Keep track of which block is being requested (and to be requested) from
each peer, and limit the number of blocks in-flight per peer. In addition,
detect stalled downloads, and disconnect if they persist for too long.
This means blocks are never requested twice, and should eliminate duplicate
downloads during synchronization.
In case the total number of orphan blocks in memory exceeds a limit
(currently set to 750), a random orphan block (which is not
depended on by another orphan block) is dropped. This means it will
need to be downloaded again, but it won't consume memory until then.
This changes the block processing logic from "try to atomically switch
to a new block" to a continuous "(dis)connect a block, aiming for the
assumed best chain".
This means the smallest atomic operations on the chainstate become
individual block connections or disconnections, instead of entire
reorganizations. It may mean that we try to reorganize to one block,
fail, and rereorganize again to the old block. This is slower, but
doesn't require unbounded RAM.
It also means that a ConnectBlock which fails may be no longer called
from the ProcessBlock which knows which node sent it. To deal with that,
a mapBlockSource is kept, and invalid blocks cause asynchronous "reject"
messages and banning (if necessary).
caca6aa Make some globals in main non-public. (Pieter Wuille)
85eb2ce Do not use the redundant BestInvalidWork record in the block database. (Pieter Wuille)
As block index entries have a flag for marking invalid blocks, the
'best invalid work' information can be derived from there. In addition,
remove the global from main.h
Instead of explicitly testing for the presence of any output, and
dealing with this case specially, just interpret it as an empty
CCoins.
The case previously caught using the HaveCoins check, is now handled
by the generic outs != outsBlock test.
This seems to cause problems on recent clang, and looks totally
redundant and unused.
The const_iterator version is identical to the vector::const_iterator
one (which is a typedef thereof). Marking it private (instead of
removing) compiles fine, so this version is effectively unused even.
This required some code movement (what was CWalletTx::AcceptToMemoryPool
doing in main?), and adding a few explicit includes that used to be
implicit through init.h.
SendMessages() tries to acquire a cs_main lock now, but this isn't nessecary
for much of its functionality. Move those parts out of the locked section,
so they can always be performed, and we hold cs_main for a shorter time.
This removes a few unused CBlockLocator methods, and moves the
construction and fork-finding logic to CChain (which can do these
more efficiently, as it has a height-indexable chain available).
It also makes CBlockLocator independent from the validation code.
971bb3e Added ping time measurement. New RPC "ping" command to request ping. Implemented "pong" message handler. New "pingtime" field in getpeerinfo, to provide results to user. New "pingwait" field, to show pings still in flight, to better see newly lagging peers. (Josh Lehan)
Instead of building a full copy of a CTransaction being signed, and
then modifying bits and pieces until its fits the form necessary
for computing the signature hash, use a wrapper serializer that
only serializes the necessary bits on-the-fly.
This makes it easier to see which data is actually being hash,
reduces load on the heap, and also marginally improves performances
(around 3-4us/sigcheck here). The performance improvements are much
larger for large transactions, though.
The old implementation of SignatureHash is moved to a unit tests,
to test whether the old and new algorithm result in the same value
for randomly-constructed transactions.
a02ddf9 Added GNU/kFreeBSD kernel name (TARGET_OS)
8487468 CondVar::SignalAll was broken, leading to deadlocks on Windows builds. http://code.google.com/p/leveldb/issues/detail?id=149
f6d84d1 Allow files to be opened for reading multiple times
cb8e3f7 Checking whether closing succeeds
d5317e8 Print actual Win32 error that occurred on file creation failure.
907f308 Port leveldb to MinGW32
9def2bf Mingw support for Windows LevelDB port
0a7b074 Pre-Vista leveldb::port::InitOnce implementation
31a2b09 Native Windows LevelDB port
058a035 Remove Snappy support
5bd76dc Release leveldb 1.12
7b094f1 Release leveldb 1.11
28dad91 Release leveldb 1.10
514c943 Make DB::Open fail if sst files are missing.
d84c825 Fix corruption bug found and analyzed by dhruba@gmail.com
ea2e919 added utility to dump leveldb files
REVERT: ae6c262 Merge branch 'leveldb' into ripple-fork
REVERT: 28fa222 Looks like a bit more delay is needed to smooth the latency.
