Re: [PATCH v1] devres: Fix devm_krealloc() allocating memory with wrong size

From: Greg KH
Date: Sun Jun 16 2024 - 06:17:41 EST


On Sun, Jun 16, 2024 at 06:10:02PM +0800, quic_zijuhu wrote:
> On 6/16/2024 6:02 PM, Greg KH wrote:
> > On Sun, Jun 16, 2024 at 05:54:15PM +0800, Zijun Hu wrote:
> >> Kernel API devm_krealloc() calls alloc_dr() with wrong argument
> >> @total_new_size, and it will cause more memory to be allocated
> >> than required, fixed by using @new_size as alloc_dr()'s argument.
> >>
> >> Fixes: f82485722e5d ("devres: provide devm_krealloc()")
> >> Signed-off-by: Zijun Hu <quic_zijuhu@xxxxxxxxxxx>
> >> ---
> >> drivers/base/devres.c | 2 +-
> >> 1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/base/devres.c b/drivers/base/devres.c
> >> index 3df0025d12aa..1cf87ddf1f55 100644
> >> --- a/drivers/base/devres.c
> >> +++ b/drivers/base/devres.c
> >> @@ -898,7 +898,7 @@ void *devm_krealloc(struct device *dev, void *ptr, size_t new_size, gfp_t gfp)
> >> * taking the lock as most probably the caller uses GFP_KERNEL.
> >> */
> >> new_dr = alloc_dr(devm_kmalloc_release,
> >> - total_new_size, gfp, dev_to_node(dev));
> >> + new_size, gfp, dev_to_node(dev));
> >
> > Wait, are you sure? Did you properly test this? What is the actual
> > savings and how do you handle the fact that you are now not allocating
> > the space reserved for the devres structure itself?
> >
> sure even if i did not test it.
> beware that alloc_dr() will call check_dr_size() to reserve space for
> devres structure.
>
> it will reserve the devres structure two times if using wrong
> @total_new_size. it is obvious wrong.

Ah, I missed that, that is tricky. Please document this better when you
resubmit a new version.

And of course, please test it.

thanks,

greg k-h