Some uses of non-threadsafe `strerror` have snuck into the code since
they were removed in #4152. Add a wrapper `SysErrorString` for
thread-safe strerror alternatives and replace all uses of `strerror`
with this.
Base32/base64 are mechanisms for encoding binary data. That they'd
decode to a string is just bizarre. The fact that they'd do that
based on the type of input arguments even more so.
a62e84438d27ee6213219fe2c233e58814fcbb5d fuzz: add `SplitString` fuzz target (MarcoFalke)
4fad7e46d94a0fdee4ff917e81360d7ae6bd8110 test: add unit tests for `SplitString` helper (Kiminuo)
9cc8e876e412056ed22d364538f0da3d5d71946d refactor: introduce single-separator split helper `SplitString` (Sebastian Falbesoner)
Pull request description:
This PR adds a simple string split helper `SplitString` that takes use of the spanparsing `Split` function that was first introduced in #13697 (commit fe8a7dcd78cfeedc9a7c705e91384f793822912b). This enables to replace most calls to `boost::split`, in the cases where only a single separator character is used. Note that while previous attempts to replace `boost::split` were controversial (e.g. #13751), this one has a trivial implementation: it merely uses an internal helper (that is unit tested and in regular use with output descriptiors) and converts its result from spans to strings. As a drawback though, not all `boost::split` instances can be tackled.
As a possible optimization, one could return a vector of `std::string_view`s (available since C++17) instead of strings, to avoid copies. This would need more carefulness on the caller sites though, to avoid potential lifetime issues, and it's probably not worth it, considering that none of the places where strings are split are really performance-critical.
ACKs for top commit:
martinus:
Code review ACK a62e84438d27ee6213219fe2c233e58814fcbb5d. Ran all tests. I also like that with `boost::split` it was not obvious that the resulting container was cleared, and with `SplitString` API that's obvious.
Tree-SHA512: 10cb22619ebe46831b1f8e83584a89381a036b54c88701484ac00743e2a62cfe52c9f3ecdbb2d0815e536c99034558277cc263600ec3f3588b291c07eef8ed24
ee02c8bd9aedad8cfd3c2618035fe275da025fb9 util/check: Add CHECK_NONFATAL identity function, NONFATAL_UNREACHABLE AND UNREACHABLE macros (Aurèle Oulès)
Pull request description:
This PR replaces the macro `CHECK_NONFATAL` with an identity function.
I simplified the usage of `CHECK_NONFATAL` where applicable in `src/rpc`.
This function is useful in sanity checks for RPC and command-line interfaces.
Context: https://github.com/bitcoin/bitcoin/pull/24804#discussion_r846182474.
Also adds `UNREACHABLE_NONFATAL` macro.
ACKs for top commit:
jonatack:
ACK ee02c8bd9aedad8cfd3c2618035fe275da025fb9
MarcoFalke:
ACK ee02c8bd9aedad8cfd3c2618035fe275da025fb9 🍨
Tree-SHA512: 3cba09223cd7b22e62fe5d0b46c4a024c1d9957d4268ba6d3fb07fcc0a5854fc0886bb3266184e6a7df5df91373b3e84edd6adf6999c4e934aeef8c043b01aa2
3ae7791bcaa88f5c68592673b8926ee807242ce7 refactor: use Span in random.* (pasta)
Pull request description:
~This PR does two things~
1. use a Span<unsigned char> for GetRandBytes and GetStrongRandBytes
~2. make GetRand a template for which any integral type can be used, where the default behavior is to return a random integral up to the max of the integral unless a max is provided.
