eBPF Map size is larger than shared memory size, cause `boost::interprocess::bad_alloc`

[janetat]

问题

1. 共享内存是静态分配的，大小为20M。
2. 共享内存里保存着bpftime用到的数据结构，包括eBPF Maps。
3. 如果eBPF Maps数据不断增长，超过共享内存大小怎么办？

期望

共享内存可以动态增长吗？

[Officeyutong]

There is a PR to allow defining shared memory size at runtime, but it's not finished yet https://github.com/eunomia-bpf/bpftime/pull/285

[janetat]

https://github.com/eunomia-bpf/bpftime/pull/285 这是通过命令行的方式传递共享内存大小，也是静态、固定的。

期望是达到Max之后，还能自动地增加共享内存么？

[Officeyutong]

285 这是通过命令行的方式传递共享内存大小，也是静态、固定的。

期望是达到Max之后，还能自动地增加共享内存么？

It's not supported now. But since boost::interprocess provided such ability, we may add it later

   //!Tries to resize the managed shared memory object so that we have
   //!room for more objects.
   //!
   //!This function is not synchronized so no other thread or process should
   //!be reading or writing the file
   static bool grow(const char *shmname, size_type extra_bytes)
   {
      return base_t::template grow
         <basic_managed_shared_memory>(shmname, extra_bytes);
   }

[yunwei37]

Now you can use flag like BPFTIME_SHM_MEMORY_MB to control the size of the shared memory. For example,

SPDLOG_LEVEL=Debug BPFTIME_SHM_MEMORY_MB=50 LD_PRELOAD=/home/yunwei37/bpftime/build/runtime/syscall-server/libbpftime-syscall-server.so ./example/malloc/malloc

It can set the shared memory size to 50MB. We may use grow in the future.