bpf: Allow bpf_map_lookup_elem() on an xskmap
authorJonathan Lemon <jonathan.lemon@gmail.com>
Thu, 6 Jun 2019 20:59:40 +0000 (13:59 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 11 Jun 2019 06:31:26 +0000 (23:31 -0700)
Currently, the AF_XDP code uses a separate map in order to
determine if an xsk is bound to a queue.  Instead of doing this,
have bpf_map_lookup_elem() return a xdp_sock.

Rearrange some xdp_sock members to eliminate structure holes.

Remove selftest - will be added back in later patch.

Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
include/linux/bpf.h
include/net/xdp_sock.h
include/uapi/linux/bpf.h
kernel/bpf/verifier.c
kernel/bpf/xskmap.c
net/core/filter.c
tools/testing/selftests/bpf/verifier/prevent_map_lookup.c

index e5a309e..1fe137a 100644 (file)
@@ -280,6 +280,7 @@ enum bpf_reg_type {
        PTR_TO_TCP_SOCK,         /* reg points to struct tcp_sock */
        PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */
        PTR_TO_TP_BUFFER,        /* reg points to a writable raw tp's buffer */
+       PTR_TO_XDP_SOCK,         /* reg points to struct xdp_sock */
 };
 
 /* The information passed from prog-specific *_is_valid_access
@@ -727,6 +728,13 @@ 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);
+bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
+                                 struct bpf_insn_access_aux *info);
+u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
+                                   const struct bpf_insn *si,
+                                   struct bpf_insn *insn_buf,
+                                   struct bpf_prog *prog,
+                                   u32 *target_size);
 
 /* Return map's numa specified by userspace */
 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr)
index d074b6d..ae0f368 100644 (file)
@@ -58,11 +58,11 @@ struct xdp_sock {
        struct xdp_umem *umem;
        struct list_head flush_node;
        u16 queue_id;
-       struct xsk_queue *tx ____cacheline_aligned_in_smp;
-       struct list_head list;
        bool zc;
        /* Protects multiple processes in the control path */
        struct mutex mutex;
+       struct xsk_queue *tx ____cacheline_aligned_in_smp;
+       struct list_head list;
        /* Mutual exclusion of NAPI TX thread and sendmsg error paths
         * in the SKB destructor callback.
         */
index 7c6aef2..ae0907d 100644 (file)
@@ -3083,6 +3083,10 @@ struct bpf_sock_tuple {
        };
 };
 
+struct bpf_xdp_sock {
+       __u32 queue_id;
+};
+
 #define XDP_PACKET_HEADROOM 256
 
 /* User return codes for XDP prog type.
index 5c2cb5b..8d17863 100644 (file)
@@ -334,7 +334,8 @@ static bool type_is_sk_pointer(enum bpf_reg_type type)
 {
        return type == PTR_TO_SOCKET ||
                type == PTR_TO_SOCK_COMMON ||
-               type == PTR_TO_TCP_SOCK;
+               type == PTR_TO_TCP_SOCK ||
+               type == PTR_TO_XDP_SOCK;
 }
 
 static bool reg_type_may_be_null(enum bpf_reg_type type)
@@ -406,6 +407,7 @@ static const char * const reg_type_str[] = {
        [PTR_TO_TCP_SOCK]       = "tcp_sock",
        [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null",
        [PTR_TO_TP_BUFFER]      = "tp_buffer",
+       [PTR_TO_XDP_SOCK]       = "xdp_sock",
 };
 
