Re: avoid W|X mappings in libffi
Jeremie Courreges-Anglas wrote: > Stuart Henderson writes: > > > This disables PROT_EXEC mappings in libffi (and thus python). > > I'm running with it in a bulk build with the "mandatory W^X" > > printfs that are going into snapshots and haven't triggered > > them yet, building python itself (done 2.7 and 3.4 so far) > > or in the ~200 py-* and py3-* things that have built already > > (I would have had a whole stack by now otherwise). > > > > There are a lot of test failures when this diff is used. > > Can anyone figure out if they're anything to worry about? > > So I think the problem is simply that since the closure isn't allocated > with PROT_EXEC permissions, you can't... execute it. > > I'm kinda discovering libffi so sorry in advance if I'm mistaken, but it > seems that the logical steps are always: > - ffi_closure_alloc > - ffi_prep_closure_loc > - ffi_closure_free > > ffi_prep_closure_loc being the function that actually writes machine > code intended to be executed later. I think we could tweak > ffi_prep_closure_loc so that it adjusts the permissions after having > done its work: move from PROT_READ|PROT_WRITE (what we get now with the > closures.c diff) to PROT_READ|PROT_EXEC. > > ffi_prep_closure is just a stub that calls ffi_prep_closure_loc, only > the latter can handle different addresses for writable and executable > user pages. That seems to be used for situations when ffi_closure_alloc > calls dlmmap_locked under the hood. > > Here's a tentative diff, I'm not sure whether it will be the best > solution in the end, but it doesn't seem to be worse than the current > patch: > [...] > Index: patches/patch-src_x86_ffi64_c > === > RCS file: patches/patch-src_x86_ffi64_c > diff -N patches/patch-src_x86_ffi64_c > --- /dev/null 1 Jan 1970 00:00:00 - > +++ patches/patch-src_x86_ffi64_c 23 May 2016 11:46:59 - > @@ -0,0 +1,27 @@ > +$OpenBSD$ > +--- src/x86/ffi64.c.orig Mon May 23 13:42:49 2016 > src/x86/ffi64.c Mon May 23 13:42:56 2016 > +@@ -27,6 +27,9 @@ > +DEALINGS IN THE SOFTWARE. > +--- > */ > + > ++#include > ++#include > ++ > + #include > + #include > + > +@@ -563,6 +566,13 @@ ffi_prep_closure_loc (ffi_closure* closure, > + closure->cif = cif; > + closure->fun = fun; > + closure->user_data = user_data; > ++ > ++ /* remove WRITE access on the closure */ > ++ if (mprotect(closure, sizeof(*closure), PROT_READ) == -1) > ++ return FFI_OK; > ++ /* ensure we have EXECUTE access on the closure */ > ++ if (mprotect(closure, sizeof(*closure), PROT_READ|PROT_EXEC) == -1) > ++ return FFI_OK; The second PROT_READ|PROT_EXEC mprotect should be enough. Is ffi_closure_alloc() guaranteed to allocate a separate page, i.e. does every closure get its own page? Because if a page can contain multiple ffi_closures, they would all get mapped to PROT_READ|PROT_EXEC at once => any later call of calling ffi_prep_closure_loc for those other closures would cause a segfault. > + return FFI_OK; > + } > > > -- > jca | PGP : 0x1524E7EE / 5135 92C1 AD36 5293 2BDF DDCC 0DFA 74AE 1524 E7EE >
devel/llvm: make clang build working static PIE programs
clang currently links against /usr/lib/crt0.o when building
a binary with the -static -pie options. That makes such
programs segfault on startup (gcc 4.9 seems to have the same
problem).
When building a static PIE program, we need to link against
/usr/lib/rcrt0.o, as shown below. Bitrig has this part also
in their local llvm tree. I'll try to get it into upstream for
clang 3.9 as well.
