lkml.org 
[lkml]   [2004]   [Aug]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patches in this message
/
From
Subject[PATCH] Oops when loading a stripped module
Date
I've stripped a module and tried to load it (I know it's meaningless, but it 
was for testing; I wanted to strip debug symbols). And to my surprise, the
kernel Oopsed. Well, not a security threat (you are root if you can load
modules), but anyway not nice, even because it dies while holding
module_mutex, so that it's nomore possible even to cat /proc/modules!
The oops is attached, together with the disassembly of the apply_alternatives
function, for deconding the EIP value: I think it's in the memcpy() call.
I've got a panic in the same situation even on UML, which defines a no-op
apply_alternatives(); I've not a stack trace for it, anyway. But anyway the
fix is to return -ENOEXEC for stripped objects; the attached patch (both
attachment and inline) should do this work (I'm not an ELF expert, but I
gave a quick look to the objdump and libbfd source code).

Anyway, some more checks should be added: an empty SHT_SYMTAB section could
probably cause the same thing (and not even root must be able to Oops the
kernel anyhow); and a malformed object could cause a lot of problems.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade_spam@yahoo.it>
---

vanilla-linux-2.6.7-SKAS-paolo/kernel/module.c | 10 +++++++++-
1 files changed, 9 insertions(+), 1 deletion(-)

diff -puN kernel/module.c~reject-stripped-module kernel/module.c
--- vanilla-linux-2.6.7-SKAS/kernel/module.c~reject-stripped-module 2004-08-19
11:34:11.000000000 +0200
+++ vanilla-linux-2.6.7-SKAS-paolo/kernel/module.c 2004-08-19
15:03:31.927895368 +0200
@@ -1456,7 +1456,8 @@ static struct module *load_module(void _
secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
sechdrs[0].sh_addr = 0;

- /* And these should exist, but gcc whinges if we don't init them */
+ /* And these should exist, but somebody could have stripped
+ * the module (yes, I did)*/
symindex = strindex = 0;

for (i = 1; i < hdr->e_shnum; i++) {
@@ -1481,6 +1482,13 @@ static struct module *load_module(void _
#endif
}

+ /*Oh, somebody fsck'ed it. Please, remember root != God ! */
+ if (symindex == 0) {
+ printk(KERN_ERR "Trying to load stripped module!!\n");
+ err = -ENOEXEC;
+ goto free_hdr;
+ }
+
modindex = find_sec(hdr, sechdrs, secstrings,
".gnu.linkonce.this_module");
if (!modindex) {
_
Unable to handle kernel NULL pointer dereference at virtual address 00000000
printing eip:
c0109c5d
*pde = 00000000
Oops: 0000 [#1]
PREEMPT
Modules linked in: tun ext3 jbd loop binfmt_misc snd_seq_oss
snd_seq_midi_event snd_seq snd_pcm_oss snd_mixer_oss snd_intel8x0
snd_ac97_codec snd_pcm snd_timer snd_page_alloc gameport snd_mpu401_uart
snd_rawmidi snd_seq_device snd soundcore af_packet ipv6 ide_cd cdrom ds
yenta_socket pcmcia_core eth1394 tg3 ohci1394 ieee1394 nls_iso8859_15
nls_cp850 vfat fat pcspkr non_fatal thermal fan button battery ac powernow_k8
processor psmouse amd64_agp agpgart sd_mod usb_storage scsi_mod dm_mod evdev
tsdev usbhid usbmouse ehci_hcd ohci_hcd usbcore rtc
CPU: 0
EIP: 0060:[apply_alternatives+125/224] Not tainted
EIP: 0060:[<c0109c5d>] Not tainted
EFLAGS: 00210202 (2.6.7-sysemu-fix-mprotect)
EIP is at apply_alternatives+0x7d/0xe0
eax: 00000003 ebx: e0fb3e48 ecx: 00000000 edx: e0fb3fc7
esi: 00000000 edi: 00005f37 ebp: c032edc0 esp: c977bee8
ds: 007b es: 007b ss: 0068
Process insmod (pid: 16714, threadinfo=c977a000 task=df7f30b0)
Stack: c032edc0 e0fb3fc7 e0f8afd0 c02f3cbf e0f8ad74 c02f3cae c011599e e0f8ace2
e0f8adc8 e0f61000 e0f8adc8 e0fb6480 000003c0 e0f61000 c012fe3d 00000000
e0f8ace2 e0fb6480 00030002 00000000 000003c0 c2153928 e0dc0000 e0fb6480
Call Trace:
[module_finalize+142/192] module_finalize+0x8e/0xc0
[<c011599e>] module_finalize+0x8e/0xc0
[load_module+2109/2496] load_module+0x83d/0x9c0
[<c012fe3d>] load_module+0x83d/0x9c0
[sys_init_module+96/592] sys_init_module+0x60/0x250
[<c0130020>] sys_init_module+0x60/0x250
[filp_close+79/144] filp_close+0x4f/0x90
[<c014ecdf>] filp_close+0x4f/0x90
[sysenter_past_esp+82/113] sysenter_past_esp+0x52/0x71
[<c0103f89>] sysenter_past_esp+0x52/0x71
Code: 66 a5 a8 01 74 01 a4 0f b6 43 0a 0f b6 53 09 29 c2 89 c5 85

