Using a cuckoo function

I was trying to track down a problem in Java where shared memory was not being accessed as it should. This was too early in the JVM startup for trace – so how to debug it?

One option I tried was using a cuckoo function. The Java code uses the C run time function shmget() — Get a shared memory segment. I wanted to write a stub function to display parameters, and then call the system version of shmget. To do this I had to get the Java program to use my cuckoo version of shmget.

My program

void wto( char * msg) { 
// time_t ltime; 
// time(&ltime); 
//printf("   the time is %s", ctime(&ltime)); 
  struct __cons_msg cmsg; 
  int rc; 
  int cmsg_cmd = 0; 
  /* fill in the __cons_msg structure */ 
  cmsg.__format.__f1.__msg = msg; 
  cmsg.__format.__f1.__msg_length = strlen(msg); 
  rc = __console(&cmsg,NULL,&cmsg_cmd); 
} 
// this has the same signature as the run time code
cpshmget( 
key_t key, size_t size, int shmflg) 
{ 
  char buffer[200]; 
  int i, j; 
  j = sprintf(buffer, ">shmget key %8.8x sz %i flg 0x%8.8x\0",  key,size,shmflg); 
  wto(buffer); 
  int myresult =  shmget(key,size,shmflg); 
//sprintf(buffer,"<shmget1 r %i e %i e2 0x%8.8x\0",myresult, errno, __errno2()); 
//wto(buffer); 
//myresult =  shmget(key,size,shmflg); 
  return  myresult; 
} 

I compiled it using a shell script (I found it easier than a make file)

name2=cp 
p1=" -DNDEBUG -O3 -qarch=10 -qlanglvl=extc99 -q64" 
p2="-Wc,DLL -D_XOPEN_SOURCE_EXTENDED -D_POSIX_THREADS" 
p2="-D_XOPEN_SOURCE_EXTENDED -D_POSIX_THREADS" 
p3="-D_OPEN_SYS_FILE_EXT                     -qstrict          " 
p4="-Wa,asa,goff -qgonumber -qenum=int" 
p5="-I." 
p6=""
p7="-Wc,ASM,SHOWINC,ASMLIB(//'SYS1.MACLIB') " 
p8="-Wc,LIST(c.lst),SOURCE,NOWARN64,XREF -Wa,LIST,RENT" 
/bin/xlc $p1 $p2 $p3 $p4 $p5 $p6 $p7 $p8  -c $name2.c -o $name2.o 
                                                                                      

It is used as an object .o code – not as an executable module.

Bind it to the Java code

I wanted to cuckoo the module libj9prt29.so .

I saved a copy of the original module libj9prt29.so as libj9prt29.so.orig. I could then include this into the binder, and not worry about any of my old code staying around.

When I needed “a clean run”, without any of my code, I copied libj9prt29.so.orig over libj9prt29.so .

The bind job

//IBMBIND JOB 1,MSGCLASS=H 
// SET BPARM='CASE=MIXED,SIZE=(900K,124K),XREF,RMODE=ANY,CALL=YES' 
// SET CPARM='MAP,DYNAM=DLL,LIST=ALL,AMODE=64,TERM=YES' 
//BIND   EXEC PGM=IEWL,REGION=0M,PARM='&CPARM,&BPARM' 
//SYSLIB   DD  DSN=CEE.SCEEBND2,DISP=SHR 
//         DD  DSN=CBC.SCCNOBJ,DISP=SHR 
//         DD  DSN=SYS1.CSSLIB,DISP=SHR 
//Z8921    DD  DSN=CEE.SCEELIB(CELQS003),DISP=SHR 
//SYSLMOD DD PATH='/usr/lpp/java/J17.0_64/lib/default/', 
//            PATHDISP=(KEEP,KEEP) 
//SYSPRINT DD  SYSOUT=* 
//SYSLIN   DD   * 
 ORDER CELQSTRT 
 ENTRY CELQSTRT 
 INCLUDE /u/tmp/java/cp.o 
 change shmget(cpshmget) 
 INCLUDE /usr/lpp/java/J17.0_64/lib/default/libj9prt29.so.orig 
 include Z8921 
 name  libj9prt29.so 
// 

Notes:

• You specify SYSLMOD as a directory and specify the “name” later. This was not obvious from the documentation
• INCLUDE /u/tmp/java/cp.o copies in my code
• change shmget(cpshmget) says any code which is included after this, cuckoo the entry point shmget to point to cpshmget
• INCLUDE /usr/lpp/java/J17.0_64/lib/default/libj9prt29.so.orig include the original module
• include Z8921 = CEE.SCEELIB(CELQS003) contains statements like IMPORT CODE64,CELQV003,’shmget’,280 which defines where shmget is really defined
• name libj9prt29.so save it as this name in //SYSLMOD

This worked well – but…

I originally had

  wto(buffer); 
  int myresult =  shmget(key,size,shmflg); 
  sprintf(buffer,"<shmget1 r %i e %i e2 0x%8.8x\0",myresult, errno, __errno2()); 
  wto(buffer); 
//myresult =  shmget(key,size,shmflg); 

But the wto function, which invokes __console(), sets errno and __errno2. The values passed back to Java were the values from the __console function, not the shmget function.

Issuing the myresult = shmget(key,size,shmflg); a second time set the errno and __errno2 – but if the request was to create, or delete, issuing it a second time gave a different error code. For example if “delete” was specified – the first time worked, the second time gave “not found”, and this confused the Java program calling this function.