Skip to contentSkip to conversation

[2.6] binfmt_elf bug (exposed by klibc).

PATCH Paweł Sikora 4 participants 10 messages Oct 7, 2005 – Oct 8, 2005 2 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi All,

I've a simple program called empty.c.

$ cat empty.c

int main(int argc, char* argv[])

{
    return 0;
}
$ cat empty410.s
        .file   "empty.c"
        .text
        .p2align 4,,15

.globl main
.type main, @function
main:
xorl %eax, %eax
ret
.size main, .-main
.ident "GCC: (GNU) 4.1.0 20050922 (experimental)"
.section .note.GNU-stack,"", at progbits

binutils-2.15.94.0.2.2 produces for this code empty .data and .bss sections:

Program Header:

    PHDR off    0x00000034 vaddr 0x08048034 paddr 0x08048034 align 2**2
         filesz 0x000000c0 memsz 0x000000c0 flags r-x
  INTERP off    0x000000f4 vaddr 0x080480f4 paddr 0x080480f4 align 2**0
         filesz 0x0000002a memsz 0x0000002a flags r--
    LOAD off    0x00000000 vaddr 0x08048000 paddr 0x08048000 align 2**12
         filesz 0x00000123 memsz 0x00000123 flags r-x
    LOAD off    0x00000124 vaddr 0x08049124 paddr 0x08049124 align 2**12
         filesz 0x00000000 memsz 0x00000000 flags rw-
   STACK off    0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
         filesz 0x00000000 memsz 0x00000000 flags rwx
PAX_FLAGS off    0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
         filesz 0x00000000 memsz 0x00000000 flags --- 2800

Sections:

Idx Name          Size      VMA       LMA       File off  Algn
  0 .interp       0000002a  080480f4  080480f4  000000f4  2**0

CONTENTS, ALLOC, LOAD, READONLY, DATA
1 .text 00000003 08048120 08048120 00000120 2**4
CONTENTS, ALLOC, LOAD, READONLY, CODE
2 .data 00000000 08049124 08049124 00000124 2**2
CONTENTS, ALLOC, LOAD, DATA
3 .bss 00000000 08049124 08049124 00000124 2**2
ALLOC

and /lib/klibc-*.so interpeter works fine.

binutils-2.16.91.0.3 doesn't produce useless empty sections:

Program Header:

    PHDR off    0x00000034 vaddr 0x08048034 paddr 0x08048034 align 2**2
         filesz 0x000000a0 memsz 0x000000a0 flags r-x
  INTERP off    0x000000f4 vaddr 0x080480f4 paddr 0x080480f4 align 2**0
         filesz 0x0000002a memsz 0x0000002a flags r--
    LOAD off    0x00000000 vaddr 0x08048000 paddr 0x08048000 align 2**12
         filesz 0x00000123 memsz 0x00000123 flags r-x
   STACK off    0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
         filesz 0x00000000 memsz 0x00000000 flags rwx
PAX_FLAGS off    0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
         filesz 0x00000000 memsz 0x00000000 flags --- 2800

Sections:

Idx Name          Size      VMA       LMA       File off  Algn
  0 .interp       0000002a  080480f4  080480f4  000000f4  2**0

CONTENTS, ALLOC, LOAD, READONLY, DATA
1 .text 00000003 08048120 08048120 00000120 2**4
CONTENTS, ALLOC, LOAD, READONLY, CODE

...and klibc's loader doesn't reach crt0.o:_start().

It receives sigsegv from kernel. The load_elf_binary() calls padzero() for nonexistent .bss mapping.

Attached workaround fixes problem. Could anybody look into this?

Regards,
Paweł.

Signature · 3 lines
The only thing necessary for the triumph of evil
  is for good men to do nothing.
                                           - Edmund Burke
fs/binfmt_elf.c +1 -1
@@ -905,7 +905,7 @@ static int load_elf_binary(struct linux_
 		send_sig(SIGKILL, current, 0);
 		goto out_free_dentry;
 	}
-	if (padzero(elf_bss)) {
+	if ((elf_bss != elf_brk) && padzero(elf_bss)) {
 		send_sig(SIGSEGV, current, 0);
 		retval = -EFAULT; /* Nobody gets to see this, but.. */
 		goto out_free_dentry;
#1 lore raw
Quoted 14 lines · Paweł Sikora

I've a simple program called empty.c.

