RFR: JDK-8202884: SA: Attach/detach might fail on Linux if debugee application create/destroy threads during attaching

Jini George jini.george at oracle.com
Thu Dec 13 03:41:22 UTC 2018


Thanks, JC! Will add the comments.

Thanks,
- Jini.

On 12/13/2018 12:09 AM, JC Beyler wrote:
> Hi Jini,
> 
> Fair enough, thanks for the explanation. It makes sense to me. I imagine 
> that it is a conservative approach, ie can't find the information then 
> assume still live.
> 
> Perhaps a small comment above the 'X'/'Z' test saying that the threads 
> are considered to be dead then?
> And for the return perhaps say: the thread is considered live if the 
> state was not 'X'/'Z' or not found?
> 
> But those are really nits and no need to send another webrev for me!
> 
> Thanks!
> Jc
> 
> 
> 
> On Wed, Dec 12, 2018 at 10:27 AM Jini George <jini.george at oracle.com 
> <mailto:jini.george at oracle.com>> wrote:
> 
> 
>     On 12/12/2018 10:47 PM, JC Beyler wrote:
>      > Hi Jini,
>      >
>      > Should your return not be return !found_state instead of false here:
>      >
>      > 257   if (!found_state) {
>      > 258     // Assuming the thread exists.
>      > 259     print_error("Could not find the 'State:' string in the
>      > /proc/%d/status file\n", pid);
>      > 260   }
>      > 261   fclose (fp);
>      > 262   return false;
>      >
>      > In your webrev.00 it was the case but now, you always return the
>     process
>      > does exist even if you have not found it.
> 
>     I referred to the gdb sources to check what is done under this
>     scenario,
>     and in gdb, it is assumed that if the line beginning with 'State:' is
>     not found, the thread is alive. But to be frank, I don't know under
>     what
>     circumstances will we ever encounter such a scenario. Let me know if
>     you
>     don't agree with this.
> 
>      > cf:
>      >
>     http://cr.openjdk.java.net/~jgeorge/8202884/webrev.00/src/jdk.hotspot.agent/linux/native/libsaproc/ps_proc.c.udiff.html
>      > vs
>      >
>     http://cr.openjdk.java.net/~jgeorge/8202884/webrev.02/src/jdk.hotspot.agent/linux/native/libsaproc/ps_proc.c.udiff.html
>      >
>      > Tiny nit: no need to check *space if you are using
>     isspace(*space) right
>      > after :)
> 
>     Will change this. Thanks!
> 
>      >
>      > Apart from the return question, the webrev looks good to me :-)
>      > Jc
> 
>     Thank you again!
>     Jini
> 
>      >
>      >
>      > On Wed, Dec 12, 2018 at 4:15 AM Jini George
>     <jini.george at oracle.com <mailto:jini.george at oracle.com>
>      > <mailto:jini.george at oracle.com <mailto:jini.george at oracle.com>>>
>     wrote:
>      >
>      >     Thank you very much for looking into this, JC!
>      >
>      >     I have a revised webrev addressing your comments at:
>      >
>      > http://cr.openjdk.java.net/~jgeorge/8202884/webrev.02/index.html
>      >
>      >     Requesting one more review for this. My comments inline:
>      >
>      >     On 12/12/2018 2:53 AM, JC Beyler wrote:
>      >      > Hi Jini,
>      >      >
>      >      > I saw a few nits:
>      >      >
>      >
>     http://cr.openjdk.java.net/~jgeorge/8202884/webrev.00/src/jdk.hotspot.agent/linux/native/libsaproc/libproc_impl.h.udiff.html
>      >      >    -> The comments are in the third person normally it
>     seems so
>      >     it would
>      >      > become (I also removed the s from threads):
>      >      >
>      >      > +// deletes a thread from the thread list
>      >     Done.
>      >      >
>      >      >
>      >
>     http://cr.openjdk.java.net/~jgeorge/8202884/webrev.00/src/jdk.hotspot.agent/linux/native/libsaproc/libproc_impl.c.udiff.html
>      >      >    -> You added two empty lines it seems that could be removed
>      >     Done.
>      >      >
>      >      >
>      >      >
>      >
>     http://cr.openjdk.java.net/~jgeorge/8202884/webrev.00/src/jdk.hotspot.agent/linux/native/libsaproc/ps_proc.c.udiff.html
>      >      >    -> Is there a real reason to have both enums? We could
