Skip to content

"SocketReactor::addEventHandler" and "SocketReactor::removeEventHandler" must protect the access to "NotifierPtr pNotifier" #1709

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 10, 2017

Conversation

sebastien-guay
Copy link

@sebastien-guay sebastien-guay commented May 1, 2017

This change is related to this issue: #1708

Those two methods already use FastMutex::ScopedLock lock(_mutex), but
the scope is not large enough to protect "NotifierPtr pNotifier" that is
accessed by calling SocketNotifier::addObserver() and
SocketNotifier::removeObserver().

It is mentioned in SocketReator.h that it is safe to call
addEventHandler() and removeEventHandler() from another thread while the
SocketReactor is running. My current use of the SocketReactor
encountered an issue where the SocketNotifier::_events has been
corrupted by a concurent write access done by
SocketReactor::addEventHandler() and
SocketReactor::removeEventHandler().
The call stack show that the SocketReactor::addEventHandler is stuck in
a while loop in gcc/libstdc++/tree.cc Rb_tree_insert_and_rebalance()

I clearly see in my logs that it happened while my
SocketConnector::unregisterConnector() and
SocketConnector::registerConnector() were called by two different
threads.

#0 0x00a80a7b in std::_Rb_tree_insert_and_rebalance () from
/usr/lib/libstdc++.so.6
#1 0x06ccb430 in std::_Rb_tree<Poco::Net::SocketNotification*,
Poco::Net::SocketNotification*,
std::_IdentityPoco::Net::SocketNotification*,
std::lessPoco::Net::SocketNotification*,
std::allocatorPoco::Net::SocketNotification* >::_M_insert
(this=0xac75dc90, __x=0x0, __p=0xac7ed0c8, __v=@0xb5fb0c40) at
/usr/lib/gcc/i386-redhat-linux/3.4.6/../../../../include/c++/3.4.6/bits/stl_tree.h:816
#2 0x06ccb15d in std::_Rb_tree<Poco::Net::SocketNotification*,
Poco::Net::SocketNotification*,
std::_IdentityPoco::Net::SocketNotification*,
std::lessPoco::Net::SocketNotification*,
std::allocatorPoco::Net::SocketNotification* >::insert_equal
(this=0xac75dc90, __v=@0xb5fb0c40) at
/usr/lib/gcc/i386-redhat-linux/3.4.6/../../../../include/c++/3.4.6/bits/stl_tree.h:858
#3 0x06ccad86 in std::multiset<Poco::Net::SocketNotification*,
std::lessPoco::Net::SocketNotification*,
std::allocatorPoco::Net::SocketNotification* >::insert (this=0xac75dc90,
__x=@0xb5fb0c40)
at
/usr/lib/gcc/i386-redhat-linux/3.4.6/../../../../include/c++/3.4.6/bits/stl_multiset.h:306
#4 0x06cca72b in Poco::Net::SocketNotifier::addObserver
(this=0xac75dc70, pReactor=0x8d27958, observer=@0xb5fb0cd0) at
src/SocketNotifier.cpp:45
#5 0x06cd060f in Poco::Net::SocketReactor::addEventHandler
(this=0x8d27958, socket=@0xac7f5a74, observer=@0xb5fb0cd0) at
src/SocketReactor.cpp:178

SocketReactor::run() is already protecting the access to the
SocketNotifier.

…er" must protect the access to "NotifierPtr pNotifier"

Those two methods already use FastMutex::ScopedLock lock(_mutex), but
the scope is not large enough to protect "NotifierPtr pNotifier" that is
accessed by calling SocketNotifier::addObserver() and
SocketNotifier::removeObserver().

It is mentioned in SocketReator.h that it is safe to call
addEventHandler() and removeEventHandler() from another thread while the
SocketReactor is running. My current use of the SocketReactor
encountered an issue where the SocketNotifier::_events has been
corrupted by a concurent write access done by
SocketReactor::addEventHandler() and
SocketReactor::removeEventHandler().
The call stack show that the SocketReactor::addEventHandler is stuck in
a while loop in gcc/libstdc++/tree.cc Rb_tree_insert_and_rebalance()

I clearly see in my logs that it happened while my
SocketConnector::unregisterConnector() and
SocketConnector::registerConnector() were called by two different
threads.

#0 0x00a80a7b in std::_Rb_tree_insert_and_rebalance () from
/usr/lib/libstdc++.so.6
pocoproject#1 0x06ccb430 in std::_Rb_tree<Poco::Net::SocketNotification*,
Poco::Net::SocketNotification*,
std::_IdentityPoco::Net::SocketNotification*,
std::lessPoco::Net::SocketNotification*,
std::allocatorPoco::Net::SocketNotification* >::_M_insert
(this=0xac75dc90, __x=0x0, __p=0xac7ed0c8, __v=@0xb5fb0c40) at
/usr/lib/gcc/i386-redhat-linux/3.4.6/../../../../include/c++/3.4.6/bits/stl_tree.h:816
pocoproject#2 0x06ccb15d in std::_Rb_tree<Poco::Net::SocketNotification*,
Poco::Net::SocketNotification*,
std::_IdentityPoco::Net::SocketNotification*,
std::lessPoco::Net::SocketNotification*,
std::allocatorPoco::Net::SocketNotification* >::insert_equal
(this=0xac75dc90, __v=@0xb5fb0c40) at
/usr/lib/gcc/i386-redhat-linux/3.4.6/../../../../include/c++/3.4.6/bits/stl_tree.h:858
pocoproject#3 0x06ccad86 in std::multiset<Poco::Net::SocketNotification*,
std::lessPoco::Net::SocketNotification*,
std::allocatorPoco::Net::SocketNotification* >::insert (this=0xac75dc90,
__x=@0xb5fb0c40)
at
/usr/lib/gcc/i386-redhat-linux/3.4.6/../../../../include/c++/3.4.6/bits/stl_multiset.h:306
pocoproject#4 0x06cca72b in Poco::Net::SocketNotifier::addObserver
(this=0xac75dc70, pReactor=0x8d27958, observer=@0xb5fb0cd0) at
src/SocketNotifier.cpp:45
pocoproject#5 0x06cd060f in Poco::Net::SocketReactor::addEventHandler
(this=0x8d27958, socket=@0xac7f5a74, observer=@0xb5fb0cd0) at
src/SocketReactor.cpp:178

SocketReactor::run() is already protecting the access to the
SocketNotifier.
@aleks-f aleks-f merged commit a9982e6 into pocoproject:develop Aug 10, 2017
@romanj85
Copy link

Hi, this wasn't successfully merged into branch, because CI tests failed. So it isn't fixed in Release 1.8.0 (2017-11-10) as you mentioned in Changelog (https://raw.githubusercontent.com/pocoproject/poco/poco-1.9.0-release/CHANGELOG). But I saw in branch pocoproject:develop that it is completely reworked the way how to store and notify events...

@aleks-f
Copy link
Member

aleks-f commented Jul 19, 2018

Socket notifier was not thread-safe to start with (see #2345), and there were other major changes to reactor (most notably, using PollSet to improve performance), fix will be in 1.9.1

For future contributions, keep in mind that develop (future 2.0) and 1.x are different lines of development, so any fix you want to see in 1.x releases should also be submitted there. we try to keep it synced, but sometimes things slip and not everything is easy or straightforward to move between branches

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants