Re: [PATCH] mm: Make failslab writable again

[Vlastimil Babka]

On 9/20/22 10:20, Alexander Atanasov wrote:
> In (060807f841ac mm, slub: make remaining slub_debug related attributes
> read-only failslab) it was made RO.

"read-only) failslab was made RO" ?

> I think it became a collateral victim to the other two options
> (sanity_checks and trace) for which the reasons are perfectly valid.

The commit also mentioned that modifying the flags is not protected in any
way, see below.

> Here is why:
>  - sanity_checks and trace are slab internal debug options,
>    failslab is used for fault injection.
>  - for fault injections, which by presumption are random, it
>    does not matter if it is not set atomically. You need to
>    set atleast one more option to trigger fault injection.
>  - in a testing scenario you may need to change it at runtime
>    example: module loading - you test all allocations limited
>    by the space option. Then you move to test only your module's
>    own slabs.
>  - when set by command line flags it effectively disables all
>    cache merges.
> 
> Cc: Vlastimil Babka <vbabka@xxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> Cc: Roman Gushchin <guro@xxxxxx>
> Cc: Christoph Lameter <cl@xxxxxxxxx>
> Cc: Jann Horn <jannh@xxxxxxxxxx>
> Cc: Vijayanand Jitta <vjitta@xxxxxxxxxxxxxx>
> Cc: David Rientjes <rientjes@xxxxxxxxxx>
> Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> Cc: Pekka Enberg <penberg@xxxxxxxxxx>
> Link: http://lkml.kernel.org/r/20200610163135.17364-5-vbabka@xxxxxxx
> 
> Signed-off-by: Alexander Atanasov <alexander.atanasov@xxxxxxxxxxxxx>
> ---
>  Documentation/mm/slub.rst |  2 ++
>  mm/slub.c                 | 14 +++++++++++++-
>  2 files changed, 15 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/mm/slub.rst b/Documentation/mm/slub.rst
> index 43063ade737a..86837073a39e 100644
> --- a/Documentation/mm/slub.rst
> +++ b/Documentation/mm/slub.rst
> @@ -116,6 +116,8 @@ options from the ``slub_debug`` parameter translate to the following files::
>  	T	trace
>  	A	failslab
>  
> +failslab file is writable, so writing 1 or 0 will enable or disable
> +the option at runtime. Write returns -EINVAL if cache is an alias.
>  Careful with tracing: It may spew out lots of information and never stop if
>  used on the wrong slab.
>  
> diff --git a/mm/slub.c b/mm/slub.c
> index 862dbd9af4f5..7c15d312e0fb 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -5617,7 +5617,19 @@ static ssize_t failslab_show(struct kmem_cache *s, char *buf)
>  {
>  	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
>  }
> -SLAB_ATTR_RO(failslab);
> +
> +static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
> +				size_t length)
> +{
> +	if (s->refcount > 1)
> +		return -EINVAL;
> +
> +	s->flags &= ~SLAB_FAILSLAB;
> +	if (buf[0] == '1')
> +		s->flags |= SLAB_FAILSLAB;

Could we at least use a temporary variable to set up the final value and
then do a WRITE_ONCE() to s->flags, so the compiler is not allowed to do
some funky stuff? Assuming this is really the only place where we modify
s->flags during runtime, so we can't miss other updates due to RMW.

> +	return length;
> +}
> +SLAB_ATTR(failslab);
>  #endif
>  
>  static ssize_t shrink_show(struct kmem_cache *s, char *buf)
> 
> base-commit: 80e78fcce86de0288793a0ef0f6acf37656ee4cf