boost::interprocess message queue Race Condition on creation?

Question

I am trying to debug sporadic access violations that occur inside a boost::interprocess message queue. (access violation reading an address in the shared memory region).

Environment: boost 1.54, VC++2010. Occurs in both Debug & Release builds.

It always occurs on or about line 854 (in case of reception) in message_queue.hpp: Comments were added by me

      recvd_size     = top_msg.len; // top_msg points to invalid location

Or line 756 (in case of sending)

BOOST_ASSERT(free_msg_hdr.priority == 0); // free_msg_hdr points to invalid location

It appears as though this is related to the message queue creation. If a message queue is created "properly" (i.e. without the possible race condition), the error never occurs. Otherwise it might occur on timed_receive() or timed_send() on the queue at seemingly random times.

I came up with a short example that represents the problem: Unfortunately I cannot run it on Coliru, since it requires two processes. One has to be started without any parameters, the second with any single parameter. After a number of runs, one of the processes will crash in message_queue.

#include <iostream>
#include <boost/interprocess/ipc/message_queue.hpp>
#include <boost/thread.hpp>
#include <boost/assert.hpp>
#include <boost/date_time.hpp>

using namespace boost::interprocess;
using namespace boost::posix_time;
using boost::posix_time::microsec_clock; // microsec_clock is ambiguous between boost::posix_time and boost::interprocess. What are the odds?

int main(int argc, wchar_t** argv)
{
    while(true)
    {
        int proc = 0;
        message_queue* queues[2] = {NULL, NULL};
        std::string names[] = {"msgq0", "msgq1"};
        if(1 == argc)
        {
            proc = 0;
            message_queue::remove(names[0].c_str());
            if(NULL != queues[0]) { delete queues[0]; queues[0] = NULL; }
            queues[0] = new message_queue(open_or_create, names[0].c_str(), 128, 10240);

            bool bRet = false;
            do
            {
                try
                {
                    if(NULL != queues[1]) { delete queues[1]; queues[1] = NULL; }
                    queues[1]=new message_queue(open_only, names[1].c_str());
                    bRet = true;
                }
                catch(const interprocess_exception&)
                {
                    //boost::this_thread::sleep(boost::posix_time::milliseconds(2));
                    delete queues[1];
                    queues[1] = NULL; 
                    continue;
                }
            }while(!bRet);

        }
        else
        {
            proc = 1;
            message_queue::remove(names[1].c_str());
            if(NULL != queues[1]) { delete queues[1]; queues[1] = NULL; }
            queues[1] = new message_queue(open_or_create, names[1].c_str(), 128, 10240);

            bool bRet = false;
            do
            {
                try
                {
                    if(NULL != queues[0]) { delete queues[0]; queues[0] = NULL; }
                    queues[0]=new message_queue(open_only, names[0].c_str());
                    bRet = true;
                }
                catch(const interprocess_exception&)
                {
                    //boost::this_thread::sleep(boost::posix_time::milliseconds(2));
                    delete queues[0];
                    queues[0] = NULL;
                    continue;
                }
            }while(!bRet);
        }

        long long nCnt = 0;
        for(int i = 0; i < 1; ++i)
        {
            if(proc)
            {
                std::string sOut;
                sOut = "Proc1 says: Hello ProcA " + std::to_string(nCnt) + " ";
                sOut.resize(10230, ':');
                for(int n = 0; n < 3; ++n)
                {
                    queues[1]->timed_send(sOut.data(), sOut.size(), 0, ptime(boost::posix_time::microsec_clock::universal_time()) + milliseconds(1));
                }

                bool bMessage = false;
                for(int n = 0; n < 3; ++n)
                {
                    size_t nRec; unsigned int nPrio;
                    std::string sIn; sIn.resize(10240);
                    bMessage = queues[0]->timed_receive(&sIn[0], 10240, nRec, nPrio, ptime(boost::posix_time::microsec_clock::universal_time()) + milliseconds(1));
                    if(bMessage)
                    {
                        sIn.resize(nRec);
                        //std::cout << sIn << " ";
                    }
                }
                if(bMessage)
                {
                    //std::cout << std::endl;
                }
            }
            else
            {
                std::string sOut;
                sOut = "Proc0 says: Hello Procccccccdadae4325a " + std::to_string(nCnt);
                sOut.resize(10240, '.');
                for(int n = 0; n < 3; ++n)
                {
                    queues[0]->timed_send(sOut.data(), sOut.size(), 0, ptime(boost::posix_time::microsec_clock::universal_time()) + milliseconds(1));
                }

                bool bMessage = false;
                for(int n = 0; n < 3; ++n)
                {
                    size_t nRec; unsigned int nPrio;
                    std::string sIn; sIn.resize(10240);
                    bMessage = queues[1]->timed_receive(&sIn[0], 10240, nRec, nPrio, ptime(boost::posix_time::microsec_clock::universal_time()) + milliseconds(1));
                    if(bMessage)
                    {
                        sIn.resize(nRec);
                        //std::cout << sIn << " ";
                    }
                }
                if(bMessage)
                {
                    //std::cout << std::endl;
                }
            }

            nCnt++;
            boost::this_thread::sleep(boost::posix_time::milliseconds(10));
        }
    }
    return 0;
}

I am still thinking I might be doing something wrong, since I cannot find anything about this problem anywhere else, and the boost libraries are normally very good.

Is there anything I might be doing wrong with the usage of the message_queue in this example?

Answer 1

I don't think that both processes using open_or_create is a supported idiom. Are you aware of this thread on the mailing list? I can't find more discussions so it looks to me like lifetime management wasn't eventually considered necessary to add.

Thus you'll need to synchronise the creation manually with boost::interprocess or possibly by having one of the processes retrying to open_only the queue until the other process creates it.