[PATCH bpf-next v4 0/6] xdp: Add devmap_hash map type

[PATCH bpf-next v4 0/6] xdp: Add devmap_hash map type

[Toke Høiland-Jørgensen]

This series adds a new map type, devmap_hash, that works like the existing
devmap type, but using a hash-based indexing scheme. This is useful for the use
case where a devmap is indexed by ifindex (for instance for use with the routing
table lookup helper). For this use case, the regular devmap needs to be sized
after the maximum ifindex number, not the number of devices in it. A hash-based
indexing scheme makes it possible to size the map after the number of devices it
should contain instead.

This was previously part of my patch series that also turned the regular
bpf_redirect() helper into a map-based one; for this series I just pulled out
the patches that introduced the new map type.

Changelog:

v4:

- Remove check_memlock parameter that was left over from an earlier patch
  series.
- Reorder struct members to avoid holes.

v3:

- Rework the split into different patches
- Use spin_lock_irqsave()
- Also add documentation and bash completion definitions for bpftool

v2:

- Split commit adding the new map type so uapi and tools changes are separate.

Changes to these patches since the previous series:

- Rebase on top of the other devmap changes (makes this one simpler!)
- Don't enforce key==val, but allow arbitrary indexes.
- Rename the type to devmap_hash to reflect the fact that it's just a hashmap now.

---

Toke Høiland-Jørgensen (6):
      include/bpf.h: Remove map_insert_ctx() stubs
      xdp: Refactor devmap allocation code for reuse
      xdp: Add devmap_hash map type for looking up devices by hashed index
      tools/include/uapi: Add devmap_hash BPF map type
      tools/libbpf_probes: Add new devmap_hash type
      tools: Add definitions for devmap_hash map type


 include/linux/bpf.h                             |   11 -
 include/linux/bpf_types.h                       |    1 
 include/trace/events/xdp.h                      |    3 
 include/uapi/linux/bpf.h                        |    1 
 kernel/bpf/devmap.c                             |  326 +++++++++++++++++++----
 kernel/bpf/verifier.c                           |    2 
 net/core/filter.c                               |    9 -
 tools/bpf/bpftool/Documentation/bpftool-map.rst |    2 
 tools/bpf/bpftool/bash-completion/bpftool       |    4 
 tools/bpf/bpftool/map.c                         |    3 
 tools/include/uapi/linux/bpf.h                  |    1 
 tools/lib/bpf/libbpf_probes.c                   |    1 
 tools/testing/selftests/bpf/test_maps.c         |   16 +
 13 files changed, 315 insertions(+), 65 deletions(-)

[PATCH bpf-next v4 1/6] include/bpf.h: Remove map_insert_ctx() stubs

[Toke Høiland-Jørgensen]

From: Toke Høiland-Jørgensen <toke@redhat.com>

When we changed the device and CPU maps to use linked lists instead of
bitmaps, we also removed the need for the map_insert_ctx() helpers to keep
track of the bitmaps inside each map. However, it seems I forgot to remove
the function definitions stubs, so remove those here.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Yonghong Song <yhs@fb.com>
---
 include/linux/bpf.h |   10 ----------
 1 file changed, 10 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 18f4cc2c6acd..bfdb54dd2ad1 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -713,7 +713,6 @@ struct xdp_buff;
 struct sk_buff;
 
 struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key);
-void __dev_map_insert_ctx(struct bpf_map *map, u32 index);
 void __dev_map_flush(struct bpf_map *map);
 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp,
 		    struct net_device *dev_rx);
@@ -721,7 +720,6 @@ int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb,
 			     struct bpf_prog *xdp_prog);
 
 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key);
-void __cpu_map_insert_ctx(struct bpf_map *map, u32 index);
 void __cpu_map_flush(struct bpf_map *map);
 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp,
 		    struct net_device *dev_rx);
@@ -801,10 +799,6 @@ static inline struct net_device  *__dev_map_lookup_elem(struct bpf_map *map,
 	return NULL;
 }
 
-static inline void __dev_map_insert_ctx(struct bpf_map *map, u32 index)
-{
-}
-
 static inline void __dev_map_flush(struct bpf_map *map)
 {
 }
@@ -834,10 +828,6 @@ struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key)
 	return NULL;
 }
 
-static inline void __cpu_map_insert_ctx(struct bpf_map *map, u32 index)
-{
-}
-
 static inline void __cpu_map_flush(struct bpf_map *map)
 {
 }

[PATCH bpf-next v4 4/6] tools/include/uapi: Add devmap_hash BPF map type

[Toke Høiland-Jørgensen]

From: Toke Høiland-Jørgensen <toke@redhat.com>

This adds the devmap_hash BPF map type to the uapi headers in tools/.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Yonghong Song <yhs@fb.com>
---
 tools/include/uapi/linux/bpf.h |    1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index f506c68b2612..f7eaec10e82e 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -134,6 +134,7 @@ enum bpf_map_type {
 	BPF_MAP_TYPE_QUEUE,
 	BPF_MAP_TYPE_STACK,
 	BPF_MAP_TYPE_SK_STORAGE,
+	BPF_MAP_TYPE_DEVMAP_HASH,
 };
 
 /* Note that tracing related programs such as

[PATCH bpf-next v4 3/6] xdp: Add devmap_hash map type for looking up devices by hashed index

[Toke Høiland-Jørgensen]

From: Toke Høiland-Jørgensen <toke@redhat.com>

A common pattern when using xdp_redirect_map() is to create a device map
where the lookup key is simply ifindex. Because device maps are arrays,
this leaves holes in the map, and the map has to be sized to fit the
largest ifindex, regardless of how many devices actually are actually
needed in the map.