$ cat empty.c

int main(int argc, char* argv[])

{
    return 0;
}
$ cat empty410.s
        .file   "empty.c"
        .text
        .p2align 4,,15
Quoted 8 lines

.globl main
.type main, @function
main:
xorl %eax, %eax
ret
.size main, .-main
.ident "GCC: (GNU) 4.1.0 20050922 (experimental)"
.section .note.GNU-stack,"", at progbits

I get a substantially different empty.s (gcc-4.0.2, Fedora rawhide on i686):

        .file   "empty.c"
        .text
        .p2align 2,,3
.globl main
        .type   main, @function
main:
        pushl   %ebp
        movl    %esp, %ebp
        subl    $8, %esp
        andl    $-16, %esp
        subl    $16, %esp
        xorl    %eax, %eax

leave
ret
.size main, .-main
.ident "GCC: (GNU) 4.0.2 20050928 (Red Hat 4.0.2-1)"
.section .note.GNU-stack,"",@progbits

Quoted 2 lines

binutils-2.15.94.0.2.2 produces for this code empty .data and .bss sections:

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools.

Signature · 4 lines
Dr. Horst H. von Brand                   User #22616 counter.li.org
Departamento de Informatica                     Fono: +56 32 654431
Universidad Tecnica Federico Santa Maria              +56 32 654239
Casilla 110-V, Valparaiso, Chile                Fax:  +56 32 797513
#2 lore raw
Quoted 12 lines · Horst von Brand
Quoted 11 lines · PaweÅ' Sikora

I've a simple program called empty.c.

$ cat empty.c

int main(int argc, char* argv[])

{
    return 0;
}
$ cat empty410.s
Quoted 12 lines
Quoted 11 lines
        .file   "empty.c"
        .text
        .p2align 4,,15

.globl main
.type main, @function
main:
xorl %eax, %eax
ret
.size main, .-main
.ident "GCC: (GNU) 4.1.0 20050922 (experimental)"
.section .note.GNU-stack,"", at progbits

With this compilation, the 'C' compiler looked and said; "you are just going to return 0..." So that's the code it generated. This compiler probably defaults to "-fomit-frame-pointer".

Quoted 14 lines

I get a substantially different empty.s (gcc-4.0.2, Fedora rawhide on i686):

       .file   "empty.c"
       .text
       .p2align 2,,3
.globl main
       .type   main, @function
main:
       pushl   %ebp
       movl    %esp, %ebp
       subl    $8, %esp
       andl    $-16, %esp
       subl    $16, %esp
       xorl    %eax, %eax
Quoted 5 lines

leave
ret
.size main, .-main
.ident "GCC: (GNU) 4.0.2 20050928 (Red Hat 4.0.2-1)"
.section .note.GNU-stack,"",@progbits

With this compiler, it assumed that you were going to do something useful so it set up a stack-frame. It also aligned the stack after it saved the intitial value in EBP. Then it cleared the return- value, reset the stack, and returned.

This compiler did not default to "-fomit-frame-pointer" so it set up a stack-frame.

Quoted 4 lines
Quoted 3 lines

binutils-2.15.94.0.2.2 produces for this code empty .data and .bss sections:

There is nothing wrong here at all.

Quoted 12 lines

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools. --
Dr. Horst H. von Brand User #22616 counter.li.org
Departamento de Informatica Fono: +56 32 654431
Universidad Tecnica Federico Santa Maria +56 32 654239
Casilla 110-V, Valparaiso, Chile Fax: +56 32 797513
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Cheers,
Dick Johnson
Penguin : Linux version 2.6.13 on an i686 machine (5589.54 BogoMips). Warning : 98.36% of all statistics are fiction.

****************************************************************
The information transmitted in this message is confidential and may be privileged. Any review, retransmission, dissemination, or other use of this information by persons or entities other than the intended recipient is prohibited. If you are not the intended recipient, please notify Analogic Corporation immediately - by replying to this message or by sending an email to [email protected] - and destroy all copies of this information, including any attachments, without reading or disclosing them.

Thank you.

#3 lore raw
In reply to Horst von Brand
Quoted 45 lines · Horst von Brand
Quoted 14 lines · Paweł Sikora

I've a simple program called empty.c.

