Initial checkin for bpf project

Copied the files from system/netd into system/bpf. No modification is
made to the files.

Test: this doesn't build since the Android.bp file is missing
Bug: 112334572

Change-Id: Ia090a429dd416e07c46766784fb3dd2037201525
This commit is contained in:
Chenbo Feng 2018-10-10 15:01:19 -07:00
parent 3d00fed6eb
commit 75b410bdf1
8 changed files with 1582 additions and 0 deletions

58
bpfloader/Android.bp Normal file
View file

@ -0,0 +1,58 @@
//
// Copyright (C) 2018 The Android Open Source Project
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
//
// bpfLoad binary
//
cc_binary {
name: "bpfloader",
defaults: ["netd_defaults"],
cflags: [
"-Wall",
"-Werror",
"-Wthread-safety",
],
sanitize: {
integer_overflow: true,
},
clang: true,
shared_libs: [
"libcutils",
"libbpf",
"libbase",
"liblog",
"libnetdutils",
],
srcs: [
"BpfLoader.cpp",
],
required: [
"bpf_kern.o",
],
}
bpf {
name: "bpf_kern.o",
srcs: ["bpf_kern.c"],
cflags: [
"-Wall",
"-Werror",
],
include_dirs: ["system/netd/libbpf/include"],
}

101
bpfloader/BpfLoader.cpp Normal file
View file

@ -0,0 +1,101 @@
/*
* Copyright (C) 2017 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef LOG_TAG
#define LOG_TAG "bpfloader"
#endif
#include <arpa/inet.h>
#include <elf.h>
#include <error.h>
#include <fcntl.h>
#include <inttypes.h>
#include <linux/bpf.h>
#include <linux/unistd.h>
#include <net/if.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <android-base/stringprintf.h>
#include <android-base/unique_fd.h>
#include <log/log.h>
#include <netdutils/Misc.h>
#include <netdutils/Slice.h>
#include "bpf/BpfUtils.h"
#include "bpf/bpf_shared.h"
using android::base::unique_fd;
using android::netdutils::Slice;
#define BPF_PROG_PATH "/system/etc/bpf"
#define BPF_PROG_SRC BPF_PROG_PATH "/bpf_kern.o"
#define CLEANANDEXIT(ret, mapPatterns) \
do { \
for (int i = 0; i < mapPatterns.size(); i++) { \
if (mapPatterns[i].fd > -1) { \
close(mapPatterns[i].fd); \
} \
} \
return ret; \
} while (0)
using android::bpf::BpfMapInfo;
using android::bpf::BpfProgInfo;
int main() {
const std::vector<BpfMapInfo> mapPatterns = {
BpfMapInfo(COOKIE_TAG_MAP, COOKIE_TAG_MAP_PATH),
BpfMapInfo(UID_COUNTERSET_MAP, UID_COUNTERSET_MAP_PATH),
BpfMapInfo(APP_UID_STATS_MAP, APP_UID_STATS_MAP_PATH),
BpfMapInfo(UID_STATS_MAP, UID_STATS_MAP_PATH),
BpfMapInfo(TAG_STATS_MAP, TAG_STATS_MAP_PATH),
BpfMapInfo(IFACE_STATS_MAP, IFACE_STATS_MAP_PATH),
BpfMapInfo(CONFIGURATION_MAP, CONFIGURATION_MAP_PATH),
BpfMapInfo(UID_OWNER_MAP, UID_OWNER_MAP_PATH),
};
for (int i = 0; i < mapPatterns.size(); i++) {
if (mapPatterns[i].fd < 0) {
ALOGE("Rerieve Map from %s failed: %d", mapPatterns[i].path.c_str(), mapPatterns[i].fd);
CLEANANDEXIT(-1, mapPatterns);
}
}
BpfProgInfo programs[] = {
{BPF_CGROUP_INET_EGRESS, BPF_EGRESS_PROG_PATH, BPF_CGROUP_EGRESS_PROG_NAME,
BPF_PROG_TYPE_CGROUP_SKB, unique_fd(-1)},
{BPF_CGROUP_INET_INGRESS, BPF_INGRESS_PROG_PATH, BPF_CGROUP_INGRESS_PROG_NAME,
BPF_PROG_TYPE_CGROUP_SKB, unique_fd(-1)},
{MAX_BPF_ATTACH_TYPE, XT_BPF_INGRESS_PROG_PATH, XT_BPF_INGRESS_PROG_NAME,
BPF_PROG_TYPE_SOCKET_FILTER, unique_fd(-1)},
{MAX_BPF_ATTACH_TYPE, XT_BPF_EGRESS_PROG_PATH, XT_BPF_EGRESS_PROG_NAME,
BPF_PROG_TYPE_SOCKET_FILTER, unique_fd(-1)},
{MAX_BPF_ATTACH_TYPE, XT_BPF_WHITELIST_PROG_PATH, XT_BPF_WHITELIST_PROG_NAME,
BPF_PROG_TYPE_SOCKET_FILTER, unique_fd(-1)},
{MAX_BPF_ATTACH_TYPE, XT_BPF_BLACKLIST_PROG_PATH, XT_BPF_BLACKLIST_PROG_NAME,
BPF_PROG_TYPE_SOCKET_FILTER, unique_fd(-1)}};
int ret = android::bpf::parseProgramsFromFile(BPF_PROG_SRC, programs, ARRAY_SIZE(programs),
mapPatterns);
CLEANANDEXIT(ret, mapPatterns);
}

60
bpfloader/bpf_kern.c Normal file
View file

@ -0,0 +1,60 @@
/*
* Copyright (C) 2018 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <linux/bpf.h>
#include "bpf_kern.h"
ELF_SEC(BPF_CGROUP_INGRESS_PROG_NAME)
int bpf_cgroup_ingress(struct __sk_buff* skb) {
return bpf_traffic_account(skb, BPF_INGRESS);
}
ELF_SEC(BPF_CGROUP_EGRESS_PROG_NAME)
int bpf_cgroup_egress(struct __sk_buff* skb) {
return bpf_traffic_account(skb, BPF_EGRESS);
}
ELF_SEC(XT_BPF_EGRESS_PROG_NAME)
int xt_bpf_egress_prog(struct __sk_buff* skb) {
uint32_t key = skb->ifindex;
bpf_update_stats(skb, IFACE_STATS_MAP, BPF_EGRESS, &key);
return BPF_MATCH;
}
ELF_SEC(XT_BPF_INGRESS_PROG_NAME)
int xt_bpf_ingress_prog(struct __sk_buff* skb) {
uint32_t key = skb->ifindex;
bpf_update_stats(skb, IFACE_STATS_MAP, BPF_INGRESS, &key);
return BPF_MATCH;
}
ELF_SEC(XT_BPF_WHITELIST_PROG_NAME)
int xt_bpf_whitelist_prog(struct __sk_buff* skb) {
uint32_t sock_uid = get_socket_uid(skb);
if (is_system_uid(sock_uid)) return BPF_MATCH;
uint8_t* whitelistMatch = find_map_entry(UID_OWNER_MAP, &sock_uid);
if (whitelistMatch) return *whitelistMatch & HAPPY_BOX_MATCH;
return BPF_NOMATCH;
}
ELF_SEC(XT_BPF_BLACKLIST_PROG_NAME)
int xt_bpf_blacklist_prog(struct __sk_buff* skb) {
uint32_t sock_uid = get_socket_uid(skb);
uint8_t* blacklistMatch = find_map_entry(UID_OWNER_MAP, &sock_uid);
if (blacklistMatch) return *blacklistMatch & PENALTY_BOX_MATCH;
return BPF_NOMATCH;
}