This patch adds a second type of device map where the key is looked up
using a hashmap, instead of being used as an array index. This allows maps
to be densely packed, so they can be smaller.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Yonghong Song <yhs@fb.com>
---
 include/linux/bpf.h        |    7 ++
 include/linux/bpf_types.h  |    1 
 include/trace/events/xdp.h |    3 -
 include/uapi/linux/bpf.h   |    1 
 kernel/bpf/devmap.c        |  194 ++++++++++++++++++++++++++++++++++++++++++++
 kernel/bpf/verifier.c      |    2 
 net/core/filter.c          |    9 ++
 7 files changed, 214 insertions(+), 3 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index bfdb54dd2ad1..f9a506147c8a 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -713,6 +713,7 @@ struct xdp_buff;
 struct sk_buff;
 
 struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key);
+struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key);
 void __dev_map_flush(struct bpf_map *map);
 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp,
 		    struct net_device *dev_rx);
@@ -799,6 +800,12 @@ static inline struct net_device  *__dev_map_lookup_elem(struct bpf_map *map,
 	return NULL;
 }
 
+static inline struct net_device  *__dev_map_hash_lookup_elem(struct bpf_map *map,
+							     u32 key)
+{
+	return NULL;
+}
+
 static inline void __dev_map_flush(struct bpf_map *map)
 {
 }
diff --git a/include/linux/bpf_types.h b/include/linux/bpf_types.h
index eec5aeeeaf92..36a9c2325176 100644
--- a/include/linux/bpf_types.h
+++ b/include/linux/bpf_types.h
@@ -62,6 +62,7 @@ BPF_MAP_TYPE(BPF_MAP_TYPE_ARRAY_OF_MAPS, array_of_maps_map_ops)
 BPF_MAP_TYPE(BPF_MAP_TYPE_HASH_OF_MAPS, htab_of_maps_map_ops)
 #ifdef CONFIG_NET
 BPF_MAP_TYPE(BPF_MAP_TYPE_DEVMAP, dev_map_ops)
+BPF_MAP_TYPE(BPF_MAP_TYPE_DEVMAP_HASH, dev_map_hash_ops)
 BPF_MAP_TYPE(BPF_MAP_TYPE_SK_STORAGE, sk_storage_map_ops)
 #if defined(CONFIG_BPF_STREAM_PARSER)
 BPF_MAP_TYPE(BPF_MAP_TYPE_SOCKMAP, sock_map_ops)
diff --git a/include/trace/events/xdp.h b/include/trace/events/xdp.h
index 68899fdc985b..8c8420230a10 100644
--- a/include/trace/events/xdp.h
+++ b/include/trace/events/xdp.h
@@ -175,7 +175,8 @@ struct _bpf_dtab_netdev {
 #endif /* __DEVMAP_OBJ_TYPE */
 
 #define devmap_ifindex(fwd, map)				\
-	((map->map_type == BPF_MAP_TYPE_DEVMAP) ?		\
+	((map->map_type == BPF_MAP_TYPE_DEVMAP ||		\
+	  map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) ?		\
 	  ((struct _bpf_dtab_netdev *)fwd)->dev->ifindex : 0)
 
 #define _trace_xdp_redirect_map(dev, xdp, fwd, map, idx)		\
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 6f68438aa4ed..e6767020ebba 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -134,6 +134,7 @@ enum bpf_map_type {
 	BPF_MAP_TYPE_QUEUE,
 	BPF_MAP_TYPE_STACK,
 	BPF_MAP_TYPE_SK_STORAGE,
+	BPF_MAP_TYPE_DEVMAP_HASH,
 };
 
 /* Note that tracing related programs such as
diff --git a/kernel/bpf/devmap.c b/kernel/bpf/devmap.c
index a0501266bdb8..b47a604b4936 100644
--- a/kernel/bpf/devmap.c
+++ b/kernel/bpf/devmap.c
@@ -37,6 +37,12 @@
  * notifier hook walks the map we know that new dev references can not be
  * added by the user because core infrastructure ensures dev_get_by_index()
  * calls will fail at this point.
+ *
+ * The devmap_hash type is a map type which interprets keys as ifindexes and
+ * indexes these using a hashmap. This allows maps that use ifindex as key to be
+ * densely packed instead of having holes in the lookup array for unused
+ * ifindexes. The setup and packet enqueue/send code is shared between the two
+ * types of devmap; only the lookup and insertion is different.
  */
 #include <linux/bpf.h>
 #include <net/xdp.h>
@@ -59,6 +65,7 @@ struct xdp_bulk_queue {
 
 struct bpf_dtab_netdev {
 	struct net_device *dev; /* must be first member, due to tracepoint */
+	struct hlist_node index_hlist;
 	struct bpf_dtab *dtab;
 	struct xdp_bulk_queue __percpu *bulkq;
 	struct rcu_head rcu;
@@ -70,11 +77,29 @@ struct bpf_dtab {
 	struct bpf_dtab_netdev **netdev_map;
 	struct list_head __percpu *flush_list;
 	struct list_head list;
+
+	/* these are only used for DEVMAP_HASH type maps */
+	struct hlist_head *dev_index_head;
+	spinlock_t index_lock;
+	unsigned int items;
 };
 
 static DEFINE_SPINLOCK(dev_map_lock);
 static LIST_HEAD(dev_map_list);
 
