Opened 11 years ago
Closed 11 years ago
#6006 closed Bugs (fixed)
Potential epoll fd leak in boost::asio::detail::epoll_reactor
Reported by: | Owned by: | chris_kohlhoff | |
---|---|---|---|
Milestone: | To Be Determined | Component: | asio |
Version: | Boost 1.40.0 | Severity: | Problem |
Keywords: | Cc: |
Description
in boost/asio/detail/epoll_reactor.hpp:
Constructor. epoll_reactor(boost::asio::io_service& io_service)
: boost::asio::detail::service_base<epoll_reactor<Own_Thread> >(io_service),
mutex_(), epoll_fd_(do_epoll_create()), ------> epoll_fd_ may leak if the following member's initialization throws an exception wait_in_progress_(false), interrupter_(), ------> may throw because of EMFILE read_op_queue_(), write_op_queue_(), except_op_queue_(), pending_cancellations_(), stop_thread_(false), thread_(0), shutdown_(false), need_epoll_wait_(true)
{
Start the reactor's internal thread only if needed. if (Own_Thread) {
boost::asio::detail::signal_blocker sb; thread_ = new boost::asio::detail::thread(
bind_handler(&epoll_reactor::call_run_thread, this));
}
Add the interrupter's descriptor to epoll. epoll_event ev = { 0, { 0 } }; ev.events = EPOLLIN | EPOLLERR; ev.data.fd = interrupter_.read_descriptor(); epoll_ctl(epoll_fd_, EPOLL_CTL_ADD, interrupter_.read_descriptor(), &ev);
}
and in 1.47.0, there is a similar situation: epoll_fd_ and timer_fd_ may leak.
Attachments (2)
Change History (5)
by , 11 years ago
Attachment: | epoll_reactor.hpp.patch_1.40.0 added |
---|
by , 11 years ago
Attachment: | fd_leak_test.cpp added |
---|
comment:1 by , 11 years ago
comment:2 by , 11 years ago
comment:3 by , 11 years ago
Resolution: | → fixed |
---|---|
Status: | new → closed |
(In [75009]) Merge from trunk:
- Make number of strand implementations configurable by defining BOOST_ASIO_STRAND_IMPLEMENTATIONS to the number.
- Programs can now define BOOST_ASIO_ENABLE_SEQUENTIAL_STRAND_ALLOCATION to switch the allocation of strand implementations to use a round-robin approach rather than hashing.
- Fix potential strand starvation issue that can occur when strand.post() is used.
- Update descriptor state allocation in kqueue_reactor to match approach used in epoll_reactor.
- Construct epoll_reactor's interrupter member first to fix exception safety issue. Fixes #6006
- Clarify that the read operation ends when the buffer is full. Fixes #5999
my patch only works, but it is not graceful enough. wrapping epoll_fd and timer_fd or other tricks?
fd_leak_test.cpp is a program reproducing the leak.