$ cat empty.c

int main(int argc, char* argv[])

{
    return 0;
}
$ cat empty410.s
        .file   "empty.c"
        .text
        .p2align 4,,15
Quoted 8 lines

.globl main
.type main, @function
main:
xorl %eax, %eax
ret
.size main, .-main
.ident "GCC: (GNU) 4.1.0 20050922 (experimental)"
.section .note.GNU-stack,"", at progbits

I get a substantially different empty.s (gcc-4.0.2, Fedora rawhide on i686):

        .file   "empty.c"
        .text
        .p2align 2,,3
.globl main
        .type   main, @function
main:
        pushl   %ebp
        movl    %esp, %ebp
        subl    $8, %esp
        andl    $-16, %esp
        subl    $16, %esp
        xorl    %eax, %eax

leave
ret
.size main, .-main
.ident "GCC: (GNU) 4.0.2 20050928 (Red Hat 4.0.2-1)"
.section .note.GNU-stack,"",@progbits

Quoted 4 lines
Quoted 2 lines

binutils-2.15.94.0.2.2 produces for this code empty .data and .bss sections:

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools.

I didn't say that is (or not) a binutils bug.
I'm only saying that kernel is killng a valid micro application.

Signature · 3 lines
The only thing necessary for the triumph of evil
  is for good men to do nothing.
                                           - Edmund Burke
#4 lore raw
Quoted 1 line · Paweł Sikora

Dnia piątek, 7 października 2005 15:46, Horst von Brand napisał:

[...]

Quoted 4 lines
Quoted 1 line

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools.

I didn't say that is (or not) a binutils bug.
I'm only saying that kernel is killng a valid micro application.

If binutils generates an invalid executable, it is not a valid application.

Signature · 4 lines
Dr. Horst H. von Brand                   User #22616 counter.li.org
Departamento de Informatica                     Fono: +56 32 654431
Universidad Tecnica Federico Santa Maria              +56 32 654239
Casilla 110-V, Valparaiso, Chile                Fax:  +56 32 797513
#5 lore raw
Quoted 5 lines · Horst von Brand
Quoted 1 line · Paweł Sikora

Dnia piątek, 7 października 2005 15:46, Horst von Brand napisał:

[...]

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools.

I didn't say that is (or not) a binutils bug.
I'm only saying that kernel is killng a valid micro application.

If binutils generates an invalid executable, it is not a valid application.

ehh, please look again at my first post :)
binutils-2.16 generates VALID app with .text/.interp and *without* .bss. kernel always calls padzero() for the .bss section inside load_elf_binary(). finally it kills a valid app. did i miss something?

Signature · 3 lines
The only thing necessary for the triumph of evil
  is for good men to do nothing.
                                           - Edmund Burke

A message that is not in the archive (sent off-list or not yet received): [email protected]

#7 lore raw
Quoted 52 lines · linux-os (Dick Johnson)
Quoted 12 lines · On Fri, 7 Oct 2005, [UTF-8] Pawe? Sikora wrote:

Dnia pitek, 7 padziernika 2005 17:33, Horst von Brand napisa:

Quoted 11 lines
Quoted 1 line · Pawe Sikora

Dnia pitek, 7 padziernika 2005 15:46, Horst von Brand napisa:

[...]

Quoted 5 lines
Quoted 2 lines

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools.

I didn't say that is (or not) a binutils bug.
I'm only saying that kernel is killng a valid micro application.

Quoted 8 lines
Quoted 2 lines

If binutils generates an invalid executable, it is not a valid application.

ehh, please look again at my first post :)
binutils-2.16 generates VALID app with .text/.interp and *without* .bss. kernel always calls padzero() for the .bss section inside load_elf_binary(). finally it kills a valid app. did i miss something?

The executable created by this:

$ cat <<EOF >xxx.S
.section	.rodata
hello:		.string	"Hello World!\n"
STRLEN = .-hello
WRITE=4
EXIT=1

.section .text
.global _start
.type _start,@function
_start:

(quoted diff) +0 -0
 	movl	$WRITE, %eax
 	movl	$1, %ebx
 	movl	$hello, %ecx
 	movl	$STRLEN, %edx
 	int	$0x80
 	movl	$EXIT, %eax
 	movl	$0, %ebx
 	int	$0x80

.end
EOF

$ as -o xxx.o xxx.S
$ ld -o xxx xxx.o
$ ./xxx
Hello World!
$
Quoted 7 lines

... does not have a .bss section. It also runs fine. The linker creates a 0 length .bss section starting at label "_end". There is no way to prevent it from happening so the kernel's zeroing the zero-length section is perfectly valid. Maybe you have executed strip and stripped out that section? If so, you no longer have a valid executable and the kernel should kill it.