+static struct hlist_head *dev_map_create_hash(void)
+{
+	int i;
+	struct hlist_head *hash;
+
+	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
+	if (hash != NULL)
+		for (i = 0; i < NETDEV_HASHENTRIES; i++)
+			INIT_HLIST_HEAD(&hash[i]);
+
+	return hash;
+}
+
 static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)
 {
 	int err, cpu;
@@ -97,6 +122,9 @@ static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)
 	cost = (u64) dtab->map.max_entries * sizeof(struct bpf_dtab_netdev *);
 	cost += sizeof(struct list_head) * num_possible_cpus();
 
+	if (attr->map_type == BPF_MAP_TYPE_DEVMAP_HASH)
+		cost += sizeof(struct hlist_head) * NETDEV_HASHENTRIES;
+
 	/* if map size is larger than memlock limit, reject it */
 	err = bpf_map_charge_init(&dtab->map.memory, cost);
 	if (err)
@@ -115,8 +143,18 @@ static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)
 	if (!dtab->netdev_map)
 		goto free_percpu;
 
+	if (attr->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
+		dtab->dev_index_head = dev_map_create_hash();
+		if (!dtab->dev_index_head)
+			goto free_map_area;
+
+		spin_lock_init(&dtab->index_lock);
+	}
+
 	return 0;
 
+free_map_area:
+	bpf_map_area_free(dtab->netdev_map);
 free_percpu:
 	free_percpu(dtab->flush_list);
 free_charge:
@@ -198,6 +236,7 @@ static void dev_map_free(struct bpf_map *map)
 
 	free_percpu(dtab->flush_list);
 	bpf_map_area_free(dtab->netdev_map);
+	kfree(dtab->dev_index_head);
 	kfree(dtab);
 }
 
@@ -218,6 +257,70 @@ static int dev_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 	return 0;
 }
 
+static inline struct hlist_head *dev_map_index_hash(struct bpf_dtab *dtab,
+						    int idx)
+{
+	return &dtab->dev_index_head[idx & (NETDEV_HASHENTRIES - 1)];
+}
+
+struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key)
+{
+	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
+	struct hlist_head *head = dev_map_index_hash(dtab, key);
+	struct bpf_dtab_netdev *dev;
+
+	hlist_for_each_entry_rcu(dev, head, index_hlist)
+		if (dev->idx == key)
+			return dev;
+
+	return NULL;
+}
+
+static int dev_map_hash_get_next_key(struct bpf_map *map, void *key,
+				    void *next_key)
+{
+	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
+	u32 idx, *next = next_key;
+	struct bpf_dtab_netdev *dev, *next_dev;
+	struct hlist_head *head;
+	int i = 0;
+
+	if (!key)
+		goto find_first;
+
+	idx = *(u32 *)key;
+
+	dev = __dev_map_hash_lookup_elem(map, idx);
+	if (!dev)
+		goto find_first;
+
+	next_dev = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&dev->index_hlist)),
+				    struct bpf_dtab_netdev, index_hlist);
+
+	if (next_dev) {
+		*next = next_dev->idx;
+		return 0;
+	}
+
+	i = idx & (NETDEV_HASHENTRIES - 1);
+	i++;
+
+ find_first:
+	for (; i < NETDEV_HASHENTRIES; i++) {
+		head = dev_map_index_hash(dtab, i);
+
+		next_dev = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
+					    struct bpf_dtab_netdev,
+					    index_hlist);
+		if (next_dev) {
+			*next = next_dev->idx;
+			return 0;
+		}
+	}
+
+	return -ENOENT;
+}
+
 static int bq_xmit_all(struct xdp_bulk_queue *bq, u32 flags,
 		       bool in_napi_ctx)
 {
@@ -373,6 +476,15 @@ static void *dev_map_lookup_elem(struct bpf_map *map, void *key)
 	return dev ? &dev->ifindex : NULL;
 }
 
+static void *dev_map_hash_lookup_elem(struct bpf_map *map, void *key)
+{
+	struct bpf_dtab_netdev *obj = __dev_map_hash_lookup_elem(map,
+								*(u32 *)key);
+	struct net_device *dev = obj ? obj->dev : NULL;
+
+	return dev ? &dev->ifindex : NULL;
+}
+
 static void dev_map_flush_old(struct bpf_dtab_netdev *dev)
 {
 	if (dev->dev->netdev_ops->ndo_xdp_xmit) {
@@ -422,6 +534,28 @@ static int dev_map_delete_elem(struct bpf_map *map, void *key)
 	return 0;
 }
 
+static int dev_map_hash_delete_elem(struct bpf_map *map, void *key)
+{
+	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
+	struct bpf_dtab_netdev *old_dev;
+	int k = *(u32 *)key;
+	unsigned long flags;
+	int ret = -ENOENT;
+
+	spin_lock_irqsave(&dtab->index_lock, flags);
+
+	old_dev = __dev_map_hash_lookup_elem(map, k);
+	if (old_dev) {
+		dtab->items--;
+		hlist_del_init_rcu(&old_dev->index_hlist);
+		call_rcu(&old_dev->rcu, __dev_map_entry_free);
+		ret = 0;
+	}
+	spin_unlock_irqrestore(&dtab->index_lock, flags);
+
+	return ret;
+}
+
 static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net,
 						    struct bpf_dtab *dtab,
 						    u32 ifindex,
@@ -502,6 +636,56 @@ static int dev_map_update_elem(struct bpf_map *map, void *key, void *value,
 				     map, key, value, map_flags);
 }
 
