Re: [PATCH bpf-next 1/2] bpf: Add open coded version of kmem_cache iterator

[Namhyung Kim <namhyung@kernel.org>]

Hello,

On Fri, Oct 18, 2024 at 11:22:00AM -0700, Martin KaFai Lau wrote:
> On 10/17/24 1:06 AM, Namhyung Kim wrote:
> > Add a new open coded iterator for kmem_cache which can be called from a
> > BPF program like below.  It doesn't take any argument and traverses all
> > kmem_cache entries.
> > 
> >    struct kmem_cache *pos;
> > 
> >    bpf_for_each(kmem_cache, pos) {
> >        ...
> >    }
> > 
> > As it needs to grab slab_mutex, it should be called from sleepable BPF
> > programs only.
> > 
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> > ---
> >   kernel/bpf/helpers.c         |  3 ++
> >   kernel/bpf/kmem_cache_iter.c | 87 ++++++++++++++++++++++++++++++++++++
> >   2 files changed, 90 insertions(+)
> > 
> > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > index 073e6f04f4d765ff..d1dfa4f335577914 100644
> > --- a/kernel/bpf/helpers.c
> > +++ b/kernel/bpf/helpers.c
> > @@ -3111,6 +3111,9 @@ BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
> >   BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
> >   BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
> >   BTF_ID_FLAGS(func, bpf_get_kmem_cache)
> > +BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
> > +BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
> > +BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
> >   BTF_KFUNCS_END(common_btf_ids)
> >   static const struct btf_kfunc_id_set common_kfunc_set = {
> > diff --git a/kernel/bpf/kmem_cache_iter.c b/kernel/bpf/kmem_cache_iter.c
> > index ebc101d7da51b57c..31ddaf452b20a458 100644
> > --- a/kernel/bpf/kmem_cache_iter.c
> > +++ b/kernel/bpf/kmem_cache_iter.c
> > @@ -145,6 +145,93 @@ static const struct bpf_iter_seq_info kmem_cache_iter_seq_info = {
> >   	.seq_ops		= &kmem_cache_iter_seq_ops,
> >   };
> > +/* open-coded version */
> > +struct bpf_iter_kmem_cache {
> > +	__u64 __opaque[1];
> > +} __attribute__((aligned(8)));
> > +
> > +struct bpf_iter_kmem_cache_kern {
> > +	struct kmem_cache *pos;
> > +} __attribute__((aligned(8)));
> > +
> > +__bpf_kfunc_start_defs();
> > +
> > +__bpf_kfunc int bpf_iter_kmem_cache_new(struct bpf_iter_kmem_cache *it)
> > +{
> > +	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
> > +
> > +	BUILD_BUG_ON(sizeof(*kit) > sizeof(*it));
> > +	BUILD_BUG_ON(__alignof__(*kit) != __alignof__(*it));
> > +
> > +	kit->pos = NULL;
> > +	return 0;
> > +}
> > +
> > +__bpf_kfunc struct kmem_cache *bpf_iter_kmem_cache_next(struct bpf_iter_kmem_cache *it)
> > +{
> > +	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
> > +	struct kmem_cache *prev = kit->pos;
> > +	struct kmem_cache *next;
> > +	bool destroy = false;
> > +
> > +	mutex_lock(&slab_mutex);
> 
> I think taking mutex_lock here should be fine since sleepable tracing prog
> should be limited to the error injection whitelist. Those functions should
> not have held the mutex afaict.
> 
> > +
> > +	if (list_empty(&slab_caches)) {
> > +		mutex_unlock(&slab_mutex);
> > +		return NULL;
> > +	}
> > +
> > +	if (prev == NULL)
> > +		next = list_first_entry(&slab_caches, struct kmem_cache, list);
> > +	else if (list_last_entry(&slab_caches, struct kmem_cache, list) == prev)
> > +		next = NULL;
> 
> At the last entry, next is NULL.
> 
> > +	else
> > +		next = list_next_entry(prev, list);
> > +
> > +	/* boot_caches have negative refcount, don't touch them */
> > +	if (next && next->refcount > 0)
> > +		next->refcount++;
> > +
> > +	/* Skip kmem_cache_destroy() for active entries */
> > +	if (prev && prev->refcount > 1)
> > +		prev->refcount--;
> > +	else if (prev && prev->refcount == 1)
> > +		destroy = true;
> > +
> > +	mutex_unlock(&slab_mutex);
> > +
> > +	if (destroy)
> > +		kmem_cache_destroy(prev);
> > +
> > +	kit->pos = next;
> 
> so kit->pos will be NULL also. Does it mean the bpf prog will be able to
> call bpf_iter_kmem_cache_next() again and re-loop from the beginning of the
> slab_caches list?

Right, I'll mark the start pos differently to prevent that.

Thanks,
Namhyung

> 
> > +	return next;
> > +}
> > +
> > +__bpf_kfunc void bpf_iter_kmem_cache_destroy(struct bpf_iter_kmem_cache *it)
> > +{
> > +	struct bpf_iter_kmem_cache_kern *kit = (void *)it;
> > +	struct kmem_cache *s = kit->pos;
> > +	bool destroy = false;
> > +
> > +	if (s == NULL)
> > +		return;
> > +
> > +	mutex_lock(&slab_mutex);
> > +
> > +	/* Skip kmem_cache_destroy() for active entries */
> > +	if (s->refcount > 1)
> > +		s->refcount--;
> > +	else if (s->refcount == 1)
> > +		destroy = true;
> > +
> > +	mutex_unlock(&slab_mutex);
> > +
> > +	if (destroy)
> > +		kmem_cache_destroy(s);
> > +}
> > +
> > +__bpf_kfunc_end_defs();
> > +
> >   static void bpf_iter_kmem_cache_show_fdinfo(const struct bpf_iter_aux_info *aux,
> >   					    struct seq_file *seq)
> >   {
>