lkml.org 
[lkml]   [2014]   [Jun]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: Segmentation fault on all golang executables
On Thu, Jun 12, 2014 at 10:20 AM, Andy Lutomirski <luto@amacapital.net> wrote:
> On Thu, Jun 12, 2014 at 9:59 AM, Kui Zhang <kuizhang@gmail.com> wrote:
>> Hello,
>>
>> This commit seem to cause Segmentation fault on any of my golang executables.
>
> Well, crap. It looks like the Go people took my sample vDSO parser...
> and broke it. WTF were they thinking? I should have noticed that
> they screwed it up when I was cc'd on this thing:
>
> https://code.google.com/p/go/source/detail?r=56ea40aac72b
>
> but I didn't. Sorry, everyone. And the issue isn't even something
> sensibly broken like trying to find vdso symbols in the symbol table.
> No, they're using the size of the SHT_DYNSYM section to deduce the
> number of entries *in the dynamic table*. This is just completely
> wrong.
>
> But even that's not all. They borrowed by error handling, so they
> should have silently failed to parse the vdso instead of crashing.
> But then they broke it completely by failing to zero the state, so
> they're just reading from initialized memory. Grr.
>
> The current offending code is here:
>
> https://code.google.com/p/go/source/browse/src/pkg/runtime/vdso_linux_amd64.c
>
> Since we don't get to break all Go executables, here are two options:
>
> 1. Try to keep the whole symbol table intact. This is annoying:
> there's a reason I removed it. The linker script doesn't know how big
> it is, so it's hard to make it compatible with the vvar table.
>
> 2. Write a dummy section table that contains a single empty section of
> type SHT_DYNSYM. Hopefully the Go runtime will then get far enough to
> fail cleanly. And they can go and write a real ELF parser or copy my
> reference parser correctly. (hpa, can you apply my patches to make
> the reference parser 32-bit clean?)

Sigh.

See attached. It seems to work for me. Can you test it? It'll hurt
performance for Go programs, but I don't feel too bad about that.

--Andy
From ad9f313026949247ed2067631b4e5c27c1935064 Mon Sep 17 00:00:00 2001
Message-Id: <ad9f313026949247ed2067631b4e5c27c1935064.1402598233.git.luto@amacapital.net>
From: Andy Lutomirski <luto@amacapital.net>
Date: Thu, 12 Jun 2014 11:31:46 -0700
Subject: [PATCH] x86,vdso: Hack to keep 64-bit Go programs working

The Go runtime has a buggy vDSO parser that currently segfaults.
This writes an empty SHT_DYNSYM entry that causes Go's runtime to
malfunction by thinking that the vDSO is empty rather than
malfunctioning by running off the end and segfaulting.

This is currently broken for big-endian build hosts. The hack
should also be disabled for x32, but I'm not sure what the right way
to do that is.

Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---
arch/x86/vdso/Makefile | 2 +-
arch/x86/vdso/vdso-fakesections.c | 20 ++++++++++++++++++++
arch/x86/vdso/vdso2c.h | 21 +++++++++++++++++----
3 files changed, 38 insertions(+), 5 deletions(-)
create mode 100644 arch/x86/vdso/vdso-fakesections.c

diff --git a/arch/x86/vdso/Makefile b/arch/x86/vdso/Makefile
index 9769df0..ffdeb5b 100644
--- a/arch/x86/vdso/Makefile
+++ b/arch/x86/vdso/Makefile
@@ -15,7 +15,7 @@ vdso-install-$(VDSO32-y) += $(vdso32-images)


# files to link into the vdso
-vobjs-y := vdso-note.o vclock_gettime.o vgetcpu.o
+vobjs-y := vdso-note.o vclock_gettime.o vgetcpu.o vdso-fakesections.o

vobjs-$(VDSOX32-y) += $(vobjx32s-compat)

diff --git a/arch/x86/vdso/vdso-fakesections.c b/arch/x86/vdso/vdso-fakesections.c
new file mode 100644
index 0000000..ef75ece
--- /dev/null
+++ b/arch/x86/vdso/vdso-fakesections.c
@@ -0,0 +1,20 @@
+/*
+ * Copyright 2014 Andy Lutomirski
+ * Subject to the GNU Public License, v.2
+ *
+ * Hack to keep broken Go programs working.
+ */
+
+#if __x86_64__ /* hack only needed for the 64-bit vDSO */
+
+#include <linux/elf.h>
+
+extern const __visible struct elf64_shdr vdso_fake_sections[];
+const __visible struct elf64_shdr vdso_fake_sections[] = {
+ {
+ .sh_type = SHT_DYNSYM,
+ .sh_entsize = sizeof(Elf64_Sym),
+ }
+};
+
+#endif
diff --git a/arch/x86/vdso/vdso2c.h b/arch/x86/vdso/vdso2c.h
index d9f6f61..43d9ab1 100644
--- a/arch/x86/vdso/vdso2c.h
+++ b/arch/x86/vdso/vdso2c.h
@@ -18,6 +18,8 @@ static void GOFUNC(void *addr, size_t len, FILE *outfile, const char *name)
const char *secstrings;
uint64_t syms[NSYMS] = {};

+ uint64_t fake_sections_value = 0, fake_sections_size = 0;
+
Elf_Phdr *pt = (Elf_Phdr *)(addr + GET_LE(&hdr->e_phoff));

/* Walk the segment table. */
@@ -84,6 +86,7 @@ static void GOFUNC(void *addr, size_t len, FILE *outfile, const char *name)
GET_LE(&symtab_hdr->sh_entsize) * i;
const char *name = addr + GET_LE(&strtab_hdr->sh_offset) +
GET_LE(&sym->st_name);
+
for (k = 0; k < NSYMS; k++) {
if (!strcmp(name, required_syms[k])) {
if (syms[k]) {
@@ -93,6 +96,13 @@ static void GOFUNC(void *addr, size_t len, FILE *outfile, const char *name)
syms[k] = GET_LE(&sym->st_value);
}
}
+
+ if (!strcmp(name, "vdso_fake_sections")) {
+ if (fake_sections_value)
+ fail("duplicate vdso_fake_sections\n");
+ fake_sections_value = GET_LE(&sym->st_value);
+ fake_sections_size = GET_LE(&sym->st_size);
+ }
}

/* Validate mapping addresses. */
@@ -112,10 +122,13 @@ static void GOFUNC(void *addr, size_t len, FILE *outfile, const char *name)
if (syms[sym_end_mapping] % 4096)
fail("end_mapping must be a multiple of 4096\n");

- /* Remove sections. */
- hdr->e_shoff = 0;
- hdr->e_shentsize = 0;
- hdr->e_shnum = 0;
+ /* Remove sections or use fakes */
+ if (fake_sections_size % sizeof(Elf_Shdr))
+ fail("vdso_fake_sections size is not a multiple of %ld\n",
+ (long)sizeof(Elf_Shdr));
+ hdr->e_shoff = fake_sections_value;
+ hdr->e_shentsize = fake_sections_value ? sizeof(Elf_Shdr) : 0;
+ hdr->e_shnum = fake_sections_size / sizeof(Elf_Shdr);
hdr->e_shstrndx = SHN_UNDEF; /* SHN_UNDEF == 0 */

if (!name) {
--
1.9.3
\
 
 \ /
  Last update: 2014-06-12 21:21    [W:0.056 / U:3.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site