Fix potential deadlock (#5124)

* 2.2.2 changed functions acquireAsync and NetworkOPsImp::recvValidation to add an item to a collection under lock, unlock, do some work, then lock again to do remove the item. It will deadlock if an exception is thrown while adding the item - before unlocking.
* Replace ScopedUnlock with scope_unlock.
This commit is contained in:
Bronek Kozicki
2024-11-06 22:22:42 +00:00
committed by GitHub
parent 8e827e32ac
commit 9e48fc0c83
4 changed files with 77 additions and 89 deletions

View File

@@ -21,6 +21,7 @@
#define RIPPLE_BASICS_SCOPE_H_INCLUDED
#include <exception>
#include <mutex>
#include <type_traits>
#include <utility>
@@ -186,6 +187,70 @@ public:
template <class EF>
scope_success(EF) -> scope_success<EF>;
/**
Automatically unlocks and re-locks a unique_lock object.
This is the reverse of a std::unique_lock object - instead of locking the
mutex for the lifetime of this object, it unlocks it.
Make sure you don't try to unlock mutexes that aren't actually locked!
This is essentially a less-versatile boost::reverse_lock.
e.g. @code
std::mutex mut;
for (;;)
{
std::unique_lock myScopedLock{mut};
// mut is now locked
... do some stuff with it locked ..
while (xyz)
{
... do some stuff with it locked ..
scope_unlock unlocker{myScopedLock};
// mut is now unlocked for the remainder of this block,
// and re-locked at the end.
...do some stuff with it unlocked ...
} // mut gets locked here.
} // mut gets unlocked here
@endcode
*/
template <class Mutex>
class scope_unlock
{
std::unique_lock<Mutex>* plock;
public:
explicit scope_unlock(std::unique_lock<Mutex>& lock) noexcept(true)
: plock(&lock)
{
assert(plock->owns_lock());
plock->unlock();
}
// Immovable type
scope_unlock(scope_unlock const&) = delete;
scope_unlock&
operator=(scope_unlock const&) = delete;
~scope_unlock() noexcept(true)
{
plock->lock();
}
};
template <class Mutex>
scope_unlock(std::unique_lock<Mutex>&) -> scope_unlock<Mutex>;
} // namespace ripple
#endif

View File

@@ -25,9 +25,11 @@
#include <xrpld/perflog/PerfLog.h>
#include <xrpl/basics/DecayingSample.h>
#include <xrpl/basics/Log.h>
#include <xrpl/basics/scope.h>
#include <xrpl/beast/container/aged_map.h>
#include <xrpl/beast/core/LexicalCast.h>
#include <xrpl/protocol/jss.h>
#include <exception>
#include <memory>
#include <mutex>
@@ -139,7 +141,7 @@ public:
if (pendingAcquires_.contains(hash))
return;
pendingAcquires_.insert(hash);
lock.unlock();
scope_unlock unlock(lock);
acquire(hash, seq, reason);
}
catch (std::exception const& e)
@@ -154,7 +156,6 @@ public:
<< "Unknown exception thrown for acquiring new inbound ledger "
<< hash;
}
lock.lock();
pendingAcquires_.erase(hash);
}