220
bpfloader/bpf_kern.h Normal file
View file

@ -0,0 +1,220 @@
/*
* Copyright (C) 2018 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* This h file together with bpf_kern.c is used for compiling the eBPF kernel
* program. To generate the bpf_kern.o file manually, use the clang prebuilt in
* this android tree to compile the files with --target=bpf options. For
* example, in system/netd/ directory, execute the following command:
* $: ANDROID_BASE_DIRECTORY/prebuilts/clang/host/linux-x86/clang-4691093/bin/clang \
* -I ANDROID_BASE_DIRECTORY/bionic/libc/kernel/uapi/ \
* -I ANDROID_BASE_DIRECTORY/system/netd/bpfloader/ \
* -I ANDROID_BASE_DIRECTORY/bionic/libc/kernel/android/uapi/ \
* -I ANDROID_BASE_DIRECTORY/bionic/libc/include \
* -I ANDROID_BASE_DIRECTORY/system/netd/libbpf/include \
* --target=bpf -O2 -c bpfloader/bpf_kern.c -o bpfloader/bpf_kern.o
*/
#include <linux/bpf.h>
#include <linux/if.h>
#include <linux/if_ether.h>
#include <linux/in.h>
#include <linux/in6.h>
#include <linux/ip.h>
#include <linux/ipv6.h>
#include <stdbool.h>
#include <stdint.h>
#include "bpf/bpf_shared.h"
#define ELF_SEC(NAME) __attribute__((section(NAME), used))
struct uid_tag {
uint32_t uid;
uint32_t tag;
};
struct stats_key {
uint32_t uid;
uint32_t tag;
uint32_t counterSet;
uint32_t ifaceIndex;
};
struct stats_value {
uint64_t rxPackets;
uint64_t rxBytes;
uint64_t txPackets;
uint64_t txBytes;
};
/* helper functions called from eBPF programs written in C */
static void* (*find_map_entry)(uint64_t map, void* key) = (void*)BPF_FUNC_map_lookup_elem;
static int (*write_to_map_entry)(uint64_t map, void* key, void* value,
uint64_t flags) = (void*)BPF_FUNC_map_update_elem;
static int (*delete_map_entry)(uint64_t map, void* key) = (void*)BPF_FUNC_map_delete_elem;
static uint64_t (*get_socket_cookie)(struct __sk_buff* skb) = (void*)BPF_FUNC_get_socket_cookie;
static uint32_t (*get_socket_uid)(struct __sk_buff* skb) = (void*)BPF_FUNC_get_socket_uid;
static int (*bpf_skb_load_bytes)(struct __sk_buff* skb, int off, void* to,
int len) = (void*)BPF_FUNC_skb_load_bytes;
// This is defined for cgroup bpf filter only.
#define BPF_PASS 1
#define BPF_DROP 0
// This is used for xt_bpf program only.
#define BPF_NOMATCH 0
#define BPF_MATCH 1
#define BPF_EGRESS 0
#define BPF_INGRESS 1
#define IP_PROTO_OFF offsetof(struct iphdr, protocol)
#define IPV6_PROTO_OFF offsetof(struct ipv6hdr, nexthdr)
#define IPPROTO_IHL_OFF 0
#define TCP_FLAG_OFF 13
#define RST_OFFSET 2
static __always_inline int is_system_uid(uint32_t uid) {
return (uid <= MAX_SYSTEM_UID) && (uid >= MIN_SYSTEM_UID);
}
static __always_inline inline void bpf_update_stats(struct __sk_buff* skb, uint64_t map,
int direction, void *key) {
struct stats_value* value;
value = find_map_entry(map, key);
if (!value) {
struct stats_value newValue = {};
write_to_map_entry(map, key, &newValue, BPF_NOEXIST);
value = find_map_entry(map, key);
}
if (value) {
if (direction == BPF_EGRESS) {
__sync_fetch_and_add(&value->txPackets, 1);
__sync_fetch_and_add(&value->txBytes, skb->len);
} else if (direction == BPF_INGRESS) {
__sync_fetch_and_add(&value->rxPackets, 1);
__sync_fetch_and_add(&value->rxBytes, skb->len);
}
}
}
static inline bool skip_owner_match(struct __sk_buff* skb) {
int offset = -1;
int ret = 0;
if (skb->protocol == ETH_P_IP) {
offset = IP_PROTO_OFF;
uint8_t proto, ihl;
uint16_t flag;
ret = bpf_skb_load_bytes(skb, offset, &proto, 1);
if (!ret) {
if (proto == IPPROTO_ESP) {
return true;
} else if (proto == IPPROTO_TCP) {
ret = bpf_skb_load_bytes(skb, IPPROTO_IHL_OFF, &ihl, 1);
ihl = ihl & 0x0F;
ret = bpf_skb_load_bytes(skb, ihl * 4 + TCP_FLAG_OFF, &flag, 1);
if (ret == 0 && (flag >> RST_OFFSET & 1)) {
return true;
}
}
}
} else if (skb->protocol == ETH_P_IPV6) {
offset = IPV6_PROTO_OFF;
uint8_t proto;
ret = bpf_skb_load_bytes(skb, offset, &proto, 1);
if (!ret) {
if (proto == IPPROTO_ESP) {
return true;
} else if (proto == IPPROTO_TCP) {
uint16_t flag;
ret = bpf_skb_load_bytes(skb, sizeof(struct ipv6hdr) + TCP_FLAG_OFF, &flag, 1);
if (ret == 0 && (flag >> RST_OFFSET & 1)) {
return true;
}
}
}
}
return false;
}
static __always_inline BpfConfig getConfig() {
uint32_t mapSettingKey = CONFIGURATION_KEY;
BpfConfig* config = find_map_entry(CONFIGURATION_MAP, &mapSettingKey);
if (!config) {
// Couldn't read configuration entry. Assume everything is disabled.
return DEFAULT_CONFIG;
}
return *config;
}
static inline int bpf_owner_match(struct __sk_buff* skb, uint32_t uid) {
if (skip_owner_match(skb)) return BPF_PASS;
if ((uid <= MAX_SYSTEM_UID) && (uid >= MIN_SYSTEM_UID)) return BPF_PASS;
BpfConfig enabledRules = getConfig();
if (!enabledRules) {
return BPF_PASS;
}
uint8_t* uidEntry = find_map_entry(UID_OWNER_MAP, &uid);
uint8_t uidRules = uidEntry ? *uidEntry : 0;
if ((enabledRules & DOZABLE_MATCH) && !(uidRules & DOZABLE_MATCH)) {
return BPF_DROP;
}
if ((enabledRules & STANDBY_MATCH) && (uidRules & STANDBY_MATCH)) {
return BPF_DROP;
}
if ((enabledRules & POWERSAVE_MATCH) && !(uidRules & POWERSAVE_MATCH)) {
return BPF_DROP;
}
return BPF_PASS;
}
static __always_inline inline int bpf_traffic_account(struct __sk_buff* skb, int direction) {
uint32_t sock_uid = get_socket_uid(skb);
int match = bpf_owner_match(skb, sock_uid);
if ((direction == BPF_EGRESS) && (match == BPF_DROP)) {
// If an outbound packet is going to be dropped, we do not count that
// traffic.
return match;
}
uint64_t cookie = get_socket_cookie(skb);
struct uid_tag* utag = find_map_entry(COOKIE_TAG_MAP, &cookie);
uint32_t uid, tag;
if (utag) {
uid = utag->uid;
tag = utag->tag;
} else {
uid = sock_uid;
tag = 0;
}
struct stats_key key = {.uid = uid, .tag = tag, .counterSet = 0, .ifaceIndex = skb->ifindex};
uint8_t* counterSet = find_map_entry(UID_COUNTERSET_MAP, &uid);
if (counterSet) key.counterSet = (uint32_t)*counterSet;
if (tag) {
bpf_update_stats(skb, TAG_STATS_MAP, direction, &key);
}
key.tag = 0;
bpf_update_stats(skb, UID_STATS_MAP, direction, &key);
bpf_update_stats(skb, APP_UID_STATS_MAP, direction, &uid);
return match;
}

