Re: kernel BUGs when removing largish files with the SLOB allocator

[David Howells <dhowells@redhat.com>]

Matt Mackall <mpm@selenic.com> wrote:

> Not sure yet. There's only one user in nommu.c that shouldn't just be
> changed to ksize() that I can see, and that's the one in
> show_process_blocks(). That could test for VM_MAPPED_COPY and keep its
> hands off otherwise. 

Hmmm...  You're right.  However, note binfmt_elf_fdpic().  This calls ksize()
but should really call kobjsize().  It should not assume that the allocation
it's been given is of any particular type.  IIRC ksize() changed purpose at
some point.

> I can imagine situations where ->mmap returns pointers to something
> that's statically allocated anyway (XIP?), where kobjsize doesn't
> really make sense.

ramfs for example.  See get_unmapped_area() hooks, not mmap() hooks.

> Also, looks like the WARN_ON_SLACK code has rotten, result isn't
> defined in that function. Change it to base, perhaps?

Yeah.  It might be worth ditching it entirely too.

David