Technology that deeply understands the principle of MMAP and makes big manufacturers love it

Like wechat MMKV Components 、 US mission Logan Components , And the log module of wechat xlog, Why do big factories prefer it ？ What magic does he have ？ I think the main reasons are as follows ：

• Cross platform ,C++ To write , Can support multiple platforms
• Cross process , Through file sharing, multiple processes can share memory , Implement process communication
• High performance , Realize zero copy of user space and kernel space , Fast and memory saving
• High stability , Page break protector , Implemented by the operating system , Stability is conceivable

Function introduction

void* mmap(void* addr, size_t length, int prot, int flags, int fd, off_t offset);

• addr Represents the starting address of the mapped virtual memory ;
• length Represents the mapping length ;
• prot Describes the access rights of this new memory area ;
• flags Describes the type of the mapping ;
• fd Represents the file descriptor ;
• offset Represents the offset value in the file .

mmap The power of , It can be configured according to parameters , Used to create shared memory , So as to improve the file mapping area IO efficiency , Realization IO Zero copy , Next, we will talk about zero copy technology , In contrast , These functions are mainly determined by three parameters , Let's explain one by one

prot

Four situations are as follows ：

• PROT_EXEC, Represents that the memory map has executable permissions , It can be regarded as a code segment , It usually stores CPU Executable machine code
• PROT_READ, Represents that the memory map is readable
• PROT_WRITE, Represents that the memory map is writable
• PROT_NONE, Represents that the memory map cannot be accessed

flags

The representative ones are as follows ：

• MAP_SHARED, Create a shared mapping area
• MAP_PRIVATE, Create a private mapping area
• MAP_ANONYMOUS, Create an anonymous mapping area , This situation only needs to be passed in -1 that will do
• MAP_FIXED, When the operating system uses addr When memory mapping for the starting address , If it is found that the length or permission requirements cannot be met , Mapping will fail , If the MAP_FIXED, Then the system will find other suitable areas to map

fd

When parameters fd It's not equal to 0 when , Memory mapping will be associated with the file , If it is equal to 0, Will become anonymous mapping , here flags Must be MAP_ANONYMOUS

Application scenarios

One mmap There are so many functions , From applying for memory allocation to loading dynamic libraries , Then to interprocess communication , It's really omnipotent , Strong enough to make people fall to the ground . Here are four situations , Take a parent-child process communication that I am most concerned about as an example , Implement a simple parent-child process communication logic , After all, the purpose of our study is to apply , How can theory alone be called a qualified blog ？

The parent and child processes share memory

#include <iostream>
#include <sys/wait.h>
#include <unistd.h>
#include <sys/mman.h>

int main() {
    pid_t c_pid = fork();

    char* shm = (char*)mmap(nullptr, 4096, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0);

    if (c_pid == -1) {
        perror("fork");
        exit(EXIT_FAILURE);
    } else if (c_pid > 0) {
        printf("parent process pid: %d\n", getpid());
        sprintf(shm, "%s", "hello, my child");
        printf("parent process got a message: %s\n", shm);
        wait(nullptr);
    } else {
        printf("child process pid: %d\n", getpid());
        sprintf(shm, "%s", "hello, father.");
        printf("child process got a message: %s\n", shm);
        exit(EXIT_SUCCESS);
    }

    return EXIT_SUCCESS;
}

After running, print as follows

parent process pid: 87799
parent process got a message: hello, my child
child process pid: 87800
child process got a message: hello, father.

Process finished with exit code 0

use mmap Created an anonymous shared memory area ,fd Pass in **-1 and MAP_ANONYMOUS Configure to implement anonymous mapping , Use MAP_SHARED** Create a shared area , Use fork Function create subprocess , In this way, sub process communication is realized , adopt sprintf Write the formatted data to the shared memory .

Cross process communication is achieved through a few lines of code , It's so simple , Such a powerful thing , Is there any support behind it ？ With questions, let's go on to find out .

MMAP The patron saint behind

Speaking of MMAP The patron saint of , Home page learn about the next memory page ： In page virtual memory , The virtual storage space and physical main storage space will be divided into pages of fixed size , Memory allocation for threads is also in pages . such as ： The size of the page is 4K, that 4GB Storage space is needed 4GB/4KB=1M Bar record , That is to say 100 More than a 4KB Page of , In memory pages , When the user reads and writes files , The kernel will request a memory page and file to read and write , Pictured ：

At this time, if there is no data in the memory page , An interrupt mechanism will occur , It is called page break , This interruption is MMAP The patron saint of , Why do you say that ？ We know mmap After function call , The process virtual address space is only established at the time of allocation , There is no physical memory allocated for virtual memory , When accessing these virtual memory without mapping relationship ,CPU The loading instruction found that the code segment is missing , It triggers the page missing interrupt , After interruption , The kernel checks the region of the virtual address , Memory mapping found , You can calculate the file offset through the virtual memory address , Locate the page of the file corresponding to the page missing in memory , Boot disk by kernel IO, Load the corresponding page from disk into memory . Final protection mmap It can go on smoothly , Selfless dedication . Understand the missing page interrupt , Let's talk more mmap Memory allocation principle in four scenarios

Four scene allocation principles

The above is a simple principle summary , There is no detailed expansion , If you are interested, you can check the information yourself .

summary

This sharing , It mainly introduces mmap Four application scenarios , The communication between parent and child processes is verified by an example , And go deep mmap Find its protector , And deeply understand mmap In four scenarios , How is the operating system organized and allocated , Through the understanding of these , After you mmap The practical application has a better theoretical basis , According to different needs , Different performance requirements , Choose the most appropriate implementation .

author ：i The headmaster
link ：https://juejin.cn/post/7119116943256190990