+static int __dev_map_hash_update_elem(struct net *net, struct bpf_map *map,
+				     void *key, void *value, u64 map_flags)
+{
+	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
+	struct bpf_dtab_netdev *dev, *old_dev;
+	u32 ifindex = *(u32 *)value;
+	u32 idx = *(u32 *)key;
+	unsigned long flags;
+
+	if (unlikely(map_flags > BPF_EXIST || !ifindex))
+		return -EINVAL;
+
+	old_dev = __dev_map_hash_lookup_elem(map, idx);
+	if (old_dev && (map_flags & BPF_NOEXIST))
+		return -EEXIST;
+
+	dev = __dev_map_alloc_node(net, dtab, ifindex, idx);
+	if (IS_ERR(dev))
+		return PTR_ERR(dev);
+
+	spin_lock_irqsave(&dtab->index_lock, flags);
+
+	if (old_dev) {
+		hlist_del_rcu(&old_dev->index_hlist);
+	} else {
+		if (dtab->items >= dtab->map.max_entries) {
+			spin_unlock_irqrestore(&dtab->index_lock, flags);
+			call_rcu(&dev->rcu, __dev_map_entry_free);
+			return -E2BIG;
+		}
+		dtab->items++;
+	}
+
+	hlist_add_head_rcu(&dev->index_hlist,
+			   dev_map_index_hash(dtab, idx));
+	spin_unlock_irqrestore(&dtab->index_lock, flags);
+
+	if (old_dev)
+		call_rcu(&old_dev->rcu, __dev_map_entry_free);
+
+	return 0;
+}
+
+static int dev_map_hash_update_elem(struct bpf_map *map, void *key, void *value,
+				   u64 map_flags)
+{
+	return __dev_map_hash_update_elem(current->nsproxy->net_ns,
+					 map, key, value, map_flags);
+}
+
 const struct bpf_map_ops dev_map_ops = {
 	.map_alloc = dev_map_alloc,
 	.map_free = dev_map_free,
@@ -512,6 +696,16 @@ const struct bpf_map_ops dev_map_ops = {
 	.map_check_btf = map_check_no_btf,
 };
 
+const struct bpf_map_ops dev_map_hash_ops = {
+	.map_alloc = dev_map_alloc,
+	.map_free = dev_map_free,
+	.map_get_next_key = dev_map_hash_get_next_key,
+	.map_lookup_elem = dev_map_hash_lookup_elem,
+	.map_update_elem = dev_map_hash_update_elem,
+	.map_delete_elem = dev_map_hash_delete_elem,
+	.map_check_btf = map_check_no_btf,
+};
+
 static int dev_map_notification(struct notifier_block *notifier,
 				ulong event, void *ptr)
 {
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index a2e763703c30..231b9e22827c 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -3460,6 +3460,7 @@ static int check_map_func_compatibility(struct bpf_verifier_env *env,
 			goto error;
 		break;
 	case BPF_MAP_TYPE_DEVMAP:
+	case BPF_MAP_TYPE_DEVMAP_HASH:
 		if (func_id != BPF_FUNC_redirect_map &&
 		    func_id != BPF_FUNC_map_lookup_elem)
 			goto error;
@@ -3542,6 +3543,7 @@ static int check_map_func_compatibility(struct bpf_verifier_env *env,
 		break;
 	case BPF_FUNC_redirect_map:
 		if (map->map_type != BPF_MAP_TYPE_DEVMAP &&
+		    map->map_type != BPF_MAP_TYPE_DEVMAP_HASH &&
 		    map->map_type != BPF_MAP_TYPE_CPUMAP &&
 		    map->map_type != BPF_MAP_TYPE_XSKMAP)
 			goto error;
diff --git a/net/core/filter.c b/net/core/filter.c
index 47f6386fb17a..619130b346af 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -3517,7 +3517,8 @@ static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
 	int err;
 
 	switch (map->map_type) {
-	case BPF_MAP_TYPE_DEVMAP: {
+	case BPF_MAP_TYPE_DEVMAP:
+	case BPF_MAP_TYPE_DEVMAP_HASH: {
 		struct bpf_dtab_netdev *dst = fwd;
 
 		err = dev_map_enqueue(dst, xdp, dev_rx);
@@ -3554,6 +3555,7 @@ void xdp_do_flush_map(void)
 	if (map) {
 		switch (map->map_type) {
 		case BPF_MAP_TYPE_DEVMAP:
+		case BPF_MAP_TYPE_DEVMAP_HASH:
 			__dev_map_flush(map);
 			break;
 		case BPF_MAP_TYPE_CPUMAP:
@@ -3574,6 +3576,8 @@ static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
 	switch (map->map_type) {
 	case BPF_MAP_TYPE_DEVMAP:
 		return __dev_map_lookup_elem(map, index);
+	case BPF_MAP_TYPE_DEVMAP_HASH:
+		return __dev_map_hash_lookup_elem(map, index);
 	case BPF_MAP_TYPE_CPUMAP:
 		return __cpu_map_lookup_elem(map, index);
 	case BPF_MAP_TYPE_XSKMAP:
@@ -3655,7 +3659,8 @@ static int xdp_do_generic_redirect_map(struct net_device *dev,
 	ri->tgt_value = NULL;
 	WRITE_ONCE(ri->map, NULL);
 
-	if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
+	if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
+	    map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
 		struct bpf_dtab_netdev *dst = fwd;
 
 		err = dev_map_generic_redirect(dst, skb, xdp_prog);

[PATCH bpf-next v4 2/6] xdp: Refactor devmap allocation code for reuse

[Toke Høiland-Jørgensen]