(gdb) disassemble apply_alternatives
Dump of assembler code for function apply_alternatives:
0xc0109be0 <apply_alternatives+0>: push %ebp
0xc0109be1 <apply_alternatives+1>: mov %eax,%ecx
0xc0109be3 <apply_alternatives+3>: xor %ebp,%ebp
0xc0109be5 <apply_alternatives+5>: push %edi
0xc0109be6 <apply_alternatives+6>: push %esi
0xc0109be7 <apply_alternatives+7>: push %ebx
0xc0109be8 <apply_alternatives+8>: sub $0x8,%esp
0xc0109beb <apply_alternatives+11>: mov 0xc032ee24,%eax
0xc0109bf0 <apply_alternatives+16>: mov %edx,0x4(%esp,1)
0xc0109bf4 <apply_alternatives+20>: movl $0xc032ed80,(%esp,1)
0xc0109bfb <apply_alternatives+27>: test %eax,%eax
0xc0109bfd <apply_alternatives+29>: js 0xc0109c25
<apply_alternatives+69>
0xc0109bff <apply_alternatives+31>: nop
0xc0109c00 <apply_alternatives+32>: bt %eax,0xc032ec0c
0xc0109c07 <apply_alternatives+39>: sbb %eax,%eax
0xc0109c09 <apply_alternatives+41>: test %eax,%eax
0xc0109c0b <apply_alternatives+43>: jne 0xc0109c1b
<apply_alternatives+59>
0xc0109c0d <apply_alternatives+45>: inc %ebp
0xc0109c0e <apply_alternatives+46>: mov 0xc032ee24(,%ebp,8),%eax
0xc0109c15 <apply_alternatives+53>: test %eax,%eax
0xc0109c17 <apply_alternatives+55>: jns 0xc0109c00
<apply_alternatives+32>
0xc0109c19 <apply_alternatives+57>: jmp 0xc0109c25
<apply_alternatives+69>
0xc0109c1b <apply_alternatives+59>: mov 0xc032ee28(,%ebp,8),%ebp
0xc0109c22 <apply_alternatives+66>: mov %ebp,(%esp,1)
0xc0109c25 <apply_alternatives+69>: cmp 0x4(%esp,1),%ecx
0xc0109c29 <apply_alternatives+73>: mov %ecx,%ebx
0xc0109c2b <apply_alternatives+75>: jae 0xc0109cac
<apply_alternatives+204>
---Type <return> to continue, or q <return> to quit---
0xc0109c2d <apply_alternatives+77>: lea 0x0(%esi),%esi
0xc0109c30 <apply_alternatives+80>: movzbl 0x8(%ebx),%eax
0xc0109c34 <apply_alternatives+84>: bt %eax,0xc032ec0c
0xc0109c3b <apply_alternatives+91>: sbb %eax,%eax
0xc0109c3d <apply_alternatives+93>: test %eax,%eax
0xc0109c3f <apply_alternatives+95>: je 0xc0109ca3
<apply_alternatives+195>
0xc0109c41 <apply_alternatives+97>: movzbl 0xa(%ebx),%eax
0xc0109c45 <apply_alternatives+101>: cmp 0x9(%ebx),%al
0xc0109c48 <apply_alternatives+104>: ja 0xc0109cb4
<apply_alternatives+212>
0xc0109c4a <apply_alternatives+106>: movzbl %al,%eax
0xc0109c4d <apply_alternatives+109>: mov (%ebx),%edi
0xc0109c4f <apply_alternatives+111>: mov 0x4(%ebx),%esi
0xc0109c52 <apply_alternatives+114>: mov %eax,%ecx
0xc0109c54 <apply_alternatives+116>: shr $0x2,%ecx
0xc0109c57 <apply_alternatives+119>: repz movsl %ds:(%esi),%es:(%edi)
0xc0109c59 <apply_alternatives+121>: test $0x2,%al
0xc0109c5b <apply_alternatives+123>: je 0xc0109c5f
<apply_alternatives+127>
0xc0109c5d <apply_alternatives+125>: movsw %ds:(%esi),%es:(%edi)
0xc0109c5f <apply_alternatives+127>: test $0x1,%al
0xc0109c61 <apply_alternatives+129>: je 0xc0109c64
<apply_alternatives+132>
0xc0109c63 <apply_alternatives+131>: movsb %ds:(%esi),%es:(%edi)
0xc0109c64 <apply_alternatives+132>: movzbl 0xa(%ebx),%eax
0xc0109c68 <apply_alternatives+136>: movzbl 0x9(%ebx),%edx
0xc0109c6c <apply_alternatives+140>: sub %eax,%edx
0xc0109c6e <apply_alternatives+142>: mov %eax,%ebp
0xc0109c70 <apply_alternatives+144>: test %edx,%edx
0xc0109c72 <apply_alternatives+146>: jle 0xc0109ca3
<apply_alternatives+195>
0xc0109c74 <apply_alternatives+148>: mov (%esp,1),%ecx
---Type <return> to continue, or q <return> to quit---
0xc0109c77 <apply_alternatives+151>: cmp $0x9,%edx
0xc0109c7a <apply_alternatives+154>: mov (%ebx),%edi
0xc0109c7c <apply_alternatives+156>: mov $0x8,%eax
0xc0109c81 <apply_alternatives+161>: cmovl %edx,%eax
0xc0109c84 <apply_alternatives+164>: mov (%ecx,%eax,4),%esi
0xc0109c87 <apply_alternatives+167>: add %ebp,%edi
0xc0109c89 <apply_alternatives+169>: mov %eax,%ecx
0xc0109c8b <apply_alternatives+171>: shr $0x2,%ecx
0xc0109c8e <apply_alternatives+174>: repz movsl %ds:(%esi),%es:(%edi)
0xc0109c90 <apply_alternatives+176>: test $0x2,%al
0xc0109c92 <apply_alternatives+178>: je 0xc0109c96
<apply_alternatives+182>
0xc0109c94 <apply_alternatives+180>: movsw %ds:(%esi),%es:(%edi)
0xc0109c96 <apply_alternatives+182>: test $0x1,%al
0xc0109c98 <apply_alternatives+184>: je 0xc0109c9b
<apply_alternatives+187>
0xc0109c9a <apply_alternatives+186>: movsb %ds:(%esi),%es:(%edi)
0xc0109c9b <apply_alternatives+187>: add %eax,%ebp
0xc0109c9d <apply_alternatives+189>: sub %eax,%edx
0xc0109c9f <apply_alternatives+191>: test %edx,%edx
0xc0109ca1 <apply_alternatives+193>: jg 0xc0109c74
<apply_alternatives+148>
0xc0109ca3 <apply_alternatives+195>: add $0xc,%ebx
0xc0109ca6 <apply_alternatives+198>: cmp 0x4(%esp,1),%ebx
0xc0109caa <apply_alternatives+202>: jb 0xc0109c30
<apply_alternatives+80>
0xc0109cac <apply_alternatives+204>: add $0x8,%esp
0xc0109caf <apply_alternatives+207>: pop %ebx
0xc0109cb0 <apply_alternatives+208>: pop %esi
0xc0109cb1 <apply_alternatives+209>: pop %edi
0xc0109cb2 <apply_alternatives+210>: pop %ebp
0xc0109cb3 <apply_alternatives+211>: ret
---Type <return> to continue, or q <return> to quit---
0xc0109cb4 <apply_alternatives+212>: ud2a
0xc0109cb6 <apply_alternatives+214>: adc $0x2f214004,%eax
0xc0109cbb <apply_alternatives+219>: shr $0x8c,%bl
0xc0109cbe <apply_alternatives+222>: nop
0xc0109cbf <apply_alternatives+223>: nop
End of assembler dump.
--
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729




