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, 59 insertions, 44 deletions

diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index 3905d4bc5b80..b76828f23b49 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -227,7 +227,7 @@ static int alloc_extra_elems(struct bpf_htab *htab)
 }
 
 /* Called from syscall */
-static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
+static int htab_map_alloc_check(union bpf_attr *attr)
 {
 	bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 		       attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
@@ -241,9 +241,6 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
 	int numa_node = bpf_map_attr_numa_node(attr);
-	struct bpf_htab *htab;
-	int err, i;
-	u64 cost;
 
 	BUILD_BUG_ON(offsetof(struct htab_elem, htab) !=
 		     offsetof(struct htab_elem, hash_node.pprev));
@@ -254,40 +251,68 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 		/* LRU implementation is much complicated than other
 		 * maps.  Hence, limit to CAP_SYS_ADMIN for now.
 		 */
-		return ERR_PTR(-EPERM);
+		return -EPERM;
 
 	if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK)
 		/* reserved bits should not be used */
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	if (!lru && percpu_lru)
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	if (lru && !prealloc)
-		return ERR_PTR(-ENOTSUPP);
+		return -ENOTSUPP;
 
 	if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
+
+	/* check sanity of attributes.
+	 * value_size == 0 may be allowed in the future to use map as a set
+	 */
+	if (attr->max_entries == 0 || attr->key_size == 0 ||
+	    attr->value_size == 0)
+		return -EINVAL;
+
+	if (attr->key_size > MAX_BPF_STACK)
+		/* eBPF programs initialize keys on stack, so they cannot be
+		 * larger than max stack size
+		 */
+		return -E2BIG;
+
+	if (attr->value_size >= KMALLOC_MAX_SIZE -
+	    MAX_BPF_STACK - sizeof(struct htab_elem))
+		/* if value_size is bigger, the user space won't be able to
+		 * access the elements via bpf syscall. This check also makes
+		 * sure that the elem_size doesn't overflow and it's
+		 * kmalloc-able later in htab_map_update_elem()
+		 */
+		return -E2BIG;
+
+	return 0;
+}
+
+static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
+{
+	bool percpu = (attr->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
+		       attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
+	bool lru = (attr->map_type == BPF_MAP_TYPE_LRU_HASH ||
+		    attr->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH);
+	/* percpu_lru means each cpu has its own LRU list.
+	 * it is different from BPF_MAP_TYPE_PERCPU_HASH where
+	 * the map's value itself is percpu.  percpu_lru has
+	 * nothing to do with the map's value.
+	 */
+	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
+	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
+	struct bpf_htab *htab;
+	int err, i;
+	u64 cost;
 
 	htab = kzalloc(sizeof(*htab), GFP_USER);
 	if (!htab)
 		return ERR_PTR(-ENOMEM);
 
-	/* mandatory map attributes */
-	htab->map.map_type = attr->map_type;
-	htab->map.key_size = attr->key_size;
-	htab->map.value_size = attr->value_size;
-	htab->map.max_entries = attr->max_entries;
-	htab->map.map_flags = attr->map_flags;
-	htab->map.numa_node = numa_node;
-
-	/* check sanity of attributes.
-	 * value_size == 0 may be allowed in the future to use map as a set
-	 */
-	err = -EINVAL;
-	if (htab->map.max_entries == 0 || htab->map.key_size == 0 ||
-	    htab->map.value_size == 0)
-		goto free_htab;
+	bpf_map_init_from_attr(&htab->map, attr);
 
 	if (percpu_lru) {
 		/* ensure each CPU's lru list has >=1 elements.
@@ -304,22 +329,6 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 	/* hash table size must be power of 2 */
 	htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
 
-	err = -E2BIG;
-	if (htab->map.key_size > MAX_BPF_STACK)
-		/* eBPF programs initialize keys on stack, so they cannot be
-		 * larger than max stack size
-		 */
-		goto free_htab;
-
-	if (htab->map.value_size >= KMALLOC_MAX_SIZE -
-	    MAX_BPF_STACK - sizeof(struct htab_elem))
-		/* if value_size is bigger, the user space won't be able to
-		 * access the elements via bpf syscall. This check also makes
-		 * sure that the elem_size doesn't overflow and it's
-		 * kmalloc-able later in htab_map_update_elem()
-		 */
-		goto free_htab;
-
 	htab->elem_size = sizeof(struct htab_elem) +
 			  round_up(htab->map.key_size, 8);
 	if (percpu)
@@ -327,6 +336,7 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 	else
 		htab->elem_size += round_up(htab->map.value_size, 8);
 
+	err = -E2BIG;
 	/* prevent zero size kmalloc and check for u32 overflow */
 	if (htab->n_buckets == 0 ||
 	    htab->n_buckets > U32_MAX / sizeof(struct bucket))
@@ -1143,6 +1153,7 @@ static void htab_map_free(struct bpf_map *map)
 }
 
 const struct bpf_map_ops htab_map_ops = {
+	.map_alloc_check = htab_map_alloc_check,
 	.map_alloc = htab_map_alloc,
 	.map_free = htab_map_free,
 	.map_get_next_key = htab_map_get_next_key,
@@ -1153,6 +1164,7 @@ const struct bpf_map_ops htab_map_ops = {
 };
 
 const struct bpf_map_ops htab_lru_map_ops = {
+	.map_alloc_check = htab_map_alloc_check,
 	.map_alloc = htab_map_alloc,
 	.map_free = htab_map_free,
 	.map_get_next_key = htab_map_get_next_key,
@@ -1236,6 +1248,7 @@ int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
 }
 
 const struct bpf_map_ops htab_percpu_map_ops = {
+	.map_alloc_check = htab_map_alloc_check,
 	.map_alloc = htab_map_alloc,
 	.map_free = htab_map_free,
 	.map_get_next_key = htab_map_get_next_key,
@@ -1245,6 +1258,7 @@ const struct bpf_map_ops htab_percpu_map_ops = {
 };
 
 const struct bpf_map_ops htab_lru_percpu_map_ops = {
+	.map_alloc_check = htab_map_alloc_check,
 	.map_alloc = htab_map_alloc,
 	.map_free = htab_map_free,
 	.map_get_next_key = htab_map_get_next_key,
@@ -1253,11 +1267,11 @@ const struct bpf_map_ops htab_lru_percpu_map_ops = {
 	.map_delete_elem = htab_lru_map_delete_elem,
 };
 
-static struct bpf_map *fd_htab_map_alloc(union bpf_attr *attr)
+static int fd_htab_map_alloc_check(union bpf_attr *attr)
 {
 	if (attr->value_size != sizeof(u32))
-		return ERR_PTR(-EINVAL);
-	return htab_map_alloc(attr);
+		return -EINVAL;
+	return htab_map_alloc_check(attr);
 }
 
 static void fd_htab_map_free(struct bpf_map *map)
@@ -1328,7 +1342,7 @@ static struct bpf_map *htab_of_map_alloc(union bpf_attr *attr)
 	if (IS_ERR(inner_map_meta))
 		return inner_map_meta;
 
-	map = fd_htab_map_alloc(attr);
+	map = htab_map_alloc(attr);
 	if (IS_ERR(map)) {
 		bpf_map_meta_free(inner_map_meta);
 		return map;
@@ -1372,6 +1386,7 @@ static void htab_of_map_free(struct bpf_map *map)
 }
 
 const struct bpf_map_ops htab_of_maps_map_ops = {
+	.map_alloc_check = fd_htab_map_alloc_check,
 	.map_alloc = htab_of_map_alloc,
 	.map_free = htab_of_map_free,
 	.map_get_next_key = htab_map_get_next_key,