From: Toke Høiland-Jørgensen <toke@redhat.com>

The subsequent patch to add a new devmap sub-type can re-use much of the
initialisation and allocation code, so refactor it into separate functions.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Yonghong Song <yhs@fb.com>
---
 kernel/bpf/devmap.c |  136 +++++++++++++++++++++++++++++++--------------------
 1 file changed, 83 insertions(+), 53 deletions(-)

diff --git a/kernel/bpf/devmap.c b/kernel/bpf/devmap.c
index d83cf8ccc872..a0501266bdb8 100644
--- a/kernel/bpf/devmap.c
+++ b/kernel/bpf/devmap.c
@@ -60,9 +60,9 @@ struct xdp_bulk_queue {
 struct bpf_dtab_netdev {
 	struct net_device *dev; /* must be first member, due to tracepoint */
 	struct bpf_dtab *dtab;
-	unsigned int bit;
 	struct xdp_bulk_queue __percpu *bulkq;
 	struct rcu_head rcu;
+	unsigned int idx; /* keep track of map index for tracepoint */
 };
 
 struct bpf_dtab {
@@ -75,28 +75,21 @@ struct bpf_dtab {
 static DEFINE_SPINLOCK(dev_map_lock);
 static LIST_HEAD(dev_map_list);
 
-static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
+static int dev_map_init_map(struct bpf_dtab *dtab, union bpf_attr *attr)
 {
-	struct bpf_dtab *dtab;
 	int err, cpu;
 	u64 cost;
 
-	if (!capable(CAP_NET_ADMIN))
-		return ERR_PTR(-EPERM);
-
 	/* check sanity of attributes */
 	if (attr->max_entries == 0 || attr->key_size != 4 ||
 	    attr->value_size != 4 || attr->map_flags & ~DEV_CREATE_FLAG_MASK)
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	/* Lookup returns a pointer straight to dev->ifindex, so make sure the
 	 * verifier prevents writes from the BPF side
 	 */
 	attr->map_flags |= BPF_F_RDONLY_PROG;
 
-	dtab = kzalloc(sizeof(*dtab), GFP_USER);
-	if (!dtab)
-		return ERR_PTR(-ENOMEM);
 
 	bpf_map_init_from_attr(&dtab->map, attr);
 
@@ -107,9 +100,7 @@ static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
 	/* if map size is larger than memlock limit, reject it */
 	err = bpf_map_charge_init(&dtab->map.memory, cost);
 	if (err)
-		goto free_dtab;
-
-	err = -ENOMEM;
+		return -EINVAL;
 
 	dtab->flush_list = alloc_percpu(struct list_head);
 	if (!dtab->flush_list)
@@ -124,19 +115,38 @@ static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
 	if (!dtab->netdev_map)
 		goto free_percpu;
 
-	spin_lock(&dev_map_lock);
-	list_add_tail_rcu(&dtab->list, &dev_map_list);
-	spin_unlock(&dev_map_lock);
-
-	return &dtab->map;
+	return 0;
 
 free_percpu:
 	free_percpu(dtab->flush_list);
 free_charge:
 	bpf_map_charge_finish(&dtab->map.memory);
-free_dtab:
-	kfree(dtab);
-	return ERR_PTR(err);
+	return -ENOMEM;
+}
+
+static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
+{
+	struct bpf_dtab *dtab;
+	int err;
+
+	if (!capable(CAP_NET_ADMIN))
+		return ERR_PTR(-EPERM);
+
+	dtab = kzalloc(sizeof(*dtab), GFP_USER);
+	if (!dtab)
+		return ERR_PTR(-ENOMEM);
+
+	err = dev_map_init_map(dtab, attr);
+	if (err) {
+		kfree(dtab);
+		return ERR_PTR(err);
+	}
+
+	spin_lock(&dev_map_lock);
+	list_add_tail_rcu(&dtab->list, &dev_map_list);
+	spin_unlock(&dev_map_lock);
+
+	return &dtab->map;
 }
 
 static void dev_map_free(struct bpf_map *map)
@@ -235,7 +245,7 @@ static int bq_xmit_all(struct xdp_bulk_queue *bq, u32 flags,
 out:
 	bq->count = 0;
 
-	trace_xdp_devmap_xmit(&obj->dtab->map, obj->bit,
+	trace_xdp_devmap_xmit(&obj->dtab->map, obj->idx,
 			      sent, drops, bq->dev_rx, dev, err);
 	bq->dev_rx = NULL;
 	__list_del_clearprev(&bq->flush_node);
@@ -412,17 +422,52 @@ static int dev_map_delete_elem(struct bpf_map *map, void *key)
 	return 0;
 }
 
-static int dev_map_update_elem(struct bpf_map *map, void *key, void *value,
-				u64 map_flags)
+static struct bpf_dtab_netdev *__dev_map_alloc_node(struct net *net,
+						    struct bpf_dtab *dtab,
+						    u32 ifindex,
+						    unsigned int idx)
 {
-	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
-	struct net *net = current->nsproxy->net_ns;
 	gfp_t gfp = GFP_ATOMIC | __GFP_NOWARN;
+	struct bpf_dtab_netdev *dev;
+	struct xdp_bulk_queue *bq;
+	int cpu;
+
+	dev = kmalloc_node(sizeof(*dev), gfp, dtab->map.numa_node);
+	if (!dev)
+		return ERR_PTR(-ENOMEM);
+
+	dev->bulkq = __alloc_percpu_gfp(sizeof(*dev->bulkq),
+					sizeof(void *), gfp);
+	if (!dev->bulkq) {
+		kfree(dev);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	for_each_possible_cpu(cpu) {
+		bq = per_cpu_ptr(dev->bulkq, cpu);
+		bq->obj = dev;
+	}
+
+	dev->dev = dev_get_by_index(net, ifindex);
+	if (!dev->dev) {
+		free_percpu(dev->bulkq);
+		kfree(dev);
+		return ERR_PTR(-EINVAL);
+	}
+
+	dev->idx = idx;
+	dev->dtab = dtab;
+
+	return dev;
+}
+
+static int __dev_map_update_elem(struct net *net, struct bpf_map *map,
+				 void *key, void *value, u64 map_flags)
+{
+	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
 	struct bpf_dtab_netdev *dev, *old_dev;
 	u32 ifindex = *(u32 *)value;
-	struct xdp_bulk_queue *bq;
 	u32 i = *(u32 *)key;
-	int cpu;
 