This simplifies a lot of code from `GetRand(std::numeric_limits<uint64_t>::max()` -> `GetRand<uint64_t>()`~
MarcoFalke this was inspired by your comment here: https://github.com/bitcoin/bitcoin/pull/24185#issuecomment-1025514263 about using Span, so hopefully I'll be able to get this PR done and merged 😂
~Also, if requested I could revert the `GetRand(std::numeric_limits<uint64_t>::max()` -> `GetRand<uint64_t>()` related changes if it ends up causing too many conflicts~
ACKs for top commit:
laanwj:
Thank you! Code review re-ACK 3ae7791bcaa88f5c68592673b8926ee807242ce7
Tree-SHA512: 12375a83b68b288916ba0de81cfcab4aac14389a66a36811ae850427435eb67dd55e47df9ac3ec47db4e214f4330139e548bec815fff8a3f571484ea558dca79
a2c4a7acd1dfb2fb7e3c9dac6b3d8c9354b2e0a6 net: use Sock::SetSockOpt() instead of standalone SetSocketNoDelay() (Vasil Dimov)
d65b6c3fb9cdd41fa53bc76a7b8f49aaa089b0bc net: use Sock::SetSockOpt() instead of setsockopt() (Vasil Dimov)
184e56d6683d05fc84f5153cfff83a2e32883556 net: add new method Sock::SetSockOpt() that wraps setsockopt() (Vasil Dimov)
Pull request description:
_This is a piece of #21878, chopped off to ease review._
Add a `virtual` (thus mockable) method `Sock::SetSockOpt()` that wraps the system `setsockopt()`.
Convert the standalone `SetSocketNoDelay()` function to a `virtual` (thus mockable) method `Sock::SetNoDelay()`.
This will help avoid syscalls during testing and to mock them to return whatever is suitable for the tests.
ACKs for top commit:
laanwj:
Code review ACK a2c4a7acd1dfb2fb7e3c9dac6b3d8c9354b2e0a6
jonatack:
ACK a2c4a7acd1dfb2fb7e3c9dac6b3d8c9354b2e0a6 change since last review is folding `Sock::SetNoDelay()` into the callers
Tree-SHA512: 3e2b016c1e4128317a28c17dc9b30472949e1ac3b071b2697c6d30cbcc830df1ee4392a4e23b2ea1ab4e3fb0f59ef450e2a4f3c1df3d8c803dd081652b6c7387
This helper uses spanparsing::Split internally and enables to replace
all calls to boost::split where only a single separator is passed.
Co-authored-by: Martin Ankerl <Martin.Ankerl@gmail.com>
Co-authored-by: MarcoFalke <falke.marco@gmail.com>
2ef47ba6c57a12840499a13908ab61aefca6cb55 util/check: stop using lambda for Assert/Assume (Anthony Towns)
7c9fe25c16d48b53a61fa2f6ff77eaf8820cb1f6 wallet: move Assert() check into constructor (Anthony Towns)
Pull request description:
Using a lambda creates a couple of odd namespacing issues, in particular making clang's thread safety analysis less helpful, and confusing gcc when calling member functions. Fix this by not using a lambda.
Fixes#21596Fixes#24654
ACKs for top commit:
MarcoFalke:
ACK 2ef47ba6c57a12840499a13908ab61aefca6cb55 🚢
jonatack:
Tested re-ACK 2ef47ba6c57a12840499a13908ab61aefca6cb55
Tree-SHA512: 4bdbf3215f3d14472df0552362c5eebe8b7eea2d0928a8a41109edd4e0c5f95de6f8220eb2fee8506874e352c003907faf5ef344174795939306a618157b1bae
f05a4cdf5a0363e1c12f00c034afb60e7ea0c775 util: Add inotify_rm_watch to syscall sandbox (AllowFileSystem) (Hennadii Stepanov)
Pull request description:
This PR fixes the current master (3297f5c11c72dd83479ff8335e047555e3f8cb3b) when running `bitcoin-qt` on Ubuntu 22.04 and quitting:
```
$ ./src/qt/bitcoin-qt -signet -sandbox=log-and-abort
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
ERROR: The syscall "inotify_rm_watch" (syscall number 255) is not allowed by the syscall sandbox in thread "main". Please report.
terminate called without an active exception
Aborted (core dumped)
```
Also see https://github.com/bitcoin/bitcoin/pull/24659#discussion_r835747166
ACKs for top commit:
fanquake:
ACK f05a4cdf5a0363e1c12f00c034afb60e7ea0c775 - checked that qt is using this in it's filesystem watcher code.