What????????????
Do you suggest that stripped executables are invalid?

$ cat xxx.S
.section        .text
.global         _start
.type           _start,@function
_start:
        movl    $1, %eax
        movl    $0, %ebx
        int     $0x80
.end
$ as xxx.S -o xxx.o; ld xxx.o -o xxx -s; objdump -x xxx

xxx: file format elf32-i386
xxx
architecture: i386, flags 0x00000102:
EXEC_P, D_PAGED
start address 0x08048094

Program Header:

    LOAD off    0x00000000 vaddr 0x08048000 paddr 0x08048000 align 2**12
         filesz 0x000000a0 memsz 0x000000a0 flags r-x
PAX_FLAGS off   0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
         filesz 0x00000000 memsz 0x00000000 flags --- 2800

Sections:

Idx Name          Size      VMA       LMA       File off  Algn
  0 .text         0000000c  08048094  08048094  00000094  2**2

CONTENTS, ALLOC, LOAD, READONLY, CODE

We have a pure executable, no .data/.rodata/.bss/etc.

On 2.6.14rc3-git6 this executable doesn't work:

$ strace ./xxx
execve("./xxx", ["./xxx"], [/* 24 vars */]) = -1 EFAULT (Bad address)
--- SIGSEGV (Segmentation fault) @ 0 (0) ---
+++ killed by SIGSEGV +++

With hacked kernel it works pretty fine:

$ strace ./xxx
execve("./xxx", ["./xxx"], [/* 24 vars */]) = 0
_exit(0)
Signature · 3 lines
The only thing necessary for the triumph of evil
  is for good men to do nothing.
                                           - Edmund Burke

A message that is not in the archive (sent off-list or not yet received): [email protected]

#9 lore raw
Quoted 61 lines · linux-os (Dick Johnson)
Quoted 60 lines · On Fri, 7 Oct 2005, [UTF-8] Pawe? Sikora wrote:

Dnia pitek, 7 padziernika 2005 18:16, linux-os (Dick Johnson) napisa:

Quoted 58 lines · On Fri, 7 Oct 2005, [UTF-8] Pawe? Sikora wrote:
Quoted 12 lines

Dnia pitek, 7 padziernika 2005 17:33, Horst von Brand napisa:

Quoted 11 lines
Quoted 1 line · Pawe Sikora

Dnia pitek, 7 padziernika 2005 15:46, Horst von Brand napisa:

[...]

Quoted 5 lines
Quoted 2 lines

binutils-2.16.91.0.2-4 doesn't. It looks like you are using broken tools.

I didn't say that is (or not) a binutils bug.
I'm only saying that kernel is killng a valid micro application.

Quoted 7 lines
Quoted 2 lines

If binutils generates an invalid executable, it is not a valid application.

ehh, please look again at my first post :)
binutils-2.16 generates VALID app with .text/.interp and *without* .bss. kernel always calls padzero() for the .bss section inside load_elf_binary(). finally it kills a valid app. did i miss something?

The executable created by this:

$ cat <<EOF >xxx.S
.section	.rodata
hello:		.string	"Hello World!\n"
STRLEN = .-hello
WRITE=4
EXIT=1

.section .text
.global _start
.type _start,@function
_start:

(quoted diff) +0 -0
 	movl	$WRITE, %eax
 	movl	$1, %ebx
 	movl	$hello, %ecx
 	movl	$STRLEN, %edx
 	int	$0x80
 	movl	$EXIT, %eax
 	movl	$0, %ebx
 	int	$0x80

.end
EOF

$ as -o xxx.o xxx.S
$ ld -o xxx xxx.o
$ ./xxx
Hello World!
$

... does not have a .bss section. It also runs fine. The linker creates a 0 length .bss section starting at label "_end". There is no way to prevent it from happening so the kernel's zeroing the zero-length section is perfectly valid. Maybe you have executed strip and stripped out that section? If so, you no longer have a valid executable and the kernel should kill it.

Quoted 4 lines
Quoted 2 lines

What????????????
Do you suggest that stripped executables are invalid?

No, not if you don't strip out sections that are required.

Please tell me what requires these useless zero-sized .data/.bss sections? Kernel design or something else? (e.g. some standard?)
Maybe H.J.Lu will know better as we are.

