Unrotate MP4 from some smartphone

/*
 * $ gcc -Wall -O3 -o unrotate unrotate.c
 * $ ./unrotate input.mp4 output.mp4
 */

#include <stdio.h>
#include <inttypes.h>

#include <sys/mman.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>

#include <string.h>

static int fd_in  = -1;
static int fd_out = -1;
static struct stat st;
static const uint8_t *buf_in  = NULL;
static uint8_t *buf_out = NULL;

static void mmap_close(void)
{
    if (buf_out     ) munmap(buf_out, st.st_size);
    if (buf_in      ) munmap((void*)buf_in, st.st_size);
    if (fd_out != -1) close(fd_out);
    if (fd_in  != -1) close(fd_in);
}

static int mmap_open(char *fname_in, char *fname_out)
{
    int offset;

    /* Open and mmap input file */
    fd_in = open(fname_in, O_RDONLY);
    if (fd_in == -1) {
        fprintf(stderr, "could not open input file '%s'\n", fname_in);
        goto end;
    }
    if (fstat(fd_in, &st) == -1) {
        fprintf(stderr, "could not stat input file '%s'\n", fname_in);
        goto end;
    }
    buf_in = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd_in, 0);
    if (buf_in == MAP_FAILED) {
        fprintf(stderr, "could not map input file '%s'\n", fname_in);
        goto end;
    }
    /* Open and mmap output file, with same size as input file */
    fd_out = open(fname_out, O_RDWR | O_CREAT | O_TRUNC, 0666);
    if (fd_out == -1) {
        fprintf(stderr, "could not open output file '%s'\n", fname_out);
        goto end;
    }
    lseek(fd_out, st.st_size-1, SEEK_SET);
    offset = write(fd_out, "", 1);
    lseek(fd_out, 0, SEEK_SET);
    buf_out = mmap(NULL, st.st_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd_out, 0);
    if (buf_out == MAP_FAILED) {
        fprintf(stderr, "could not map output file '%s'\n", fname_out);
        goto end;
    }

    return st.st_size;

end:
    mmap_close();
    return 0;
}

static uint32_t read_be32(const uint8_t *p)
{
    return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | (p[3]);
}

static void write_be16(uint8_t *p, uint32_t val)
{
    p[0] = (val >> 8) & 0xff;
    p[1] = (val >> 0) & 0xff;
}

int main(int argc, char *argv[])
{
    unsigned int file_size;
    unsigned int i;
    int ok = 0;

    if (argc < 3) {
        fprintf(stderr, "usage: %s <input> <output>\n", argv[0]);
        return -1;
    }

    file_size = mmap_open(argv[1], argv[2]);
    if (!file_size)
        return -1;

    for (i = 0; i < 4; i++)
        buf_out[i] = buf_in[i];
    for (     ; !ok && i < file_size-(4+104); i++) {
        if ((*(uint32_t *)&buf_in[i]) == 0x64686b74) {
            uint32_t tkhd_size = read_be32(&buf_in[i-4]);
            uint8_t version = buf_in[i+4];
            uint8_t offset = (version ? 14 : 11) << 2;

            memcpy(&buf_out[i], &buf_in[i], tkhd_size);

            if (read_be32(&buf_in[i+offset+36]) && read_be32(&buf_in[i+offset+40])) {
                write_be16(&buf_out[i+offset   ], 1);
                write_be16(&buf_out[i+offset+ 4], 0);
                write_be16(&buf_out[i+offset+ 8], 0);
                write_be16(&buf_out[i+offset+12], 0);
                write_be16(&buf_out[i+offset+16], 1);
                write_be16(&buf_out[i+offset+20], 0);
                fprintf(stderr, "video tkhd found and matrix changed\n");
                ok = 1;
            }

            i += tkhd_size-1;
        } else {
            buf_out[i] = buf_in[i];
        }
    }
    memcpy(&buf_out[i], &buf_in[i], file_size - i);

    mmap_close();

    return 0;
}