Search code examples
c++cboostshared-memoryboost-interprocess

Boost's managed_shared_memory usage in between two processes (C and C++)

I am having a Design/Implementation issue with Boost managed_shared_memory.

Problem

I have two processes: Writer and Reader.

  • Here Writer is a C++ executable which is using boost to write in shared memory. Code snapshot for writing method:

    shm_controller.reset(new boost::interprocess::managed_shared_memory( boost::interprocess::open_or_create, shmName, size));
void * addr = shm_controller->allocate(size) ;
shm_controller->deallocate(addr);

    Once I have addr pointer, I am using memcpy to write in SHM.

  • On other hand I am having Reader, which is C executable. They both are communicating via message passing application. I wrote Reader code in such way:

    • Writer is passing address, which is stored in "addr" (from above code snapshot) to Reader.
    • Once Reader receive the address and size, I am using memcpy to write SHM data from Reader's local variable.

Note 1: (This Reader is not using any boost shm API, as it's C application )
Note 2: I validated address is same on Writer and Reader (after communication).

But Reader's memcpy to read shm data is leading crash in my executable (Reader).

Questions

  1. Can we use boost::interprocess::managed_shared_memory in C ??
  2. Is there any implementation problem in above section. (can't a Reader access shm of Writer via pointer)
  3. Is there a way to solve this C and C++ limitation like any wrapper implementation etc.
Solution

  • Q. Can we use boost::interprocess::managed_shared_memory in C ??

    No. Technically, you can, of course (C is turing complete), but it's not supported.

    Q. Is there any implementation problem in above section. (can't a reader access shm of writer via pointer)

    Yes, you cannot usefully pass a pointer from one process to another, because they have isolated address space.

    Q. Is there a way to solve this C and C++ limitation like any wrapper implementation etc.

    The usual approach is to pass data among processes as part of the message. Use your message passing library (http://man7.org/linux/man-pages/man7/mq_overview.7.html, http://zeromq.org/, etc.).

    Note: This also avoids trouble with synchronizing access to shared memory, that you haven't even talked about. This gives me the idea you didn't think of it

    The next best idea would be to use raw shm API on both sides and pass the offset into the shared-memory area across processes. You could still use BIP, if you limit yourself to e.g. shared_memory_object and e.g. offset_ptr