stackleak: don't modify ctl_table argument

Commit Message

Thomas Weißschuh May 3, 2024, 1:44 p.m. UTC

Sysctl handlers are not supposed to modify the ctl_table passed to them.
Adapt the logic to work with a temporary
variable, similar to how it is done in other parts of the kernel.

This is also a prerequisite to enforce the immutability of the argument
through the callbacks prototy.

Fixes: 964c9dff0091 ("stackleak: Allow runtime disabling of kernel stack erasing")
Cc: [e-mail]
Acked-by: Kees Cook <[e-mail]>
Reviewed-by: Luis Chamberlain <[e-mail]>
Signed-off-by: Thomas Weißschuh <[e-mail]>
---
This was split out of my sysctl-const-handler series [0].

As that series will take some more time, submit the patch on its own,
as it is a generic bugfix that is valuable on its own.
And I can get it out of my books.

Changelog in contrast to the patch in the series:
* Reword commit message to remove strong relation to the constification
* Cc stable

[0] https://lore.kernel.org/lkml/[e-mail]/

Cc: Joel Granados <[e-mail]>
---
 kernel/stackleak.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)


---
base-commit: f03359bca01bf4372cf2c118cd9a987a5951b1c8
change-id: 20240503-sysctl-const-stackleak-af3e67bc65b0

Best regards,

Comments

Tycho Andersen May 3, 2024, 1:58 p.m. UTC | #1

On Fri, May 03, 2024 at 03:44:09PM +0200, Thomas Weißschuh wrote:
> Sysctl handlers are not supposed to modify the ctl_table passed to them.
> Adapt the logic to work with a temporary
> variable, similar to how it is done in other parts of the kernel.
> 
> This is also a prerequisite to enforce the immutability of the argument
> through the callbacks prototy.
> 
> Fixes: 964c9dff0091 ("stackleak: Allow runtime disabling of kernel stack erasing")

Reviewed-by: Tycho Andersen <[e-mail]>

Tycho

Kees Cook May 3, 2024, 5:55 p.m. UTC | #2

On Fri, May 03, 2024 at 03:44:09PM +0200, Thomas Weißschuh wrote:
> Sysctl handlers are not supposed to modify the ctl_table passed to them.
> Adapt the logic to work with a temporary
> variable, similar to how it is done in other parts of the kernel.
> 
> This is also a prerequisite to enforce the immutability of the argument
> through the callbacks prototy.
> 
> Fixes: 964c9dff0091 ("stackleak: Allow runtime disabling of kernel stack erasing")
> Cc: [e-mail]

I realize I've already Acked, but does this actually need to be CCed
to stable?

> Acked-by: Kees Cook <[e-mail]>
> Reviewed-by: Luis Chamberlain <[e-mail]>
> Signed-off-by: Thomas Weißschuh <[e-mail]>
> ---
> This was split out of my sysctl-const-handler series [0].
> 
> As that series will take some more time, submit the patch on its own,
> as it is a generic bugfix that is valuable on its own.
> And I can get it out of my books.
> 
> Changelog in contrast to the patch in the series:
> * Reword commit message to remove strong relation to the constification
> * Cc stable
> 
> [0] https://lore.kernel.org/lkml/[e-mail]/
> 
> Cc: Joel Granados <[e-mail]>
> ---
>  kernel/stackleak.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/stackleak.c b/kernel/stackleak.c
> index 34c9d81eea94..b292e5ca0b7d 100644
> --- a/kernel/stackleak.c
> +++ b/kernel/stackleak.c
> @@ -27,10 +27,11 @@ static int stack_erasing_sysctl(struct ctl_table *table, int write,
>  	int ret = 0;
>  	int state = !static_branch_unlikely(&stack_erasing_bypass);
>  	int prev_state = state;
> +	struct ctl_table tmp = *table;
>  
> -	table->data = &state;
> -	table->maxlen = sizeof(int);
> -	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
> +	tmp.data = &state;
> +	tmp.maxlen = sizeof(int);
> +	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
>  	state = !!state;
>  	if (ret || !write || state == prev_state)
>  		return ret;

I can pick this up; thanks!

-Kees

> 
> ---
> base-commit: f03359bca01bf4372cf2c118cd9a987a5951b1c8
> change-id: 20240503-sysctl-const-stackleak-af3e67bc65b0
> 
> Best regards,
> -- 
> Thomas Weißschuh <[e-mail]>
>

Thomas Weißschuh May 3, 2024, 6:24 p.m. UTC | #3

May 3, 2024 19:55:37 Kees Cook <[e-mail]>:

> On Fri, May 03, 2024 at 03:44:09PM +0200, Thomas Weißschuh wrote:
>> Sysctl handlers are not supposed to modify the ctl_table passed to them.
>> Adapt the logic to work with a temporary
>> variable, similar to how it is done in other parts of the kernel.
>>
>> This is also a prerequisite to enforce the immutability of the argument
>> through the callbacks prototy.
>>
>> Fixes: 964c9dff0091 ("stackleak: Allow runtime disabling of kernel stack erasing")
>> Cc: [e-mail]
>
> I realize I've already Acked, but does this actually need to be CCed
> to stable?

You acked it without the Cc stable.
I shouldn't have kept your Ack, sorry.

Feel free to drop the Cc, it shouldn't be critical.
I suspect the bots will pick it up anyways.