We get Oopses when loading stripped modules... and they are useless.
I hope this patch will detect them.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade_spam@yahoo.it>
---

vanilla-linux-2.6.7-SKAS-paolo/kernel/module.c | 10 +++++++++-
1 files changed, 9 insertions(+), 1 deletion(-)

diff -puN kernel/module.c~reject-stripped-module kernel/module.c
--- vanilla-linux-2.6.7-SKAS/kernel/module.c~reject-stripped-module 2004-08-19 11:34:11.000000000 +0200
+++ vanilla-linux-2.6.7-SKAS-paolo/kernel/module.c 2004-08-19 15:03:31.927895368 +0200
@@ -1456,7 +1456,8 @@ static struct module *load_module(void _
secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
sechdrs[0].sh_addr = 0;

- /* And these should exist, but gcc whinges if we don't init them */
+ /* And these should exist, but somebody could have stripped
+ * the module (yes, I did)*/
symindex = strindex = 0;

for (i = 1; i < hdr->e_shnum; i++) {
@@ -1481,6 +1482,13 @@ static struct module *load_module(void _
#endif
}

+ /*Oh, somebody fsck'ed it. Please, remember root != God ! */
+ if (symindex == 0) {
+ printk(KERN_ERR "Trying to load stripped module!!\n");
+ err = -ENOEXEC;
+ goto free_hdr;
+ }
+
modindex = find_sec(hdr, sechdrs, secstrings,
".gnu.linkonce.this_module");
if (!modindex) {
_
\
 
 \ /
  Last update: 2005-03-22 14:05    [W:0.052 / U:0.056 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site