bpf, xdp: Make bpf_redirect_map() a map operation
authorBjörn Töpel <bjorn.topel@intel.com>
Mon, 8 Mar 2021 11:29:06 +0000 (12:29 +0100)
committerDaniel Borkmann <daniel@iogearbox.net>
Wed, 10 Mar 2021 00:06:34 +0000 (01:06 +0100)
Currently the bpf_redirect_map() implementation dispatches to the
correct map-lookup function via a switch-statement. To avoid the
dispatching, this change adds bpf_redirect_map() as a map
operation. Each map provides its bpf_redirect_map() version, and
correct function is automatically selected by the BPF verifier.

A nice side-effect of the code movement is that the map lookup
functions are now local to the map implementation files, which removes
one additional function call.

Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210308112907.559576-2-bjorn.topel@gmail.com
include/linux/bpf.h
include/linux/filter.h
include/net/xdp_sock.h
kernel/bpf/cpumap.c
kernel/bpf/devmap.c
kernel/bpf/verifier.c
net/core/filter.c
net/xdp/xskmap.c

index c931bc9..a25730e 100644 (file)
@@ -118,6 +118,9 @@ struct bpf_map_ops {
                                           void *owner, u32 size);
        struct bpf_local_storage __rcu ** (*map_owner_storage_ptr)(void *owner);
 
+       /* Misc helpers.*/
+       int (*map_redirect)(struct bpf_map *map, u32 ifindex, u64 flags);
+
        /* map_meta_equal must be implemented for maps that can be
         * used as an inner map.  It is a runtime check to ensure
         * an inner map can be inserted to an outer map.
@@ -1450,9 +1453,9 @@ struct btf *bpf_get_btf_vmlinux(void);
 /* Map specifics */
 struct xdp_buff;
 struct sk_buff;
+struct bpf_dtab_netdev;
+struct bpf_cpu_map_entry;
 
-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_flush(void);
 int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp,
                    struct net_device *dev_rx);
@@ -1462,7 +1465,6 @@ int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb,
                             struct bpf_prog *xdp_prog);
 bool dev_map_can_have_prog(struct bpf_map *map);
 
-struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key);
 void __cpu_map_flush(void);
 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp,
                    struct net_device *dev_rx);
@@ -1593,17 +1595,6 @@ static inline int bpf_obj_get_user(const char __user *pathname, int flags)
        return -EOPNOTSUPP;
 }
 
-static inline struct net_device  *__dev_map_lookup_elem(struct bpf_map *map,
-                                                      u32 key)
-{
-       return NULL;
-}
-
-static inline struct net_device  *__dev_map_hash_lookup_elem(struct bpf_map *map,
-                                                            u32 key)
-{
-       return NULL;
-}
 static inline bool dev_map_can_have_prog(struct bpf_map *map)
 {
        return false;
@@ -1615,6 +1606,7 @@ static inline void __dev_flush(void)
 
 struct xdp_buff;
 struct bpf_dtab_netdev;
+struct bpf_cpu_map_entry;
 
 static inline
 int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp,
@@ -1639,12 +1631,6 @@ static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst,
        return 0;
 }
 
-static inline
-struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key)
-{
-       return NULL;
-}
-
 static inline void __cpu_map_flush(void)
 {
 }
index 3b00fc9..008691f 100644 (file)
@@ -1472,4 +1472,31 @@ static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
 }
 #endif /* IS_ENABLED(CONFIG_IPV6) */
 
+static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex, u64 flags,
+                                                 void *lookup_elem(struct bpf_map *map, u32 key))
+{
+       struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+
+       /* Lower bits of the flags are used as return code on lookup failure */
+       if (unlikely(flags > XDP_TX))
+               return XDP_ABORTED;
+
+       ri->tgt_value = lookup_elem(map, ifindex);
+       if (unlikely(!ri->tgt_value)) {
+               /* If the lookup fails we want to clear out the state in the
+                * redirect_info struct completely, so that if an eBPF program
+                * performs multiple lookups, the last one always takes
+                * precedence.
+                */
+               WRITE_ONCE(ri->map, NULL);
+               return flags;
+       }
+
+       ri->flags = flags;
+       ri->tgt_index = ifindex;
+       WRITE_ONCE(ri->map, map);
+
+       return XDP_REDIRECT;
+}
+
 #endif /* __LINUX_FILTER_H__ */
index cc17bc9..9c0722c 100644 (file)
@@ -80,19 +80,6 @@ int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp);
 int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp);
 void __xsk_map_flush(void);
 