 	if (unlikely(map_flags > BPF_EXIST))
 		return -EINVAL;
@@ -434,31 +479,9 @@ static int dev_map_update_elem(struct bpf_map *map, void *key, void *value,
 	if (!ifindex) {
 		dev = NULL;
 	} else {
-		dev = kmalloc_node(sizeof(*dev), gfp, map->numa_node);
-		if (!dev)
-			return -ENOMEM;
-
-		dev->bulkq = __alloc_percpu_gfp(sizeof(*dev->bulkq),
-						sizeof(void *), gfp);
-		if (!dev->bulkq) {
-			kfree(dev);
-			return -ENOMEM;
-		}
-
-		for_each_possible_cpu(cpu) {
-			bq = per_cpu_ptr(dev->bulkq, cpu);
-			bq->obj = dev;
-		}
-
-		dev->dev = dev_get_by_index(net, ifindex);
-		if (!dev->dev) {
-			free_percpu(dev->bulkq);
-			kfree(dev);
-			return -EINVAL;
-		}
-
-		dev->bit = i;
-		dev->dtab = dtab;
+		dev = __dev_map_alloc_node(net, dtab, ifindex, i);
+		if (IS_ERR(dev))
+			return PTR_ERR(dev);
 	}
 
 	/* Use call_rcu() here to ensure rcu critical sections have completed
@@ -472,6 +495,13 @@ static int dev_map_update_elem(struct bpf_map *map, void *key, void *value,
 	return 0;
 }
 
+static int dev_map_update_elem(struct bpf_map *map, void *key, void *value,
+			       u64 map_flags)
+{
+	return __dev_map_update_elem(current->nsproxy->net_ns,
+				     map, key, value, map_flags);
+}
+
 const struct bpf_map_ops dev_map_ops = {
 	.map_alloc = dev_map_alloc,
 	.map_free = dev_map_free,

[PATCH bpf-next v4 5/6] tools/libbpf_probes: Add new devmap_hash type

[Toke Høiland-Jørgensen]

From: Toke Høiland-Jørgensen <toke@redhat.com>

This adds the definition for BPF_MAP_TYPE_DEVMAP_HASH to libbpf_probes.c in
tools/lib/bpf.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Yonghong Song <yhs@fb.com>
---
 tools/lib/bpf/libbpf_probes.c |    1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/lib/bpf/libbpf_probes.c b/tools/lib/bpf/libbpf_probes.c
index ace1a0708d99..4b0b0364f5fc 100644
--- a/tools/lib/bpf/libbpf_probes.c
+++ b/tools/lib/bpf/libbpf_probes.c
@@ -244,6 +244,7 @@ bool bpf_probe_map_type(enum bpf_map_type map_type, __u32 ifindex)
 	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
 	case BPF_MAP_TYPE_HASH_OF_MAPS:
 	case BPF_MAP_TYPE_DEVMAP:
+	case BPF_MAP_TYPE_DEVMAP_HASH:
 	case BPF_MAP_TYPE_SOCKMAP:
 	case BPF_MAP_TYPE_CPUMAP:
 	case BPF_MAP_TYPE_XSKMAP:

[PATCH bpf-next v4 6/6] tools: Add definitions for devmap_hash map type

[Toke Høiland-Jørgensen]

From: Toke Høiland-Jørgensen <toke@redhat.com>

This adds selftest and bpftool updates for the devmap_hash map type.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
---
 tools/bpf/bpftool/Documentation/bpftool-map.rst |    2 +-
 tools/bpf/bpftool/bash-completion/bpftool       |    4 ++--
 tools/bpf/bpftool/map.c                         |    3 ++-
 tools/testing/selftests/bpf/test_maps.c         |   16 ++++++++++++++++
 4 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/tools/bpf/bpftool/Documentation/bpftool-map.rst b/tools/bpf/bpftool/Documentation/bpftool-map.rst
index 490b4501cb6e..61d1d270eb5e 100644
--- a/tools/bpf/bpftool/Documentation/bpftool-map.rst
+++ b/tools/bpf/bpftool/Documentation/bpftool-map.rst
@@ -46,7 +46,7 @@ MAP COMMANDS
 |	*TYPE* := { **hash** | **array** | **prog_array** | **perf_event_array** | **percpu_hash**
 |		| **percpu_array** | **stack_trace** | **cgroup_array** | **lru_hash**
 |		| **lru_percpu_hash** | **lpm_trie** | **array_of_maps** | **hash_of_maps**
-|		| **devmap** | **sockmap** | **cpumap** | **xskmap** | **sockhash**
+|		| **devmap** | **devmap_hash** | **sockmap** | **cpumap** | **xskmap** | **sockhash**
 |		| **cgroup_storage** | **reuseport_sockarray** | **percpu_cgroup_storage**
 |		| **queue** | **stack** }
 