276
libbpf/BpfMapTest.cpp Normal file
View file

@ -0,0 +1,276 @@
/*
* Copyright (C) 2018 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <fstream>
#include <iostream>
#include <string>
#include <vector>
#include <fcntl.h>
#include <inttypes.h>
#include <linux/inet_diag.h>
#include <linux/sock_diag.h>
#include <net/if.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <unistd.h>
#include <gtest/gtest.h>
#include <android-base/stringprintf.h>
#include <android-base/strings.h>
#include <netdutils/MockSyscalls.h>
#include "bpf/BpfMap.h"
#include "bpf/BpfNetworkStats.h"
#include "bpf/BpfUtils.h"
using ::testing::_;
using ::testing::ByMove;
using ::testing::Invoke;
using ::testing::Return;
using ::testing::StrictMock;
using ::testing::Test;
namespace android {
namespace bpf {
using base::unique_fd;
using netdutils::StatusOr;
constexpr uint32_t TEST_MAP_SIZE = 10;
constexpr uint32_t TEST_KEY1 = 1;
constexpr uint32_t TEST_VALUE1 = 10;
constexpr const char PINNED_MAP_PATH[] = "/sys/fs/bpf/testMap";
class BpfMapTest : public testing::Test {
protected:
BpfMapTest() {}
int mMapFd;
void SetUp() {
if (!access(PINNED_MAP_PATH, R_OK)) {
EXPECT_EQ(0, remove(PINNED_MAP_PATH));
}
mMapFd = createMap(BPF_MAP_TYPE_HASH, sizeof(uint32_t), sizeof(uint32_t), TEST_MAP_SIZE,
BPF_F_NO_PREALLOC);
}
void TearDown() {
if (!access(PINNED_MAP_PATH, R_OK)) {
EXPECT_EQ(0, remove(PINNED_MAP_PATH));
}
close(mMapFd);
}
void checkMapInvalid(BpfMap<uint32_t, uint32_t>& map) {
EXPECT_FALSE(map.isValid());
EXPECT_EQ(-1, map.getMap().get());
EXPECT_TRUE(map.getPinnedPath().empty());
}
void checkMapValid(BpfMap<uint32_t, uint32_t>& map) {
EXPECT_LE(0, map.getMap().get());
EXPECT_TRUE(map.isValid());
}
void writeToMapAndCheck(BpfMap<uint32_t, uint32_t>& map, uint32_t key, uint32_t value) {
ASSERT_TRUE(isOk(map.writeValue(key, value, BPF_ANY)));
uint32_t value_read;
ASSERT_EQ(0, findMapEntry(map.getMap(), &key, &value_read));
checkValueAndStatus(value, value_read);
}
void checkValueAndStatus(uint32_t refValue, StatusOr<uint32_t> value) {
ASSERT_TRUE(isOk(value.status()));
ASSERT_EQ(refValue, value.value());
}
void populateMap(uint32_t total, BpfMap<uint32_t, uint32_t>& map) {
for (uint32_t key = 0; key < total; key++) {
uint32_t value = key * 10;
EXPECT_TRUE(isOk(map.writeValue(key, value, BPF_ANY)));
}
}
void expectMapEmpty(BpfMap<uint32_t, uint32_t>& map) {
auto isEmpty = map.isEmpty();
ASSERT_TRUE(isOk(isEmpty));
ASSERT_TRUE(isEmpty.value());
}
};
TEST_F(BpfMapTest, constructor) {
BpfMap<uint32_t, uint32_t> testMap1;
checkMapInvalid(testMap1);
BpfMap<uint32_t, uint32_t> testMap2(mMapFd);
checkMapValid(testMap2);
EXPECT_TRUE(testMap2.getPinnedPath().empty());
BpfMap<uint32_t, uint32_t> testMap3(BPF_MAP_TYPE_HASH, TEST_MAP_SIZE, BPF_F_NO_PREALLOC);
checkMapValid(testMap3);
EXPECT_TRUE(testMap3.getPinnedPath().empty());
}
TEST_F(BpfMapTest, basicHelpers) {
BpfMap<uint32_t, uint32_t> testMap(mMapFd);
uint32_t key = TEST_KEY1;
uint32_t value_write = TEST_VALUE1;
writeToMapAndCheck(testMap, key, value_write);
StatusOr<uint32_t> value_read = testMap.readValue(key);
checkValueAndStatus(value_write, value_read);
StatusOr<uint32_t> key_read = testMap.getFirstKey();
checkValueAndStatus(key, key_read);
ASSERT_TRUE(isOk(testMap.deleteValue(key)));
ASSERT_GT(0, findMapEntry(testMap.getMap(), &key, &value_read));
ASSERT_EQ(ENOENT, errno);
}
TEST_F(BpfMapTest, reset) {
BpfMap<uint32_t, uint32_t> testMap;
testMap.reset(mMapFd);
uint32_t key = TEST_KEY1;
uint32_t value_write = TEST_VALUE1;
writeToMapAndCheck(testMap, key, value_write);
testMap.reset();
checkMapInvalid(testMap);
unique_fd invalidFd(mMapFd);
ASSERT_GT(0, findMapEntry(invalidFd, &key, &value_write));
ASSERT_EQ(EBADF, errno);
}
TEST_F(BpfMapTest, moveConstructor) {
BpfMap<uint32_t, uint32_t> testMap1(mMapFd);
BpfMap<uint32_t, uint32_t> testMap2;
