[dm-devel] [git pull] device mapper changes for 6.1

[Mikulas Patocka]

On Thu, 20 Oct 2022, Christoph Hellwig wrote:

> On Tue, Oct 18, 2022 at 02:48:01PM -0400, Mike Snitzer wrote:
> > > That really does not sound like a good idea at all.  And it does not
> > > seem to have any MM or core maintainer signoffs.
> > 
> > Sorry, I should've solicited proper review more loudly.
> > 
> > But if you have a specific concern with how DM is looking to use
> > is_vmalloc_or_module_addr() please say what that is.
> 
> If I understand the patch correct it tries to use it to detect if
> a string is a string global.  Besides being nasty API abuse I can't
> see how this would even work if DM is built in.

It works for built-in DM.

You have "kfree_const" that detects if the string points to to .rodata 
with "is_kernel_rodata". Unfortunatelly, is_kernel_rodata doesn't detect 
if the string points to some modules's rodata, so we need an extra check 
for that.

So, the logic is:
if (!is_vmalloc_or_module_addr(ptr) && !is_kernel_rodata(ptr)) kfree(ptr);

I thought about modifying is_kernel_rodata to detect module's rodata as 
well, but it wouldn't work because kstrdup_const uses it and there would 
be crash possibility:
ptr = kstrdup_const(modules_rodata); unload_module(); use "ptr";

Mikulas