Re: [PATCH 0/5] kstrdup optimization

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 13 Jan 2015 16:10:57 -0800 Craig Milo Rogers <rogers@isi.edu> wrote:

> > As kfree_const() has the exact same signature as kfree(), the risk of
> > accidentally passing pointers returned from kstrdup_const() to kfree() seems
> > high, which may lead to memory corruption if the pointer doesn't point to
> > allocated memory.
> ...
> >> To verify if the source is in .rodata function checks if the address is between
> >> sentinels __start_rodata, __end_rodata. I guess it should work with all
> >> architectures.
> 
> 	kfree() could also check if the region being freed is in .rodata, and
> ignore the call; kfree_const() would not be needed.  If making this check all
> the time leads to a significant decrease in performance (numbers needed here),
> another option is to keep kfree_const() but add a check to kfree(), when
> compiled for debugging, that issues a suitable complaint if the region being
> freed is in .rodata.
> 

Adding overhead to kfree() would be a show-stopper - it's a real
hotpath.

kstrdup_const() is only used in a small number of places.  Just don't
screw it up.


btw, I have vague memories that gcc used to put some strings into .text
under some circumstances.