
`fclose(3)` may fail to flush the previously written data to disk, thus a failing `fclose(3)` is as serious as a failing `fwrite(3)`.
Previously the code ignored `fclose(3)` failures. This PR improves that by changing all users of `AutoFile` that use it to write data to explicitly close the file and handle a possible error.
---
Other alternatives are:
1. `fflush(3)` after each write to the file (and throw if it fails from the `AutoFile::write()` method) and hope that `fclose(3)` will then always succeed. Assert that it succeeds from the destructor 🙄. Will hurt performance.
2. Throw nevertheless from the destructor. Exception within the exception in C++ I think results in terminating the program without a useful message.
3. (this is implemented in the latest incarnation of this PR) Redesign `AutoFile` so that its destructor cannot fail. Adjust _all_ its users 😭. For example, if the file has been written to, then require the callers to explicitly call the `AutoFile::fclose()` method before the object goes out of scope. In the destructor, as a sanity check, assume/assert that this is indeed the case. Defeats the purpose of a RAII wrapper for `FILE*` which automatically closes the file when it goes out of scope and there are a lot of users of `AutoFile`.
4. Pass a new callback function to the `AutoFile` constructor which will be called from the destructor to handle `fclose()` errors, as described in https://github.com/bitcoin/bitcoin/pull/29307#issuecomment-2243842400. My thinking is that if that callback is going to only log a message, then we can log the message directly from the destructor without needing a callback. If the callback is going to do more complicated error handling then it is easier to do that at the call site by directly calling `AutoFile::fclose()` instead of getting the `AutoFile` object out of scope (so that its destructor is called) and inspecting for side effects done by the callback (e.g. set a variable to indicate a failed `fclose()`).
ACKs for top commit:
l0rinc:
ACK c10e382d2a
achow101:
ACK c10e382d2a
hodlinator:
re-ACK c10e382d2a
Tree-SHA512: 3994ca57e5b2b649fc84f24dad144173b7500fc0e914e06291d5c32fbbf8d2b1f8eae0040abd7a5f16095ddf4e11fe1636c6092f49058cda34f3eb2ee536d7ba
There is no way to report a close error from `AutoFile` destructor.
Such an error could be serious if the file has been written to because
it may mean the file is now corrupted (same as if write fails).
So, change all users of `AutoFile` that use it to write data to
explicitly close the file and handle a possible error.
Eliminate one SHA‑256 double‑hash computation of the header per block read by reusing the hash for:
* proof‑of‑work verification;
* (optional) integrity check against the supplied hash.
The obfuscation (XOR) operations are currently done byte-by-byte during serialization. Buffering the reads will enable batching the obfuscation operations later.
Different operating systems handle file caching differently, so reading larger batches (and processing them from memory) is measurably faster, likely because of fewer native fread calls and reduced lock contention.
Note that `ReadRawBlock` doesn't need buffering since it already reads the whole block directly.
Unlike `ReadBlockUndo`, the new `ReadBlock` implementation delegates to `ReadRawBlock`, which uses more memory than a buffered alternative but results in slightly simpler code and a small performance increase (~0.4%). This approach also clearly documents that `ReadRawBlock` is a logical subset of `ReadBlock` functionality.
The current implementation, which iterates over a fixed-size buffer, provides a more general alternative to Cory Fields' solution of reading the entire block size in advance.
Buffer sizes were selected based on benchmarking to ensure the buffered reader produces performance similar to reading the whole block into memory. Smaller buffers were slower, while larger ones showed diminishing returns.
------
> macOS Sequoia 15.3.1
> C++ compiler .......................... Clang 19.1.7
> cmake -B build -DBUILD_BENCH=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ && cmake --build build -j$(nproc) && build/bin/bench_bitcoin -filter='ReadBlockBench' -min-time=10000
Before:
| ns/op | op/s | err% | total | benchmark
|--------------------:|--------------------:|--------:|----------:|:----------
| 2,271,441.67 | 440.25 | 0.1% | 11.00 | `ReadBlockBench`
After:
| ns/op | op/s | err% | total | benchmark
|--------------------:|--------------------:|--------:|----------:|:----------
| 1,738,971.29 | 575.05 | 0.2% | 10.97 | `ReadBlockBench`
------
> Ubuntu 24.04.2 LTS
> C++ compiler .......................... GNU 13.3.0
> cmake -B build -DBUILD_BENCH=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ && cmake --build build -j$(nproc) && build/bin/bench_bitcoin -filter='ReadBlockBench' -min-time=20000
Before:
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 6,895,987.11 | 145.01 | 0.0% | 71,055,269.86 | 23,977,374.37 | 2.963 | 5,074,828.78 | 0.4% | 22.00 | `ReadBlockBench`
After:
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 5,771,882.71 | 173.25 | 0.0% | 65,741,889.82 | 20,453,232.33 | 3.214 | 3,971,321.75 | 0.3% | 22.01 | `ReadBlockBench`
Co-authored-by: maflcko <6399679+maflcko@users.noreply.github.com>
Co-authored-by: Ryan Ofsky <ryan@ofsky.org>
Co-authored-by: Martin Leitner-Ankerl <martin.ankerl@gmail.com>
Co-authored-by: Cory Fields <cory-nospam-@coryfields.com>
Made every OpenBlockFile#fReadOnly value explicit.
