[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20190530105206.GR16415@port70.net>
Date: Thu, 30 May 2019 12:52:07 +0200
From: Szabolcs Nagy <nsz@...t70.net>
To: musl@...ts.openwall.com
Cc: Rich Felker <dalias@...c.org>
Subject: Re: [PATCH] fix musl-gcc.specs.sh to correctly handle
-static-pie
* Fangrui Song <i@...kray.me> [2019-05-30 14:20:44 +0800]:
> On 2019-05-29, Rich Felker wrote:
> > On Tue, May 28, 2019 at 10:19:18PM +0200, Ferdi265 wrote:
> > > Hello,
> > >
> > > Today I wanted to use musl-gcc to build -static-pie binaries. I noticed
> > > that this does not work (musl-libc is still linked as a shared library
> > > and is also still requested as an interpreter).
> > >
> > > Looking into the musl-gcc.specs file the bug was obvious: rcrt1.o was
> > > not used as a startfile, and neither -no-dynamic-linker nor -static were
> > > passed to the linker.
> > >
> > > This patch fixes this by actually using rcrt1.o and passing the linker
> > > options when -static-pie is given.
> > >
> > > I don't have much experience with the specifics of gcc .spec files and
> > > which options need to be passed, but this seems to work with all
> > > variations of -shared, -static, and -static-pie that I've tried.
> > >
> > > Here (https://github.com/Ferdi265/musl) is the repository with the patch
> > > on GitHub, and I've also attached the patch below.
> > >
> > > Greetings,
> > > Ferdinand "Ferdi265" Bachmann
> > >
> > > --- PATCH BELOW ---
> > >
> > > From 070bce8f7e508a951d3b65da227b2fca3a65f37b Mon Sep 17 00:00:00 2001
> > > From: Ferdinand Bachmann <theferdi265@...il.com>
> > > Date: Tue, 28 May 2019 21:53:25 +0200
> > > Subject: [PATCH] fix musl-gcc.specs.sh to correctly handle -static-pie
> > >
> > > ---
> > > tools/musl-gcc.specs.sh | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/tools/musl-gcc.specs.sh b/tools/musl-gcc.specs.sh
> > > index 30492574..7206cb25 100644
> > > --- a/tools/musl-gcc.specs.sh
> > > +++ b/tools/musl-gcc.specs.sh
> > > @@ -17,13 +17,13 @@ cat <<EOF
> > > libgcc.a%s %:if-exists(libgcc_eh.a%s)
> > >
> > > *startfile:
> > > -%{!shared: $libdir/Scrt1.o} $libdir/crti.o crtbeginS.o%s
> > > +%{static-pie: $libdir/rcrt1.o} %{!static-pie: %{!shared:
> > > $libdir/Scrt1.o}} $libdir/crti.o crtbeginS.o%s
> > >
> > > *endfile:
> > > crtendS.o%s $libdir/crtn.o
> > >
> > > *link:
> > > --dynamic-linker $ldso -nostdlib %{shared:-shared} %{static:-static}
> > > %{rdynamic:-export-dynamic}
> > > +%{static-pie:-no-dynamic-linker -static} %{!static-pie:-dynamic-linker
> > > $ldso} -nostdlib %{shared:-shared} %{static:-static}
> > > %{rdynamic:-export-dynamic}
> > >
> > > *esp_link:
> > >
> >
> > This looks okay-ish, but could be improved. GCC spec syntax doesn't
> > require separate static-pie and !static-pie conditions; it supports
> > "else" with notation:
> >
> > %{static-pie:-no-dynamic-linker -static;-dynamic-linker $ldso}
> >
> > There's also the question of whether we should make -static -pie work
> > like it's intended to by us, or like gcc does it upstream (ignoring
> > -pie). I think it's hard to duplicate the behavior we want since it
> > depends on knowing if the compiler was built as default-pie, so the
> > way you've done it is probably the best we can easily do, and at least
> > non-broken.
> >
> > Anyone else have comments on this?
>
> gcc -dumpspecs says:
>
> %{static-pie:-static -pie --no-dynamic-linker -z text}
>
> Do we want to specify -z text?
musl does not really support text rels so
arguably -z text should be always passed.
but since it's just a diagnostic option it
is not strictly necessary, just userfriendly.
>
> In ld.bfd and gold, -z notext is the default. If text relocations are needed:
>
> -z notext: allow and emit DF_TEXTREL. DF_TEXTREL is not emitted if there is no text relocation.
> -z text: error
> In lld, -z text is the default (this change is a no-op).
>
> -z text: error on text relocations
> -z notext: allow text relocations, and emit DF_TEXTREL no matter whether text relocations exist.
>
>
> While comparing the differences, I also came across this rule:
>
> %{!static|static-pie:--eh-frame-hdr}
>
> > It means "pass --eh-frame-hdr to ld if -static isn't used or -static-pie
> > is used since -static-pie needs PT_GNU_EH_FRAME segment.
>
> This is related to PT_GNU_EH_FRAME (it is produced if .eh_frame_hdr exists) and various unwinders.
> I'm not particularly clear about the internals. Some related links:
allows unwinding across or within an elf module
without load-time registration of the module with
the unwinder (registration implies single unwinder
instance with global data).
(with eh-frame-hdr an unwinder can find the data
using dl_iterate_phdr, no registration is needed)
i think the difference between static and static-pie
exists because different compiler startup code
(crtbeginT.o vs crtbeginS.o) gets linked in, the first
uses the registry (__register_frame_info) the second
assumes eh-frame-hdr.
>
> https://reviews.llvm.org/D19029 FreeBSD folk said LLVM's libunwind needed this on ARM
> I think this piece of code is relevant
> https://github.com/llvm-mirror/libunwind/blob/master/src/AddressSpace.hpp#L532
> https://reviews.llvm.org/D43203 clang -static adds --eh-frame-hdr unconditionally
> https://www.openwall.com/lists/musl/2014/06/16/2 Re: gcc'c crtstuff.c: a musl-related experience
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