testMap2 = std::move(testMap1);
uint32_t key = TEST_KEY1;
checkMapInvalid(testMap1);
uint32_t value = TEST_VALUE1;
writeToMapAndCheck(testMap2, key, value);
}
TEST_F(BpfMapTest, pinnedToPath) {
BpfMap<uint32_t, uint32_t> testMap1(mMapFd);
EXPECT_OK(testMap1.pinToPath(PINNED_MAP_PATH));
EXPECT_EQ(0, access(PINNED_MAP_PATH, R_OK));
EXPECT_EQ(0, testMap1.getPinnedPath().compare(PINNED_MAP_PATH));
BpfMap<uint32_t, uint32_t> testMap2(mapRetrieve(PINNED_MAP_PATH, 0));
checkMapValid(testMap2);
uint32_t key = TEST_KEY1;
uint32_t value = TEST_VALUE1;
writeToMapAndCheck(testMap1, key, value);
StatusOr<uint32_t> value_read = testMap2.readValue(key);
checkValueAndStatus(value, value_read);
}
TEST_F(BpfMapTest, SetUpMap) {
BpfMap<uint32_t, uint32_t> testMap1;
EXPECT_OK(testMap1.getOrCreate(TEST_MAP_SIZE, PINNED_MAP_PATH, BPF_MAP_TYPE_HASH));
EXPECT_EQ(0, access(PINNED_MAP_PATH, R_OK));
checkMapValid(testMap1);
EXPECT_EQ(0, testMap1.getPinnedPath().compare(PINNED_MAP_PATH));
BpfMap<uint32_t, uint32_t> testMap2;
EXPECT_OK(testMap2.getOrCreate(TEST_MAP_SIZE, PINNED_MAP_PATH, BPF_MAP_TYPE_HASH));
checkMapValid(testMap2);
EXPECT_EQ(0, testMap2.getPinnedPath().compare(PINNED_MAP_PATH));
uint32_t key = TEST_KEY1;
uint32_t value = TEST_VALUE1;
writeToMapAndCheck(testMap1, key, value);
StatusOr<uint32_t> value_read = testMap2.readValue(key);
checkValueAndStatus(value, value_read);
}
TEST_F(BpfMapTest, iterate) {
BpfMap<uint32_t, uint32_t> testMap(mMapFd);
populateMap(TEST_MAP_SIZE, testMap);
int totalCount = 0;
int totalSum = 0;
const auto iterateWithDeletion = [&totalCount, &totalSum](const uint32_t& key,
BpfMap<uint32_t, uint32_t>& map) {
EXPECT_GE((uint32_t)TEST_MAP_SIZE, key);
totalCount++;
totalSum += key;
return map.deleteValue(key);
};
EXPECT_OK(testMap.iterate(iterateWithDeletion));
EXPECT_EQ((int)TEST_MAP_SIZE, totalCount);
EXPECT_EQ(((1 + TEST_MAP_SIZE - 1) * (TEST_MAP_SIZE - 1)) / 2, (uint32_t)totalSum);
expectMapEmpty(testMap);
}
TEST_F(BpfMapTest, iterateWithValue) {
BpfMap<uint32_t, uint32_t> testMap(mMapFd);
populateMap(TEST_MAP_SIZE, testMap);
int totalCount = 0;
int totalSum = 0;
const auto iterateWithDeletion = [&totalCount, &totalSum](const uint32_t& key,
const uint32_t& value,
BpfMap<uint32_t, uint32_t>& map) {
EXPECT_GE((uint32_t)TEST_MAP_SIZE, key);
EXPECT_EQ(value, key * 10);
totalCount++;
totalSum += value;
return map.deleteValue(key);
};
EXPECT_OK(testMap.iterateWithValue(iterateWithDeletion));
EXPECT_EQ((int)TEST_MAP_SIZE, totalCount);
EXPECT_EQ(((1 + TEST_MAP_SIZE - 1) * (TEST_MAP_SIZE - 1)) * 5, (uint32_t)totalSum);
expectMapEmpty(testMap);
}
TEST_F(BpfMapTest, mapIsEmpty) {
BpfMap<uint32_t, uint32_t> testMap(mMapFd);
expectMapEmpty(testMap);
uint32_t key = TEST_KEY1;
uint32_t value_write = TEST_VALUE1;
writeToMapAndCheck(testMap, key, value_write);
auto isEmpty = testMap.isEmpty();
ASSERT_TRUE(isOk(isEmpty));
ASSERT_FALSE(isEmpty.value());
ASSERT_TRUE(isOk(testMap.deleteValue(key)));
ASSERT_GT(0, findMapEntry(testMap.getMap(), &key, &value_write));
ASSERT_EQ(ENOENT, errno);
expectMapEmpty(testMap);
int entriesSeen = 0;
EXPECT_OK(testMap.iterate(
[&entriesSeen](const unsigned int&,
const BpfMap<unsigned int, unsigned int>&) -> netdutils::Status {
entriesSeen++;
return netdutils::status::ok;
}));
EXPECT_EQ(0, entriesSeen);
EXPECT_OK(testMap.iterateWithValue(
[&entriesSeen](const unsigned int&, const unsigned int&,
const BpfMap<unsigned int, unsigned int>&) -> netdutils::Status {
entriesSeen++;
return netdutils::status::ok;
}));
EXPECT_EQ(0, entriesSeen);
}
TEST_F(BpfMapTest, mapClear) {
BpfMap<uint32_t, uint32_t> testMap(mMapFd);
populateMap(TEST_MAP_SIZE, testMap);
auto isEmpty = testMap.isEmpty();
ASSERT_TRUE(isOk(isEmpty));
ASSERT_FALSE(isEmpty.value());
ASSERT_TRUE(isOk(testMap.clear()));
expectMapEmpty(testMap);
}
} // namespace bpf
} // namespace android