View File

@@ -46,10 +46,12 @@
#include <xrpl/basics/UptimeClock.h>
#include <xrpl/basics/contract.h>
#include <xrpl/basics/safe_cast.h>
#include <xrpl/basics/scope.h>
#include <xrpl/protocol/BuildInfo.h>
#include <xrpl/protocol/HashPrefix.h>
#include <xrpl/protocol/digest.h>
#include <xrpl/resource/Fees.h>
#include <algorithm>
#include <cassert>
#include <chrono>
@@ -60,86 +62,6 @@
namespace ripple {
namespace {
//==============================================================================
/**
Automatically unlocks and re-locks a unique_lock object.
This is the reverse of a std::unique_lock object - instead of locking the
mutex for the lifetime of this object, it unlocks it.
Make sure you don't try to unlock mutexes that aren't actually locked!
This is essentially a less-versatile boost::reverse_lock.
e.g. @code
std::mutex mut;
for (;;)
{
std::unique_lock myScopedLock{mut};
// mut is now locked
... do some stuff with it locked ..
while (xyz)
{
... do some stuff with it locked ..
ScopedUnlock unlocker{myScopedLock};
// mut is now unlocked for the remainder of this block,
// and re-locked at the end.
...do some stuff with it unlocked ...
} // mut gets locked here.
} // mut gets unlocked here
@endcode
*/
template <class MutexType>
class ScopedUnlock
{
std::unique_lock<MutexType>& lock_;
public:
/** Creates a ScopedUnlock.
As soon as it is created, this will unlock the unique_lock, and
when the ScopedLock object is deleted, the unique_lock will
be re-locked.
Make sure this object is created and deleted by the same thread,
otherwise there are no guarantees what will happen! Best just to use it
as a local stack object, rather than creating on the heap.
*/
explicit ScopedUnlock(std::unique_lock<MutexType>& lock) : lock_(lock)
{
assert(lock_.owns_lock());
lock_.unlock();
}
ScopedUnlock(ScopedUnlock const&) = delete;
ScopedUnlock&
operator=(ScopedUnlock const&) = delete;
/** Destructor.
The unique_lock will be locked after the destructor is called.
Make sure this object is created and deleted by the same thread,
otherwise there are no guarantees what will happen!
*/
~ScopedUnlock() noexcept(false)
{
lock_.lock();
}
};
} // namespace
// Don't catch up more than 100 ledgers (cannot exceed 256)
static constexpr int MAX_LEDGER_GAP{100};
@@ -1336,7 +1258,7 @@ LedgerMaster::findNewLedgersToPublish(
auto valLedger = mValidLedger.get();
std::uint32_t valSeq = valLedger->info().seq;
ScopedUnlock sul{sl};
scope_unlock sul{sl};
try
{
for (std::uint32_t seq = pubSeq; seq <= valSeq; ++seq)
@@ -1882,7 +1804,7 @@ LedgerMaster::fetchForHistory(
InboundLedger::Reason reason,
std::unique_lock<std::recursive_mutex>& sl)
{
ScopedUnlock sul{sl};
scope_unlock sul{sl};
if (auto hash = getLedgerHashForHistory(missing, reason))
{
assert(hash->isNonZero());
@@ -2052,7 +1974,7 @@ LedgerMaster::doAdvance(std::unique_lock<std::recursive_mutex>& sl)
for (auto const& ledger : pubLedgers)
{
{
ScopedUnlock sul{sl};
scope_unlock sul{sl};
JLOG(m_journal.debug())
<< "tryAdvance publishing seq " << ledger->info().seq;
setFullLedger(ledger, true, true);
@@ -2061,7 +1983,7 @@ LedgerMaster::doAdvance(std::unique_lock<std::recursive_mutex>& sl)
setPubLedger(ledger);
{
ScopedUnlock sul{sl};
scope_unlock sul{sl};
app_.getOPs().pubLedger(ledger);
}
}

View File

@@ -53,6 +53,7 @@
#include <xrpl/basics/UptimeClock.h>
#include <xrpl/basics/mulDiv.h>
#include <xrpl/basics/safe_cast.h>
#include <xrpl/basics/scope.h>
#include <xrpl/beast/rfc2616.h>
#include <xrpl/beast/utility/rngfill.h>
#include <xrpl/crypto/RFC1751.h>
@@ -2310,7 +2311,7 @@ NetworkOPsImp::recvValidation(
bypassAccept = BypassAccept::yes;
else
pendingValidations_.insert(val->getLedgerHash());
lock.unlock();
scope_unlock unlock(lock);
handleNewValidation(app_, val, source, bypassAccept, m_journal);
}
catch (std::exception const& e)
@@ -2327,10 +2328,9 @@ NetworkOPsImp::recvValidation(
}
if (bypassAccept == BypassAccept::no)
{
lock.lock();
pendingValidations_.erase(val->getLedgerHash());
lock.unlock();
}
lock.unlock();
pubValidation(val);