Re: [PATCH bpf-next 1/3] bpf: support cloning sk storage on accept()

[Stanislav Fomichev <sdf@fomichev.me>]

On 08/08, Martin Lau wrote:
> On Wed, Aug 07, 2019 at 08:47:18AM -0700, Stanislav Fomichev wrote:
> > Add new helper bpf_sk_storage_clone which optionally clones sk storage
> > and call it from bpf_sk_storage_clone. Reuse the gap in
> > bpf_sk_storage_elem to store clone/non-clone flag.
> > 
> > Cc: Martin KaFai Lau <kafai@fb.com>
> > Signed-off-by: Stanislav Fomichev <sdf@google.com>
> > ---
> >  include/net/bpf_sk_storage.h |  10 ++++
> >  include/uapi/linux/bpf.h     |   1 +
> >  net/core/bpf_sk_storage.c    | 102 +++++++++++++++++++++++++++++++++--
> >  net/core/sock.c              |   9 ++--
> >  4 files changed, 115 insertions(+), 7 deletions(-)
> > 
> > diff --git a/include/net/bpf_sk_storage.h b/include/net/bpf_sk_storage.h
> > index b9dcb02e756b..8e4f831d2e52 100644
> > --- a/include/net/bpf_sk_storage.h
> > +++ b/include/net/bpf_sk_storage.h
> > @@ -10,4 +10,14 @@ void bpf_sk_storage_free(struct sock *sk);
> >  extern const struct bpf_func_proto bpf_sk_storage_get_proto;
> >  extern const struct bpf_func_proto bpf_sk_storage_delete_proto;
> >  
> > +#ifdef CONFIG_BPF_SYSCALL
> > +int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk);
> > +#else
> > +static inline int bpf_sk_storage_clone(const struct sock *sk,
> > +				       struct sock *newsk)
> > +{
> > +	return 0;
> > +}
> > +#endif
> > +
> >  #endif /* _BPF_SK_STORAGE_H */
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index 4393bd4b2419..00459ca4c8cf 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -2931,6 +2931,7 @@ enum bpf_func_id {
> >  
> >  /* BPF_FUNC_sk_storage_get flags */
> >  #define BPF_SK_STORAGE_GET_F_CREATE	(1ULL << 0)
> > +#define BPF_SK_STORAGE_GET_F_CLONE	(1ULL << 1)
> It is only used in bpf_sk_storage_get().
> What if the elem is created from bpf_fd_sk_storage_update_elem()
> i.e. from the syscall API ?
> 
> What may be the use case for a map to have both CLONE and non-CLONE
> elements?  If it is not the case, would it be better to add
> BPF_F_CLONE to bpf_attr->map_flags?
I didn't think about putting it on the map itself since the API
is on a per-element, but it does make sense. I can't come up
with a use-case for a per-element selective clone/non-clone.
Thanks, will move to the map itself.

> >  
> >  /* Mode for BPF_FUNC_skb_adjust_room helper. */
> >  enum bpf_adj_room_mode {
> > diff --git a/net/core/bpf_sk_storage.c b/net/core/bpf_sk_storage.c
> > index 94c7f77ecb6b..b6dea67965bc 100644
> > --- a/net/core/bpf_sk_storage.c
> > +++ b/net/core/bpf_sk_storage.c
> > @@ -12,6 +12,9 @@
> >  
> >  static atomic_t cache_idx;
> >  
> > +#define BPF_SK_STORAGE_GET_F_MASK	(BPF_SK_STORAGE_GET_F_CREATE | \
> > +					 BPF_SK_STORAGE_GET_F_CLONE)
> > +
> >  struct bucket {
> >  	struct hlist_head list;
> >  	raw_spinlock_t lock;
> > @@ -66,7 +69,8 @@ struct bpf_sk_storage_elem {
> >  	struct hlist_node snode;	/* Linked to bpf_sk_storage */
> >  	struct bpf_sk_storage __rcu *sk_storage;
> >  	struct rcu_head rcu;
> > -	/* 8 bytes hole */
> > +	u8 clone:1;
> > +	/* 7 bytes hole */
> >  	/* The data is stored in aother cacheline to minimize
> >  	 * the number of cachelines access during a cache hit.
> >  	 */
> > @@ -509,7 +513,7 @@ static int sk_storage_delete(struct sock *sk, struct bpf_map *map)
> >  	return 0;
> >  }
> >  
> > -/* Called by __sk_destruct() */
> > +/* Called by __sk_destruct() & bpf_sk_storage_clone() */
> >  void bpf_sk_storage_free(struct sock *sk)
> >  {
> >  	struct bpf_sk_storage_elem *selem;
> > @@ -739,19 +743,106 @@ static int bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
> >  	return err;
> >  }
> >  
> > +static struct bpf_sk_storage_elem *
> > +bpf_sk_storage_clone_elem(struct sock *newsk,
> > +			  struct bpf_sk_storage_map *smap,
> > +			  struct bpf_sk_storage_elem *selem)
> > +{
> > +	struct bpf_sk_storage_elem *copy_selem;
> > +
> > +	copy_selem = selem_alloc(smap, newsk, NULL, true);
> > +	if (!copy_selem)
> > +		return ERR_PTR(-ENOMEM);
> nit.
> may be just return NULL as selem_alloc() does.
Sounds good.