415
libbpf/BpfUtils.cpp Normal file
View file

@ -0,0 +1,415 @@
/*
* Copyright (C) 2017 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#define LOG_TAG "BpfUtils"
#include <elf.h>
#include <inttypes.h>
#include <linux/bpf.h>
#include <linux/if_ether.h>
#include <linux/in.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/utsname.h>
#include <sstream>
#include <string>
#include <android-base/properties.h>
#include <android-base/stringprintf.h>
#include <android-base/unique_fd.h>
#include <netdutils/MemBlock.h>
#include <netdutils/Slice.h>
#include <netdutils/StatusOr.h>
#include "bpf/BpfUtils.h"
using android::base::GetUintProperty;
using android::base::StringPrintf;
using android::base::unique_fd;
using android::netdutils::MemBlock;
using android::netdutils::Slice;
using android::netdutils::statusFromErrno;
using android::netdutils::StatusOr;
constexpr size_t LOG_BUF_SIZE = 65536;
namespace android {
namespace bpf {
/* The bpf_attr is a union which might have a much larger size then the struct we are using, while
* The inline initializer only reset the field we are using and leave the reset of the memory as
* is. The bpf kernel code will performs a much stricter check to ensure all unused field is 0. So
* this syscall will normally fail with E2BIG if we don't do a memset to bpf_attr.
*/
bool operator==(const StatsKey& lhs, const StatsKey& rhs) {
return ((lhs.uid == rhs.uid) && (lhs.tag == rhs.tag) && (lhs.counterSet == rhs.counterSet) &&
(lhs.ifaceIndex == rhs.ifaceIndex));
}
bool operator==(const UidTag& lhs, const UidTag& rhs) {
return ((lhs.uid == rhs.uid) && (lhs.tag == rhs.tag));
}
bool operator==(const StatsValue& lhs, const StatsValue& rhs) {
return ((lhs.rxBytes == rhs.rxBytes) && (lhs.txBytes == rhs.txBytes) &&
(lhs.rxPackets == rhs.rxPackets) && (lhs.txPackets == rhs.txPackets));
}
int bpf(int cmd, Slice bpfAttr) {
return syscall(__NR_bpf, cmd, bpfAttr.base(), bpfAttr.size());
}
int createMap(bpf_map_type map_type, uint32_t key_size, uint32_t value_size, uint32_t max_entries,
uint32_t map_flags) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.map_type = map_type;
attr.key_size = key_size;
attr.value_size = value_size;
attr.max_entries = max_entries;
attr.map_flags = map_flags;
return bpf(BPF_MAP_CREATE, Slice(&attr, sizeof(attr)));
}
int writeToMapEntry(const base::unique_fd& map_fd, void* key, void* value, uint64_t flags) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.map_fd = map_fd.get();
attr.key = ptr_to_u64(key);
attr.value = ptr_to_u64(value);
attr.flags = flags;
return bpf(BPF_MAP_UPDATE_ELEM, Slice(&attr, sizeof(attr)));
}
int findMapEntry(const base::unique_fd& map_fd, void* key, void* value) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.map_fd = map_fd.get();
attr.key = ptr_to_u64(key);
attr.value = ptr_to_u64(value);
return bpf(BPF_MAP_LOOKUP_ELEM, Slice(&attr, sizeof(attr)));
}
int deleteMapEntry(const base::unique_fd& map_fd, void* key) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.map_fd = map_fd.get();
attr.key = ptr_to_u64(key);
return bpf(BPF_MAP_DELETE_ELEM, Slice(&attr, sizeof(attr)));
}
int getNextMapKey(const base::unique_fd& map_fd, void* key, void* next_key) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.map_fd = map_fd.get();
attr.key = ptr_to_u64(key);
attr.next_key = ptr_to_u64(next_key);
return bpf(BPF_MAP_GET_NEXT_KEY, Slice(&attr, sizeof(attr)));
}
int getFirstMapKey(const base::unique_fd& map_fd, void* firstKey) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.map_fd = map_fd.get();
attr.key = 0;
attr.next_key = ptr_to_u64(firstKey);
return bpf(BPF_MAP_GET_NEXT_KEY, Slice(&attr, sizeof(attr)));
}
int bpfProgLoad(bpf_prog_type prog_type, Slice bpf_insns, const char* license,
uint32_t kern_version, Slice bpf_log) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.prog_type = prog_type;
attr.insns = ptr_to_u64(bpf_insns.base());
attr.insn_cnt = bpf_insns.size() / sizeof(struct bpf_insn);
attr.license = ptr_to_u64((void*)license);
attr.log_buf = ptr_to_u64(bpf_log.base());
attr.log_size = bpf_log.size();
attr.log_level = DEFAULT_LOG_LEVEL;
attr.kern_version = kern_version;
int ret = bpf(BPF_PROG_LOAD, Slice(&attr, sizeof(attr)));
if (ret < 0) {
std::string prog_log = netdutils::toString(bpf_log);
std::istringstream iss(prog_log);
for (std::string line; std::getline(iss, line);) {
ALOGE("%s", line.c_str());
}
}
return ret;
}
int bpfFdPin(const base::unique_fd& map_fd, const char* pathname) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.pathname = ptr_to_u64((void*)pathname);
attr.bpf_fd = map_fd.get();
return bpf(BPF_OBJ_PIN, Slice(&attr, sizeof(attr)));
}
int mapRetrieve(const char* pathname, uint32_t flag) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.pathname = ptr_to_u64((void*)pathname);
attr.file_flags = flag;
return bpf(BPF_OBJ_GET, Slice(&attr, sizeof(attr)));
}
int attachProgram(bpf_attach_type type, uint32_t prog_fd, uint32_t cg_fd) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.target_fd = cg_fd;
attr.attach_bpf_fd = prog_fd;
attr.attach_type = type;
return bpf(BPF_PROG_ATTACH, Slice(&attr, sizeof(attr)));
}
int detachProgram(bpf_attach_type type, uint32_t cg_fd) {
bpf_attr attr;
memset(&attr, 0, sizeof(attr));
attr.target_fd = cg_fd;
attr.attach_type = type;
return bpf(BPF_PROG_DETACH, Slice(&attr, sizeof(attr)));
}
uint64_t getSocketCookie(int sockFd) {
uint64_t sock_cookie;
socklen_t cookie_len = sizeof(sock_cookie);
int res = getsockopt(sockFd, SOL_SOCKET, SO_COOKIE, &sock_cookie, &cookie_len);
if (res < 0) {
res = -errno;
ALOGE("Failed to get socket cookie: %s\n", strerror(errno));
errno = -res;
// 0 is an invalid cookie. See sock_gen_cookie.
return NONEXISTENT_COOKIE;
}
return sock_cookie;
}
bool hasBpfSupport() {
struct utsname buf;
int kernel_version_major;
int kernel_version_minor;
uint64_t api_level = GetUintProperty<uint64_t>("ro.product.first_api_level", 0);
if (api_level == 0) {
ALOGE("Cannot determine initial API level of the device");
