OSDN Git Service

bpf: allow eBPF programs to use maps
authorAlexei Starovoitov <ast@plumgrid.com>
Fri, 14 Nov 2014 01:36:49 +0000 (17:36 -0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 18 Nov 2014 18:44:00 +0000 (13:44 -0500)
expose bpf_map_lookup_elem(), bpf_map_update_elem(), bpf_map_delete_elem()
map accessors to eBPF programs

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/bpf.h
include/uapi/linux/bpf.h
kernel/bpf/Makefile
kernel/bpf/helpers.c [new file with mode: 0644]

index 51e9242..75e94ea 100644 (file)
@@ -133,4 +133,9 @@ struct bpf_prog *bpf_prog_get(u32 ufd);
 /* verify correctness of eBPF program */
 int bpf_check(struct bpf_prog *fp, union bpf_attr *attr);
 
+/* verifier prototypes for helper functions called from eBPF programs */
+extern struct bpf_func_proto bpf_map_lookup_elem_proto;
+extern struct bpf_func_proto bpf_map_update_elem_proto;
+extern struct bpf_func_proto bpf_map_delete_elem_proto;
+
 #endif /* _LINUX_BPF_H */
index 0d662fe..4a3d0f8 100644 (file)
@@ -158,6 +158,9 @@ union bpf_attr {
  */
 enum bpf_func_id {
        BPF_FUNC_unspec,
+       BPF_FUNC_map_lookup_elem, /* void *map_lookup_elem(&map, &key) */
+       BPF_FUNC_map_update_elem, /* int map_update_elem(&map, &key, &value, flags) */
+       BPF_FUNC_map_delete_elem, /* int map_delete_elem(&map, &key) */
        __BPF_FUNC_MAX_ID,
 };
 
index 72ec98b..a5ae60f 100644 (file)
@@ -1,5 +1,5 @@
 obj-y := core.o
-obj-$(CONFIG_BPF_SYSCALL) += syscall.o verifier.o hashtab.o arraymap.o
+obj-$(CONFIG_BPF_SYSCALL) += syscall.o verifier.o hashtab.o arraymap.o helpers.o
 ifdef CONFIG_TEST_BPF
 obj-$(CONFIG_BPF_SYSCALL) += test_stub.o
 endif
diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
new file mode 100644 (file)
index 0000000..9e3414d
--- /dev/null
@@ -0,0 +1,89 @@
+/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+#include <linux/bpf.h>
+#include <linux/rcupdate.h>
+
+/* If kernel subsystem is allowing eBPF programs to call this function,
+ * inside its own verifier_ops->get_func_proto() callback it should return
+ * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
+ *
+ * Different map implementations will rely on rcu in map methods
+ * lookup/update/delete, therefore eBPF programs must run under rcu lock
+ * if program is allowed to access maps, so check rcu_read_lock_held in
+ * all three functions.
+ */
+static u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
+{
+       /* verifier checked that R1 contains a valid pointer to bpf_map
+        * and R2 points to a program stack and map->key_size bytes were
+        * initialized
+        */
+       struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
+       void *key = (void *) (unsigned long) r2;
+       void *value;
+
+       WARN_ON_ONCE(!rcu_read_lock_held());
+
+       value = map->ops->map_lookup_elem(map, key);
+
+       /* lookup() returns either pointer to element value or NULL
+        * which is the meaning of PTR_TO_MAP_VALUE_OR_NULL type
+        */
+       return (unsigned long) value;
+}
+
+struct bpf_func_proto bpf_map_lookup_elem_proto = {
+       .func = bpf_map_lookup_elem,
+       .gpl_only = false,
+       .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
+       .arg1_type = ARG_CONST_MAP_PTR,
+       .arg2_type = ARG_PTR_TO_MAP_KEY,
+};
+
+static u64 bpf_map_update_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
+{
+       struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
+       void *key = (void *) (unsigned long) r2;
+       void *value = (void *) (unsigned long) r3;
+
+       WARN_ON_ONCE(!rcu_read_lock_held());
+
+       return map->ops->map_update_elem(map, key, value, r4);
+}
+
+struct bpf_func_proto bpf_map_update_elem_proto = {
+       .func = bpf_map_update_elem,
+       .gpl_only = false,
+       .ret_type = RET_INTEGER,
+       .arg1_type = ARG_CONST_MAP_PTR,
+       .arg2_type = ARG_PTR_TO_MAP_KEY,
+       .arg3_type = ARG_PTR_TO_MAP_VALUE,
+       .arg4_type = ARG_ANYTHING,
+};
+
+static u64 bpf_map_delete_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
+{
+       struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
+       void *key = (void *) (unsigned long) r2;
+
+       WARN_ON_ONCE(!rcu_read_lock_held());
+
+       return map->ops->map_delete_elem(map, key);
+}
+
+struct bpf_func_proto bpf_map_delete_elem_proto = {
+       .func = bpf_map_delete_elem,
+       .gpl_only = false,
+       .ret_type = RET_INTEGER,
+       .arg1_type = ARG_CONST_MAP_PTR,
+       .arg2_type = ARG_PTR_TO_MAP_KEY,
+};