[Vm-dev] script to add internal plugins ?

Sun May 14 21:38:52 UTC 2017

Hi Ben,

On Sat, May 13, 2017 at 9:22 AM, Ben Coman <btc at openinworld.com> wrote:

>
> On Sat, May 13, 2017 at 12:00 PM, Eliot Miranda <eliot.miranda at gmail.com>
> wrote:
> >
> > Hi Sophie,
> >
> > On Thu, May 11, 2017 at 8:04 AM, Sophie Kaleba <sophie.kaleba at gmail.com>
> wrote:
> >>
> >>
> >> Update :
> >>
> >> I did add the plugins to the file. That was the easy part, it went well.
> >>
> >> Then it got trickier when I tried to actually build the (Pharo) VM
> (build.linux32x86/pharo.cog.spur/build.itimerheatbeat) with my brand new
> plugins :
> >>
> >> - at first, I had an error : "can't find PharoV50.sources".
>
> You're the fifth newcomer compiling the VM that I've seen bump into
> this problem.
> The fix PR has been available since last October...
> https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/79
>
>
> >>
> >> (I cloned the opensmalltalk-vm directory,
> >>  ran updateSCCSVersions
> >> and then went in the "build.linux32x86/pharo.cog.spur/build.itimerheatbeat"
> and ran ./mvm -f )
> >>
> >> Manually adding a "sources" directory containing the missing file
> solved the problem
> >> Did I miss something on that part?
> >>
> >>
> >> -OK, now, the VM is built with this small hack.
> >>
> >> When I try to use the VMProfiler (which works fine using the latest
> squeak VM from bintray), I get a segmentation fault, with this message :
> >>
> >> *** stack smashing detected ***: ./squeak terminated
> >> output file stack is full.
> >
> >
> > I wonder if this is some checking feature supported by the C compiler
> you're using.  I expect that it is either disabled or not available in the
> VMs built on binary.  I say this because there is no occurrence of the
> string 'stack smashing detected' in our sources and so expect it is being
> produced by the C runtime on your version of linux.
>
> The best description I found in a quick search is...
> http://wiki.osdev.org/Stack_Smashing_Protector
>
> This implies a canary variable is being overwritten, which implies(??)
> a legitimate smashed-stack, with the culprit perhaps being some other
> protection mechanism like address space layout randomisation.
>
> Here is a detailed explanation including sample assembly code on the
> second page.
> http://www.drdobbs.com/security/anatomy-of-a-stack-smashing-attack-and-h/
> 240001832
>
> >
> > The stack organization in the VM is a little unusual.  The VM reserves
> space for a set of "stack pages" upon which Smalltalk activations exist.
> The space for the stack pages is allocated via alloca on the C stack at
> startup.  It could be that this strange organization, which means that at
> some times, when the VM is executing C, the stack pointer is at the bottom
> end of the C stack, but when it is executing JITted Smalltalk code the
> stack pointer is somewhere in the middle of the C stack.  I expect that the
> C runtime sees this happening and misinterprets it as an error.
>
> @Eliot, could ASLR cause assumptions about this memory layout to be
> violated?
>

I don't think so.  Given that the Stack Zone is alloca'ed on the C stack it
effectively stays in the same relative position to the C stack.

> It could be related to EBP register "position independent executables"
> http://forum.world.st/IMPORTANT-GCC-6-generates-position-independent-
> executables-by-default-on-Linux-td4935173.html

That's an interesting thought.  There is code executed at start-up (see
platforms/Cross/vm/sqCogStackAlignment.h: extern sqInt
cFramePointerInUse; platforms//unix/vm/sqUnixMain.c:isCFramePointerInUse())
t6hat attempts to find out if the frame pointer is in use and there-fore
whether the glue code between the JIT and the C run-time should save and
restore the frame pointer or not.

This article mentions both Stack Smashing and EBP register but I can't
> exactly tie them together.
> https://users.ece.cmu.edu/~vsekar/Teaching/Fall16/18487-
> f16/reading/Makowski_2011_Smashing%20the%20Stack%20in%202011.pdf
>
> @Sophie,  Could you report your OS and compiler version here? (sorry
> if you've already done this elsewhere)
>
> cheers -ben
>

Sophie, Ben, David, regarding the stack smashing, I did notice that there
is a variable intended to stop recursive error reporting within the sigsegv
handler (which also handles SIGBUS and SIGILL) that was not being set when
an error occurred.  This would have indeed causes an infinite recursion if
the error reporting was, for example, fooled by the use of EBP as a general
purpose register into crashing while trying to report an error.  I
committed the fix, and fixed the Mac VM.  But we should still investigate
exactly what the recursive crash is.  BTW, the variable is

oscogvm$ grep inFault platforms/unix/vm/* platforms/iOS/vm/Common/Classes/*
platforms/unix/vm/sqUnixMain.c:static int inFault = 0;
platforms/unix/vm/sqUnixMain.c: if (!inFault) {
platforms/unix/vm/sqUnixMain.c: inFault = 1;
platforms/iOS/vm/Common/Classes/sqSqueakMainApp.m:static int inFault = 0;
platforms/iOS/vm/Common/Classes/sqSqueakMainApp.m: if (!inFault) {
platforms/iOS/vm/Common/Classes/sqSqueakMainApp.m: inFault = 1;