-static inline struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map,
-                                                    u32 key)
-{
-       struct xsk_map *m = container_of(map, struct xsk_map, map);
-       struct xdp_sock *xs;
-
-       if (key >= map->max_entries)
-               return NULL;
-
-       xs = READ_ONCE(m->xsk_map[key]);
-       return xs;
-}
-
 #else
 
 static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
@@ -109,12 +96,6 @@ static inline void __xsk_map_flush(void)
 {
 }
 
-static inline struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map,
-                                                    u32 key)
-{
-       return NULL;
-}
-
 #endif /* CONFIG_XDP_SOCKETS */
 
 #endif /* _LINUX_XDP_SOCK_H */
index 5d1469d..7352d41 100644 (file)
@@ -563,7 +563,7 @@ static void cpu_map_free(struct bpf_map *map)
        kfree(cmap);
 }
 
-struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key)
+static void *__cpu_map_lookup_elem(struct bpf_map *map, u32 key)
 {
        struct bpf_cpu_map *cmap = container_of(map, struct bpf_cpu_map, map);
        struct bpf_cpu_map_entry *rcpu;
@@ -600,6 +600,11 @@ static int cpu_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
        return 0;
 }
 
+static int cpu_map_redirect(struct bpf_map *map, u32 ifindex, u64 flags)
+{
+       return __bpf_xdp_redirect_map(map, ifindex, flags, __cpu_map_lookup_elem);
+}
+
 static int cpu_map_btf_id;
 const struct bpf_map_ops cpu_map_ops = {
        .map_meta_equal         = bpf_map_meta_equal,
@@ -612,6 +617,7 @@ const struct bpf_map_ops cpu_map_ops = {
        .map_check_btf          = map_check_no_btf,
        .map_btf_name           = "bpf_cpu_map",
        .map_btf_id             = &cpu_map_btf_id,
+       .map_redirect           = cpu_map_redirect,
 };
 
 static void bq_flush_to_queue(struct xdp_bulk_queue *bq)
index 85d9d1b..f7f4244 100644 (file)
@@ -258,7 +258,7 @@ static int dev_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
        return 0;
 }
 
-struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key)
+static void *__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);
@@ -392,7 +392,7 @@ void __dev_flush(void)
  * update happens in parallel here a dev_put wont happen until after reading the
  * ifindex.
  */
-struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key)
+static void *__dev_map_lookup_elem(struct bpf_map *map, u32 key)
 {
        struct bpf_dtab *dtab = container_of(map, struct bpf_dtab, map);
        struct bpf_dtab_netdev *obj;
@@ -735,6 +735,16 @@ static int dev_map_hash_update_elem(struct bpf_map *map, void *key, void *value,
                                         map, key, value, map_flags);
 }
 
+static int dev_map_redirect(struct bpf_map *map, u32 ifindex, u64 flags)
+{
+       return __bpf_xdp_redirect_map(map, ifindex, flags, __dev_map_lookup_elem);
+}
+
+static int dev_hash_map_redirect(struct bpf_map *map, u32 ifindex, u64 flags)
+{
+       return __bpf_xdp_redirect_map(map, ifindex, flags, __dev_map_hash_lookup_elem);
+}
+
 static int dev_map_btf_id;
 const struct bpf_map_ops dev_map_ops = {
        .map_meta_equal = bpf_map_meta_equal,
@@ -747,6 +757,7 @@ const struct bpf_map_ops dev_map_ops = {
        .map_check_btf = map_check_no_btf,
        .map_btf_name = "bpf_dtab",
        .map_btf_id = &dev_map_btf_id,
+       .map_redirect = dev_map_redirect,
 };
 
 static int dev_map_hash_map_btf_id;
@@ -761,6 +772,7 @@ const struct bpf_map_ops dev_map_hash_ops = {
        .map_check_btf = map_check_no_btf,
        .map_btf_name = "bpf_dtab",
        .map_btf_id = &dev_map_hash_map_btf_id,
+       .map_redirect = dev_hash_map_redirect,
 };
 
 static void dev_map_hash_remove_netdev(struct bpf_dtab *dtab,
index 9fe90ce..97eb0b2 100644 (file)
@@ -5582,7 +5582,8 @@ record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta,
            func_id != BPF_FUNC_map_push_elem &&
            func_id != BPF_FUNC_map_pop_elem &&
            func_id != BPF_FUNC_map_peek_elem &&
-           func_id != BPF_FUNC_for_each_map_elem)
+           func_id != BPF_FUNC_for_each_map_elem &&
+           func_id != BPF_FUNC_redirect_map)
                return 0;
 
