Fix bugs in vty error handlingex25b

Advance version to 0.99.20ex25b.

Bug fixes:

  * when closing a vty, if close() returned an error, the vtys would
    fall into a loop trying to recover.

    This was found on FreeBSD, which will return ECONNRESET on close(),
    which is not standard POSIX behaviour.

  * when closing a vty in response to EPIPE or ECONNRESET, managed to
    leave the vty mutex locked.

    This stopped the Routing Engine *dead* on the next CLI command
    that required the Routing Engine.  SIGTERM/SIGINT would not
    stop bgpd -- a SIGKILL would be required.

Other changes:

  * toned down the error reporting of EPIPE in the vty -- so no
    longer looks like an error...

      ...closing a vty by "exit" command is logged as "closed"

      ...closing a vty in response to the client closing their end of
         the connection is logged as "terminated: terminal closed",
         and no longer logs an EPIPE error.

  * changed error reporting on close() and shutdown() for vty, so
    that nothing is logged if an i/o error has already logged...

    ...so that redundant error messages are suppressed.

  * applied slightly finer jittering to bgpd timers.

  * work in progress on new vtysh.

1 files changed, 4 insertions, 5 deletions

diff --git a/lib/qrand.c b/lib/qrand.c
index 402bb432..a395b92c 100644
--- a/lib/qrand.c
+++ b/lib/qrand.c
@@ -32,17 +32,16 @@
  *
  * If range == 1, returns 0 every time !
  */
-extern int
-qrand(qrand_seq seq, int range)
+extern uint
+qrand(qrand_seq seq, uint range)
 {
   uint64_t  r ;
 
-  r = seq->last ^ 3141592653 ;
-  r = ((r * 2650845021) + 5) & 0xFFFFFFFF ;     /* see Knuth    */
+  r = ((seq->last * 2650845021) + 5) & 0xFFFFFFFF ;     /* see Knuth    */
   seq->last = r ;
 
   if (range == 0)
     return r >> 1 ;
   else
-    return (r % range) ;
+    return (r * range) >> 32 ;
 } ;