Bug 220943

From gdb

There is absolutely no warranty for GDB.  Type "show warranty" for details.
This GDB was configured as "amd64-marcel-freebsd"...
(gdb) run 
Starting program: /usr/local/bin/valgrind 

Program terminated with signal SIGSEGV, Segmentation fault.

Then i compiled with WITH_DEBUG=yes but result is same

Valgrind need root permits :-)

Comment 3 Bryan Drewery 2017-08-10 17:43:04 UTC

Valgrind does not need root!

Hi , do you need some logs or any information about this problem ?

It looks like this just happened to me on FreeBSD 11, with a custom kernel config and the installer-provided hardening options enabled.  I enabled unprivileged debugging (sysctl security.bsd.unprivileged_proc_debug=1) and rebuilt my kernel to include FreeBSD 10 compatibility (options COMPAT_FREEBSD10), then valgrind started working better.  A reconstructed session looks like this:

% sudo sysctl security.bsd.unprivileged_proc_debug
security.bsd.unprivileged_proc_debug: 0
% valgrind ./sample
zsh: segmentation fault  valgrind ./sample
% sudo sysctl security.bsd.unprivileged_proc_debug=1
security.bsd.unprivileged_proc_debug: 0 -> 1
% valgrind ./sample
==5611== Memcheck, a memory error detector
==5611== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==5611== Using Valgrind-3.10.1 and LibVEX; rerun with -h for copyright info
==5611== Command: ./sample
==5611==
zsh: invalid system call  valgrind ./sample
% truss valgrind ./sample
(Lots of output elided.)
pread(0x3,0x4028f0d70,0x2000,0x63e000)     = 8192 (0x2000)
close(3)           = 0 (0x0)
compat10.pipe()          ERR#78 'Function not implemented'
SIGNAL 12 (SIGSYS)
process killed, signal = 12
(Kernel rebuild, install, and reboot elided.)
% valgrind ./sample
==67066== Memcheck, a memory error detector
==67066== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==67066== Using Valgrind-3.10.1 and LibVEX; rerun with -h for copyright info
==67066== Command: ./sample
==67066== 
(Output elided.)
==67066== 
==67066== HEAP SUMMARY:
==67066==     in use at exit: 0 bytes in 0 blocks
==67066==   total heap usage: 0 allocs, 0 frees, 0 bytes allocated
==67066== 
==67066== All heap blocks were freed -- no leaks are possible
==67066== 
==67066== For counts of detected and suppressed errors, rerun with: -v
==67066== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)

Created attachment 185695 [details]
Mention additional requirements in pkg-message

In case it helps, I've included a patch to pkg-message that mentions what I found.

I tested % sudo sysctl security.bsd.unprivileged_proc_debug=1, yes this change is  a functional change.

Comment 8 Bryan Drewery 2017-10-18 17:25:52 UTC

(In reply to Brad Sliger from comment #5)
> It looks like this just happened to me on FreeBSD 11, with a custom kernel
> config and the installer-provided hardening options enabled.  I enabled
> unprivileged debugging (sysctl security.bsd.unprivileged_proc_debug=1) and
> rebuilt my kernel to include FreeBSD 10 compatibility (options
> COMPAT_FREEBSD10), then valgrind started working better.  A reconstructed
> session looks like this:
> 
> % sudo sysctl security.bsd.unprivileged_proc_debug
> security.bsd.unprivileged_proc_debug: 0
> % valgrind ./sample
> zsh: segmentation fault  valgrind ./sample
> % sudo sysctl security.bsd.unprivileged_proc_debug=1
> security.bsd.unprivileged_proc_debug: 0 -> 1
> % valgrind ./sample
> ==5611== Memcheck, a memory error detector

valgrind really shouldn't be crashing here. We should show a message about
the issue.

Thanks for the investigation, I'll write up a patch to fix the
crash and show a message.

I ran into this today on stable/11 (r330260).  I tried three separate times, including a final time via "sudo valgrind ...".

security.bsd.unprivileged_proc_debug was 1 (default), but my custom kernel intentionally omits "options COMPAT_FREEBSD10".

The crashes I got were not SIGSEGV (11) but SIGSYS (12) as demonstrated in c#5 (it's easy to overlook that line).  Kernel logged these as such:

pid 33167 (memcheck-amd64-free), uid 1000: exited on signal 12 (core dumped)
pid 33245 (memcheck-amd64-free), uid 0: exited on signal 12 (core dumped)
pid 33244 (sudo), uid 0: exited on signal 12

Just noting all this here as a form of "this is still a problem 7 months later".

Comment 10 Bryan Drewery 2018-03-06 18:19:41 UTC

(In reply to Jeremy Chadwick from comment #9)
> I ran into this today on stable/11 (r330260).  I tried three separate times,
> including a final time via "sudo valgrind ...".
> 
> security.bsd.unprivileged_proc_debug was 1 (default), but my custom kernel
> intentionally omits "options COMPAT_FREEBSD10".
> 
> The crashes I got were not SIGSEGV (11) but SIGSYS (12) as demonstrated in
> c#5 (it's easy to overlook that line).  Kernel logged these as such:
> 
> pid 33167 (memcheck-amd64-free), uid 1000: exited on signal 12 (core dumped)
> pid 33245 (memcheck-amd64-free), uid 0: exited on signal 12 (core dumped)
> pid 33244 (sudo), uid 0: exited on signal 12
> 
> Just noting all this here as a form of "this is still a problem 7 months
> later".

Thanks for the reminder. I don't consider this high priority since
COMPAT_FREEBSD10 is all-but-required in general IMO.  But I do consider
this a bug worth fixing.

Comment 11 Walter Schwarzenfeld 2018-03-07 16:25:11 UTC

Possible duplicate of Bug #209886.

Comment 12 Walter Schwarzenfeld 2019-03-10 11:48:51 UTC

Assign to new maintainer.

Hello, 
and result this bug ?

ping

(In reply to Martin FIlla from comment #14)

Can you try the code in my github repo (https://github.com/paulfloyd/freebsd_valgrind)?

I'll add a check for the security.bsd.unprivileged_proc_debug soon.

This is now fixed in my github repo(https://github.com/paulfloyd/freebsd_valgrind)

With the sysctl set

root@euler:/usr/include # sysctl security.bsd.unprivileged_proc_debug=1
security.bsd.unprivileged_proc_debug: 0 -> 1


paulf> ./vg-in-place --tool=none ./leakcpp.clang++
==2894== Nulgrind, the minimal Valgrind tool
==2894== Copyright (C) 2002-2017, and GNU GPL'd, by Nicholas Nethercote.
==2894== Using Valgrind-3.16.0.GIT and LibVEX; rerun with -h for copyright info
==2894== Command: ./leakcpp.clang++
==2894== 
==2894== 
euler:/usr/home/paulf/scratch/valgrind

With the sysctl cleared

paulf> ./vg-in-place --tool=none ./leakcpp.clang++
--6443:0:    main Valgrind: FATAL:
--6443:0:    main security.bsd.unprivileged_proc_debug sysctl is 0.
--6443:0:    main    Set this sysctl with
--6443:0:    main    'sysctl security.bsd.unprivileged_proc_debug=1'.
--6443:0:    main    Cannot continue.
euler:/usr/home/paulf/scratch/valgrind

This has to be very early in the tool execution, before the address space manager starts.

This should now be fixed in devel/valgrind-devel and this item can be closed.

Comment 18 Kubilay Kocak 2020-09-20 09:42:45 UTC

^Triage: Unless a specific change (usually commit) can be identified that resolved the issue, OBE is more appropriate.