valgrind on FreeBSD?

Sun Oct 9 14:21:48 UTC 2011

On Sun, Oct 09, 2011 at 04:18:48PM +0200, V??clav Zeman wrote:
> Jeremy Chadwick wrote, On 9.10.2011 16:11:
> > On Sun, Oct 09, 2011 at 03:48:57PM +0200, V??clav Zeman wrote:
> >> V??clav Zeman wrote, On 9.10.2011 15:25:
> >>> Bakul Shah wrote, On 6.10.2011 8:40:
> >>>> On Wed, 05 Oct 2011 23:06:04 +0200 =?UTF-8?B?VsOhY2xhdiBaZW1hbg==?= <v.haisman at sh.cvut.cz>  wrote:
> >>>>> Hi.
> >>>>>
> >>>>> No matter what I try, valgrind on 7.3-STABLE is giving me this, both Valgrind
> >>>>> ports:
> >>>>>
> >>>>> valgrind: Startup or configuration error:
> >>>>>    Can't establish current working directory at startup
> >>>>> valgrind: Unable to start up properly.  Giving up.
> >>>>>
> >>>>> What do I need to do to make it work?
> >>>>
> >>>> Try running valgrind under ktrace (& view with kdump). That
> >>>> will tell you what directory it is trying to access or what
> >>>> syscall fails and why.
> >>> Hi.
> >>>
> >>> So I have done that and more. I have first updated from 7.3 to 8.2 (RELENG_8
> >>> actually). I have not managed to recompile all of the installed Ports yet,
> >>> but I made sure to recompile valgrind and its dependencies. The same thing
> >>> has happened!
> >>>
> >>> As I have said, I have done the ktrace and here is the interesting bit:
> >>>
> >>>  78028 valgrind NAMI  "/usr/local/lib/valgrind/memcheck-amd64-freebsd"
> >>>  78028 memcheck-amd64-free RET   execve 0
> >>>  78028 memcheck-amd64-free CALL  getpid
> >>>  78028 memcheck-amd64-free RET   getpid 78028/0x130cc
> >>>  78028 memcheck-amd64-free CALL
> >>> __sysctl(0x39a91450,0x4,0x389a3800,0x39a91468,0,0)
> >>>  78028 memcheck-amd64-free SCTL  "kern.proc.vmmap.78028"
> >>>  78028 memcheck-amd64-free RET   __sysctl 0
> >>>  78028 memcheck-amd64-free CALL
> >>> mmap(0x400009000,0x400000,PROT_READ|PROT_WRITE|PROT_EXEC,MAP_PRIVATE|MAP_FIXED|MAP_ANON,0xffffffffffffffff,0)
> >>>  78028 memcheck-amd64-free RET   mmap 17179906048/0x400009000
> >>>  78028 memcheck-amd64-free CALL  getrlimit(RLIMIT_DATA,0x39e6a780)
> >>>  78028 memcheck-amd64-free RET   getrlimit 0
> >>>  78028 memcheck-amd64-free CALL  setrlimit(RLIMIT_DATA,0x39a919e0)
> >>>  78028 memcheck-amd64-free RET   setrlimit 0
> >>>  78028 memcheck-amd64-free CALL  getrlimit(RLIMIT_STACK,0x39e6a790)
> >>>  78028 memcheck-amd64-free RET   getrlimit 0
> >>>  78028 memcheck-amd64-free CALL  __getcwd(0x3882d700,0x3ff)
> >>>  78028 memcheck-amd64-free NAMI  ".."
> >>>  78028 memcheck-amd64-free RET   __getcwd -1 errno 2 No such file or directory
> >>>  78028 memcheck-amd64-free CALL  write(0x2,0x3830b060,0x6c)
> >>>  78028 memcheck-amd64-free GIO   fd 2 wrote 108 bytes
> >>>        "valgrind: Startup or configuration error:
> >>>         valgrind:    Can't establish current working directory at startup
> >>>        "
> >>>  78028 memcheck-amd64-free RET   write 108/0x6c
> >>>  78028 memcheck-amd64-free CALL  write(0x2,0x3830b060,0x33)
> >>>  78028 memcheck-amd64-free GIO   fd 2 wrote 51 bytes
> >>>        "valgrind: Unable to start up properly.  Giving up.
> >>>        "
> >>>  78028 memcheck-amd64-free RET   write 51/0x33
> >>>  78028 memcheck-amd64-free CALL  exit(0x1)
> >>>
