Automicity and Race Conditions

• He first justifies some of the special flags to open.
• Consider two processes that want to open a file exclusively
• Assume two processes are working as the same service.
  • A process will be the lead process if no log file exists.
  • A process will be a worker process if a log file exists.
  • Algorithm

    attempt to open a file read only
    if fail (the file does not exist)
        open the file in create mode
        become lead process
    else
        become worker process
    	

  • The problem is really the two open calls.
  • See the figure on page 91.
  • If O_CREAT | O_EXCL are specified
    • The file creation process is atomic
      • the process of file creation will not be interpreted.
    • We get a 0 if the file was created, and a -1 if not.
  • This also explains the O_APPEND flag.
  • As opposed to seeking to the end of the file and writing.
• I can't stand not giving real demonstrations so.....
• fork
  • You should read chapter 24.
  • create a new process.
  • WARNING
    • The system call we about to discuss is dangerous.
      • You can cripple a system if this gets away from you.
    • Some cautions
      • Please do not put forks inside of a loop or recursive call yet.
      • Please DO NOT develop fork code on cslab 100 or 101.
      • Please do use cslab 161-179 but be careful
    • If you have a fork bomb
      • Please DO try to stop it.
      • If you can try killall -u username
      • If you can't stop it, send email to me (danbennett360@gmail.com) and Chris Loveland (cloveland@edinboro.edu).
      • PLEASE DO NOT TRY TO HIDE THIS.
  • #include <unistd.h>
  • pid_t fork(void);
    • Creates an identical copy of a process except
      • the return value is different.
        • Child process return is a 0.
        • Parent process gets the pid of child process.
        • -1 on error and no child process.
      • file descriptors opened with O_CLOEXEC are closed.
      • The process has a unique pid
      • CPU time counters are reset.
      • Signals are cleared
      • Other things that we will discuss as time goes by.
• Useful now are getpid and getppid.
  • sys/types.h and unistd.h
  • pid_t getpid(void);
  • pid_t getppid(void);
  • return the process id and the parent process id.
• And sleep and usleep
  • unistd.h
  • unsigned int sleep(unsigned int seconds);
  • man 3 sleep
  • Either returns 0 if the full time has passed or the amount of time remaining.
  • int usleep(useconds_t usec);
  • Returns 0 on success or -1 on error.
  • Expects time between 0 and 1000000
• fork is somewhat expensive, so a fast version is available.
  • pid_t vfork(void);
  • Designed to be called before exec*
  • We can ignore this for now.