Solaris 8 linker patch 109147-21 / 109148-21 breaks runtime linker

Question

The "Sun ONE Studio 8, Compiler Collection Early Access" release requires the
installation of a linker patch (109147-21 sparc / 109148-21 x86).
It seems that the solaris 8 linker patch breaks the runtime linker audit feature
for multithreaded applications.
I can reproduce the problem like this (using $LD_PROFILE, which makes
use of the runtime linker audit feature):
% cat thr.c
#define _REENTRANT
#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
void*
thr_main(void*arg)
printf("hello, world!\n");
int
main(int argc, char **argv)
pthread_t tid;
void *rval;
pthread_create(&tid, NULL, thr_main, NULL);
pthread_join(tid, &rval);
exit(0);
% cc -o thr thr.c -lpthread
% thr
hello, world!
OK, fine so far...
Now we try to run the application under rtld audit control:
% env LD_PROFILE=libc.so.1 thr
< the multithreaded application hangs >
% env LD_PROFILE=libc.so.1 LD_LIBRARY_PATH=/usr/lib/lwp thr
< the multithreaded application hangs >
(on a S8 system with 109148-14 installed, this works without hanging!!)
A "pstack `pgrep thr`" command shows multiple LWPs, all blocked
waiting in an lwp_mutex_lock() call, all trying to lock the same
mutex - which happens to be the dynamic linker's global 'rtldlock'.
% pstack `pgrep thr`
7631: thr
----------------- lwp# 1 --------------------------------
dfbd6b5f lwp_mutex_lock (dfbf1618)
dfbd69f6 lwpmutex_lock (dfbf1618) + 2a
dfbc8739 rt_mutex_lock (dfbf1618) + 19
dfbcd575 enter () + 29
dfbd2193 elf_bndr (dfa94eac, 50) + 23
dfbc3339 elf_rtbndr () + 5
----------------- lwp# 2 / thread# 2 --------------------
dfbd6b5f lwp_mutex_lock (dfbf1618)
dfbd69f6 lwpmutex_lock (dfbf1618) + 2a
dfbc8739 rt_mutex_lock (dfbf1618) + 19
dfbcd575 enter () + 29
dfbd2193 elf_bndr (dfa94eac, 238) + 23
dfbc3339 elf_rtbndr () + 5
----------------- lwp# 3 --------------------------------
dfbd6b5f lwp_mutex_lock (dfbf1618)
dfbd69f6 lwpmutex_lock (dfbf1618) + 2a
dfbc8739 rt_mutex_lock (dfbf1618) + 19
dfbcd575 enter () + 29
dfbd4257 auditpltenter () + bb
dfbd431a audit_pltenter (dfa94eac, dfa60414, df92131c, 48, df8a4cd0, df921318) + 7a
dfbc32a2 elf_plt_trace (3, ffffffff, df8a4d44) + 52
df8f435e scsetup (ffffffff, 1) + 7a
df8ea13e lwpstart () + 22
-------------------------- thread# 1 --------------------
df8e961b _swtch   (0) + 3b3
df8eb806 _dopreempt (0) + ea
df8ed247 _sigon   () + 37
df8f41d0 bindclear (1) + 48
dfbc8799 rt_bind_clear (1) + 19
dfbcd87c leave (dfbef0f4) + e8
dfbca8f7 call_init (df8d0288, 1) + 10f
dfbca339 setup (80470fc, 80471e4, 0, 8047ff2, 1000, dfbc17af, dfbf06e4, dfbc0000, ...) + cd5
dfbd3393 _setup   () + 2f7
dfbc323a rtboot () + 58
0804725c ???????? ()
-------------------------- thread# 3 --------------------
df8e7958 threadstart ()
% dbx - `pgrep thr`
Reading -
Reading ld.so.1
Reading libpthread.so.1
Reading libc.so.1
Reading libdl.so.1
Reading libthread.so.1
Reading ldprof.so.1
Reading libmapmalloc.so.1
detected a multithreaded program
Attached to process 7631 with 3 LWPs
t@3 (l@null) stopped in threadstart at 0xdf8e7958
0xdf8e7958: threadstart : call threadstart+0x5 <0xdf8e795d>
(dbx) x 0xdfbf1618 /4X
dbx: warning: unknown language, 'c' assumed
0xdfbf1618: rtldlock : 0x00000000 0x00000000 0x00000000 0x01010000
(dbx) q
detaching from process 7631
Is this a known bug?
Are there patches to fix this issue?

Answer 1

I think this is bugid 4804328.
Synopsis: auditing of threaded applications results in deadlock
It doesn't look like the next patch is avilable
yet, but the bug seems to be on track for being
fixed in patchid 109147-22 (the next rev of the
patch you mention. There should be new patches
for Solaris 8 sparc/intel and Solaris 9 sparc/intel.