[PATCH] mm/mmap: undo ->mmap() when mas_preallocate() fails

Mike Kravetz posted 1 patch 3 years, 5 months ago
mm/mmap.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
[PATCH] mm/mmap: undo ->mmap() when mas_preallocate() fails
Posted by Mike Kravetz 3 years, 5 months ago
A memory leak in hugetlb_reserve_pages was reported in [1].  The root
cause was traced to an error path in mmap_region when mas_preallocate()
fails.  In this case, the vma is freed after a successful call to
filesystem specific mmap.  The hugetlbfs mmap routine may allocate data
structures pointed to by m_private_data.  These need to be cleaned up
by the hugetlb vm_ops->close() routine.

The same issue was addressed by commit deb0f6562884 ("mm/mmap: undo
->mmap() when arch_validate_flags() fails") for the arch_validate_flags()
test.  Go to the same close_and_free_vma label if mas_preallocate() fails.

[1] https://lore.kernel.org/linux-mm/CAKXUXMxf7OiCwbxib7MwfR4M1b5+b3cNTU7n5NV9Zm4967=FPQ@mail.gmail.com/

Fixes: d4af56c5c7c6 ("mm: start tracking VMAs with maple tree")
Reported-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
---
 mm/mmap.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/mmap.c b/mm/mmap.c
index bf2122af94e7..3c9890e443a3 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2681,7 +2681,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
 	if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
 		error = -ENOMEM;
 		if (file)
-			goto unmap_and_free_vma;
+			goto close_and_free_vma;
 		else
 			goto free_vma;
 	}
-- 
2.37.3
Re: [PATCH] mm/mmap: undo ->mmap() when mas_preallocate() fails
Posted by Liam Howlett 3 years, 5 months ago
This is for sure necessary.

Thanks Mike & Lukas.

Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>

* Mike Kravetz <mike.kravetz@oracle.com> [221017 22:49]:
> A memory leak in hugetlb_reserve_pages was reported in [1].  The root
> cause was traced to an error path in mmap_region when mas_preallocate()
> fails.  In this case, the vma is freed after a successful call to
> filesystem specific mmap.  The hugetlbfs mmap routine may allocate data
> structures pointed to by m_private_data.  These need to be cleaned up
> by the hugetlb vm_ops->close() routine.
> 
> The same issue was addressed by commit deb0f6562884 ("mm/mmap: undo
> ->mmap() when arch_validate_flags() fails") for the arch_validate_flags()
> test.  Go to the same close_and_free_vma label if mas_preallocate() fails.
> 
> [1] https://lore.kernel.org/linux-mm/CAKXUXMxf7OiCwbxib7MwfR4M1b5+b3cNTU7n5NV9Zm4967=FPQ@mail.gmail.com/
> 
> Fixes: d4af56c5c7c6 ("mm: start tracking VMAs with maple tree")
> Reported-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
> Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
> ---
>  mm/mmap.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/mmap.c b/mm/mmap.c
> index bf2122af94e7..3c9890e443a3 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -2681,7 +2681,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
>  	if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
>  		error = -ENOMEM;
>  		if (file)
> -			goto unmap_and_free_vma;
> +			goto close_and_free_vma;
>  		else
>  			goto free_vma;
>  	}
> -- 
> 2.37.3
>