Replaced hard-coded values in ReadRawBlock with STORAGE_HEADER_BYTES.
Changed `STORAGE_HEADER_BYTES` and `UNDO_DATA_DISK_OVERHEAD` to `uint32_t` to avoid casts.
Also added `LIFETIMEBOUND` to the `AutoFile` parameter of `BufferedFile`, which stores a reference to the underlying `AutoFile`, allowing Clang to emit warnings if the referenced `AutoFile` might be destroyed while `BufferedFile` still exists.
Without this attribute, code with lifetime violations wouldn't trigger compiler warnings.
Co-authored-by: maflcko <6399679+maflcko@users.noreply.github.com>
Renames the constant to be less verbose and better reflect its purpose:
it represents the size of the storage header that precedes serialized block data on disk,
not to be confused with a block's own header.
-BEGIN VERIFY SCRIPT-
git grep -q "STORAGE_HEADER_BYTES" $(git ls-files) && echo "Error: Target name STORAGE_HEADER_BYTES already exists in the codebase" && exit 1
sed -i 's/BLOCK_SERIALIZATION_HEADER_SIZE/STORAGE_HEADER_BYTES/g' $(git grep -l 'BLOCK_SERIALIZATION_HEADER_SIZE')
-END VERIFY SCRIPT-
Reorganized error handling in block-related operations by grouping related operations together within the same scope.
In `ReadBlockUndo()` and `ReadBlock()`, moved all deserialization operations, comments and checksum verification inside a single try/catch block for cleaner error handling.
In `WriteBlockUndo()`, consolidated hash calculation and data writing operations within a common block to better express their logical relationship.
0cdddeb224 kernel: Move block tree db open to BlockManager constructor (TheCharlatan)
7fbb1bc44b kernel: Move block tree db open to block manager (TheCharlatan)
57ba59c0cd refactor: Remove redundant reindex check (TheCharlatan)
Pull request description:
Before this change the block tree db was needlessly re-opened during startup when loading a completed snapshot. Improve this by letting the block manager open it on construction. This also simplifies the test code a bit.
The change was initially motivated to make it easier for users of the kernel library to instantiate a BlockManager that may be used to read data from disk without loading the block index into a cache.
ACKs for top commit:
maflcko:
re-ACK 0cdddeb224🏪
achow101:
ACK 0cdddeb224
mzumsande:
re-ACK 0cdddeb224
Tree-SHA512: fe3d557a725367e549e6a0659f64259cfef6aaa565ec867d9a177be0143ff18a2c4a20dd57e35e15f97cf870df476d88c05b03b6a7d9e8d51c568d9eda8947ef
2656a5658c tests: add a test for the new blocksdir lock (Cory Fields)
bdc0a68e67 init: lock blocksdir in addition to datadir (Cory Fields)
cabb2e5c24 refactor: introduce a more general LockDirectories for init (Cory Fields)
1db331ba76 init: allow a new xor key to be written if the blocksdir is newly created (Cory Fields)
Pull request description:
This probably should've been included in #12653 when `-blocksdir` was introduced. Credit TheCharlatan for noticing that it's missing.
This guards against 2 processes running with separate datadirs but the same blocksdir. I didn't add `walletdir` as I assume sqlite has us covered there.
It's not likely to happen currently, but may be more relevant in the future with applications using the kernel. Note that the kernel does not currently do any dir locking, but it should.
ACKs for top commit:
maflcko:
review ACK 2656a5658c 🏼
kevkevinpal:
ACK [2656a56](2656a5658c)
achow101:
ACK 2656a5658c
tdb3:
Code review and light test ACK 2656a5658c
Tree-SHA512: 3ba17dc670126adda104148e14d1322ea4f67d671c84aaa9c08c760ef778ca1936832c0dc843cd6367e09939f64c6f0a682b0fa23a5967e821b899dff1fff961
Make the block db open RAII style by calling it in the BlockManager
constructor.
Before this change the block tree db was needlessly re-opened during
startup when loading a completed snapshot. Improve this by letting the
block manager open it on construction. This also simplifies the test
code a bit.
The change was initially motivated to make it easier for users of the
kernel library to instantiate a BlockManager that may be used to read
data from disk without loading the block index into a cache.