> >>> Now what? Why would the __getcwd call be failing with "No such file or
> >>> directory"?
> >>>
> >> It is the nullfs!
> >>
> >> I have /home mounted using nullfs to /usr/home:
> >>
> >> /usr/home               /home                   nullfs  rw,multilabel,acls
> >>    0 0
> >>
> >> When I run valgrind from the /usr based directory, it works:
> >>
> >> shell::wilx:/usr/home/users/wilx/tmp/yttool> valgrind --tool=memcheck ./yttool
> >> ==34679== Memcheck, a memory error detector
> >> ==34679== Copyright (C) 2002-2010, and GNU GPL'd, by Julian Seward et al.
> >> ==34679== Using Valgrind-3.6.1 and LibVEX; rerun with -h for copyright info
> >> ==34679== Command: ./yttool
> >> ==34679==
> >> ==34679==
> >> ==34679== HEAP SUMMARY:
> >> ==34679==     in use at exit: 20,395 bytes in 119 blocks
> >> ==34679==   total heap usage: 6,719 allocs, 6,600 frees, 716,787 bytes allocated
> >> ==34679==
> >> ==34679== LEAK SUMMARY:
> >> ==34679==    definitely lost: 0 bytes in 0 blocks
> >> ==34679==    indirectly lost: 0 bytes in 0 blocks
> >> ==34679==      possibly lost: 134 bytes in 4 blocks
> >> ==34679==    still reachable: 20,261 bytes in 115 blocks
> >> ==34679==         suppressed: 0 bytes in 0 blocks
> >> ==34679== Rerun with --leak-check=full to see details of leaked memory
> >> ==34679==
> >> ==34679== For counts of detected and suppressed errors, rerun with: -v
> >> ==34679== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
> >>
> >> But when I run it from the nullfs mount, it fails:
> >>
> >> shell::wilx:/usr/home/users/wilx/tmp/yttool> cd $HOME/tmp/yttool
> >> shell::wilx:~/tmp/yttool> valgrind --tool=memcheck ./yttool
> >> valgrind: Startup or configuration error:
> >> valgrind:    Can't establish current working directory at startup
> >> valgrind: Unable to start up properly.  Giving up.
> > 
> > Amazing how userland utilities behave differently depending upon the
> > underlying filesystem type, eh?  Good thing I asked what your underlying
> > filesystem types were.  Don't ever think that "it'll all just work".
> > :-)
> > 
> > I believe there are other issues/stipulations with nullfs (some have
> > been reported over the years), so I'm not too surprised by this issue.
> > I have no idea who currently maintains nullfs(5) either; it looks like a
> > major group effort given the committers who have touched it in the past
> > few years:
> > 
> > http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/fs/nullfs/
> > 
> > I'm CC'ing Kostik (kib@) as he might have some ideas.
> > 
> > If this isn't a known issue, please file a PR for the issue with
> > nullfs(5).  The issue is not within valgrind, so the PR should not be
> > for that.
> I have filled a PR: <http://www.freebsd.org/cgi/query-pr.cgi?pr=161424>
Nullfs VNTOCNP implementation has a known deficiency.
Working on the item is in my TODO list.

> 
> > 
> > As for a workaround: is there some reason you can't just use "ln -s
> > /usr/home /home" and solve the problem?
> None. I remember using nullfs for /home instead of the link because I just
> liked that it never has shown /usr there and also because it seemed cooler. :)
> 
> -- 
> VZ
> 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20111009/4c832498/attachment.pgp