[Bug 200138] [PATCH] Fixed race condition incurred in libthr by some foreign malloc libraries

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=200138

            Bug ID: 200138
           Summary: [PATCH] Fixed race condition incurred in libthr by
                    some foreign malloc libraries
           Product: Base System
           Version: 10.1-STABLE
          Hardware: Any
                OS: Any
            Status: New
          Keywords: patch
          Severity: Affects Only Me
          Priority: ---
         Component: threads
          Assignee: freebsd-threads at FreeBSD.org
          Reporter: yuri at rawbw.com
          Keywords: patch

Created attachment 156674
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=156674&action=edit
patch

While investigating the message coming from "Thread %p has exited with %d
leftover", I found that _pthread_setspecific is entered twice with the same
TCB, _get_curthread() returned the same thread pointer. Second invocation was
incurred by the google tcmalloc library (devel/google-perftools port).

> (gdb) bt
> #0  yuri_enter (serno=60) at /usr/src/lib/libthr/thread/thr_spec.c:60
> #1  0x0000000804e23e89 in _pthread_setspecific (userkey=3, value=0x740a080) at /usr/src/lib/libthr/thread/thr_spec.c:228
> #2  0x000000080408b95c in ?? () from /usr/local/lib/qt4/libQtCore.so.4
> #3  0x0000000804e1da3a in thread_start (curthread=0x73f6800) at /usr/src/lib/libthr/thread/thr_create.c:288
> #4  0x0000000000000000 in ?? ()
> Backtrace stopped: Cannot access memory at address 0x7fffdf9fb000


> (gdb) bt
> #0  yuri_enter (serno=61) at /usr/src/lib/libthr/thread/thr_spec.c:60
> #1  0x0000000804e23e89 in _pthread_setspecific (userkey=2, value=0x7a9520) at /usr/src/lib/libthr/thread/thr_spec.c:228
> #2  0x00000008008ff4de in perftools_pthread_setspecific(int, void*) () from /usr/local/lib/libtcmalloc.so.4
> #3  0x00000008008f2322 in tcmalloc::ThreadCache::CreateCacheIfNecessary() () from /usr/local/lib/libtcmalloc.so.4
> #4  0x00000008008ea666 in ?? () from /usr/local/lib/libtcmalloc.so.4
> #5  0x00000008008ffd92 in tc_calloc () from /usr/local/lib/libtcmalloc.so.4
> #6  0x0000000804e243a1 in pthread_key_allocate_data () at /usr/src/lib/libthr/thread/thr_spec.c:212
> #7  0x0000000804e23f19 in _pthread_setspecific (userkey=3, value=0x740a080) at /usr/src/lib/libthr/thread/thr_spec.c:231
> #8  0x000000080408b95c in ?? () from /usr/local/lib/qt4/libQtCore.so.4
> #9  0x0000000804e1da3a in thread_start (curthread=0x73f6800) at /usr/src/lib/libthr/thread/thr_create.c:288
> #10 0x0000000000000000 in ?? ()

This causes some memory loss, and stray message about leftover TLS data.

The attached patch corrects for such potential problem by double-checking the
pthread's "specific" array after allocating it, and frees the newly allocated
copy if it happens to be already allocated.

Please keep the comment, because I think these lines may appear very confusing
and wrong to people without the knowledge of this specific situation.

-- 
You are receiving this mail because:
You are the assignee for the bug.