A subsequent commit will add a .lock file to this dir at startup, meaning that
the blocksdir is never empty by the time the xor key is being read/written.
Ignore all hidden files when determining if this is the first run.
When the behavior was changes in a previous commit (caching `GetSerializeSize` and avoiding `AutoFile.tell`), (static)asserts were added to make sure the behavior was kept - to make sure reviewers and CI validates it.
We can safely remove them now.
Logs were also slightly modernized since they were trivial to do.
Co-authored-by: Anthony Towns <aj@erisian.com.au>
Co-authored-by: Hodlinator <172445034+hodlinator@users.noreply.github.com>
For consistency `UNDO_DATA_DISK_OVERHEAD` was also extracted to avoid the constant's ambiguity.
Asserts were added to help with the review - they are removed in the next commit.
Co-authored-by: Ryan Ofsky <ryan@ofsky.org>
Similarly, `WriteBlockToDisk` wasn't really extracting a meaningful subset of the `SaveBlockToDisk` functionality, it's tied closely to the only caller (needs the header size twice, recalculated block serializes size, returns multiple branches, mutates parameter).
The inlined code should only differ in these parts (modernization will be done in other commits):
* renamed `blockPos` to `pos` in `SaveBlockToDisk` to match the parameter name;
* changed `return false` to `return FlatFilePos()`.
Also removed remaining references to `SaveBlockToDisk`.
Co-authored-by: Ryan Ofsky <ryan@ofsky.org>
`UndoWriteToDisk` wasn't really extracting a meaningful subset of the `WriteUndoDataForBlock` functionality, it's tied closely to the only caller (needs the header size twice, recalculated undo serializes size, returns multiple branches, modifies parameter, needs documentation).
The inlined code should only differ in these parts (modernization will be done in other commits):
* renamed `_pos` to `pos` in `WriteUndoDataForBlock` to match the parameter name;
* inlined `hashBlock` parameter usage into `hasher << block.pprev->GetBlockHash()`;
* changed `return false` to `return FatalError`;
* capitalize comment.
Co-authored-by: Ryan Ofsky <ryan@ofsky.org>
a2955f0979 validation: Use span for ImportBlocks paths (TheCharlatan)
20515ea3f5 validation: Use span for CalculateClaimedHeadersWork (TheCharlatan)
52575e96e7 validation: Use span for ProcessNewBlockHeaders (TheCharlatan)
Pull request description:
Makes it friendlier for potential future users of the kernel library if they do not store the headers in a std::vector, but can guarantee contiguous memory.
Take this opportunity to also change the argument of ImportBlocks previously taking a `std::vector` to a `std::span`.
ACKs for top commit:
stickies-v:
re-ACK a2955f0979 - no changes except further walking the ~file~ path of modernizing variable names.
maflcko:
ACK a2955f0979🕑
achow101:
ACK a2955f0979
danielabrozzoni:
ACK a2955f0979
Tree-SHA512: 8b07f4ad26e270b65600d1968cd78847b85caca5bfbb83fd9860389f26656b1d9a40b85e0990339f50403d18cedcd2456990054f3b8b0bedce943e50222d2709
Makes it friendlier for potential future users of the kernel library if
they do not store the headers in a std::vector, but can guarantee
contiguous memory.
fa895c7283 mingw: Document mode wbx workaround (MarcoFalke)
fa359255fe Add -blocksxor boolean option (MarcoFalke)
fa7f7ac040 Return XOR AutoFile from BlockManager::Open*File() (MarcoFalke)
Pull request description:
Currently the *.dat files in the blocksdir store the data received from remote peers as-is. This may be problematic when a program other than Bitcoin Core tries to interpret them by accident. For example, an anti-virus program or other program may scan them and move them into quarantine, or delete them, or corrupt them. This may cause Bitcoin Core to fail a reorg, or fail to reply to block requests (via P2P, RPC, REST, ...).
Fix this, similar to https://github.com/bitcoin/bitcoin/pull/6650, by rolling a random XOR pattern over the dat files when writing or reading them.
Obviously this can only protect against programs that accidentally and unintentionally are trying to mess with the dat files. Any program that intentionally wants to mess with the dat files can still trivially do so.
The XOR pattern is only applied when the blocksdir is freshly created, and there is an option to disable it (on creation), so that people can disable it, if needed.
ACKs for top commit:
achow101:
ACK fa895c7283
TheCharlatan:
Re-ACK fa895c7283
hodlinator:
ACK fa895c7283
Tree-SHA512: c92a6a717da83bc33a9b8671a779eeefde2c63b192362ba1d71e6535ee31d08e2802b74acc908345197de9daac6930e4771595ee25b09acd5a67f7ea34854720