> > +
> > +	if (map_value_has_spin_lock(&smap->map))
> > +		copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
> > +				      SDATA(selem)->data, true);
> > +	else
> > +		copy_map_value(&smap->map, SDATA(copy_selem)->data,
> > +			       SDATA(selem)->data);
> > +
> > +	return copy_selem;
> > +}
> > +
> > +int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
> > +{
> > +	struct bpf_sk_storage *new_sk_storage = NULL;
> > +	struct bpf_sk_storage *sk_storage;
> > +	struct bpf_sk_storage_elem *selem;
> > +	int ret;
> > +
> > +	RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
> > +
> > +	rcu_read_lock();
> > +	sk_storage = rcu_dereference(sk->sk_bpf_storage);
> > +
> > +	if (!sk_storage || hlist_empty(&sk_storage->list))
> > +		goto out;
> > +
> > +	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
> > +		struct bpf_sk_storage_map *smap;
> > +		struct bpf_sk_storage_elem *copy_selem;
> > +
> > +		if (!selem->clone)
> > +			continue;
> > +
> > +		smap = rcu_dereference(SDATA(selem)->smap);
> > +		if (!smap)
> smap should not be NULL.
I see; you never set it back to NULL and we are guaranteed that the
map is still around due to rcu. Removed.

> > +			continue;
> > +
> > +		copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
> > +		if (IS_ERR(copy_selem)) {
> > +			ret = PTR_ERR(copy_selem);
> > +			goto err;
> > +		}
> > +
> > +		if (!new_sk_storage) {
> > +			ret = sk_storage_alloc(newsk, smap, copy_selem);
> > +			if (ret) {
> > +				kfree(copy_selem);
> > +				atomic_sub(smap->elem_size,
> > +					   &newsk->sk_omem_alloc);
> > +				goto err;
> > +			}
> > +
> > +			new_sk_storage = rcu_dereference(copy_selem->sk_storage);
> > +			continue;
> > +		}
> > +
> > +		raw_spin_lock_bh(&new_sk_storage->lock);
> > +		selem_link_map(smap, copy_selem);
> Unlike the existing selem-update use-cases in bpf_sk_storage.c,
> the smap->map.refcnt has not been held here.  Reading the smap
> is fine.  However, adding a new selem to a deleting smap is an issue.
> Hence, I think bpf_map_inc_not_zero() should be done first.
In this case, I should probably do it after smap = rcu_deref()?

> > +		__selem_link_sk(new_sk_storage, copy_selem);
> > +		raw_spin_unlock_bh(&new_sk_storage->lock);
> > +	}
> > +
> > +out:
> > +	rcu_read_unlock();
> > +	return 0;
> > +
> > +err:
> > +	rcu_read_unlock();
> > +
> > +	bpf_sk_storage_free(newsk);
> > +	return ret;
> > +}
> > +
> >  BPF_CALL_4(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
> >  	   void *, value, u64, flags)
> >  {
> >  	struct bpf_sk_storage_data *sdata;
> >  
> > -	if (flags > BPF_SK_STORAGE_GET_F_CREATE)
> > +	if (flags & ~BPF_SK_STORAGE_GET_F_MASK)
> > +		return (unsigned long)NULL;
> > +
> > +	if ((flags & BPF_SK_STORAGE_GET_F_CLONE) &&
> > +	    !(flags & BPF_SK_STORAGE_GET_F_CREATE))
> >  		return (unsigned long)NULL;
> >  
> >  	sdata = sk_storage_lookup(sk, map, true);
> >  	if (sdata)
> >  		return (unsigned long)sdata->data;
> >  
> > -	if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
> > +	if ((flags & BPF_SK_STORAGE_GET_F_CREATE) &&
> >  	    /* Cannot add new elem to a going away sk.
> >  	     * Otherwise, the new elem may become a leak
> >  	     * (and also other memory issues during map
> > @@ -762,6 +853,9 @@ BPF_CALL_4(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
> >  		/* sk must be a fullsock (guaranteed by verifier),
> >  		 * so sock_gen_put() is unnecessary.
> >  		 */
> > +		if (!IS_ERR(sdata))
> > +			SELEM(sdata)->clone =
> > +				!!(flags & BPF_SK_STORAGE_GET_F_CLONE);
> >  		sock_put(sk);
> >  		return IS_ERR(sdata) ?
> >  			(unsigned long)NULL : (unsigned long)sdata->data;
> > diff --git a/net/core/sock.c b/net/core/sock.c
> > index d57b0cc995a0..f5e801a9cea4 100644
> > --- a/net/core/sock.c
> > +++ b/net/core/sock.c
> > @@ -1851,9 +1851,12 @@ struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
> >  			goto out;
> >  		}
> >  		RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
> > -#ifdef CONFIG_BPF_SYSCALL
> > -		RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
> > -#endif
> > +
> > +		if (bpf_sk_storage_clone(sk, newsk)) {
> > +			sk_free_unlock_clone(newsk);
> > +			newsk = NULL;
> > +			goto out;
> > +		}
> >  
> >  		newsk->sk_err	   = 0;
> >  		newsk->sk_err_soft = 0;
> > -- 
> > 2.22.0.770.g0f2c4a37fd-goog
> >