>> Acked-by: Kees Cook <[e-mail]>
>> Reviewed-by: Luis Chamberlain <[e-mail]>
>> Signed-off-by: Thomas Weißschuh <[e-mail]>
>> ---
>> This was split out of my sysctl-const-handler series [0].
>>
>> As that series will take some more time, submit the patch on its own,
>> as it is a generic bugfix that is valuable on its own.
>> And I can get it out of my books.
>>
>> Changelog in contrast to the patch in the series:
>> * Reword commit message to remove strong relation to the constification
>> * Cc stable
>>
>> [0] https://lore.kernel.org/lkml/[e-mail]/
>>
>> Cc: Joel Granados <[e-mail]>
>> ---
>> kernel/stackleak.c | 7 ++++---
>> 1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/kernel/stackleak.c b/kernel/stackleak.c
>> index 34c9d81eea94..b292e5ca0b7d 100644
>> --- a/kernel/stackleak.c
>> +++ b/kernel/stackleak.c
>> @@ -27,10 +27,11 @@ static int stack_erasing_sysctl(struct ctl_table *table, int write,
>>     int ret = 0;
>>     int state = !static_branch_unlikely(&stack_erasing_bypass);
>>     int prev_state = state;
>> +   struct ctl_table tmp = *table;
>>
>> -   table->data = &state;
>> -   table->maxlen = sizeof(int);
>> -   ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
>> +   tmp.data = &state;
>> +   tmp.maxlen = sizeof(int);
>> +   ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
>>     state = !!state;
>>     if (ret || !write || state == prev_state)
>>         return ret;
>
> I can pick this up; thanks!

Thanks!

Kees Cook May 3, 2024, 7:31 p.m. UTC | #4

On Fri, May 03, 2024 at 03:44:09PM +0200, Thomas Weißschuh wrote:
> Sysctl handlers are not supposed to modify the ctl_table passed to them.
> Adapt the logic to work with a temporary
> variable, similar to how it is done in other parts of the kernel.
> 
> This is also a prerequisite to enforce the immutability of the argument
> through the callbacks prototy.
                        ^^^^^^^

Was this supposed to be "prototype"? I couldn't quite figure out what
was meant there; the sentence reads fine to me without the word there at
all. :)

> 
> Fixes: 964c9dff0091 ("stackleak: Allow runtime disabling of kernel stack erasing")
> Cc: [e-mail]
> Acked-by: Kees Cook <[e-mail]>
> Reviewed-by: Luis Chamberlain <[e-mail]>
> Signed-off-by: Thomas Weißschuh <[e-mail]>
> ---
> This was split out of my sysctl-const-handler series [0].
> 
> As that series will take some more time, submit the patch on its own,
> as it is a generic bugfix that is valuable on its own.
> And I can get it out of my books.
> 
> Changelog in contrast to the patch in the series:
> * Reword commit message to remove strong relation to the constification
> * Cc stable
> 
> [0] https://lore.kernel.org/lkml/[e-mail]/
> 
> Cc: Joel Granados <[e-mail]>
> ---
>  kernel/stackleak.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/stackleak.c b/kernel/stackleak.c
> index 34c9d81eea94..b292e5ca0b7d 100644
> --- a/kernel/stackleak.c
> +++ b/kernel/stackleak.c
> @@ -27,10 +27,11 @@ static int stack_erasing_sysctl(struct ctl_table *table, int write,
>  	int ret = 0;
>  	int state = !static_branch_unlikely(&stack_erasing_bypass);
>  	int prev_state = state;
> +	struct ctl_table tmp = *table;
>  
> -	table->data = &state;
> -	table->maxlen = sizeof(int);
> -	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
> +	tmp.data = &state;
> +	tmp.maxlen = sizeof(int);
> +	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);

In looking at this yet again, I can't figure out why maxlen is being
set. It starts its life as sizeof(int):

static struct ctl_table stackleak_sysctls[] = {
        {
                .procname       = "stack_erasing",
                .data           = NULL,
                .maxlen         = sizeof(int),

-Kees

>  	state = !!state;
>  	if (ret || !write || state == prev_state)
>  		return ret;
> 
> ---
> base-commit: f03359bca01bf4372cf2c118cd9a987a5951b1c8
> change-id: 20240503-sysctl-const-stackleak-af3e67bc65b0
> 
> Best regards,
> -- 
> Thomas Weißschuh <[e-mail]>
>

Kees Cook May 3, 2024, 7:35 p.m. UTC | #5

On Fri, 03 May 2024 15:44:09 +0200, Thomas Weißschuh wrote:
> Sysctl handlers are not supposed to modify the ctl_table passed to them.
> Adapt the logic to work with a temporary
> variable, similar to how it is done in other parts of the kernel.
> 
> This is also a prerequisite to enforce the immutability of the argument
> through the callbacks prototy.
> 
> [...]

Applied to for-next/hardening, thanks!

[1/1] stackleak: don't modify ctl_table argument
      https://git.kernel.org/kees/c/0e148d3cca0d

Take care,

Patch

diff --git a/kernel/stackleak.c b/kernel/stackleak.c
index 34c9d81eea94..b292e5ca0b7d 100644
--- a/kernel/stackleak.c
+++ b/kernel/stackleak.c
@@ -27,10 +27,11 @@  static int stack_erasing_sysctl(struct ctl_table *table, int write,
 	int ret = 0;
 	int state = !static_branch_unlikely(&stack_erasing_bypass);
 	int prev_state = state;
+	struct ctl_table tmp = *table;
 
-	table->data = &state;
-	table->maxlen = sizeof(int);
-	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
+	tmp.data = &state;
+	tmp.maxlen = sizeof(int);
+	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 	state = !!state;
 	if (ret || !write || state == prev_state)
 		return ret;