Tree-SHA512: 9c7920a25422cd3a040bc1cbc487c12c3dc2b91358c3757f1030d6a1ff12c18c688a8e5b7466f683da88a5e4f5f15d442975660022d706e47021253c24c58f4a
532c64a7264dd3c7329e8839547837c57da7dbe8 build: Fix Boost.Process test for Boost 1.78 (Hennadii Stepanov)
Pull request description:
Rebased #24415 with Luke's suggestion.
Fixes#24413.
ACKs for top commit:
hebasto:
ACK 532c64a7264dd3c7329e8839547837c57da7dbe8, tested on Mac mini (M1, 2020) + macOS Monterey 12.3 (21E230).
Tree-SHA512: 74f779695f6bbc45a2b7341a1402f747cc0d433d74825c7196cb9f156db0c0299895365f01665bd0bff12a8ebb5ea33a29b9a52f5eac0007ec35d1dca6544705
a84650ebd5ac2cbb49f14eb7c98736a3f8215bf1 util: Fix ReadBinaryFile reading beyond maxsize (klementtan)
Pull request description:
Currently `ReadBinaryFile` will read beyond `maxsize` if `maxsize` is not a multiple of `128` (size of buffer)
This is due to `fread` being called with `count = 128` instead of `count = min(128, maxsize - retval.size()` at every iteration
The following unit test will fail:
```cpp
BOOST_AUTO_TEST_CASE(util_ReadWriteFile)
{
fs::path tmpfolder = m_args.GetDataDirBase();
fs::path tmpfile = tmpfolder / "read_binary.dat";
std::string expected_text(300,'c');
{
std::ofstream file{tmpfile};
file << expected_text;
}
{
// read half the contents in file
auto [valid, text] = ReadBinaryFile(tmpfile, expected_text.size() / 2);
BOOST_CHECK_EQUAL(text.size(), 150);
}
}
```
Error:
```
test/util_tests.cpp:2593: error: in "util_tests/util_ReadWriteFile": check text.size() == 150 has failed [256 != 150]
```
ACKs for top commit:
laanwj:
Code review ACK a84650ebd5ac2cbb49f14eb7c98736a3f8215bf1
theStack:
Code-review ACK a84650ebd5ac2cbb49f14eb7c98736a3f8215bf1
Tree-SHA512: 752eebe58bc2102dec199b6775f8c3304d899f0ce36d6a022a58e27b076ba945ccd572858b19137b769effd8c6de73a9277f641be24dfb17657fb7173ea0eda0
Fix GUI startup crash reported by Rspigler in
https://github.com/bitcoin/bitcoin/issues/24457 that happens if
settings.json contains an integer value for any of the configuration
options which GUI settings can currently clash with (-dbcache, -par,
-spendzeroconfchange, -signer, -upnp, -natpmp, -listen, -server, -proxy,
-proxy, -onion, -onion, -lang, and -prune).
Fix is a one-line change in ArgsManager::GetArg.
Let GetPathArg method be used more places for path arguments that have
default values, like "-settings" and BITCOIN_SETTINGS_FILENAME in the
next commit.
Also:
- Fix negated argument handling. Return path{} not path{"0"} when path
argument is negated.
- Add new tests for default and negated cases
- Move GetPathArg() method declaration next to GetArg() declarations.
The two methods are close substitutes for each other, so this should
help keep them consistent and make them more discoverable.
fa1b89a6bdbab50bdb0504782afd4bb3375d1b57 scripted-diff: Rename nReadPos to m_read_pos in streams.h (MarcoFalke)
fa56c79df91e5d87533af38b64f4f4148a48a276 Make CDataStream work properly on 64-bit systems (MarcoFalke)
fab02f799194c75af7def3a2ab45c443b75de230 streams: Fix read-past-the-end and integer overflows (MarcoFalke)
Pull request description:
This is a follow-up to commit e26b62093ae21e89ed7d36a24a6b863f38ec631d with the following fixes:
* Fix unsigned integer overflow in `ignore()`, when `nReadPos` wraps.
