bpf: offload: add map offload infrastructure

BPF map offload follow similar path to program offload.  At creation
time users may specify ifindex of the device on which they want to
create the map.  Map will be validated by the kernel's
.map_alloc_check callback and device driver will be called for the
actual allocation.  Map will have an empty set of operations
associated with it (save for alloc and free callbacks).  The real
device callbacks are kept in map->offload->dev_ops because they
have slightly different signatures.  Map operations are called in
process context so the driver may communicate with HW freely,
msleep(), wait() etc.

Map alloc and free callbacks are muxed via existing .ndo_bpf, and
are always called with rtnl lock held.  Maps and programs are
guaranteed to be destroyed before .ndo_uninit (i.e. before
unregister_netdev() returns).  Map callbacks are invoked with
bpf_devs_lock *read* locked, drivers must take care of exclusive
locking if necessary.

All offload-specific branches are marked with unlikely() (through
bpf_map_is_dev_bound()), given that branch penalty will be
negligible compared to IO anyway, and we don't want to penalize
SW path unnecessarily.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

3 files changed, 226 insertions, 13 deletions

diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
index cdd1e19..453785f 100644
--- a/kernel/bpf/offload.c
+++ b/kernel/bpf/offload.c
@@ -24,11 +24,13 @@
 #include <linux/rtnetlink.h>
 #include <linux/rwsem.h>
 
-/* Protects bpf_prog_offload_devs and offload members of all progs.
+/* Protects bpf_prog_offload_devs, bpf_map_offload_devs and offload members
+ * of all progs.
  * RTNL lock cannot be taken when holding this lock.
  */
 static DECLARE_RWSEM(bpf_devs_lock);
 static LIST_HEAD(bpf_prog_offload_devs);
+static LIST_HEAD(bpf_map_offload_devs);
 
 static int bpf_dev_offload_check(struct net_device *netdev)
 {
@@ -250,11 +252,186 @@ int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
 const struct bpf_prog_ops bpf_offload_prog_ops = {
 };
 
