Re: parted being killed by SIGKILL (1.4.5, 1.4.6)

[Andrew Clausen]

Thomas Roelz wrote:
> 
> Hello,
> 
> I just tried to shrink a FAT partition and got parted killed
> by the kernel. I tried this with 1.4.5 and 1.4.6. Same result.
> In the attachment is a corresponding strace output. Anyone
> encountered similar problems? Might this be a memory problem?

Yes, I already know about this one :-(

I think it won't happen with linux 2.4 (because it mmap()s on
read() - I believe 2.2 doesn't.  This is important for the FATs
- i.e. the FATs are BIG, and if we can mmap() that's quite
a bit of memory saved...)

Basically: neither malloc(3) nor brk(2) allocate pages.  Writing
to the page does.  Therefore, allocation doesn't fail at
malloc() or brk() time, but at write time.  If the kernel can't
allocate any pages, then it kills a big process (parted).

So, I guess we need to wait for 2.4 to become popular!

Also, we can probably do some tricks to make the new fat
be "largely" mmap()ed...

I really need to learn more VM stuff!  (Living with Rik van
Riel for a month helped, though ;-)

Andrew Clausen