* Fix unsigned integer overflow in `read()`, when `nReadPos` wraps.
* Fix read-past-the-end in `read()`, when `nReadPos` wraps.
This shouldn't be remote-exploitable, because it requires a stream of more than 1GB of size. However, it might be exploitable if the attacker controls the datadir (I haven't checked).
A unit test for the overflow in `ignore()` looks like following. It is left as an excercise to the reader to replace `foo.ignore(7)` with the appropriate call to `read()` to reproduce the overflow and read-error in `read()`.
```diff
diff --git a/src/test/coins_tests.cpp b/src/test/coins_tests.cpp
index 922fd8e513..ec6ea93919 100644
--- a/src/test/coins_tests.cpp
+++ b/src/test/coins_tests.cpp
@@ -534,6 +534,20 @@ BOOST_AUTO_TEST_CASE(ccoins_serialization)
} catch (const std::ios_base::failure&) {
}
+ CDataStream foo{0, 0};
+ auto size{std::numeric_limits<uint32_t>::max()};
+ foo.resize(size);
+ BOOST_CHECK_EQUAL(foo.size(), size);
+ foo.ignore(std::numeric_limits<int32_t>::max());
+ size -= std::numeric_limits<int32_t>::max();
+ BOOST_CHECK_EQUAL(foo.size(), size);
+ foo.ignore(std::numeric_limits<int32_t>::max());
+ size -= std::numeric_limits<int32_t>::max();
+ BOOST_CHECK_EQUAL(foo.size(), size);
+ BOOST_CHECK_EQUAL(foo.size(), 1);
+ foo.ignore(7); // Should overflow, as the size is only 1
+ BOOST_CHECK_EQUAL(foo.size(), uint32_t(1 - 7));
+
// Very large scriptPubKey (3*10^9 bytes) past the end of the stream
CDataStream tmp(SER_DISK, CLIENT_VERSION);
uint64_t x = 3000000000ULL;
```
ACKs for top commit:
klementtan:
Code Review ACK fa1b89a6bdbab50bdb0504782afd4bb3375d1b57:
Tree-SHA512: 67f0a1baafe88eaf1dc844ac55b638d5cf168a18c945e3bf7a2cb03c9a5976674a8e3af2487d8a2c3eae21e5c0e7a519c8b16ee7f104934442e2769d100660e9
dc01cbc538765f64326bca30952c83e3862d0d54 test: Add fs_tests/rename unit test (Hennadii Stepanov)
d4999d40b9bd04dc20111aaaa6ed2d3db1a5caf9 util: Revert back MoveFileExW call for MinGW-w64 (Hennadii Stepanov)
Pull request description:
Unfortunately, bitcoin/bitcoin#24308 introduced a [regression](https://github.com/bitcoin/bitcoin/pull/24308#issuecomment-1037259386) for mingw builds.
The root of the problem is a broken implementation of [`std::filesystem::rename`](https://en.cppreference.com/w/cpp/filesystem/rename). In particular, the expected behavior
> If `old_p` is a non-directory file, then `new_p` must be ... existing non-directory file: `new_p` _is first deleted_...
fails with the "File exists" error.
This PR reverts back the `MoveFileExW` call, and adds the [suggested](https://github.com/bitcoin/bitcoin/pull/24308#pullrequestreview-878832906) unit test.
ACKs for top commit:
vasild:
ACK dc01cbc538765f64326bca30952c83e3862d0d54
Tree-SHA512: c8e5a98844cfa32bec0ad67a1aaa58fe2efd0c5474d3e83490211985b110f83245758a742dcaa0a933a192ab66a7f11807e0c53ae69260b7dd02fc99f6d03849
Use std::filesystem::rename() instead of std::rename(). We rely on the
destination to be overwritten if it exists, but std::rename()'s behavior
is implementation-defined in this case.
