[PATCH] mm/mremap: fix uninitialized return code

Arnd Bergmann posted 1 patch 11 months, 1 week ago
mm/mremap.c | 1 +
1 file changed, 1 insertion(+)
[PATCH] mm/mremap: fix uninitialized return code
Posted by Arnd Bergmann 11 months, 1 week ago
From: Arnd Bergmann <arnd@arndb.de>

The 'err' variable is set in a conditinal branch and is not
set otherwise:

mm/mremap.c:1017:7: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
 1017 |                 if (vma->vm_start != old_addr)

Set it to zero before the initial value is set.

Fixes: 3129f7896afb ("mm/mremap: initial refactor of move_vma()")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 mm/mremap.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mm/mremap.c b/mm/mremap.c
index 456849b9e7bd..9c51a2360d84 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -1014,6 +1014,7 @@ static unsigned long prep_move_vma(struct vma_remap_struct *vrm)
 		return -ENOMEM;
 
 	if (vma->vm_ops && vma->vm_ops->may_split) {
+		err = 0;
 		if (vma->vm_start != old_addr)
 			err = vma->vm_ops->may_split(vma, old_addr);
 		if (!err && vma->vm_end != old_addr + old_len)
-- 
2.39.5
Re: [PATCH] mm/mremap: fix uninitialized return code
Posted by Yosry Ahmed 11 months, 1 week ago
On Wed, Mar 05, 2025 at 06:27:56PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The 'err' variable is set in a conditinal branch and is not
> set otherwise:
> 
> mm/mremap.c:1017:7: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
>  1017 |                 if (vma->vm_start != old_addr)
> 
> Set it to zero before the initial value is set.
> 
> Fixes: 3129f7896afb ("mm/mremap: initial refactor of move_vma()")

I think Andrew already fixed this:
https://lore.kernel.org/lkml/20250304151556.635d9041a7ca36f1960fe664@linux-foundation.org/

> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  mm/mremap.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/mm/mremap.c b/mm/mremap.c
> index 456849b9e7bd..9c51a2360d84 100644
> --- a/mm/mremap.c
> +++ b/mm/mremap.c
> @@ -1014,6 +1014,7 @@ static unsigned long prep_move_vma(struct vma_remap_struct *vrm)
>  		return -ENOMEM;
>  
>  	if (vma->vm_ops && vma->vm_ops->may_split) {
> +		err = 0;
>  		if (vma->vm_start != old_addr)
>  			err = vma->vm_ops->may_split(vma, old_addr);
>  		if (!err && vma->vm_end != old_addr + old_len)
> -- 
> 2.39.5
> 
>
Re: [PATCH] mm/mremap: fix uninitialized return code
Posted by Lorenzo Stoakes 11 months, 1 week ago
On Wed, Mar 05, 2025 at 06:27:56PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
>
> The 'err' variable is set in a conditinal branch and is not
> set otherwise:
>
> mm/mremap.c:1017:7: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
>  1017 |                 if (vma->vm_start != old_addr)
>
> Set it to zero before the initial value is set.
>
> Fixes: 3129f7896afb ("mm/mremap: initial refactor of move_vma()")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Thanks for this, however this has already been fixed via Andrew doing a
fix-patch which should land in mm-unstable,-next soon (possibly tmr?).

This is entirely my bad btw, for some reason my compiler locally didn't
catch this even with CONFIG_WERROR on which _really_ surprises me...

> ---
>  mm/mremap.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/mm/mremap.c b/mm/mremap.c
> index 456849b9e7bd..9c51a2360d84 100644
> --- a/mm/mremap.c
> +++ b/mm/mremap.c
> @@ -1014,6 +1014,7 @@ static unsigned long prep_move_vma(struct vma_remap_struct *vrm)
>  		return -ENOMEM;
>
>  	if (vma->vm_ops && vma->vm_ops->may_split) {
> +		err = 0;
>  		if (vma->vm_start != old_addr)
>  			err = vma->vm_ops->may_split(vma, old_addr);
>  		if (!err && vma->vm_end != old_addr + old_len)
> --
> 2.39.5
>
Re: [PATCH] mm/mremap: fix uninitialized return code
Posted by Arnd Bergmann 11 months, 1 week ago
On Wed, Mar 5, 2025, at 18:29, Lorenzo Stoakes wrote:
> On Wed, Mar 05, 2025 at 06:27:56PM +0100, Arnd Bergmann wrote:
>> From: Arnd Bergmann <arnd@arndb.de>
>>
>> The 'err' variable is set in a conditinal branch and is not
>> set otherwise:
>>
>> mm/mremap.c:1017:7: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
>>  1017 |                 if (vma->vm_start != old_addr)
>>
>> Set it to zero before the initial value is set.
>>
>> Fixes: 3129f7896afb ("mm/mremap: initial refactor of move_vma()")
>> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
>
> Thanks for this, however this has already been fixed via Andrew doing a
> fix-patch which should land in mm-unstable,-next soon (possibly tmr?).
>
> This is entirely my bad btw, for some reason my compiler locally didn't
> catch this even with CONFIG_WERROR on which _really_ surprises me...

Unfortunately gcc never shows warnings about conditionally uninitialized
variables. There is a -Wmaybe-uninitialized, but that has so many
false positives that it is completely useless and it is turned
off globally.

      Arnd
Re: [PATCH] mm/mremap: fix uninitialized return code
Posted by Lorenzo Stoakes 11 months, 1 week ago
On Wed, Mar 05, 2025 at 07:25:38PM +0100, Arnd Bergmann wrote:
> On Wed, Mar 5, 2025, at 18:29, Lorenzo Stoakes wrote:
> > On Wed, Mar 05, 2025 at 06:27:56PM +0100, Arnd Bergmann wrote:
> >> From: Arnd Bergmann <arnd@arndb.de>
> >>
> >> The 'err' variable is set in a conditinal branch and is not
> >> set otherwise:
> >>
> >> mm/mremap.c:1017:7: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
> >>  1017 |                 if (vma->vm_start != old_addr)
> >>
> >> Set it to zero before the initial value is set.
> >>
> >> Fixes: 3129f7896afb ("mm/mremap: initial refactor of move_vma()")
> >> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> >
> > Thanks for this, however this has already been fixed via Andrew doing a
> > fix-patch which should land in mm-unstable,-next soon (possibly tmr?).
> >
> > This is entirely my bad btw, for some reason my compiler locally didn't
> > catch this even with CONFIG_WERROR on which _really_ surprises me...
>
> Unfortunately gcc never shows warnings about conditionally uninitialized
> variables. There is a -Wmaybe-uninitialized, but that has so many
> false positives that it is completely useless and it is turned
> off globally.

OK I notice by building with LLVM=1 this is immediately reported locally,
so from now on I'll make sure to do a clang build too.

At least there is some means by which I can sensibly determine this!

Cheers!

>
>       Arnd