diff --git a/tools/bpf/bpftool/bash-completion/bpftool b/tools/bpf/bpftool/bash-completion/bpftool
index c8f42e1fcbc9..6b961a5ed100 100644
--- a/tools/bpf/bpftool/bash-completion/bpftool
+++ b/tools/bpf/bpftool/bash-completion/bpftool
@@ -489,8 +489,8 @@ _bpftool()
                                 perf_event_array percpu_hash percpu_array \
                                 stack_trace cgroup_array lru_hash \
                                 lru_percpu_hash lpm_trie array_of_maps \
-                                hash_of_maps devmap sockmap cpumap xskmap \
-                                sockhash cgroup_storage reuseport_sockarray \
+                                hash_of_maps devmap devmap_hash sockmap cpumap \
+                                xskmap sockhash cgroup_storage reuseport_sockarray \
                                 percpu_cgroup_storage queue stack' -- \
                                                    "$cur" ) )
                             return 0
diff --git a/tools/bpf/bpftool/map.c b/tools/bpf/bpftool/map.c
index 5da5a7311f13..bfbbc6b4cb83 100644
--- a/tools/bpf/bpftool/map.c
+++ b/tools/bpf/bpftool/map.c
@@ -37,6 +37,7 @@ const char * const map_type_name[] = {
 	[BPF_MAP_TYPE_ARRAY_OF_MAPS]		= "array_of_maps",
 	[BPF_MAP_TYPE_HASH_OF_MAPS]		= "hash_of_maps",
 	[BPF_MAP_TYPE_DEVMAP]			= "devmap",
+	[BPF_MAP_TYPE_DEVMAP_HASH]		= "devmap_hash",
 	[BPF_MAP_TYPE_SOCKMAP]			= "sockmap",
 	[BPF_MAP_TYPE_CPUMAP]			= "cpumap",
 	[BPF_MAP_TYPE_XSKMAP]			= "xskmap",
@@ -1271,7 +1272,7 @@ static int do_help(int argc, char **argv)
 		"       TYPE := { hash | array | prog_array | perf_event_array | percpu_hash |\n"
 		"                 percpu_array | stack_trace | cgroup_array | lru_hash |\n"
 		"                 lru_percpu_hash | lpm_trie | array_of_maps | hash_of_maps |\n"
-		"                 devmap | sockmap | cpumap | xskmap | sockhash |\n"
+		"                 devmap | devmap_hash | sockmap | cpumap | xskmap | sockhash |\n"
 		"                 cgroup_storage | reuseport_sockarray | percpu_cgroup_storage }\n"
 		"       " HELP_SPEC_OPTIONS "\n"
 		"",
diff --git a/tools/testing/selftests/bpf/test_maps.c b/tools/testing/selftests/bpf/test_maps.c
index 5443b9bd75ed..e1f1becda529 100644
--- a/tools/testing/selftests/bpf/test_maps.c
+++ b/tools/testing/selftests/bpf/test_maps.c
@@ -508,6 +508,21 @@ static void test_devmap(unsigned int task, void *data)
 	close(fd);
 }
 
+static void test_devmap_hash(unsigned int task, void *data)
+{
+	int fd;
+	__u32 key, value;
+
+	fd = bpf_create_map(BPF_MAP_TYPE_DEVMAP_HASH, sizeof(key), sizeof(value),
+			    2, 0);
+	if (fd < 0) {
+		printf("Failed to create devmap_hash '%s'!\n", strerror(errno));
+		exit(1);
+	}
+
+	close(fd);
+}
+
 static void test_queuemap(unsigned int task, void *data)
 {
 	const int MAP_SIZE = 32;
@@ -1684,6 +1699,7 @@ static void run_all_tests(void)
 	test_arraymap_percpu_many_keys();
 
 	test_devmap(0, NULL);
+	test_devmap_hash(0, NULL);
 	test_sockmap(0, NULL);
 
 	test_map_large();

Re: [PATCH bpf-next v4 3/6] xdp: Add devmap_hash map type for looking up devices by hashed index

[Jesper Dangaard Brouer]

On Mon, 22 Jul 2019 13:52:48 +0200
Toke Høiland-Jørgensen <toke@redhat.com> wrote:

> +static inline struct hlist_head *dev_map_index_hash(struct bpf_dtab *dtab,
> +						    int idx)
> +{
> +	return &dtab->dev_index_head[idx & (NETDEV_HASHENTRIES - 1)];
> +}

It is good for performance that our "hash" function is simply an AND
operation on the idx.  We want to keep it this way.

I don't like that you are using NETDEV_HASHENTRIES, because the BPF map
infrastructure already have a way to specify the map size (struct
bpf_map_def .max_entries).  BUT for performance reasons, to keep the
AND operation, we would need to round up the hash-array size to nearest
power of 2 (or reject if user didn't specify a power of 2, if we want
to "expose" this limit to users).

> +struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key)
> +{
> +	struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
> +	struct hlist_head *head = dev_map_index_hash(dtab, key);
> +	struct bpf_dtab_netdev *dev;
> +
> +	hlist_for_each_entry_rcu(dev, head, index_hlist)
> +		if (dev->idx == key)
> +			return dev;
> +
> +	return NULL;
> +}

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer

Re: [PATCH bpf-next v4 3/6] xdp: Add devmap_hash map type for looking up devices by hashed index

[Toke Høiland-Jørgensen]

Jesper Dangaard Brouer <brouer@redhat.com> writes:

> On Mon, 22 Jul 2019 13:52:48 +0200
> Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>
>> +static inline struct hlist_head *dev_map_index_hash(struct bpf_dtab *dtab,
>> +						    int idx)
>> +{
>> +	return &dtab->dev_index_head[idx & (NETDEV_HASHENTRIES - 1)];
>> +}
>
> It is good for performance that our "hash" function is simply an AND
> operation on the idx.  We want to keep it this way.
>
> I don't like that you are using NETDEV_HASHENTRIES, because the BPF map
> infrastructure already have a way to specify the map size (struct
> bpf_map_def .max_entries).  BUT for performance reasons, to keep the
> AND operation, we would need to round up the hash-array size to nearest
> power of 2 (or reject if user didn't specify a power of 2, if we want
> to "expose" this limit to users).

