End of Chapter 4

• The fact that a process is handed open file descriptors 0,1,2 is interesting.
  • Did you ever wonder how cmd > outfile is done?
  • Take a look at smallCat.cpp.
  • Normally this is done before the exec, but we don't know that.
  • We will see more of this later.
  • But

    // prog > afile < bfile
    fork()
     // in the child process
     close(STDOUT_FILENO);
     open afile
     close (STDIN_FILENO);
     open bfile
     exec prog 

• The same thing happens with the | command.
  • And we will see that later too.
• lseek
  • needs unistd.h
  • off_t lseek(int fd, off_t offset, int whence);
  • fd is an open file descriptor
  • offset is a position in bytes into the file.
  • whence is a symbolic constant
    • SEEK_SET move to the value of offset. (absolute seek)
    • SEEK_CUR move to the current position + offset
    • SEEK_END move to the end of the file + offset
      • This is like iterators, the end is one past the last byte.
      • lseek(fd, 0, SEEK_END) will add a new byte to the file.
      • lseek(fd, -1, SEEK_END) will change the last byte in the file..
  • Since 3.1 the linux kernel supports
    • SEEK_DATA : find the next position >= offset that contains data.
    • SEEK_HOLE : find the next position >= offset that contains a hole.
      • if offset is in a hole, the position is set to offset.
      • if there is no hole past offset, position is set to the end of the file.
  • Lseek returns the current position, or -1 on fail.
• Holes?
  • Yes, if you seek past the end of a file, then write, it will create a hole.
  • Holes are filled with 0 (null, 00000000, not '0');
  • Surprisingly holes do not take up space on the file system.
    • Except, remember, segments are allocated in a block, so any data written within a block will take up that block.
  • Some file system types do not support blocks.
• Look at seeker.cpp