api_level = GetUintProperty<uint64_t>("ro.build.version.sdk", 0);
}
int ret = uname(&buf);
if (ret) {
return false;
}
char dummy;
ret = sscanf(buf.release, "%d.%d%c", &kernel_version_major, &kernel_version_minor, &dummy);
if (ret >= 2 && ((kernel_version_major > 4) ||
(kernel_version_major == 4 && kernel_version_minor >= 9))) {
// Check if the device is shipped originally with android P.
return api_level >= MINIMUM_API_REQUIRED;
}
return false;
}
int loadAndPinProgram(BpfProgInfo* prog, Slice progBlock) {
// Program doesn't exist. Try to load it.
char bpf_log_buf[LOG_BUF_SIZE];
Slice bpfLog = Slice(bpf_log_buf, sizeof(bpf_log_buf));
prog->fd.reset(bpfProgLoad(prog->loadType, progBlock, "Apache 2.0", 0, bpfLog));
if (prog->fd < 0) {
int ret = -errno;
ALOGE("load %s failed: %s", prog->name, strerror(errno));
return ret;
}
if (prog->attachType == BPF_CGROUP_INET_EGRESS || prog->attachType == BPF_CGROUP_INET_INGRESS) {
unique_fd cg_fd(open(CGROUP_ROOT_PATH, O_DIRECTORY | O_RDONLY | O_CLOEXEC));
if (cg_fd < 0) {
int ret = -errno;
ALOGE("Failed to open the cgroup directory");
return ret;
}
int ret = android::bpf::attachProgram(prog->attachType, prog->fd, cg_fd);
if (ret) {
ret = -errno;
ALOGE("%s attach failed: %s", prog->name, strerror(errno));
return ret;
}
}
if (prog->path) {
int ret = android::bpf::bpfFdPin(prog->fd, prog->path);
if (ret) {
ret = -errno;
ALOGE("Pin %s as file %s failed: %s", prog->name, prog->path, strerror(errno));
return ret;
}
}
return 0;
}
int extractAndLoadProg(BpfProgInfo* prog, Elf64_Shdr* sectionPtr, Slice fileContents,
const std::vector<BpfMapInfo>& mapPatterns) {
uint64_t progSize = (uint64_t) sectionPtr->sh_size;
Slice progSection = take(drop(fileContents, sectionPtr->sh_offset), progSize);
if (progSection.size() < progSize) {
ALOGE("programSection out of bound");
return -EINVAL;
}
MemBlock progCopy(progSection);
if (progCopy.get().size() != progSize) {
ALOGE("program cannot be extracted");
return -EINVAL;
}
Slice remaining = progCopy.get();
while (remaining.size() >= MAP_CMD_SIZE) {
// Scan the program, examining all possible places that might be the start of a
// map load operation (i.e., all bytes of value MAP_LD_CMD_HEAD).
// In each of these places, check whether it is the start of one of the patterns
// we want to replace, and if so, replace it.
Slice mapHead = findFirstMatching(remaining, MAP_LD_CMD_HEAD);
if (mapHead.size() < MAP_CMD_SIZE) break;
bool replaced = false;
for (const auto& pattern : mapPatterns) {
if (!memcmp(mapHead.base(), pattern.search.data(), MAP_CMD_SIZE)) {
memcpy(mapHead.base(), pattern.replace.data(), MAP_CMD_SIZE);
replaced = true;
break;
}
}
remaining = drop(mapHead, replaced ? MAP_CMD_SIZE : sizeof(uint8_t));
}
if (!(prog->path) || access(prog->path, R_OK) == -1) {
return loadAndPinProgram(prog, progCopy.get());
}
return 0;
}
int parsePrograms(Slice fileContents, BpfProgInfo* programs, size_t size,
const std::vector<BpfMapInfo>& mapPatterns) {
Slice elfHeader = take(fileContents, sizeof(Elf64_Ehdr));
if (elfHeader.size() < sizeof(Elf64_Ehdr)) {
ALOGE("bpf fileContents does not have complete elf header");
return -EINVAL;
}
Elf64_Ehdr* elf = (Elf64_Ehdr*) elfHeader.base();
// Find section names string table. This is the section whose index is e_shstrndx.
if (elf->e_shstrndx == SHN_UNDEF) {
ALOGE("cannot locate namesSection\n");
return -EINVAL;
}
size_t totalSectionSize = (elf->e_shnum) * sizeof(Elf64_Shdr);
Slice sections = take(drop(fileContents, elf->e_shoff), totalSectionSize);
if (sections.size() < totalSectionSize) {
ALOGE("sections corrupted");
return -EMSGSIZE;
}
Slice namesSection =
take(drop(sections, elf->e_shstrndx * sizeof(Elf64_Shdr)), sizeof(Elf64_Shdr));
if (namesSection.size() != sizeof(Elf64_Shdr)) {
ALOGE("namesSection corrupted");
return -EMSGSIZE;
}
size_t strTabOffset = ((Elf64_Shdr*) namesSection.base())->sh_offset;
size_t strTabSize = ((Elf64_Shdr*) namesSection.base())->sh_size;
Slice strTab = take(drop(fileContents, strTabOffset), strTabSize);
if (strTab.size() < strTabSize) {
ALOGE("string table out of bound\n");
return -EMSGSIZE;
}
for (int i = 0; i < elf->e_shnum; i++) {
Slice section = take(drop(sections, i * sizeof(Elf64_Shdr)), sizeof(Elf64_Shdr));
if (section.size() < sizeof(Elf64_Shdr)) {
ALOGE("section %d is out of bound, section size: %zu, header size: %zu, total size: "
"%zu",
i, section.size(), sizeof(Elf64_Shdr), sections.size());
return -EBADF;
}
Elf64_Shdr* sectionPtr = (Elf64_Shdr*) section.base();
Slice nameSlice = drop(strTab, sectionPtr->sh_name);
if (nameSlice.size() == 0) {
ALOGE("nameSlice out of bound, i: %d, strTabSize: %zu, sh_name: %u", i, strTabSize,
sectionPtr->sh_name);
return -EBADF;
}
for (size_t i = 0; i < size; i++) {
BpfProgInfo* prog = programs + i;
if (!strcmp((char*) nameSlice.base(), prog->name)) {
int ret = extractAndLoadProg(prog, sectionPtr, fileContents, mapPatterns);
if (ret) return ret;
}
}
}
// Check all the program struct passed in to make sure they all have a valid fd.
for (size_t i = 0; i < size; i++) {
BpfProgInfo* prog = programs + i;
if (access(prog->path, R_OK) == -1) {
ALOGE("Load program %s failed", prog->name);
return -EINVAL;
}
}
return 0;
}
int parseProgramsFromFile(const char* path, BpfProgInfo* programs, size_t size,
const std::vector<BpfMapInfo>& mapPatterns) {
unique_fd fd(open(path, O_RDONLY));
int ret;
if (fd < 0) {
ret = -errno;
ALOGE("Failed to open %s program: %s", path, strerror(errno));
return ret;
}
struct stat stat;
if (fstat(fd.get(), &stat)) {
ret = -errno;
ALOGE("Failed to get file (%s) size: %s", path, strerror(errno));
return ret;
}
off_t fileLen = stat.st_size;
char* baseAddr =
(char*) mmap(NULL, fileLen, PROT_READ, MAP_PRIVATE | MAP_POPULATE, fd.get(), 0);
if (baseAddr == MAP_FAILED) {
ALOGE("Failed to map the program (%s) into memory: %s", path, strerror(errno));
ret = -errno;
return ret;
}
ret = parsePrograms(Slice(baseAddr, fileLen), programs, size, mapPatterns);
munmap(baseAddr, fileLen);
return ret;
}
} // namespace bpf
} // namespace android