--- tools/clang/lib/Driver/Tools.cpp.orig Fri Feb 12 23:51:41 2016
+++ tools/clang/lib/Driver/Tools.cppSun Jun 26 20:24:44 2016
@@ -7600,6 +7600,10 @@ void openbsd::Linker::ConstructJob(Compilation &C, con
if (Args.hasArg(options::OPT_pg))
CmdArgs.push_back(
Args.MakeArgString(getToolChain().GetFilePath("gcrt0.o")));
+ else if (Args.hasArg(options::OPT_static) &&
+ !Args.hasArg(options::OPT_nopie))
+CmdArgs.push_back(
+Args.MakeArgString(getToolChain().GetFilePath("rcrt0.o")));
else
CmdArgs.push_back(
Args.MakeArgString(getToolChain().GetFilePath("crt0.o")));
Full diff against the ports tree below.
ok?
Index: Makefile
===
RCS file: /cvs/ports/devel/llvm/Makefile,v
retrieving revision 1.114
diff -u -p -r1.114 Makefile
--- Makefile24 May 2016 07:53:23 - 1.114
+++ Makefile27 Jun 2016 17:24:32 -
@@ -11,7 +11,7 @@ COMMENT = modular, fast C/C++/ObjC compi
LLVM_V = 3.8.0
DISTNAME = llvm-${LLVM_V}.src
PKGNAME = llvm-${LLVM_V}
-REVISION = 1
+REVISION = 2
CATEGORIES = devel
DISTFILES =llvm-${LLVM_V}.src${EXTRACT_SUFX} \
cfe-${LLVM_V}.src${EXTRACT_SUFX}
Index: patches/patch-tools_clang_lib_Driver_Tools_cpp
===
RCS file: /cvs/ports/devel/llvm/patches/patch-tools_clang_lib_Driver_Tools_cpp,v
retrieving revision 1.30
diff -u -p -r1.30 patch-tools_clang_lib_Driver_Tools_cpp
--- patches/patch-tools_clang_lib_Driver_Tools_cpp 24 May 2016 07:53:23
- 1.30
+++ patches/patch-tools_clang_lib_Driver_Tools_cpp 27 Jun 2016 17:24:32
-
@@ -1,6 +1,5 @@
-$OpenBSD: patch-tools_clang_lib_Driver_Tools_cpp,v 1.30 2016/05/24 07:53:23
ajacoutot Exp $
tools/clang/lib/Driver/Tools.cpp.orig Fri Feb 12 17:51:41 2016
-+++ tools/clang/lib/Driver/Tools.cpp Tue May 17 14:45:22 2016
+--- tools/clang/lib/Driver/Tools.cpp.orig Fri Feb 12 23:51:41 2016
tools/clang/lib/Driver/Tools.cpp Sun Jun 26 20:24:44 2016
@@ -78,7 +78,7 @@ static const char *getSparcAsmModeForCPU(StringRef Nam
.Case("niagara2", "-Av9b")
.Case("niagara3", "-Av9d")
@@ -10,7 +9,18 @@ $OpenBSD: patch-tools_clang_lib_Driver_T
} else {
return llvm::StringSwitch(Name)
.Case("v8", "-Av8")
-@@ -7611,15 +7611,17 @@ void openbsd::Linker::ConstructJob(Compilation &C, con
+@@ -7600,6 +7600,10 @@ void openbsd::Linker::ConstructJob(Compilation &C, con
+ if (Args.hasArg(options::OPT_pg))
+ CmdArgs.push_back(
+ Args.MakeArgString(getToolChain().GetFilePath("gcrt0.o")));
++ else if (Args.hasArg(options::OPT_static) &&
++ !Args.hasArg(options::OPT_nopie))
++CmdArgs.push_back(
++Args.MakeArgString(getToolChain().GetFilePath("rcrt0.o")));
+ else
+ CmdArgs.push_back(
+ Args.MakeArgString(getToolChain().GetFilePath("crt0.o")));
+@@ -7611,15 +7615,17 @@ void openbsd::Linker::ConstructJob(Compilation &C, con
}
}
Re: devel/llvm: make clang build working static PIE programs
ping?
