Two reasons for this change:
1. Need to always use boost::thread's sleep, even on Windows, so the
sleeps can be interrupted (prior code used Windows' built-in Sleep).
2. I always forgot what units the old Sleep took.
Create a boost::thread_group object at the qt/bitcoind main-loop level
that will hold pointers to all the main-loop threads.
This will replace the vnThreadsRunning[] array.
For testing, ported the BitcoinMiner threads to use its
own boost::thread_group.
This fixes test_bitcoin failures on openbsd reported by dhill on IRC.
On some systems rand() is a simple LCG over 2^31 and so it produces
an even-odd sequence. ApproximateBestSubset was only using the least
significant bit and so every run of the iterative solver would be the
same for some inputs, resulting in some pretty dumb decisions.
Using something other than the least significant bit would paper over
the issue but who knows what other way a system's rand() might get us
here. Instead we use an internal RNG with a period of something like
2^60 which is well behaved. This also makes it possible to make the
selection deterministic for the tests, if we wanted to implement that.
Initialize the OutputDebugStringF mutex and file pointer using
boost::call_once, to be thread-safe.
Make the return value of OutputDebugStringF really be the number of
characters written (*printf() semantics).
Declare the fReopenDebugLog flag volatile, since it is changed from
a signal handler.
And don't declare OutputDebugStringF() as inline.
When the coin database is out of date with the block database, the
best block in it is automatically switched to. This reconnection
process can take time, so allow it to be interrupted.
This also stops block connection as soon as shutdown is requested,
leading to a faster shutdown.
Introduce a AllocateFileRange() function in util, which wipes or
at least allocates a given range of a file. It can be overriden
by more efficient OS-dependent versions if necessary.
Block and undo files are now allocated in chunks of 16 and 1 MiB,
respectively.
NOTE: These thread names are visible in gdb when using 'info threads'.
Additionally both 'top' and 'ps' show these names *unless* told to
display the command-line instead of task name.
Signed-off-by: Giel van Schijndel <me@mortis.eu>
Prior to this change, each TX typically generated 3+ debug messages,
askfor tx 8644cc97480ba1537214 0
sending getdata: tx 8644cc97480ba1537214
askfor tx 8644cc97480ba1537214 1339640761000000
askfor tx 8644cc97480ba1537214 1339640881000000
CTxMemPool::accept() : accepted 8644cc9748 (poolsz 6857)
After this change, there is only one message for each valid TX received
CTxMemPool::accept() : accepted 22a73c5d8c (poolsz 42)
and two messages for each orphan tx received
ERROR: FetchInputs() : 673dc195aa mempool Tx prev not found 1e439346fc
stored orphan tx 673dc195aa (mapsz 19)
The -debugnet option, or its superset -debug, will restore the full debug
output.
The best log rotation method formerly available was to configure
logrotate with the copytruncate option. As described in the logrotate
documentation, "there is a very small time slice between copying the
file and truncating it, so some logging data might be lost".
By sending SIGHUP to the server process, one can now reopen the debug
log file without losing any data.
This commit removes the dependency of serialize.h on PROTOCOL_VERSION,
and makes this parameter required instead of implicit. This is much saner,
as it makes the places where changing a version number can have an
influence obvious.
Reference miner exists for testnet-in-a-box type situations, and as a
reference. We don't care enough about highly optimized internal
mining to keep workarounds like this.
Where possible, use boost::filesystem::path instead of std::string or
char* for filenames. This avoids a lot of manual string tinkering, in
favor of path::operator/.
GetDataDir is also reworked significantly, it now only keeps two cached
directory names (the network-specific data dir, and the root data dir),
which are decided through a parameter instead of pre-initialized global
variables.
Finally, remove the "upgrade from 0.1.5" case where a debug.log in the
current directory has to be removed.
This commit simplifies the locking system: CCriticalSection becomes a
simple typedef for boost::interprocess::interprocess_recursive_mutex,
and CCriticalBlock and CTryCriticalBlock are replaced by a templated
CMutexLock, which wraps boost::interprocess::scoped_lock.
By making the lock type a template parameter, some critical sections
can now be changed to non-recursive locks, which support waiting via
condition variables. These are implemented in CWaitableCriticalSection
and WAITABLE_CRITICAL_BLOCK.
CWaitableCriticalSection is a wrapper for a different Boost mutex,
which supports waiting/notification via condition variables. This
should enable us to remove much of the used polling code. Important
is that this mutex is not recursive, so functions that perform the
locking must not call eachother.
Because boost::interprocess::scoped_lock does not support assigning
and copying, I had to revert to the older CRITICAL_BLOCK macros that
use a nested for loop instead of a simple if.
2^31 milliseconds is only about 25 days. Also clamps Sleep() to 10 years,
because it currently sleeps for 0 seconds when the sleep time would cross
2^31 seconds since the epoch. Hopefully boost will be fixed by 2028.
This introduces CNetAddr and CService, respectively wrapping an
(IPv6) IP address and an IP+port combination. This functionality used
to be part of CAddress, which also contains network flags and
connection attempt information. These extra fields are however not
always necessary.
These classes, along with logic for creating connections and doing
name lookups, are moved to netbase.{h,cpp}, which does not depend on
headers.h.
Furthermore, CNetAddr is mostly IPv6-ready, though IPv6
functionality is not yet enabled for the application itself.
Remembering all time samples makes nTimeOffset slow to respond to
system clock corrections. For instance, I start my node with a system
clock that's 30 minutes slow and run it for a few days. During that
time, I accumulate 10,000 offset samples with a median of 1800
seconds. Now I correct my system clock. Without this change, my node
must collect another 10,000 samples before nTimeOffset is correct
again. With this change, I must only accumulate 100 samples to
correct the offset.
Storing unlimited time samples also allows an attacker with many IP
addresses (ex, a large botnet) to perform a memory exhaustion attack
against Bitcoin nodes. The attacker sends a version message from each
IP to his target, consuming more of the target's memory each time.
Time samples are small, so this attack might be impractical under the
old code, but it's impossible with the new code.
SecureString is identical to std::string except with secure_allocator
substituting for std::allocator. This makes casting between them
impossible, so converting between the two at API boundaries requires
calling ::c_str() for now.
Replaced all occurrences of #if* __WXMSW__ with WIN32,
and all occurrences of __WXMAC_OSX__ with MAC_OSX, and made
sure those are defined appropriately in the makefile and bitcoin-qt.pro.