284
libbpf/include/bpf/BpfMap.h Normal file
View file

@ -0,0 +1,284 @@
/*
* Copyright (C) 2018 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef BPF_BPFMAP_H
#define BPF_BPFMAP_H
#include <linux/bpf.h>
#include <android-base/stringprintf.h>
#include <android-base/unique_fd.h>
#include <utils/Log.h>
#include "bpf/BpfUtils.h"
#include "netdutils/Status.h"
#include "netdutils/StatusOr.h"
namespace android {
namespace bpf {
// This is a class wrapper for eBPF maps. The eBPF map is a special in-kernel
// data structure that stores data in <Key, Value> pairs. It can be read/write
// from userspace by passing syscalls with the map file descriptor. This class
// is used to generalize the procedure of interacting with eBPF maps and hide
// the implementation detail from other process. Besides the basic syscalls
// wrapper, it also provides some useful helper functions as well as an iterator
// nested class to iterate the map more easily.
//
// NOTE: A kernel eBPF map may be accessed by both kernel and userspace
// processes at the same time. Or if the map is pinned as a virtual file, it can
// be obtained by multiple eBPF map class object and accessed concurrently.
// Though the map class object and the underlying kernel map are thread safe, it
// is not safe to iterate over a map while another thread or process is deleting
// from it. In this case the iteration can return duplicate entries.
template <class Key, class Value>
class BpfMap {
public:
BpfMap<Key, Value>() : mMapFd(-1){};
BpfMap<Key, Value>(int fd) : mMapFd(fd){};
BpfMap<Key, Value>(bpf_map_type map_type, uint32_t max_entries, uint32_t map_flags) {
int map_fd = createMap(map_type, sizeof(Key), sizeof(Value), max_entries, map_flags);
if (map_fd < 0) {
mMapFd.reset(-1);
} else {
mMapFd.reset(map_fd);
}
}
netdutils::Status pinToPath(const std::string path) {
int ret = bpfFdPin(mMapFd, path.c_str());
if (ret) {
return netdutils::statusFromErrno(errno,
base::StringPrintf("pin to %s failed", path.c_str()));
}
mPinnedPath = path;
return netdutils::status::ok;
}
netdutils::StatusOr<Key> getFirstKey() const {
Key firstKey;
if (getFirstMapKey(mMapFd, &firstKey)) {
return netdutils::statusFromErrno(
errno, base::StringPrintf("Get firstKey map %d failed", mMapFd.get()));
}
return firstKey;
}
netdutils::StatusOr<Key> getNextKey(const Key& key) const {
Key nextKey;
if (getNextMapKey(mMapFd, const_cast<Key*>(&key), &nextKey)) {
return netdutils::statusFromErrno(
errno, base::StringPrintf("Get next key of map %d failed", mMapFd.get()));
}
return nextKey;
}
netdutils::Status writeValue(const Key& key, const Value& value, uint64_t flags) {
if (writeToMapEntry(mMapFd, const_cast<Key*>(&key), const_cast<Value*>(&value), flags)) {
return netdutils::statusFromErrno(
errno, base::StringPrintf("write to map %d failed", mMapFd.get()));
}
return netdutils::status::ok;
}
netdutils::StatusOr<Value> readValue(const Key key) const {
Value value;
if (findMapEntry(mMapFd, const_cast<Key*>(&key), &value)) {
return netdutils::statusFromErrno(
errno, base::StringPrintf("read value of map %d failed", mMapFd.get()));
}
return value;
}
netdutils::Status deleteValue(const Key& key) {
if (deleteMapEntry(mMapFd, const_cast<Key*>(&key))) {
return netdutils::statusFromErrno(
errno, base::StringPrintf("delete entry from map %d failed", mMapFd.get()));
}
return netdutils::status::ok;
}
// Function that tries to get map from a pinned path, if the map doesn't
// exist yet, create a new one and pinned to the path.
netdutils::Status getOrCreate(const uint32_t maxEntries, const char* path,
const bpf_map_type mapType);
// Iterate through the map and handle each key retrieved based on the filter
// without modification of map content.
netdutils::Status iterate(
const std::function<netdutils::Status(const Key& key, const BpfMap<Key, Value>& map)>&
filter) const;
// Iterate through the map and get each <key, value> pair, handle each <key,
// value> pair based on the filter without modification of map content.
netdutils::Status iterateWithValue(
const std::function<netdutils::Status(const Key& key, const Value& value,
const BpfMap<Key, Value>& map)>& filter) const;
// Iterate through the map and handle each key retrieved based on the filter
netdutils::Status iterate(
const std::function<netdutils::Status(const Key& key, BpfMap<Key, Value>& map)>& filter);
// Iterate through the map and get each <key, value> pair, handle each <key,
// value> pair based on the filter.
netdutils::Status iterateWithValue(
const std::function<netdutils::Status(const Key& key, const Value& value,
BpfMap<Key, Value>& map)>& filter);
const base::unique_fd& getMap() const { return mMapFd; };
const std::string getPinnedPath() const { return mPinnedPath; };
// Move constructor
void operator=(BpfMap<Key, Value>&& other) noexcept {
mMapFd = std::move(other.mMapFd);
if (!other.mPinnedPath.empty()) {
mPinnedPath = other.mPinnedPath;
} else {
mPinnedPath.clear();
}
other.reset();
}
void reset(int fd = -1) {
mMapFd.reset(fd);
mPinnedPath.clear();
}
bool isValid() const { return mMapFd != -1; }
// It is only safe to call this method if it is guaranteed that nothing will concurrently
