Skip to contentSkip to conversation

System reset on Kernel 2.4.10

DISCUSSION Petr Vandrovec 5 participants 6 messages Oct 2, 2001 – Oct 3, 2001

Subsystems (2)
lore
Share & tools
lore
#0 lore raw
Quoted 6 lines · VDA

V> Straced vmlinux does not reboot.
V> Kernel: 2.4.10+ext3+preempt

Well... sometimes it reboots too.
Once it rebooted ~10 mins after strace (system was at zero load). Also it rebooted after two strace's in succession.

Look at fs/binfmt_elf.c, at line 642 (in -ac2). There is

error = elf_map(....)

but nobody bothers with checking error value, it even tries it to use as an offset if stars are in wrong constellation. If you could add these lines below the call:

if ((unsigned long)error >= (unsigned long)(-256)) {
set_fs(old_fs);
printk(KERN_DEBUG "Something went wrong with elf_map()\n"); kfree(elf_phdata);
send_sig(SIGSEGV, current, 0);
return 0;
}

and then report results...
Petr Vandrovec
(not willing to test it myself)

#1 lore raw
Quoted 24 lines · Petr Vandrovec
Quoted 6 lines · VDA

V> Straced vmlinux does not reboot.
V> Kernel: 2.4.10+ext3+preempt

Well... sometimes it reboots too.
Once it rebooted ~10 mins after strace (system was at zero load). Also it rebooted after two strace's in succession.

Look at fs/binfmt_elf.c, at line 642 (in -ac2). There is

error = elf_map(....)

but nobody bothers with checking error value, it even tries it to use as an offset if stars are in wrong constellation. If you could add these lines below the call:

if ((unsigned long)error >= (unsigned long)(-256)) {
set_fs(old_fs);
printk(KERN_DEBUG "Something went wrong with elf_map()\n"); kfree(elf_phdata);
send_sig(SIGSEGV, current, 0);
return 0;
}

Quoted 1 line

and then report results...

Well, I was not able to trigger reboot with unpatched kernel. With patched one behavior looks same to me, except that elf_map went wrong is printed by kernel.

I was not able to find where problem could be with unpatched kernel, but arguments passed to do_brk(), set into mm->start_brk, {start,end}_code and so on looks very suspicious... But as on my system it does not crash neither with nor without patch below, I leave answer on someone else.

Btw, my system is 2.4.10-ac2, SMP PIII, compiled with Debian 2.95.4.
Petr Vandrovec [email protected]

fs/binfmt_elf.c renamed from fs/binfmt_elf.c.xx +7 -1
@@ -640,7 +640,13 @@
 		}
 
 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt, elf_prot, elf_flags);
-
+		if ((unsigned long)error >= (unsigned long)(-256)) {
+			set_fs(old_fs);
+			printk(KERN_DEBUG "elf_map went wrong\n");
+			kfree(elf_phdata);
+			send_sig(SIGSEGV, current, 0);
+			return 0;
+		}
 		if (!load_addr_set) {
 			load_addr_set = 1;
 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
#2 lore raw
In reply to Petr Vandrovec
Quoted 5 lines · Petr Vandrovec

I was not able to find where problem could be with unpatched kernel, but arguments passed to do_brk(), set into mm->start_brk, {start,end}_code and so on looks very suspicious... But as on my system it does not crash neither with nor without patch below, I leave answer on someone else.

Well, your patch does seem to fix it:

$ /usr/src/linux-2.4.10/vmlinux
Segmentation fault
$ dmesg
...
elf_map error code: -22

Cheers,

MikaL

#3 lore raw
In reply to Petr Vandrovec
Quoted 6 lines · Petr Vandrovec

I was not able to find where problem could be with unpatched kernel, but arguments passed to do_brk(), set into mm->start_brk, {start,end}_code and so on looks very suspicious... But as on my system it does not crash neither with nor without patch below, I leave answer on someone else.

I no longer get the restart but then I also made more than this change to the kernel and tested with the new vmlinux image.

#4 lore raw
In reply to Petr Vandrovec
Quoted 5 lines · Petr Vandrovec

but nobody bothers with checking error value, it even tries it to use as an offset if stars are in wrong constellation. If you could add these lines below the call:

if ((unsigned long)error >= (unsigned long)(-256)) {

What's wrong with IS_ERR(error)?

Cheers, Andreas
--
Andreas Dilger \ "If a man ate a pound of pasta and a pound of antipasto,
\ would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/ -- Dogbert

#5 lore raw
In reply to Petr Vandrovec

Hello Petr,
Wednesday, October 03, 2001, 2:02:28 PM, you wrote:
PV> Look at fs/binfmt_elf.c, at line 642 (in -ac2). There is

PV> error = elf_map(....)

PV> but nobody bothers with checking error value, it even tries it PV> to use as an offset if stars are in wrong constellation. PV> If you could add these lines below the call:

PV> if ((unsigned long)error >= (unsigned long)(-256)) {

PV>   set_fs(old_fs);
PV>   printk(KERN_DEBUG "Something went wrong with elf_map()\n");
PV>   kfree(elf_phdata);
PV>   send_sig(SIGSEGV, current, 0);
PV>   return 0;

PV> }

PV> and then report results...

It fixes reboot for me. Now vmlinux segfaults, and I see "Something went wrong with elf_map()" in the log.

Signature · 2 lines
Best regards, VDA
mailto:[email protected]