Stefan Kempf wrote:
> clang currently links against /usr/lib/crt0.o when building
> a binary with the -static -pie options. That makes such
> programs segfault on startup (gcc 4.9 seems to have the same
> problem).
>
> When building a static PIE program, we need to link against
> /usr/lib/rcrt0.o, as shown below. Bitrig has this part also
> in their local llvm tree. I'll try to get it into upstream for
> clang 3.9 as well.
>
> --- tools/clang/lib/Driver/Tools.cpp.orig Fri Feb 12 23:51:41 2016
> +++ tools/clang/lib/Driver/Tools.cpp Sun Jun 26 20:24:44 2016
> @@ -7600,6 +7600,10 @@ void openbsd::Linker::ConstructJob(Compilation &C, con
>if (Args.hasArg(options::OPT_pg))
> CmdArgs.push_back(
> Args.MakeArgString(getToolChain().GetFilePath("gcrt0.o")));
> + else if (Args.hasArg(options::OPT_static) &&
> + !Args.hasArg(options::OPT_nopie))
> +CmdArgs.push_back(
> +Args.MakeArgString(getToolChain().GetFilePath("rcrt0.o")));
>else
> CmdArgs.push_back(
> Args.MakeArgString(getToolChain().GetFilePath("crt0.o")));
>
> Full diff against the ports tree below.
>
> ok?
>
> Index: Makefile
> ===
> RCS file: /cvs/ports/devel/llvm/Makefile,v
> retrieving revision 1.114
> diff -u -p -r1.114 Makefile
> --- Makefile 24 May 2016 07:53:23 - 1.114
> +++ Makefile 27 Jun 2016 17:24:32 -
> @@ -11,7 +11,7 @@ COMMENT = modular, fast C/C++/ObjC compi
> LLVM_V = 3.8.0
> DISTNAME = llvm-${LLVM_V}.src
> PKGNAME =llvm-${LLVM_V}
> -REVISION = 1
> +REVISION = 2
> CATEGORIES = devel
> DISTFILES = llvm-${LLVM_V}.src${EXTRACT_SUFX} \
> cfe-${LLVM_V}.src${EXTRACT_SUFX}
> Index: patches/patch-tools_clang_lib_Driver_Tools_cpp
> ===
> RCS file:
> /cvs/ports/devel/llvm/patches/patch-tools_clang_lib_Driver_Tools_cpp,v
> retrieving revision 1.30
> diff -u -p -r1.30 patch-tools_clang_lib_Driver_Tools_cpp
> --- patches/patch-tools_clang_lib_Driver_Tools_cpp24 May 2016 07:53:23
> - 1.30
> +++ patches/patch-tools_clang_lib_Driver_Tools_cpp27 Jun 2016 17:24:32
> -
> @@ -1,6 +1,5 @@
> -$OpenBSD: patch-tools_clang_lib_Driver_Tools_cpp,v 1.30 2016/05/24 07:53:23
> ajacoutot Exp $
> tools/clang/lib/Driver/Tools.cpp.origFri Feb 12 17:51:41 2016
> -+++ tools/clang/lib/Driver/Tools.cpp Tue May 17 14:45:22 2016
> +--- tools/clang/lib/Driver/Tools.cpp.origFri Feb 12 23:51:41 2016
> tools/clang/lib/Driver/Tools.cpp Sun Jun 26 20:24:44 2016
> @@ -78,7 +78,7 @@ static const char *getSparcAsmModeForCPU(StringRef Nam
> .Case("niagara2", "-Av9b")
> .Case("niagara3", "-Av9d")
> @@ -10,7 +9,18 @@ $OpenBSD: patch-tools_clang_lib_Driver_T
> } else {
> return llvm::StringSwitch(Name)
> .Case("v8", "-Av8")
> -@@ -7611,15 +7611,17 @@ void openbsd::Linker::ConstructJob(Compilation &C,
> con
> +@@ -7600,6 +7600,10 @@ void openbsd::Linker::ConstructJob(Compilation &C, con
> + if (Args.hasArg(options::OPT_pg))
> + CmdArgs.push_back(
> + Args.MakeArgString(getToolChain().GetFilePath("gcrt0.o")));
> ++ else if (Args.hasArg(options::OPT_static) &&
> ++ !Args.hasArg(options::OPT_nopie))
> ++CmdArgs.push_back(
> ++Args.MakeArgString(getToolChain().GetFilePath("rcrt0.o")));
> + else
> + CmdArgs.push_back(
> + Args.MakeArgString(getToolChain().GetFilePath("crt0.o")));
> +@@ -7611,15 +7615,17 @@ void openbsd::Linker::ConstructJob(Compilation &C,
> con
> }
> }
>
Re: fix gcc 4.9 -static
Tobias Ulmer wrote:
> Sync STARTFILE_SPEC, fixes egcc -static producing crashing programs
>
> I don't expect any fallout from this (builtin spec file change only)
Checking with gcc -v, it seems to work on amd64. gcc 4.9 and our
base gcc behave the same given various -pie/-nopie/-shared/-static/-pg
combinations.