>     have a
>      >     single
>      >      > enum it seems and not lose too much
>      >
>      >     You are right. I have done away with the WAITPID* enum.
>      >
>      >      >    -> you have a switch "
>      >      >         switch (errno) {"
>      >      >          -> Where really you could simplify the reading by
>     moving
>      >     the
>      >      > EINTR case outside with its continue
>      >      >          -> The switch could then remain as it was (though
>     you move
>      >      > print_debug to print_error)
>      >      >          -> and just return in each cases
>      >     I have changed this to:
>      >
>      >     206     } else {
>      >     207       switch (errno) {
>      >     208         case EINTR:
>      >     209           continue;
>      >     210           break;
>      >     211         case ECHILD:
>      >     212           print_debug("waitpid() failed. Child process
>     pid (%d)
>      >     does
>      >     not exist \n", pid);
>      >     213           return ATTACH_THREAD_DEAD;
>      >     214         case EINVAL:
>      >     215           print_error("waitpid() failed. Invalid options
>      >     argument.\n");
>      >     216           return ATTACH_FAIL;
>      >     217         default:
>      >     218           print_error("waitpid() failed. Unexpected error
>     %d\n",
>      >     errno);
>      >     219           return ATTACH_FAIL;
>      >     220       }
>      >     221     } // else
>      >
>      >
>      >      >
>      >      >     -> if (strncmp (buf, "State:", 6) == 0) {
>      >      >        -> You use sizeof("State:") right below; perhaps
>     you could
>      >     just
>      >      > use "  const char const state[] = "State:";" and use
>      >     sizeof(state) and
>      >      > for the string, it seems less error prone
>      >      >
>      >      >    -> A minor "bug" is here:
>      >      > +      state = buf + sizeof ("State:");
>      >      >          -> You did a strncmp above but that only assures the
>      >     start of
>      >      > the string is "State:", technically the character after
>     the ':'
>      >     is the
>      >      > but it could only be that; sizeof("State:") is 7 and not 6. So
>      >     you miss
>      >      > one character when you are skipping spaces
>      >      >          -> It was probably ok because you always had at
>     least one
>      >      > space, ie "State: "
>      >
>      >     Thanks! I have made some changes here to use a const char
>     string and a
>      >     variable to store the calculated length using strlen(). And I
>     am using
>      >     isspace() now to skip spaces since tabs could also be used as a
>      >     delimiter.
>      >
>      >      >    -> Extra space here before the '(': "sizeof (buf)"
>      >     Done.
>      >      >
>      >      > Finally your return sequence for that method could be
>     simplified to:
>      >      >
>      >      > +  if (!found_state) {
>      >      > +    print_error(" Could not find the State: string in the
>     status
>      >     file
>      >      > for pid %d\n", pid);
>      >      > +  }
>      >      > +  fclose (fp);
>      >      > +  return !found_state;
>      >
>      >     I have modified this to:
>      >
>      >     257   if (!found_state) {
>      >     258     // Assuming the thread exists.
>      >     259     print_error("Could not find the 'State:' string in the
>      >     /proc/%d/status file\n", pid);
>      >     260   }
>      >     261   fclose (fp);
>      >     262   return false;
>      >
>      >     Thank you,
>      >     Jini.
>      >
>      >      >
>      >      > Thanks!
>      >      > Jc
>      >      >
>      >      > On Tue, Dec 11, 2018 at 9:30 AM Jini George
>      >     <jini.george at oracle.com <mailto:jini.george at oracle.com>
>     <mailto:jini.george at oracle.com <mailto:jini.george at oracle.com>>
>      >      > <mailto:jini.george at oracle.com
>     <mailto:jini.george at oracle.com> <mailto:jini.george at oracle.com
