Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-1240826

fp.bugs 3902 potential race condition in the definition of Ref.get

    XMLWordPrintable

    Details

    • Subcomponent:
    • Resolved In Build:
      1.0.2
    • CPU:
      sparc
    • OS:
      solaris_2.5
    • Verification:
      Not verified

      Description

      From: ###@###.### (Rich Cohen)
      This does not look like form output to me.


      There appears to be a bug in the Ref class as defined in JDK 1.0.

      There is a potential race condition in the definition of Ref.get.
      Ref.get could return a null pointer if the object gets flushed in
      another thread. (The documentation is unclear, but it appears that
      Ref.get is not intended to return null pointers, since a null pointer
      triggers a call to reconstitute.)

      This problem can be avoided simply by returning the value of a local
      variable, rather than returning the value of a shared variable.

      I've appended the erroneous source code and an appropriate fix.

      -- Rich Cohen


      From JDK 1.0 file Ref.java:

      /*
       * @(#)Ref.java 1.17 95/09/06
       *
       */

      [...]

       /**
        * Returns a pointer to the object referenced by this Ref. If the object
        * has been thrown away by the garbage collector, it will be
        * reconstituted. This method does everything necessary to ensure that the garbage
        * collector throws things away in Least Recently Used(LRU) order. Applications should
        * never override this method. The get() method effectively caches calls to
        * reconstitute().
        */
       public Object get() {
           Object p = thing;
           if (p == null) {
               /* synchronize if thing is null, but then check again
                  in case somebody else beat us to it */
               synchronized (this) {
                   if (thing == null) {
                       p = reconstitute();
                       thing = p;
                   }
               }
           }
           priority = ++lruclock;
           /* *** Erroneous code from JDK: ***
           return thing; */
           /* *** Better code: ***/
           return p;
       }

        Attachments

          Activity

            People

            Assignee:
            tlindholsunw Timothy Lindholm (Inactive)
            Reporter:
            bhagen Benjamin Hagen (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            0 Start watching this issue

              Dates

              Created:
              Updated:
              Resolved:
              Imported:
              Indexed: