2024-04-11 11:03:42

by Guanrui Huang

[permalink] [raw]
Subject: [PATCH v2] irqchip/gic-v3-its: Fix double free on error

In its_vpe_irq_domain_alloc, when its_vpe_init() returns an error
with i > 0, its_vpe_irq_domain_free may free bitmap and vprop_page,
and then there is a double free in its_vpe_irq_domain_alloc.

Fix it by calling its_vpe_irq_domain_free directly, bitmap and
vprop_page will be freed in this function.

And check whether its_vm is equal to domain->host_data to make sure
its_vpe_irq_domain_free handle right its_vm.

Signed-off-by: Guanrui Huang <[email protected]>
---
drivers/irqchip/irq-gic-v3-its.c | 12 +++++-------
1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index fca888b36680..72c44e555c88 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -4523,6 +4523,9 @@ static int its_vpe_irq_domain_alloc(struct irq_domain *domain, unsigned int virq

BUG_ON(!vm);

+ if (vm != domain->host_data)
+ return -EINVAL;
+
bitmap = its_lpi_alloc(roundup_pow_of_two(nr_irqs), &base, &nr_ids);
if (!bitmap)
return -ENOMEM;
@@ -4561,13 +4564,8 @@ static int its_vpe_irq_domain_alloc(struct irq_domain *domain, unsigned int virq
irqd_set_resend_when_in_progress(irq_get_irq_data(virq + i));
}

- if (err) {
- if (i > 0)
- its_vpe_irq_domain_free(domain, virq, i);
-
- its_lpi_free(bitmap, base, nr_ids);
- its_free_prop_table(vprop_page);
- }
+ if (err)
+ its_vpe_irq_domain_free(domain, virq, i);

return err;
}
--
2.36.1



2024-04-11 11:19:35

by Marc Zyngier

[permalink] [raw]
Subject: Re: [PATCH v2] irqchip/gic-v3-its: Fix double free on error

On Thu, 11 Apr 2024 11:56:30 +0100,
Guanrui Huang <[email protected]> wrote:
>
> In its_vpe_irq_domain_alloc, when its_vpe_init() returns an error
> with i > 0, its_vpe_irq_domain_free may free bitmap and vprop_page,
> and then there is a double free in its_vpe_irq_domain_alloc.
>
> Fix it by calling its_vpe_irq_domain_free directly, bitmap and
> vprop_page will be freed in this function.
>
> And check whether its_vm is equal to domain->host_data to make sure
> its_vpe_irq_domain_free handle right its_vm.
>
> Signed-off-by: Guanrui Huang <[email protected]>
> ---
> drivers/irqchip/irq-gic-v3-its.c | 12 +++++-------
> 1 file changed, 5 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> index fca888b36680..72c44e555c88 100644
> --- a/drivers/irqchip/irq-gic-v3-its.c
> +++ b/drivers/irqchip/irq-gic-v3-its.c
> @@ -4523,6 +4523,9 @@ static int its_vpe_irq_domain_alloc(struct irq_domain *domain, unsigned int virq
>
> BUG_ON(!vm);
>
> + if (vm != domain->host_data)
> + return -EINVAL;
> +

How can this happen?

> bitmap = its_lpi_alloc(roundup_pow_of_two(nr_irqs), &base, &nr_ids);
> if (!bitmap)
> return -ENOMEM;
> @@ -4561,13 +4564,8 @@ static int its_vpe_irq_domain_alloc(struct irq_domain *domain, unsigned int virq
> irqd_set_resend_when_in_progress(irq_get_irq_data(virq + i));
> }
>
> - if (err) {
> - if (i > 0)
> - its_vpe_irq_domain_free(domain, virq, i);
> -
> - its_lpi_free(bitmap, base, nr_ids);
> - its_free_prop_table(vprop_page);
> - }
> + if (err)
> + its_vpe_irq_domain_free(domain, virq, i);
>
> return err;
> }

This otherwise looks reasonable.

Thanks,

M.

--
Without deviation from the norm, progress is not possible.

2024-04-11 12:09:45

by Guanrui Huang

[permalink] [raw]
Subject: Re: [PATCH v2] irqchip/gic-v3-its: Fix double free on error

if (vm != domain->host_data)

This is just a safety check. It looks unlikely happened.
After all, vm is obtained from the args. If the caller has a bug, can we check here to avoid possible problems that may arise in later code?

Thanks,
Guanrui

2024-04-11 13:01:04

by Marc Zyngier

[permalink] [raw]
Subject: Re: [PATCH v2] irqchip/gic-v3-its: Fix double free on error

On Thu, 11 Apr 2024 12:58:05 +0100,
Guanrui Huang <[email protected]> wrote:
>
> if (vm != domain->host_data)
>
> This is just a safety check. It looks unlikely happened.

It's not a safety check. It is just papering over problems. If you
don't trust the vm pointer, why do you trust the domain pointer? Why
do you trust *anything* at all?

> After all, vm is obtained from the args. If the caller has a bug,
> can we check here to avoid possible problems that may arise in later
> code?

If the caller has a bug, please fix the caller. Don't paper over it.

And you can remove the BUG_ON(!vm) which is equally useless.

Thanks,

M.

--
Without deviation from the norm, progress is not possible.