Re: [PATCH] vfio iommu type1: Fix memory leak in vfio_iommu_type1_pin_pages

From: Alex Williamson
Date: Tue Oct 20 2020 - 12:34:26 EST


On Fri, 16 Oct 2020 17:35:58 +0800
"xuxiaoyang (C)" <xuxiaoyang2@xxxxxxxxxx> wrote:

> From 099744c26513e386e707faecb3f17726e236d9bc Mon Sep 17 00:00:00 2001
> From: Xiaoyang Xu <xuxiaoyang2@xxxxxxxxxx>
> Date: Fri, 16 Oct 2020 15:32:02 +0800
> Subject: [PATCH] vfio iommu type1: Fix memory leak in
> vfio_iommu_type1_pin_pages
>
> pfn is not added to pfn_list when vfio_add_to_pfn_list fails.
> vfio_unpin_page_external will exit directly without calling
> vfio_iova_put_vfio_pfn.This will lead to a memory leak.
>
> Signed-off-by: Xiaoyang Xu <xuxiaoyang2@xxxxxxxxxx>
> ---
> drivers/vfio/vfio_iommu_type1.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
> index c255a6683f31..26f518b02c81 100644
> --- a/drivers/vfio/vfio_iommu_type1.c
> +++ b/drivers/vfio/vfio_iommu_type1.c
> @@ -640,6 +640,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
> unsigned long remote_vaddr;
> struct vfio_dma *dma;
> bool do_accounting;
> + int unlocked;
>
> if (!iommu || !user_pfn || !phys_pfn)
> return -EINVAL;
> @@ -693,7 +694,9 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
>
> ret = vfio_add_to_pfn_list(dma, iova, phys_pfn[i]);
> if (ret) {
> - vfio_unpin_page_external(dma, iova, do_accounting);
> + unlocked = put_pfn(phys_pfn[i], dma->prot);
> + if (do_accounting)
> + vfio_lock_acct(dma, -unlocked, true);
> goto pin_unwind;
> }
>

Thanks, this looks correct to me, but can also be simplified to:

diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
index defd44522319..57b068abceb9 100644
--- a/drivers/vfio/vfio_iommu_type1.c
+++ b/drivers/vfio/vfio_iommu_type1.c
@@ -693,7 +693,8 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,

ret = vfio_add_to_pfn_list(dma, iova, phys_pfn[i]);
if (ret) {
- vfio_unpin_page_external(dma, iova, do_accounting);
+ if (put_pfn(phys_pfn[i], dma->prot) && do_accounting)
+ vfio_lock_acct(dma, -1, true);
goto pin_unwind;
}

ie. we don't need a variable to track the one page we unpin that might
be accounted and we can avoid branching to vfio_lock_acct() for -0 unlocked.

We should also add a Fixes tag so this propagates back to the relevant
stable releases:

Fixes: a54eb55045ae ("vfio iommu type1: Add support for mediated devices")

I've made both of these changes in my next queue, please let me know if
I've overlooked something or there's an objection. Thanks,

Alex