Re: arch/riscv/kernel/module.c:639 process_accumulated_relocations() error: uninitialized symbol 'curr_type'.

[Dan Carpenter]

On Wed, Dec 27, 2023 at 04:59:25PM -0800, Charlie Jenkins wrote:
> On Thu, Dec 14, 2023 at 11:00:46AM +0300, Dan Carpenter wrote:
> > On Wed, Dec 13, 2023 at 11:27:02AM -0800, Charlie Jenkins wrote:
> > > > 8fd6c5142395a1 Charlie Jenkins 2023-11-01  638  			}
> > > > 8fd6c5142395a1 Charlie Jenkins 2023-11-01 @639  			reloc_handlers[curr_type].accumulate_handler(
> > > >                                                                                        ^^^^^^^^^
> > > > Can the list be empty?  Uninitialized in that case.
> > > 
> > > That's a tricky one, the list cannot be empty. Each bucket in the
> > > bucket_iter is guarunteed to have at least one rel_entry. I can probably
> > > resolve this by extracting this for loop into a do-while loop.
> > 
> > You can just ignore false positives.  It's not really a fix to change it
> > to a do-while loop.  I reviewed the do while code before reading this
> > email and I still wondered about empty lists, but just to hear that it's
> > not going to be empty is enough.  Just the email was sufficient.
> > 
> > regards,
> > dan carpenter
> > 
> 
> The fix isn't the do-while loop but rather the use after free, the
> incorrect sizeof, and incorrect error handling when
> initialize_relocation_hashtable fails. I decided to include the do-while
> code because I was already touching the surrounding code. Can you review
> [1]? If you would prefer that the do-while is reverted, I can do that,
> but it is important that the rest of the fixes are merged before 6.7 is
> released.
> 
> [1] https://lore.kernel.org/all/20231213-module_loading_fix-v1-1-da9b7c92ade5@xxxxxxxxxxxx/


Most subsystems wouldn't merge a patch like this which does so many
things at the same time...

regards,
dan carpenter