 static char slot_type_char[] = {
@@ -1363,6 +1365,7 @@ static bool is_spillable_regtype(enum bpf_reg_type type)
        case PTR_TO_SOCK_COMMON_OR_NULL:
        case PTR_TO_TCP_SOCK:
        case PTR_TO_TCP_SOCK_OR_NULL:
+       case PTR_TO_XDP_SOCK:
                return true;
        default:
                return false;
@@ -1843,6 +1846,9 @@ static int check_sock_access(struct bpf_verifier_env *env, int insn_idx,
        case PTR_TO_TCP_SOCK:
                valid = bpf_tcp_sock_is_valid_access(off, size, t, &info);
                break;
+       case PTR_TO_XDP_SOCK:
+               valid = bpf_xdp_sock_is_valid_access(off, size, t, &info);
+               break;
        default:
                valid = false;
        }
@@ -2007,6 +2013,9 @@ static int check_ptr_alignment(struct bpf_verifier_env *env,
        case PTR_TO_TCP_SOCK:
                pointer_desc = "tcp_sock ";
                break;
+       case PTR_TO_XDP_SOCK:
+               pointer_desc = "xdp_sock ";
+               break;
        default:
                break;
        }
@@ -2905,10 +2914,14 @@ static int check_map_func_compatibility(struct bpf_verifier_env *env,
         * appear.
         */
        case BPF_MAP_TYPE_CPUMAP:
-       case BPF_MAP_TYPE_XSKMAP:
                if (func_id != BPF_FUNC_redirect_map)
                        goto error;
                break;
+       case BPF_MAP_TYPE_XSKMAP:
+               if (func_id != BPF_FUNC_redirect_map &&
+                   func_id != BPF_FUNC_map_lookup_elem)
+                       goto error;
+               break;
        case BPF_MAP_TYPE_ARRAY_OF_MAPS:
        case BPF_MAP_TYPE_HASH_OF_MAPS:
                if (func_id != BPF_FUNC_map_lookup_elem)
@@ -3799,6 +3812,7 @@ static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env,
        case PTR_TO_SOCK_COMMON_OR_NULL:
        case PTR_TO_TCP_SOCK:
        case PTR_TO_TCP_SOCK_OR_NULL:
+       case PTR_TO_XDP_SOCK:
                verbose(env, "R%d pointer arithmetic on %s prohibited\n",
                        dst, reg_type_str[ptr_reg->type]);
                return -EACCES;
@@ -5038,6 +5052,9 @@ static void mark_ptr_or_null_reg(struct bpf_func_state *state,
                        if (reg->map_ptr->inner_map_meta) {
                                reg->type = CONST_PTR_TO_MAP;
                                reg->map_ptr = reg->map_ptr->inner_map_meta;
+                       } else if (reg->map_ptr->map_type ==
+                                  BPF_MAP_TYPE_XSKMAP) {
+                               reg->type = PTR_TO_XDP_SOCK;
                        } else {
                                reg->type = PTR_TO_MAP_VALUE;
                        }
@@ -6299,6 +6316,7 @@ static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur,
        case PTR_TO_SOCK_COMMON_OR_NULL:
        case PTR_TO_TCP_SOCK:
        case PTR_TO_TCP_SOCK_OR_NULL:
+       case PTR_TO_XDP_SOCK:
                /* Only valid matches are exact, which memcmp() above
                 * would have accepted
                 */
@@ -6693,6 +6711,7 @@ static bool reg_type_mismatch_ok(enum bpf_reg_type type)
        case PTR_TO_SOCK_COMMON_OR_NULL:
        case PTR_TO_TCP_SOCK:
        case PTR_TO_TCP_SOCK_OR_NULL:
+       case PTR_TO_XDP_SOCK:
                return false;
        default:
                return true;
@@ -7826,6 +7845,9 @@ static int convert_ctx_accesses(struct bpf_verifier_env *env)
                case PTR_TO_TCP_SOCK:
                        convert_ctx_access = bpf_tcp_sock_convert_ctx_access;
                        break;
+               case PTR_TO_XDP_SOCK:
+                       convert_ctx_access = bpf_xdp_sock_convert_ctx_access;
+                       break;
                default:
                        continue;
                }
index 413d75f..ef7338c 100644 (file)
@@ -152,6 +152,12 @@ void __xsk_map_flush(struct bpf_map *map)
 
 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
 {
+       WARN_ON_ONCE(!rcu_read_lock_held());
+       return __xsk_map_lookup_elem(map, *(u32 *)key);
+}
+
+static void *xsk_map_lookup_elem_sys_only(struct bpf_map *map, void *key)
+{
        return ERR_PTR(-EOPNOTSUPP);
 }
 
@@ -218,6 +224,7 @@ const struct bpf_map_ops xsk_map_ops = {
        .map_free = xsk_map_free,
        .map_get_next_key = xsk_map_get_next_key,
        .map_lookup_elem = xsk_map_lookup_elem,
+       .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only,
        .map_update_elem = xsk_map_update_elem,
        .map_delete_elem = xsk_map_delete_elem,
        .map_check_btf = map_check_no_btf,
index f2777dc..a5e4ac7 100644 (file)
@@ -5680,6 +5680,46 @@ BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
        return INET_ECN_set_ce(skb);
 }
 
+bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
+                                 struct bpf_insn_access_aux *info)
+{
+       if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
+               return false;
+
+       if (off % size != 0)
+               return false;
+
+       switch (off) {
+       default:
+               return size == sizeof(__u32);
+       }
+}
+
+u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
+                                   const struct bpf_insn *si,
+                                   struct bpf_insn *insn_buf,
+                                   struct bpf_prog *prog, u32 *target_size)
+{
+       struct bpf_insn *insn = insn_buf;
+
+#define BPF_XDP_SOCK_GET(FIELD)                                                \
+       do {                                                            \
+               BUILD_BUG_ON(FIELD_SIZEOF(struct xdp_sock, FIELD) >     \
+                            FIELD_SIZEOF(struct bpf_xdp_sock, FIELD)); \
+               *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
+                                     si->dst_reg, si->src_reg,         \
+                                     offsetof(struct xdp_sock, FIELD)); \
+       } while (0)
+
+       switch (si->off) {
+       case offsetof(struct bpf_xdp_sock, queue_id):
+               BPF_XDP_SOCK_GET(queue_id);
+               break;
+       }
+
+       return insn - insn_buf;
+}
+
 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
        .func           = bpf_skb_ecn_set_ce,
        .gpl_only       = false,
index bbdba99..da7a4b3 100644 (file)
        .prog_type = BPF_PROG_TYPE_SOCK_OPS,
 },
 {
-       "prevent map lookup in xskmap",
-       .insns = {
-       BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
-       BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
-       BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
-       BPF_LD_MAP_FD(BPF_REG_1, 0),
-       BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
-       BPF_EXIT_INSN(),
-       },
-       .fixup_map_xskmap = { 3 },
-       .result = REJECT,
-       .errstr = "cannot pass map_type 17 into func bpf_map_lookup_elem",
-       .prog_type = BPF_PROG_TYPE_XDP,
-},
-{
        "prevent map lookup in stack trace",
        .insns = {
        BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),