write syscall with mmaped buffer

I work with large files and as a result of development I had to copy large files (about 10 gigabytes) in C code.

In this regard, the question arose: how effective is the combination of map()+write() (an existing file is mapped in memory and write done to an existing file)?

Will conditional Windows/macOS/Linux/*BSD handle this in a special way to optimize copying data from one file to another

Sample code that is present in my project (assume code ran on unix-like OS and all syscalls succeed, input and output are some regular files):

int main() {
    struct stat sb;
    int fdsrc, fddst;
    void *p;

    fdsrc = open("input", O_RDONLY);
    fddst = open("output", O_WRONLY);

    fstat(fdsrc, &sb);
    p = mmap(NULL, sb.st_size, PROT_READ, MAP_SHARED, fdsrc, 0);
    write(fddst, p, sb.st_size);

    return 0;
}

Will it be as fast as copy_file_range(2) or kinda? Is there some code in Linux/*BSD sources which handles this case?

P.S.: On linux write(2) as i know never writes more than INT_MAX bytes in one call, so in my project i use full_write() wrapper which calls write(2) until all data written or one of the syscalls fails.

You need to sign in to view this answers