        if (map == NULL) {
@@ -12017,7 +12018,8 @@ static int do_misc_fixups(struct bpf_verifier_env *env)
                     insn->imm == BPF_FUNC_map_delete_elem ||
                     insn->imm == BPF_FUNC_map_push_elem   ||
                     insn->imm == BPF_FUNC_map_pop_elem    ||
-                    insn->imm == BPF_FUNC_map_peek_elem)) {
+                    insn->imm == BPF_FUNC_map_peek_elem   ||
+                    insn->imm == BPF_FUNC_redirect_map)) {
                        aux = &env->insn_aux_data[i + delta];
                        if (bpf_map_ptr_poisoned(aux))
                                goto patch_call_imm;
@@ -12059,6 +12061,9 @@ static int do_misc_fixups(struct bpf_verifier_env *env)
                                     (int (*)(struct bpf_map *map, void *value))NULL));
                        BUILD_BUG_ON(!__same_type(ops->map_peek_elem,
                                     (int (*)(struct bpf_map *map, void *value))NULL));
+                       BUILD_BUG_ON(!__same_type(ops->map_redirect,
+                                    (int (*)(struct bpf_map *map, u32 ifindex, u64 flags))NULL));
+
 patch_map_ops_generic:
                        switch (insn->imm) {
                        case BPF_FUNC_map_lookup_elem:
@@ -12085,6 +12090,10 @@ patch_map_ops_generic:
                                insn->imm = BPF_CAST_CALL(ops->map_peek_elem) -
                                            __bpf_call_base;
                                continue;
+                       case BPF_FUNC_redirect_map:
+                               insn->imm = BPF_CAST_CALL(ops->map_redirect) -
+                                           __bpf_call_base;
+                               continue;
                        }
 
                        goto patch_call_imm;
index 588b19b..183b0aa 100644 (file)
@@ -3943,22 +3943,6 @@ void xdp_do_flush(void)
 }
 EXPORT_SYMBOL_GPL(xdp_do_flush);
 
-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:
-               return __xsk_map_lookup_elem(map, index);
-       default:
-               return NULL;
-       }
-}
-
 void bpf_clear_redirect_map(struct bpf_map *map)
 {
        struct bpf_redirect_info *ri;
@@ -4112,28 +4096,7 @@ static const struct bpf_func_proto bpf_xdp_redirect_proto = {
 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
           u64, flags)
 {
-       struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
-
-       /* Lower bits of the flags are used as return code on lookup failure */
-       if (unlikely(flags > XDP_TX))
-               return XDP_ABORTED;
-
-       ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
-       if (unlikely(!ri->tgt_value)) {
-               /* If the lookup fails we want to clear out the state in the
-                * redirect_info struct completely, so that if an eBPF program
-                * performs multiple lookups, the last one always takes
-                * precedence.
-                */
-               WRITE_ONCE(ri->map, NULL);
-               return flags;
-       }
-
-       ri->flags = flags;
-       ri->tgt_index = ifindex;
-       WRITE_ONCE(ri->map, map);
-
-       return XDP_REDIRECT;
+       return map->ops->map_redirect(map, ifindex, flags);
 }
 
 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
index 113fd90..fbeb487 100644 (file)
@@ -125,6 +125,16 @@ static int xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
        return insn - insn_buf;
 }
 
+static void *__xsk_map_lookup_elem(struct bpf_map *map, u32 key)
+{
+       struct xsk_map *m = container_of(map, struct xsk_map, map);
+
+       if (key >= map->max_entries)
+               return NULL;
+
+       return READ_ONCE(m->xsk_map[key]);
+}
+
 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
 {
        WARN_ON_ONCE(!rcu_read_lock_held());
@@ -215,6 +225,11 @@ static int xsk_map_delete_elem(struct bpf_map *map, void *key)
        return 0;
 }
 
+static int xsk_map_redirect(struct bpf_map *map, u32 ifindex, u64 flags)
+{
+       return __bpf_xdp_redirect_map(map, ifindex, flags, __xsk_map_lookup_elem);
+}
+
 void xsk_map_try_sock_delete(struct xsk_map *map, struct xdp_sock *xs,
                             struct xdp_sock **map_entry)
 {
@@ -247,4 +262,5 @@ const struct bpf_map_ops xsk_map_ops = {
        .map_check_btf = map_check_no_btf,
        .map_btf_name = "xsk_map",
        .map_btf_id = &xsk_map_btf_id,
+       .map_redirect = xsk_map_redirect,
 };