> OK?
>
> ? todo.txt
> Index: Makefile
> ===
> RCS file: /home/vcs/cvs/openbsd/ports/lang/gcc/4.9/Makefile,v
> retrieving revision 1.32
> diff -u -p -r1.32 Makefile
> --- Makefile 29 Jun 2016 16:14:41 - 1.32
> +++ Makefile 9 Jul 2016 21:41:29 -
> @@ -4,8 +4,7 @@ ONLY_FOR_ARCHS = amd64 arm hppa i386 mip
> DPB_PROPERTIES = parallel
>
> V = 4.9.3
> -REVISION = 7
> -REVISION-java = 8
> +REVISION = 9
> FULL_VERSION = $V
> FULL_PKGVERSION = $V
>
> Index: patches/patch-gcc_config_alpha_openbsd_h
> ===
> RCS file:
> /home/vcs/cvs/openbsd/ports/lang/gcc/4.9/patches/patch-gcc_config_alpha_openbsd_h,v
> retrieving revision 1.1.1.1
> diff -u -p -r1.1.1.1 patch-gcc_config_alpha_openbsd_h
> --- patches/patch-gcc_config_alpha_openbsd_h 26 Jun 2014 16:30:17 -
> 1.1.1.1
> +++ patches/patch-gcc_config_alpha_openbsd_h 9 Jul 2016 21:41:29 -
> @@ -1,6 +1,6 @@
> $OpenBSD: patch-gcc_config_alpha_openbsd_h,v 1.1.1.1 2014/06/26 16:30:17
> pascal Exp $
> gcc/config/alpha/openbsd.h.orig Thu Jan 10 21:38:27 2013
> -+++ gcc/config/alpha/openbsd.h Thu Apr 18 20:58:27 2013
> +--- gcc/config/alpha/openbsd.h.orig Thu Jan 2 23:23:26 2014
> gcc/config/alpha/openbsd.h Sat Jul 9 22:13:30 2016
> @@ -19,6 +19,28 @@ along with GCC; see the file COPYING3. If not see
>
> /* Controlling the compilation driver. */
> @@ -21,11 +21,11 @@ $OpenBSD: patch-gcc_config_alpha_openbsd
> +/* As an elf system, we need crtbegin/crtend stuff. */
> +#undef STARTFILE_SPEC
> +#define STARTFILE_SPEC "\
> -+%{!shared: %{pg:gcrt0%O%s} %{!pg:%{p:gcrt0%O%s} %{!p:crt0%O%s}} \
> -+crtbegin%O%s} %{shared:crtbeginS%O%s}"
> ++%{!shared: %{pg:gcrt0%O%s} %{!pg:%{p:gcrt0%O%s} \
> ++%{!p:%{!static:crt0%O%s} %{static:%{nopie:crt0%O%s} \
> ++%{!nopie:rcrt0%O%s crtbegin%O%s} %{shared:crtbeginS%O%s}"
> +#undef ENDFILE_SPEC
> +#define ENDFILE_SPEC "%{!shared:crtend%O%s} %{shared:crtendS%O%s}"
> -+
> +
> /* run-time target specifications */
> #define TARGET_OS_CPP_BUILTINS()\
> Index: patches/patch-gcc_config_i386_openbsdelf_h
> ===
> RCS file:
> /home/vcs/cvs/openbsd/ports/lang/gcc/4.9/patches/patch-gcc_config_i386_openbsdelf_h,v
