Re: [PATCH] iommu/intel: Fix memleak in intel_irq_remapping_alloc

[Lu Baolu]

On 1/3/21 2:22 PM, dinghao.liu@xxxxxxxxxx wrote:
> > On 2021/1/3 12:08, dinghao.liu@xxxxxxxxxx wrote:
> > > > Hi,
> > > >
> > > > On 2021/1/2 17:50, Dinghao Liu wrote:
> > > > > When irq_domain_get_irq_data() or irqd_cfg() fails
> > > > > meanwhile i == 0, data allocated by kzalloc() has not
> > > > > been freed before returning, which leads to memleak.
> > > > >
> > > > > Fixes: b106ee63abccb ("irq_remapping/vt-d: Enhance Intel IR driver to support hierarchical irqdomains")
> > > > > Signed-off-by: Dinghao Liu <dinghao.liu@xxxxxxxxxx>
> > > > > ---
> > > > >     drivers/iommu/intel/irq_remapping.c | 2 ++
> > > > >     1 file changed, 2 insertions(+)
> > > > >
> > > > > diff --git a/drivers/iommu/intel/irq_remapping.c b/drivers/iommu/intel/irq_remapping.c
> > > > > index aeffda92b10b..cdaeed36750f 100644
> > > > > --- a/drivers/iommu/intel/irq_remapping.c
> > > > > +++ b/drivers/iommu/intel/irq_remapping.c
> > > > > @@ -1354,6 +1354,8 @@ static int intel_irq_remapping_alloc(struct irq_domain *domain,
> > > > >     		irq_cfg = irqd_cfg(irq_data);
> > > > >     		if (!irq_data || !irq_cfg) {
> > > > >     			ret = -EINVAL;
> > > > > +			kfree(data);
> > > > > +			data = NULL;
> > > >
> > > > Do you need to check (i == 0) here? @data will not be used anymore as it
> > > > goes to out branch, why setting it to NULL here?
> > >
> > > data will be passed to ire_data->chip_data when i == 0 and
> > > intel_free_irq_resources() will free it on failure. Thus I
> >
> > Isn't it going to "goto out_free_data"? If "i == 0", the allocated @data
> > won't be freed by intel_free_irq_resources(), hence memory leaking. Does
> > this patch aim to fix this?
> >
> > Best regards,
> > baolu
>
> Correct, this is what I mean. When i > 0, data has been passed to
> irq_data->chip_data, which will be freed in intel_free_irq_resources()
> on failure. So there is no memleak in this case. The memleak only occurs
> on failure when i == 0 (data has not been passed to irq_data->chip_data).

So how about

diff --git a/drivers/iommu/intel/irq_remapping.c 
b/drivers/iommu/intel/irq_remapping.c
index aeffda92b10b..685200a5cff0 100644
--- a/drivers/iommu/intel/irq_remapping.c
+++ b/drivers/iommu/intel/irq_remapping.c
@@ -1353,6 +1353,8 @@ static int intel_irq_remapping_alloc(struct 
irq_domain *domain,
                irq_data = irq_domain_get_irq_data(domain, virq + i);
                irq_cfg = irqd_cfg(irq_data);
                if (!irq_data || !irq_cfg) {
+                       if (!i)
+                               kfree(data);
                        ret = -EINVAL;
                        goto out_free_data;
                }

> I set data to NULL after kfree() in this patch to prevent double-free
> when the failure occurs at i > 0.

if i>0, @data has been passed and will be freed by
intel_free_irq_resources() on the failure path. No need to free or
clear, right?

Best regards,
baolu

> Regards,
> Dinghao
>
> > > set it to NULL to prevent double-free. However, if we add
> > > a check (i == 0) here, we will not need to set it to NULL.
> > > If this is better, I will resend a new patch soon.
> > >
> > > Regards,
> > > Dinghao