> >
> > I expect further that you can, and should, disable such checking.
> >
> > HTH
> >
> >
> >>
> >>
> >> I already ran into this error when I first tried to run the VMProfiler
> a few weeks ago. At that time, I was using a built Squeak VM (still
> linux32x86 , cog.spur, itimer).
> >> Seeing that I got the same problem building both pharo and squeak vm
> (and not having it with vm I haven't actually built), I guess it has
> something to do with my environment.
> >> I was told it could be caused by a near-full disk, but I still have
> 8Gb+ of free space so, well.
> >>
> >> Does somebody have any idea about what could cause this error ? It
> would actually help me narrowing down the possibilities...
> >> I am using ubuntu 15.10 (old) so I will update it later today and check
> if it solves the problem.
> >> I will also try to update some libs, just in case.
> >>
> >> In the meantime, i will make a PR with the modified plugins files
> anyway, and see if it gets rejected or not.
> >>
> >> Sophie
> >>
> >>
> >>
> >> 2017-05-11 3:18 GMT+02:00 Ben Coman <btc at openinworld.com>:
> >>>
> >>>
> >>> On Thu, May 11, 2017 at 8:08 AM, tim Rowledge <tim at rowledge.org>
> wrote:
> >>> >
> >>> >
> >>> >> On 10-05-2017, at 4:57 PM, David T. Lewis <lewis at mail.msen.com>
> wrote:
> >>> >>
> >>> >>
> >>> >> On Wed, May 10, 2017 at 02:36:20PM -0700, tim Rowledge wrote:
> >>> >>>
> >>> >>>
> >>> >>>> On 10-05-2017, at 1:47 PM, Sophie Kaleba <sophie.kaleba at gmail.com>
> wrote:
> >>> >>>>
> >>> >>>> Hi Tim,
> >>> >>>>
> >>> >>>> I hadn't thought about that, thanks ! I will make sure not to
> create any duplicate of these plugins
> >>> >>>
> >>> >>> I???ve never thought to experiment and see what happens if one
> tried to make both internal and external versions of the same plugin. I???d
> guess that it might not cause any problems in most cases.. hmm, unless on
> some platform you actually have to compile with different flags, and then
> maybe there???d be problems if the internal version were compiled with flag
> A, the ???.o??? file were left around, the external version compile process
> found it, therefore didn???t compile with flag B, thus making something
> fail?
> >>> >>>
> >>> >>
> >>> >> I have not checked in a while, but I am pretty sure that it is
> perfectly
> >>> >> ok to have both an internally compiled plugin as part of the main VM
> >>> >> executable, and also an externally compiled version of the same
> plugin
> >>> >> (a loadable module in a file separate from the main VM executable).
> In
> >>> >> that case, the external plugin should override the internal one.
> >>> >>
> >>> >> The idea is that it should be easy to make an external plugin that
> might
> >>> >> add some new behaviour, and the plugin developer should be able to
> do
> >>> >> this without waiting for somebody else to distribute a new VM.
> >>> >>
> >>> >> If this is not the case, then it certainly /should/ be. Tim, weren't
> >>> >> you one of the people who invented this loadable module stuff in the
> >>> >> first place?
> >>> >
> >>> > D’oh, of course you’re right. Oh the joys of aging…
> >>> >
> >>> > And what I *should* have said was that I haven’t experimented with
> making internal and external versions within the same build process, given
> faint concerns about the object files maybe confusing the matter if one
> version is built and (incorrect) object files are left for the next version
> make-rule to find.
> >>> > So er, something like build BlobbyPlugin.c -> BlobbyPlugin.o -> link
> to make vm with internal BlobbyPlugin followed almost immediately by build
> BlobbyPlugin.c -> ooh, already got BlobbyPlugin.o -> make library
> BlobbyPlugin.la
> >>> > I’m imagining a situation where BlobbyPlugin.o actually ought to
> have been (re)compiled with subtly different flags in order to make a
> correct library. I may be being paranoid. Or maybe the OS is actually out
> to get me...
> >>>
> >>> Remember, just because you think you're paranoid doesn't mean they're
> >>> not watching you.
>

_,,,^..^,,,_
best, Eliot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20170514/d0f927f9/attachment.html>