> retrieving revision 1.1.1.1
> diff -u -p -r1.1.1.1 patch-gcc_config_i386_openbsdelf_h
> --- patches/patch-gcc_config_i386_openbsdelf_h26 Jun 2014 16:30:17
> - 1.1.1.1
> +++ patches/patch-gcc_config_i386_openbsdelf_h9 Jul 2016 21:41:29
> -
> @@ -1,7 +1,7 @@
> $OpenBSD: patch-gcc_config_i386_openbsdelf_h,v 1.1.1.1 2014/06/26 16:30:17
> pascal Exp $
> gcc/config/i386/openbsdelf.h.origSun Sep 2 16:13:21 2012
> -+++ gcc/config/i386/openbsdelf.h Fri Oct 12 11:24:33 2012
> -@@ -97,10 +97,13 @@ along with GCC; see the file COPYING3. If not see
> +--- gcc/config/i386/openbsdelf.h.origThu Jan 2 23:23:26 2014
> gcc/config/i386/openbsdelf.h Fri Jul 8 17:18:50 2016
> +@@ -97,14 +97,17 @@ along with GCC; see the file COPYING3. If not see
> %{shared:-shared} %{R*} \
> %{static:-Bstatic} \
> %{!static:-Bdynamic} \
> @@ -10,8 +10,15 @@ $OpenBSD: patch-gcc_config_i386_openbsde
> -dynamic-linker /usr/libexec/ld.so"
>
> #undef STARTFILE_SPEC
> -+
> +-#define STARTFILE_SPEC "\
> +-%{!shared: %{pg:gcrt0%O%s} %{!pg:%{p:gcrt0%O%s} %{!p:crt0%O%s}} \
> +-crtbegin%O%s} %{shared:crtbeginS%O%s}"
> +
> +#define SUBTARGET32_DEFAULT_CPU "i486"
> - #define STARTFILE_SPEC "\
> - %{!shared: %{pg:gcrt0%O%s} %{!pg:%{p:gcrt0%O%s} %{!p:crt0%O%s}} \
> - crtbegin%O%s} %{shared:crtbeginS%O%s}"
> ++#define STARTFILE_SPEC "\
> ++%{!shared: %{pg:gcrt0%O%s} %{!pg:%{p:gcrt0%O%s} \
> ++%{!p:%{!static:crt0%O%s} %{static:%{nopie:crt0%O%s} \
> ++%{!nopie:rcrt0%O%s crtbegin%O%s} %{shared:crtbeginS%O%s}"
> + #undef ENDFILE_SPEC
> + #define ENDFILE_SPEC "%{!shared:crtend%O%s} %{shared:crtendS%O%s}"
> +
> Index: patches/patch-gcc_config_mips_openbsd_h
> ===
> RCS file:
> /home/vcs/cvs/openbsd/ports/lang/gcc/4.9/patches/patch-gcc_config_mips_openbsd_h,v
> retrieving revision 1.1.1.1
> diff -u -p -r1.1.1.1 patch-gcc_config_mips_openbsd_h
> --- patches/patch-gcc_config_mips_openbsd_h 26 Jun 2014 16:30:18 -
> 1.1.1.1
> +++ patches/patch-gcc_config_mips_openbsd_h 9 Jul 2016 21:41:29 -
> @@ -1,7 +1,7 @@
> $OpenBSD: patch-gcc_config_mips_openbsd_h,v 1.1.1.1 2014/06/26 16:30:18
> pas
