Merge branch 'bpf-nfp-map-offload'

Jakub Kicinski says:

====================
This set adds support for creating maps on networking devices.  BPF is
programs+maps, the pure program offload has been around for quite some
time, this patchset adds the map part of the equation.

Maps are allocated on the target device from the start.  There is no
host copy when map is created on the device.  Device maps are represented
by struct bpf_offloaded_map, regardless of type.  Host programs can't
access such maps, access is only possible from a program also loaded
to the same device and/or via the BPF syscall.

Offloaded programs are currently only allowed to perform lookups,
control plane is responsible for populating the maps.

For brevity only infrastructure and basic NFP patches are included.
Target device reporting, netdevsim and tests will follow up as well as
some further optimizations to the NFP code.

v2:
 - leave out the array maps, we will add them trivially later to avoid
   merge conflicts with ongoing spectere&meltdown mitigations.
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

1 files changed, 61 insertions, 10 deletions

diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 2bac0dc8baba..c691b9e972e3 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -94,18 +94,34 @@ static int check_uarg_tail_zero(void __user *uaddr,
 	return 0;
 }
 
+const struct bpf_map_ops bpf_map_offload_ops = {
+	.map_alloc = bpf_map_offload_map_alloc,
+	.map_free = bpf_map_offload_map_free,
+};
+
 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 {
+	const struct bpf_map_ops *ops;
 	struct bpf_map *map;
+	int err;
 
-	if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
-	    !bpf_map_types[attr->map_type])
+	if (attr->map_type >= ARRAY_SIZE(bpf_map_types))
+		return ERR_PTR(-EINVAL);
+	ops = bpf_map_types[attr->map_type];
+	if (!ops)
 		return ERR_PTR(-EINVAL);
 
-	map = bpf_map_types[attr->map_type]->map_alloc(attr);
+	if (ops->map_alloc_check) {
+		err = ops->map_alloc_check(attr);
+		if (err)
+			return ERR_PTR(err);
+	}
+	if (attr->map_ifindex)
+		ops = &bpf_map_offload_ops;
+	map = ops->map_alloc(attr);
 	if (IS_ERR(map))
 		return map;
-	map->ops = bpf_map_types[attr->map_type];
+	map->ops = ops;
 	map->map_type = attr->map_type;
 	return map;
 }
@@ -134,6 +150,16 @@ void bpf_map_area_free(void *area)
 	kvfree(area);
 }
 
+void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
+{
+	map->map_type = attr->map_type;
+	map->key_size = attr->key_size;
+	map->value_size = attr->value_size;
+	map->max_entries = attr->max_entries;
+	map->map_flags = attr->map_flags;
+	map->numa_node = bpf_map_attr_numa_node(attr);
+}
+
 int bpf_map_precharge_memlock(u32 pages)
 {
 	struct user_struct *user = get_current_user();
@@ -189,16 +215,25 @@ static int bpf_map_alloc_id(struct bpf_map *map)
 	return id > 0 ? 0 : id;
 }
 
-static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
+void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 {
 	unsigned long flags;
 
+	/* Offloaded maps are removed from the IDR store when their device
+	 * disappears - even if someone holds an fd to them they are unusable,
+	 * the memory is gone, all ops will fail; they are simply waiting for
+	 * refcnt to drop to be freed.
+	 */
+	if (!map->id)
+		return;
+
 	if (do_idr_lock)
 		spin_lock_irqsave(&map_idr_lock, flags);
 	else
 		__acquire(&map_idr_lock);
 
 	idr_remove(&map_idr, map->id);
+	map->id = 0;
 
 	if (do_idr_lock)
 		spin_unlock_irqrestore(&map_idr_lock, flags);
@@ -378,7 +413,7 @@ static int bpf_obj_name_cpy(char *dst, const char *src)
 	return 0;
 }
 
-#define BPF_MAP_CREATE_LAST_FIELD map_name
+#define BPF_MAP_CREATE_LAST_FIELD map_ifindex
 /* called via syscall */
 static int map_create(union bpf_attr *attr)
 {
@@ -566,8 +601,10 @@ static int map_lookup_elem(union bpf_attr *attr)
 	if (!value)
 		goto free_key;
 
-	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
-	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
+	if (bpf_map_is_dev_bound(map)) {
+		err = bpf_map_offload_lookup_elem(map, key, value);
+	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
+		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 		err = bpf_percpu_hash_copy(map, key, value);
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 		err = bpf_percpu_array_copy(map, key, value);
@@ -654,7 +691,10 @@ static int map_update_elem(union bpf_attr *attr)
 		goto free_value;
 
 	/* Need to create a kthread, thus must support schedule */
-	if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
+	if (bpf_map_is_dev_bound(map)) {
+		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
+		goto out;
+	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
 		err = map->ops->map_update_elem(map, key, value, attr->flags);
 		goto out;
 	}
@@ -731,6 +771,11 @@ static int map_delete_elem(union bpf_attr *attr)
 		goto err_put;
 	}
 
+	if (bpf_map_is_dev_bound(map)) {
+		err = bpf_map_offload_delete_elem(map, key);
+		goto out;
+	}
+
 	preempt_disable();
 	__this_cpu_inc(bpf_prog_active);
 	rcu_read_lock();
@@ -738,7 +783,7 @@ static int map_delete_elem(union bpf_attr *attr)
 	rcu_read_unlock();
 	__this_cpu_dec(bpf_prog_active);
 	preempt_enable();
-
+out:
 	if (!err)
 		trace_bpf_map_delete_elem(map, ufd, key);
 	kfree(key);
@@ -788,9 +833,15 @@ static int map_get_next_key(union bpf_attr *attr)
 	if (!next_key)
 		goto free_key;
 
+	if (bpf_map_is_dev_bound(map)) {
+		err = bpf_map_offload_get_next_key(map, key, next_key);
+		goto out;
+	}
+
 	rcu_read_lock();
 	err = map->ops->map_get_next_key(map, key, next_key);
 	rcu_read_unlock();
+out:
 	if (err)
 		goto free_next_key;