>     <mailto:jini.george at oracle.com>>>>
>      >     wrote:
>      >      >
>      >      >     Hello !
>      >      >
>      >      >     Requesting reviews for:
>      >      >
>      >      > https://bugs.openjdk.java.net/browse/JDK-8202884
>      >      >     Webrev:
>      > http://cr.openjdk.java.net/~jgeorge/8202884/webrev.00/index.html
>      >      >
>      >      >     Details:
>      >      >     For attaching to the threads in a process, we first go
>     ahead
>      >     and do a
>      >      >     ptrace attach to the main thread. Later, we use the
>     libthread_db
>      >      >     library
>      >      >     (or, in the case of being within a container, iterate
>     through the
>      >      >     /proc/<pid>/task files) to discover the threads of the
>      >     process, and add
>      >      >     them to the threads list (within SA) for this process.
>     Once,
>      >     we have
>      >      >     discovered all the threads and added these to the list of
>      >     threads, we
>      >      >     then invoke ptrace attach individually on all these
>     threads to
>      >      >     attach to
>      >      >     these. When we deal with an application where the
>     threads are
>      >     exiting
>      >      >     continuously, some of these threads might not exist by the
>      >     time we try
>      >      >     to ptrace attach to these threads. The proposed fix
>     includes the
>      >      >     following modifications to solve this.
>      >      >        1. Check the state of the threads in the thread_db
>      >     callback routine,
>      >      >     and skip if the state of the thread is TD_THR_UNKNOWN or
>      >     TD_THR_ZOMBIE.
>      >      >     SA does not try to ptrace attach to these threads and does
>      >     not include
>      >      >     these threads in the threads list.
>      >      >        2. While ptrace attaching to the thread, if
>      >     ptrace(PTRACE_ATTACH,
>      >      >     ...)
>      >      >     fails with either ESCRH or EPERM, check the state of the
>      >     thread by
>      >      >     checking if the /proc/<pid>/status file corresponding
>     to that
>      >     thread
>      >      >     exists and if so, reading in the 'State:' line of that
>     file. Skip
>      >      >     attaching to this thread and delete this thread from
>     the SA
>      >     list of
>      >      >     threads, if the thread is dead (State: X) or is a zombie
>      >     (State: Z).
>      >      >       From the /proc man page, "Current state of the
>     process. One
>      >     of "R
>      >      >     (running)", "S (sleeping)", "D (disk sleep)", "T
>     (stopped)", "T
>      >      >     (tracing
>      >      >     stop)", "Z (zombie)", or "X (dead)"."
>      >      >        3. If waitpid() on the thread is a failure, again skip
>      >     this thread
>      >      >     (delete this from SA's list of threads) instead of bailing
>      >     out if the
>      >      >     thread has exited or terminated.
>      >      >
>      >      >     Testing:
>      >      >     1. Tested by attaching and detaching multiple times to
>     a test
>      >     program
>      >      >     spawning numerous short lived threads.
>      >      >     2. The SA tests (under
>     test/hotspot/jtreg/serviceability/sa)
>      >     passed
>      >      >     with
>      >      >     100 repeats on Mach5.
>      >      >     3. No new failures and no occurrences of JDK-8202884 seen
>      >     with testing
>      >      >     the SA tests (tiers 1 to 5) on Mach5.
>      >      >
>      >      >     More details in the bug comments section.
>      >      >
>      >      >     Thank you,
>      >      >     Jini.
>      >      >
>      >      >
>      >      >
>      >      > --
>      >      >
>      >      > Thanks,
>      >      > Jc
>      >
>      >
>      >
>      > --
>      >
>      > Thanks,
>      > Jc
> 
> 
> 
> -- 
> 
> Thanks,
> Jc


More information about the serviceability-dev mailing list