diff mbox series

[5/5] arm64/vdso: Restrict splitting VVAR VMA

Message ID 20200204175913.74901-6-avagin@gmail.com (mailing list archive)
State New, archived
Headers show
Series arm64: add the time namespace support | expand

Commit Message

Andrei Vagin Feb. 4, 2020, 5:59 p.m. UTC 
Forbid splitting VVAR VMA resulting in a stricter ABI and reducing the
amount of corner-cases to consider while working further on VDSO time
namespace support.

As the offset from timens to VVAR page is computed compile-time, the pages
in VVAR should stay together and not being partically mremap()'ed.

Signed-off-by: Andrei Vagin <avagin@gmail.com>
---
 arch/arm64/kernel/vdso.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

Comments

Vincenzo Frascino Feb. 20, 2020, 12:22 p.m. UTC | #1 
Hi Andrei,

On 04/02/2020 17:59, Andrei Vagin wrote:
> Forbid splitting VVAR VMA resulting in a stricter ABI and reducing the
> amount of corner-cases to consider while working further on VDSO time
> namespace support.
> 
> As the offset from timens to VVAR page is computed compile-time, the pages
> in VVAR should stay together and not being partically mremap()'ed.
> 

I agree on the concept, but why do we need to redefine mremap?
special_mapping_mremap() (mm/mmap.c +3317) seems doing already the same thing if
we leave mremap == NULL as is.

> Signed-off-by: Andrei Vagin <avagin@gmail.com>
> ---
>  arch/arm64/kernel/vdso.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/arch/arm64/kernel/vdso.c b/arch/arm64/kernel/vdso.c
> index 2e553468b183..e6ebdc184c1e 100644
> --- a/arch/arm64/kernel/vdso.c
> +++ b/arch/arm64/kernel/vdso.c
> @@ -229,6 +229,17 @@ static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
>  	return vmf_insert_pfn(vma, vmf->address, pfn);
>  }
>  
> +static int vvar_mremap(const struct vm_special_mapping *sm,
> +		       struct vm_area_struct *new_vma)
> +{
> +	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
> +
> +	if (new_size != VVAR_NR_PAGES * PAGE_SIZE)
> +		return -EINVAL;
> +
> +	return 0;
> +}
> +
>  static int __setup_additional_pages(enum arch_vdso_type arch_index,
>  				    struct mm_struct *mm,
>  				    struct linux_binprm *bprm,
> @@ -311,6 +322,7 @@ static struct vm_special_mapping aarch32_vdso_spec[C_PAGES] = {
>  	{
>  		.name = "[vvar]",
>  		.fault = vvar_fault,
> +		.mremap = vvar_mremap,
>  	},
>  	{
>  		.name = "[vdso]",
> @@ -493,6 +505,7 @@ static struct vm_special_mapping vdso_spec[A_PAGES] __ro_after_init = {
>  	{
>  		.name	= "[vvar]",
>  		.fault = vvar_fault,
> +		.mremap = vvar_mremap,
>  	},
>  	{
>  		.name	= "[vdso]",
>
Andrei Vagin Feb. 23, 2020, 11:30 p.m. UTC | #2 
On Thu, Feb 20, 2020 at 12:22:52PM +0000, Vincenzo Frascino wrote:
> Hi Andrei,
> 
> On 04/02/2020 17:59, Andrei Vagin wrote:
> > Forbid splitting VVAR VMA resulting in a stricter ABI and reducing the
> > amount of corner-cases to consider while working further on VDSO time
> > namespace support.
> > 
> > As the offset from timens to VVAR page is computed compile-time, the pages
> > in VVAR should stay together and not being partically mremap()'ed.
> > 
> 
> I agree on the concept, but why do we need to redefine mremap?
> special_mapping_mremap() (mm/mmap.c +3317) seems doing already the same thing if
> we leave mremap == NULL as is.
> 

Hmmm. I have read the code of special_mapping_mremap() and I don't see where
it restricts splitting the vvar mapping.

Here is the code what I see in the source:

static int special_mapping_mremap(struct vm_area_struct *new_vma)
{
        struct vm_special_mapping *sm = new_vma->vm_private_data;

        if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
                return -EFAULT;

        if (sm->mremap)
                return sm->mremap(sm, new_vma);

        return 0;
}

And I have checked that without this patch, I can remap only one page of
the vvar mapping.

Thanks,
Andrei
Vincenzo Frascino Feb. 24, 2020, 10:08 a.m. UTC | #3 
Hi Andrei,

On 2/23/20 11:30 PM, Andrei Vagin wrote:
[...]

> 
> Hmmm. I have read the code of special_mapping_mremap() and I don't see where
> it restricts splitting the vvar mapping.
> 
> Here is the code what I see in the source:
> 
> static int special_mapping_mremap(struct vm_area_struct *new_vma)
> {
>         struct vm_special_mapping *sm = new_vma->vm_private_data;
> 
>         if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
>                 return -EFAULT;
> 
>         if (sm->mremap)
>                 return sm->mremap(sm, new_vma);
> 
>         return 0;
> }
> 
> And I have checked that without this patch, I can remap only one page of
> the vvar mapping.
>

I checked it a second time and I agree. The check on new_size is required in
this case.

> Thanks,
> Andrei
>
diff mbox series

Patch

diff --git a/arch/arm64/kernel/vdso.c b/arch/arm64/kernel/vdso.c
index 2e553468b183..e6ebdc184c1e 100644
--- a/arch/arm64/kernel/vdso.c
+++ b/arch/arm64/kernel/vdso.c
@@ -229,6 +229,17 @@  static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 	return vmf_insert_pfn(vma, vmf->address, pfn);
 }
 
+static int vvar_mremap(const struct vm_special_mapping *sm,
+		       struct vm_area_struct *new_vma)
+{
+	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
+
+	if (new_size != VVAR_NR_PAGES * PAGE_SIZE)
+		return -EINVAL;
+
+	return 0;
+}
+
 static int __setup_additional_pages(enum arch_vdso_type arch_index,
 				    struct mm_struct *mm,
 				    struct linux_binprm *bprm,
@@ -311,6 +322,7 @@  static struct vm_special_mapping aarch32_vdso_spec[C_PAGES] = {
 	{
 		.name = "[vvar]",
 		.fault = vvar_fault,
+		.mremap = vvar_mremap,
 	},
 	{
 		.name = "[vdso]",
@@ -493,6 +505,7 @@  static struct vm_special_mapping vdso_spec[A_PAGES] __ro_after_init = {
 	{
 		.name	= "[vvar]",
 		.fault = vvar_fault,
+		.mremap = vvar_mremap,
 	},
 	{
 		.name	= "[vdso]",