Quoted 24 lines
Quoted 24 lines
$ cat xxx.S
.section        .text
.global         _start
.type           _start,@function
_start:
       movl    $1, %eax
       movl    $0, %ebx
       int     $0x80
.end
$ as xxx.S -o xxx.o; ld xxx.o -o xxx -s; objdump -x xxx

xxx: file format elf32-i386
xxx
architecture: i386, flags 0x00000102:
EXEC_P, D_PAGED
start address 0x08048094

Program Header:

   LOAD off    0x00000000 vaddr 0x08048000 paddr 0x08048000 align 2**12
        filesz 0x000000a0 memsz 0x000000a0 flags r-x
PAX_FLAGS off  0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
        filesz 0x00000000 memsz 0x00000000 flags --- 2800
Quoted 8 lines
Quoted 6 lines

Sections:
Idx Name Size VMA LMA File off Algn
0 .text 0000000c 08048094 08048094 00000094 2**2
CONTENTS, ALLOC, LOAD, READONLY, CODE

We have a pure executable, no .data/.rodata/.bss/etc.

Your executable works fine on 2.6.14 as it should.

Hmm, it's strange. On my 2.6.14rc3-git6 + binutils-2.16.91.0.3 it doesn't work (vide -EFAULT). On 2.6.13 + binutils-2.15.94.0.2.2 it works fine (executable contains zero-sized .data + .bss sections). (btw. fs/binfmt_elf.c are the same)

Quoted 8 lines
Quoted 6 lines
$ strace ./xxx
execve("./xxx", ["./xxx"], [/* 24 vars */]) = -1 EFAULT (Bad address)
killed by SIGSEGV +++ renamed from SIGSEGV (Segmentation fault) @ 0 (0) --- +0 -0

With hacked kernel it works pretty fine:

What did you hack? Patch please.

Ugly workaround: lkml.org/lkml/2005/10/7/48/

Quoted 2 lines

Did somebody accidentally
screw up some kernel code between 2.6.13 and 2.6.14?

I think kernel elf loader doesn't handle binaries without .bss. Earlier binutils (<2.16) emits zero-sized .data/.bss and problem wasn't exposed. Modern binutils doesn't emit useless zero-sized .data/.bss sections and kernel kills these binaries.

Regards,
Paweł.

Signature · 3 lines
The only thing necessary for the triumph of evil
  is for good men to do nothing.
                                           - Edmund Burke
#10 lore raw
Quoted 7 lines · Paweł Sikora
Quoted 2 lines

Did somebody accidentally
screw up some kernel code between 2.6.13 and 2.6.14?

I think kernel elf loader doesn't handle binaries without .bss. Earlier binutils (<2.16) emits zero-sized .data/.bss and problem wasn't exposed. Modern binutils doesn't emit useless zero-sized .data/.bss sections and kernel kills these binaries.

I had this problem at some time, too. This was when I started to redo the uClibc port to vax-linux, which I started with a hand-crafted assembly file. It also crashed upon execution, though I was sure the program was technically okay.

However, I haven't looked up any paper or standard to verify either position. So I don't know for *sure* if it's legal to omit these (empty) sections.

MfG, JBG

Signature · 4 lines
Jan-Benedict Glaw       [email protected]    . +49-172-7608481             _ O _
"Eine Freie Meinung in  einem Freien Kopf    | Gegen Zensur | Gegen Krieg  _ _ O
 für einen Freien Staat voll Freier Bürger"  | im Internet! |   im Irak!   O O O
ret = do_actions((curr | FREE_SPEECH) & ~(NEW_COPYRIGHT_LAW | DRM | TCPA));
#11 lore raw
Quoted 9 lines · Jan-Benedict Glaw
Quoted 7 lines · Paweł Sikora
Quoted 2 lines

Did somebody accidentally
screw up some kernel code between 2.6.13 and 2.6.14?

I think kernel elf loader doesn't handle binaries without .bss. Earlier binutils (<2.16) emits zero-sized .data/.bss and problem wasn't exposed. Modern binutils doesn't emit useless zero-sized .data/.bss sections and kernel kills these binaries.

Quoted 4 lines

I had this problem at some time, too. This was when I started to redo the uClibc port to vax-linux, which I started with a hand-crafted assembly file. It also crashed upon execution, though I was sure the program was technically okay.

I think kernel wrongly assumes that all binaries in the world have .text and at least .data/.bss. E.g. in embedded world software are often pure, minimalistic and technically okay.

Signature · 3 lines
The only thing necessary for the triumph of evil
  is for good men to do nothing.
                                           - Edmund Burke