But do we really want the number of hash buckets to be equal to the max
number of entries? The values are not likely to be evenly distributed,
so we'll end up with big buckets if the number is small, meaning we'll
blow performance on walking long lists in each bucket.

Also, if the size is dynamic the size needs to be loaded from memory
instead of being a compile-time constant, which will presumably hurt
performance (though not sure by how much)?

-Toke

Re: [PATCH bpf-next v4 3/6] xdp: Add devmap_hash map type for looking up devices by hashed index

[Jesper Dangaard Brouer]

On Thu, 25 Jul 2019 12:32:19 +0200
Toke Høiland-Jørgensen <toke@redhat.com> wrote:

> Jesper Dangaard Brouer <brouer@redhat.com> writes:
> 
> > On Mon, 22 Jul 2019 13:52:48 +0200
> > Toke Høiland-Jørgensen <toke@redhat.com> wrote:
> >  
> >> +static inline struct hlist_head *dev_map_index_hash(struct bpf_dtab *dtab,
> >> +						    int idx)
> >> +{
> >> +	return &dtab->dev_index_head[idx & (NETDEV_HASHENTRIES - 1)];
> >> +}  
> >
> > It is good for performance that our "hash" function is simply an AND
> > operation on the idx.  We want to keep it this way.
> >
> > I don't like that you are using NETDEV_HASHENTRIES, because the BPF map
> > infrastructure already have a way to specify the map size (struct
> > bpf_map_def .max_entries).  BUT for performance reasons, to keep the
> > AND operation, we would need to round up the hash-array size to nearest
> > power of 2 (or reject if user didn't specify a power of 2, if we want
> > to "expose" this limit to users).  
> 
> But do we really want the number of hash buckets to be equal to the max
> number of entries? The values are not likely to be evenly distributed,
> so we'll end up with big buckets if the number is small, meaning we'll
> blow performance on walking long lists in each bucket.

The requested change makes it user-configurable, instead of fixed 256
entries.  I've seen production use-case with >5000 net_devices, thus
they need a knob to increase this (to avoid the list walking as you
mention).


> Also, if the size is dynamic the size needs to be loaded from memory
> instead of being a compile-time constant, which will presumably hurt
> performance (though not sure by how much)?

To counter this, the mask value which need to be loaded from memory,
needs to be placed next to some other struct member which is already in
use (at least on same cacheline, Intel have some 16 bytes access micro
optimizations, which I've never been able to measure, as its in 0.5
nanosec scale).

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer

Re: [PATCH bpf-next v4 3/6] xdp: Add devmap_hash map type for looking up devices by hashed index

[Toke Høiland-Jørgensen]

Jesper Dangaard Brouer <brouer@redhat.com> writes:

> On Thu, 25 Jul 2019 12:32:19 +0200
> Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>
>> Jesper Dangaard Brouer <brouer@redhat.com> writes:
>> 
>> > On Mon, 22 Jul 2019 13:52:48 +0200
>> > Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>> >  
>> >> +static inline struct hlist_head *dev_map_index_hash(struct bpf_dtab *dtab,
>> >> +						    int idx)
>> >> +{
>> >> +	return &dtab->dev_index_head[idx & (NETDEV_HASHENTRIES - 1)];
>> >> +}  
>> >
>> > It is good for performance that our "hash" function is simply an AND
>> > operation on the idx.  We want to keep it this way.
>> >
>> > I don't like that you are using NETDEV_HASHENTRIES, because the BPF map
>> > infrastructure already have a way to specify the map size (struct
>> > bpf_map_def .max_entries).  BUT for performance reasons, to keep the
>> > AND operation, we would need to round up the hash-array size to nearest
>> > power of 2 (or reject if user didn't specify a power of 2, if we want
>> > to "expose" this limit to users).  
>> 
>> But do we really want the number of hash buckets to be equal to the max
>> number of entries? The values are not likely to be evenly distributed,
>> so we'll end up with big buckets if the number is small, meaning we'll
>> blow performance on walking long lists in each bucket.
>
> The requested change makes it user-configurable, instead of fixed 256
> entries.  I've seen production use-case with >5000 net_devices, thus
> they need a knob to increase this (to avoid the list walking as you
> mention).

Ah, I see. That makes sense; I thought you wanted to make it smaller
(cf. the previous discussion about it being too big). Still, it seems
counter-intuitive to overload max_entries in this way.

I do see that this is what the existing hash map is also doing, though,
so I guess there is some precedence. I do wonder if we'll end up getting
bad performance from the hash being too simplistic, but I guess we can
always fix that later.

>> Also, if the size is dynamic the size needs to be loaded from memory
>> instead of being a compile-time constant, which will presumably hurt
>> performance (though not sure by how much)?
>
> To counter this, the mask value which need to be loaded from memory,
> needs to be placed next to some other struct member which is already in
> use (at least on same cacheline, Intel have some 16 bytes access micro
> optimizations, which I've never been able to measure, as its in 0.5
> nanosec scale).

In the fast path (i.e., in __xdp_map_lookup_elem) we will have already
loaded map->max_entries since it's on the same cacheline as map_type
which we use to disambiguate which function to call. So it should be
fine to just use that directly.

I'll send a new version with this change :)

-Toke