+static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
+			       enum bpf_netdev_command cmd)
+{
+	struct netdev_bpf data = {};
+	struct net_device *netdev;
+
+	ASSERT_RTNL();
+
+	data.command = cmd;
+	data.offmap = offmap;
+	/* Caller must make sure netdev is valid */
+	netdev = offmap->netdev;
+
+	return netdev->netdev_ops->ndo_bpf(netdev, &data);
+}
+
+struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
+{
+	struct net *net = current->nsproxy->net_ns;
+	struct bpf_offloaded_map *offmap;
+	int err;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return ERR_PTR(-EPERM);
+	if (attr->map_type != BPF_MAP_TYPE_HASH)
+		return ERR_PTR(-EINVAL);
+
+	offmap = kzalloc(sizeof(*offmap), GFP_USER);
+	if (!offmap)
+		return ERR_PTR(-ENOMEM);
+
+	bpf_map_init_from_attr(&offmap->map, attr);
+
+	rtnl_lock();
+	down_write(&bpf_devs_lock);
+	offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
+	err = bpf_dev_offload_check(offmap->netdev);
+	if (err)
+		goto err_unlock;
+
+	err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
+	if (err)
+		goto err_unlock;
+
+	list_add_tail(&offmap->offloads, &bpf_map_offload_devs);
+	up_write(&bpf_devs_lock);
+	rtnl_unlock();
+
+	return &offmap->map;
+
+err_unlock:
+	up_write(&bpf_devs_lock);
+	rtnl_unlock();
+	kfree(offmap);
+	return ERR_PTR(err);
+}
+
+static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap)
+{
+	WARN_ON(bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_FREE));
+	/* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */
+	bpf_map_free_id(&offmap->map, true);
+	list_del_init(&offmap->offloads);
+	offmap->netdev = NULL;
+}
+
+void bpf_map_offload_map_free(struct bpf_map *map)
+{
+	struct bpf_offloaded_map *offmap = map_to_offmap(map);
+
+	rtnl_lock();
+	down_write(&bpf_devs_lock);
+	if (offmap->netdev)
+		__bpf_map_offload_destroy(offmap);
+	up_write(&bpf_devs_lock);
+	rtnl_unlock();
+
+	kfree(offmap);
+}
+
+int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
+{
+	struct bpf_offloaded_map *offmap = map_to_offmap(map);
+	int ret = -ENODEV;
+
+	down_read(&bpf_devs_lock);
+	if (offmap->netdev)
+		ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+
+int bpf_map_offload_update_elem(struct bpf_map *map,
+				void *key, void *value, u64 flags)
+{
+	struct bpf_offloaded_map *offmap = map_to_offmap(map);
+	int ret = -ENODEV;
+
+	if (unlikely(flags > BPF_EXIST))
+		return -EINVAL;
+
+	down_read(&bpf_devs_lock);
+	if (offmap->netdev)
+		ret = offmap->dev_ops->map_update_elem(offmap, key, value,
+						       flags);
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+
+int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
+{
+	struct bpf_offloaded_map *offmap = map_to_offmap(map);
+	int ret = -ENODEV;
+
+	down_read(&bpf_devs_lock);
+	if (offmap->netdev)
+		ret = offmap->dev_ops->map_delete_elem(offmap, key);
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+
+int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
+{
+	struct bpf_offloaded_map *offmap = map_to_offmap(map);
+	int ret = -ENODEV;
+
+	down_read(&bpf_devs_lock);
+	if (offmap->netdev)
+		ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+
+bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map)
+{
+	struct bpf_offloaded_map *offmap;
+	struct bpf_prog_offload *offload;
+	bool ret;
+
+	if (!!bpf_prog_is_dev_bound(prog->aux) != !!bpf_map_is_dev_bound(map))
+		return false;
+	if (!bpf_prog_is_dev_bound(prog->aux))
+		return true;
+
+	down_read(&bpf_devs_lock);
+	offload = prog->aux->offload;
+	offmap = map_to_offmap(map);
+
+	ret = offload && offload->netdev == offmap->netdev;
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+
+static void bpf_offload_orphan_all_progs(struct net_device *netdev)
+{
+	struct bpf_prog_offload *offload, *tmp;
+
+	list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs, offloads)
+		if (offload->netdev == netdev)
+			__bpf_prog_offload_destroy(offload->prog);
+}
+
+static void bpf_offload_orphan_all_maps(struct net_device *netdev)
+{
+	struct bpf_offloaded_map *offmap, *tmp;
+
+	list_for_each_entry_safe(offmap, tmp, &bpf_map_offload_devs, offloads)
+		if (offmap->netdev == netdev)
+			__bpf_map_offload_destroy(offmap);
+}
+
 static int bpf_offload_notification(struct notifier_block *notifier,
 				    ulong event, void *ptr)
 {
 	struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
-	struct bpf_prog_offload *offload, *tmp;
 
 	ASSERT_RTNL();
 
@@ -265,11 +442,8 @@ static int bpf_offload_notification(struct notifier_block *notifier,
 			break;
 
 		down_write(&bpf_devs_lock);
-		list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs,
-					 offloads) {
-			if (offload->netdev == netdev)
-				__bpf_prog_offload_destroy(offload->prog);
-		}
+		bpf_offload_orphan_all_progs(netdev);
+		bpf_offload_orphan_all_maps(netdev);
 		up_write(&bpf_devs_lock);
 		break;
 	default:
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index a3f726b..c691b9e 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -94,6 +94,11 @@ 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;
@@ -111,6 +116,8 @@ static struct bpf_map *find_and_alloc_map(union bpf_attr *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;
@@ -208,16 +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);
@@ -397,7 +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)
 {
@@ -585,8 +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);
@@ -673,7 +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;
 	}
@@ -750,6 +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();
@@ -757,7 +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);
@@ -807,9 +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;
 
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 48b61ca..ceabb39 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -4816,6 +4816,13 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
 			return -EINVAL;
 		}
 	}
+
+	if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) &&
+	    !bpf_offload_dev_match(prog, map)) {
+		verbose(env, "offload device mismatch between prog and map\n");
+		return -EINVAL;
+	}
+
 	return 0;
 }