REVERT: a18f3e6 Tidy up JobQueue, add ripple_core module
REVERT: ab82e57 Release leveldb 1.12
REVERT: 02c6259 Release leveldb 1.11
REVERT: 5bbb544 Rate limit compactions with a 25ms pause after each complete file.
REVERT: 8c29c47 LevelDB issue 178 fix: cannot resize a level 0 compaction set
REVERT: 18b245c Added GNU/kFreeBSD kernel name (TARGET_OS)
REVERT: 8be9d12 CondVar::SignalAll was broken, leading to deadlocks on Windows builds. http://code.google.com/p/leveldb/issues/detail?id=149
REVERT: c9fc070 Upgrade LevelDB to 1.10.0, mostly for better write stall logging.
REVERT: 8215b15 Tweak to variable name to support unity build
REVERT: aca1ffc Allow files to be opened for reading multiple times
REVERT: 693a70c Checking whether closing succeeds
REVERT: 0144d04 Print actual Win32 error that occurred on file creation failure.
REVERT: 43ed517 Fix corruption bug found and analyzed by dhruba@gmail.com
REVERT: 413c74c added utility to dump leveldb files
REVERT: 96eda85 Port leveldb to MinGW32
REVERT: 0967260 Mingw support for Windows LevelDB port
REVERT: ee3f9bd Pre-Vista leveldb::port::InitOnce implementation
REVERT: f5d0a41 Native Windows LevelDB port
REVERT: 28b35f1 Remove Snappy support
git-subtree-dir: src/leveldb
git-subtree-split: a02ddf9b14d145e88185ee209ab8b01d8826663a
To fix a minor malleability found by Sergio Lerner (reported here:
https://bitcointalk.org/index.php?topic=8392.msg1245898#msg1245898)
The problem is that if (R,S) is a valid ECDSA signature for a given
message and public key, (R,-S) is also valid. Modulo N (the order
of the secp256k1 curve), this means that both (R,S) and (R,N-S) are
valid. Given that N is odd, S and N-S have a different lowest bit.
We solve the problem by forcing signatures to have an even S value,
excluding one of the alternatives.
This commit just changes the signing code to always produce even S
values, and adds a verification mode to check it. This code is not
enabled anywhere yet. Existing tests in key_tests.cpp verify that
the produced signatures are still valid.
The key refactor changed the way unencrypted private keys with compressed
public key are stored in the wallet. Apparently older versions relied on
this to verify the correctness of stored keys.
Note that earlier pre-release versions do risk creating wallets that can
not be opened by 0.8.3 and earlier.
dumpwallet: produce a dump of all keys in a wallet, in a format
compatible with Bitcoin Wallet for Android and Multibit.
importwallet: import such a dump
Compute safe lower bounds on the birth times of all wallet keys. For
pool keys or keys with metadata, the actually stored birth time is
used. For all others, the birth times are inferred from the wallet
transactions.
This function finds all keys affected by a particular output script,
supporting everything ExtractDestinations supports (pay-to-pubkey,
pay-to-pubkeyhash, multisig) and recurses into subscripts (P2SH).
In case no database exists yet, and -txindex(=1) is passed, we currently first
check whether fTxIndex differs from -txindex (and ask the user to reindex in
that case), and only afterwards initialize the database. By swapping these
around (the initialization is a no-op in case the database already exists),
we allow it to be born in txindex mode, without warning.
That also means we don't need to check -reindex anymore, as the wiping/reinit
of the databases happens before checking.
Refactor keytime:
* Key metadata is kept in a CWallet::mapKeyMetadata (std::map<CKeyId,CKeyMetadata>).
* When generating a new key, time is put in that map, and new key is written.
* AddKeyPubKey and AddCryptedKey do not take a creation time argument, but instead
pull it from that map, if it exists there.
Bugfix:
* AddKeyPubKey and AddCryptedKey in CWallet didn't override the CKeyStore
definition anymore. This is fixed, as they no longed need the nCreationTime
argument now.
Also a few related other changes:
* Metadata can be overwritten.
* Only GenerateNewKey calls GetTime(), as it's the only place where we know for
sure a key was not constructed earlier.
* When the nTimeFirstKey is known to be inaccurate, it is set to the value 1
(instead of 0, which would mean unknown).
* Use CPubKey instead of std::vector<unsigned char> where possible.
Pieter Wuille