How to crash a system programmatically

Question

If I could write a user program that would crash my OS (not my application), how would I do it?

I was thinking somehow switch my usermode program to kernel mode and cause a memory corruption. Is it possible?

Note: I am not creating a virus. Just curiosity.

To answer for your system, we'd need to know which OS you use. — entropo, Apr 29 '11 at 07:23
Do you assume your program has root or administrator privileges? Or is it running as an untrusted user program? — sarnold, Apr 29 '11 at 07:23
Write version 0.9 of a graphics driver, that ought to do it. :) — Will A, Apr 29 '11 at 07:24
I think based on the below answers, there are more known ways of crashing linux than windows...:) — SysAdmin, Apr 29 '11 at 07:35
The easy way is to pull the power cable out of the wall. To do this prgrmatically: robot.grab(powercode).pullSharply(Direction.AwayFromWall); — Martin York, Apr 29 '11 at 07:35
Yeah, but it was also a "for serious" comment... They're very different operating systems at almost every level. If you want to go beyond vague generalities, you really have to pick one; the answers are completely different. — Cody Gray - on strike, Apr 29 '11 at 07:36
@cody gray - I guess I am more interested to know Windows crashes — SysAdmin, Apr 29 '11 at 07:37
On old PET computers that was a memory location that contained the refresh rate of the screen. If you poked 255 into this location it would eventually caused the screen to overheat and catch fire. That's my kind of crash. — Martin York, Apr 29 '11 at 07:38

score 8 · Accepted Answer · answered Apr 29 '11 at 08:10

8

KeBugCheck on Windows is the documented way to get a BSOD.

You can also try deleting the root registry key (\REGISTRY) in Windows XP, using the native NT API.

answered Apr 29 '11 at 08:10

user541686

205,094
128
528
886

score 7 · Answer 2 · answered Apr 29 '11 at 07:27

7

Write and load a kernel module that calls panic() or implement equivalent thereof.

Or simply exec the shutdown or halt command or the syscall that implements it.

answered Apr 29 '11 at 07:27

Chris Stratton

39,853
6
84
117

score 6 · Answer 3 · answered Apr 29 '11 at 07:28

6

If the OS happens to be windows, create a fake driver that dereferences a NULL pointer. Crash!

answered Apr 29 '11 at 07:28

Tony The Lion

61,704
67
242
415

score 4 · Answer 4 · answered Apr 29 '11 at 07:24

4

The whole idea of an operating system is that a user program can't crash it under normal conditions. Of course you could still do something like exhaust the disk space on a partition that is used for a swap file and that would impair many operating systems or you could find a known vulnerability but there's no very easy way to reliably crash it.

answered Apr 29 '11 at 07:24

sharptooth

167,383
100
513
979

3

And we all know how well that works out in practice. ;-) – Andrew Cooper Apr 29 '11 at 07:25
@Andrew: For a second I read your name as Anderson Cooper, and I was like, =O!!! – user541686 Apr 29 '11 at 08:34
@Mehrdad: Sorry to disappoint you. ;-) – Andrew Cooper Apr 29 '11 at 10:18

entropo · Answer 5 · 2021-03-08T06:29:01.207

3

In Linux, Alt-SysRq-C will crash/restart your kernel.

In Windows, see: https://web.archive.org/web/20110513143420/http://www.dailygyan.com/2008/09/some-methods-to-crash-your-windows.html [Ed: March 8, 2021 - Switch to Archive.org link due to site going down.]

edited Mar 08 '21 at 06:29

answered Apr 29 '11 at 07:30

entropo

2,481
15
15

ublock origin detects this link as malicious – elite gamer88 Aug 12 '20 at 11:55

score 2 · Answer 6 · answered Apr 29 '11 at 07:29

2

For Windows one possibility is to write a kernel mode driver which locks some memory pages owned by a process and then terminate that process. Will result in a BSOD "Process has locked pages".

answered Apr 29 '11 at 07:29

ur.

2,890
1
18
21

score 1 · Answer 7 · answered Apr 29 '11 at 07:31

1

Linux: Even though not strictly crashing the OS, you can quite easily make it unusable by allocating lots of memory (and read/writing it for the allocation to actually become effective and make the OS swap a lot) and by forking lots of processes. "Fork bomb" is the keyword and can even be done in shell script.

answered Apr 29 '11 at 07:31

Tilman Vogel

9,337
4
33
32

2

Not just in shell script, but in shell script using cute ascii smileys! – R.. GitHub STOP HELPING ICE Apr 29 '11 at 16:47
http://stackoverflow.com/questions/991142/how-does-this-bash-fork-bomb-work – Tilman Vogel Apr 29 '11 at 17:30

score 1 · Answer 8 · answered Apr 29 '11 at 08:07

I think the reason why you want to crash the OS is relevant here. Are you trying to simulate a condition for testing, or are you just plain curious?

Here are two options if you wish to recreate, and automate, crashing, for the purpose of fault tolerance.

Run insider a virtual machine (vmware, VirtualBox) and simply kill the VM process. Alternately you can give it very low priority, drop devices, or otherwise simulate bad things.
Use servers that have a management console. This will have an API that can simply turn off the device.

The other numerous suggestions are good if you wish to crash from within the OS itself. These software crashes can help reproduce a miscreant process. A similar set of hardware related crashes could also work (such as reducing speed on a programmable fan and overheating the CPU).

The reason behind your request is actually quite important since all the different faults will yield a slightly different result.

Gui13 · Answer 9 · 2011-04-29T08:08:42.397

0

Try allocating chunks of memory until you have no free memory:

int alloced = 0;
for(;;)
{
    char *alloc = malloc(10*1024*1024); // alloc 10 MB
    if(alloc != NULL)
    {
        alloced += 10;
        // edit: you have to memset the memory otherwise the system will give it back to you next time
        memset(alloc, 0xab, 10*1024*1024);
        printf(" alloced %d MB\n", alloced);
    }
}

edit: I actually tried just right now on a 64 bits linux with 2GB of ram and 3.3GB of swap: the screen has frozen, I could allocate 4950MB of ram, but then the process was killed by the system, and linux fell back on its feet gracefully, so, no, this doesnt work :=)

edited Apr 29 '11 at 08:08

answered Apr 29 '11 at 07:30

Gui13

12,993
17
57
104

3

Virtual memory and paging means you're going to be there a while. And at least in Windows, you'll crash your application's own process *long* before you crash the OS. – Cody Gray - on strike Apr 29 '11 at 07:40
The trick is to allocate and access enough for the system to become terribly slow but not as much that the OOM killer gets after you. Then, "working" on that memory will make the system basically unusable. In my view, Linux is not very robust against that situation. – Tilman Vogel Apr 29 '11 at 16:18
You could launch a thread constantly writing thing on each memory chunk? :D – Gui13 Apr 29 '11 at 16:27

score 0 · Answer 10 · answered Apr 29 '11 at 10:11

Crash an OS using pure user-mode application means the kernel is vulnerable. If the OS is well tested, then this should not occur.

You can try BSoD Windows by attacking bugous 3rd-party drivers via sending garbage IO-CONTROLs to them.

DeviceIoControl Function (Windows) http://msdn.microsoft.com/en-us/library/aa363216(VS.85).aspx

How to crash a system programmatically

10 Answers10