Why does boost::interprocess::managed_shared_memory throw a boost::interprocess_exception upon construction?

问题

In the code below, I'm trying to initialize a managed_shared_memory object. When the constructor is invoked I see the below error message -

terminate called after throwing an instance of 'boost::interprocess::interprocess_exception' what(): boost::interprocess_exception::library_error Aborted

why is this exception being thrown? I'm running this on an ubuntu 16.04 linux OS, compiled the program using g++ 9.3.0. Boost version 1.58.0

struct test_obj {
    size_t x;
    size_t y;
    uint8_t buf[32];
    bool is_valid;
};

class shm_wrapper {
        public:
            shm_wrapper() : m_shm(
                                boost::interprocess::open_or_create, 
                                "my_shm",
                                sizeof(test_obj) )
            {};
        private:
        boost::interprocess::managed_shared_memory m_shm;
};

回答1:

It aborts because the size is insufficient for the segment manager control blocks.

sizeof(test_obj) is just 56 bytes (on my system).

If you give the segment 10 KiB, it reports 224 bytes effectively used:

Live On Coliru

#include <boost/interprocess/managed_mapped_file.hpp>
#include <boost/interprocess/managed_shared_memory.hpp>

namespace bip = boost::interprocess;

#ifdef COLIRU
    using segment_type = bip::managed_mapped_file;
#else
    using segment_type = bip::managed_shared_memory;
#endif

static constexpr size_t SegmentSize = 10 * 1024;

struct test_obj {
    size_t x;
    size_t y;
    uint8_t buf[32];
    bool is_valid;
};

class shm_wrapper {
  public:
    shm_wrapper() : m_shm(bip::open_or_create, "my_shm", SegmentSize){};

    size_t free() const { return m_shm.get_free_memory(); }

  private:
    segment_type m_shm;
};

#include <iostream>
int main() {
    std::cout << sizeof(test_obj) << std::endl;
    shm_wrapper w;
    std::cout << w.free() << "\n";
    std::cout << "Effectively used:" << (SegmentSize - w.free()) << "\n";
}

Prints

56
10016
Effectively used:224

Summary

Maybe you didn't want a segment manager with dynamic allocation features. In that case, have a look at shared_memory_object instead.

BONUS: Example Using Mapped Region

To store a "dumb object" in a fixed-size region, you don't need a segment manager. You'd use a mapped_region on a shared_memory_object (or a file_mapping).

Here's another sample Live On Coliru

#include <boost/interprocess/shared_memory_object.hpp>
#include <boost/interprocess/file_mapping.hpp>
#include <boost/interprocess/mapped_region.hpp>
#include <iostream>
namespace bip = boost::interprocess;

struct test_obj {
    size_t x;
    size_t y;
    uint8_t buf[32];
    bool is_valid;
};

#ifdef COLIRU
    #include <fstream>
    using mapping_type = bip::file_mapping;
#else
    using mapping_type = bip::shared_memory_object;
#endif

template <typename T>
class shm_wrapper {
    static_assert(std::is_trivial_v<T>);
    static_assert(std::is_standard_layout_v<T>);

#ifdef COLIRU // file mappings require more work to cater for the storage
    struct backing_t { } backing;
    backing_t ensure_file(std::string name, size_t size) {
        std::filebuf fbuf;
        fbuf.open(name, std::ios::in | std::ios::out | std::ios::app | std::ios::binary);
        //set the size, sparsely
        fbuf.pubseekoff(size-1, std::ios_base::beg);
        fbuf.sputc(0);
        fbuf.close();
        return {};
    }
  public:
    shm_wrapper()
        : backing { ensure_file("my_shm", sizeof(T)) },
          m_mappable("my_shm", bip::read_write),
          m_reg(m_mappable, bip::read_write, 0, sizeof(T))
    { }
#else
  public:
    shm_wrapper()
        : m_mappable(bip::open_or_create, "my_shm", bip::read_write),
          m_reg(m_mappable, bip::read_write, 0, sizeof(T))
    {
        m_mappable.truncate(sizeof(T));
    }
#endif

    T& get()             { return *static_cast<T*>(m_reg.get_address()); }
    T const& get() const { return *static_cast<T const*>(m_reg.get_address()); }

    auto size() const { return m_reg.get_size(); } 
    auto flush()      { return m_reg.flush();    } 

  private:
    mapping_type m_mappable;
    bip::mapped_region m_reg;
};

int main() {
    shm_wrapper<test_obj> w;
    std::cout << "x:" << w.get().x << "\n";
    w.get().x ^= 0xa7;
    return w.flush()? 0 : 1;
}

Prints, when run 4x in succession:

x:0
x:167
x:0
x:167

来源：https://stackoverflow.com/questions/62748438/why-does-boostinterprocessmanaged-shared-memory-throw-a-boostinterprocess