// iterate over the map in any process.
netdutils::Status clear() {
const auto deleteAllEntries = [](const Key& key, BpfMap<Key, Value>& map) {
netdutils::Status res = map.deleteValue(key);
if (!isOk(res) && (res.code() != ENOENT)) {
ALOGE("Failed to delete data %s\n", strerror(res.code()));
}
return netdutils::status::ok;
};
RETURN_IF_NOT_OK(iterate(deleteAllEntries));
return netdutils::status::ok;
}
netdutils::StatusOr<bool> isEmpty() const {
auto key = this->getFirstKey();
// Return error code ENOENT means the map is empty
if (!isOk(key) && key.status().code() == ENOENT) return true;
RETURN_IF_NOT_OK(key);
return false;
}
private:
base::unique_fd mMapFd;
std::string mPinnedPath;
};
template <class Key, class Value>
netdutils::Status BpfMap<Key, Value>::getOrCreate(const uint32_t maxEntries, const char* path,
bpf_map_type mapType) {
int ret = access(path, R_OK);
/* Check the pinned location first to check if the map is already there.
* otherwise create a new one.
*/
if (ret == 0) {
mMapFd = base::unique_fd(mapRetrieve(path, 0));
if (mMapFd == -1) {
reset();
return netdutils::statusFromErrno(
errno,
base::StringPrintf("pinned map not accessible or does not exist: (%s)\n", path));
}
mPinnedPath = path;
} else if (ret == -1 && errno == ENOENT) {
mMapFd = base::unique_fd(
createMap(mapType, sizeof(Key), sizeof(Value), maxEntries, BPF_F_NO_PREALLOC));
if (mMapFd == -1) {
reset();
return netdutils::statusFromErrno(errno,
base::StringPrintf("map create failed!: %s", path));
}
netdutils::Status pinStatus = pinToPath(path);
if (!isOk(pinStatus)) {
reset();
return pinStatus;
}
mPinnedPath = path;
} else {
return netdutils::statusFromErrno(
errno, base::StringPrintf("pinned map not accessible: %s", path));
}
return netdutils::status::ok;
}
template <class Key, class Value>
netdutils::Status BpfMap<Key, Value>::iterate(
const std::function<netdutils::Status(const Key& key, const BpfMap<Key, Value>& map)>& filter)
const {
netdutils::StatusOr<Key> curKey = getFirstKey();
while (isOk(curKey)) {
const netdutils::StatusOr<Key>& nextKey = getNextKey(curKey.value());
RETURN_IF_NOT_OK(filter(curKey.value(), *this));
curKey = nextKey;
}
return curKey.status().code() == ENOENT ? netdutils::status::ok : curKey.status();
}
template <class Key, class Value>
netdutils::Status BpfMap<Key, Value>::iterateWithValue(
const std::function<netdutils::Status(const Key& key, const Value& value,
const BpfMap<Key, Value>& map)>& filter) const {
netdutils::StatusOr<Key> curKey = getFirstKey();
while (isOk(curKey)) {
const netdutils::StatusOr<Key>& nextKey = getNextKey(curKey.value());
Value curValue;
ASSIGN_OR_RETURN(curValue, this->readValue(curKey.value()));
RETURN_IF_NOT_OK(filter(curKey.value(), curValue, *this));
curKey = nextKey;
}
return curKey.status().code() == ENOENT ? netdutils::status::ok : curKey.status();
}
template <class Key, class Value>
netdutils::Status BpfMap<Key, Value>::iterate(
const std::function<netdutils::Status(const Key& key, BpfMap<Key, Value>& map)>& filter) {
netdutils::StatusOr<Key> curKey = getFirstKey();
while (isOk(curKey)) {
const netdutils::StatusOr<Key>& nextKey = getNextKey(curKey.value());
RETURN_IF_NOT_OK(filter(curKey.value(), *this));
curKey = nextKey;
}
return curKey.status().code() == ENOENT ? netdutils::status::ok : curKey.status();
}
template <class Key, class Value>
netdutils::Status BpfMap<Key, Value>::iterateWithValue(
const std::function<netdutils::Status(const Key& key, const Value& value,
BpfMap<Key, Value>& map)>& filter) {
netdutils::StatusOr<Key> curKey = getFirstKey();
while (isOk(curKey)) {
const netdutils::StatusOr<Key>& nextKey = getNextKey(curKey.value());
Value curValue;
ASSIGN_OR_RETURN(curValue, this->readValue(curKey.value()));
RETURN_IF_NOT_OK(filter(curKey.value(), curValue, *this));
curKey = nextKey;
}
return curKey.status().code() == ENOENT ? netdutils::status::ok : curKey.status();
}
} // namespace bpf
} // namespace android
#endif

View file

@ -0,0 +1,168 @@
/*
* Copyright (C) 2017 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef BPF_BPFUTILS_H
#define BPF_BPFUTILS_H
#include <linux/bpf.h>
#include <linux/if_ether.h>
#include <linux/in.h>
#include <linux/unistd.h>
#include <net/if.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include "android-base/unique_fd.h"
#include "netdutils/Slice.h"
#include "netdutils/StatusOr.h"
#define BPF_PASS 1
#define BPF_DROP 0
#define ptr_to_u64(x) ((uint64_t)(uintptr_t)(x))
#define DEFAULT_LOG_LEVEL 1
#define MAP_LD_CMD_HEAD 0x18
#define ARRAY_SIZE(a) (sizeof(a) / sizeof(*(a)))
// The BPF instruction bytes that we need to replace. x is a placeholder (e.g., COOKIE_TAG_MAP).
#define BPF_MAP_SEARCH_PATTERN(x) \
{ \
0x18, 0x01, 0x00, 0x00, \
(x)[0], (x)[1], (x)[2], (x)[3], \
0x00, 0x00, 0x00, 0x00, \
(x)[4], (x)[5], (x)[6], (x)[7] \
}
// The bytes we'll replace them with. x is the actual fd number for the map at runtime.
// The second byte is changed from 0x01 to 0x11 since 0x11 is the special command used
// for bpf map fd loading. The original 0x01 is only a normal load command.
#define BPF_MAP_REPLACE_PATTERN(x) \
{ \
0x18, 0x11, 0x00, 0x00, \
(x)[0], (x)[1], (x)[2], (x)[3], \
0x00, 0x00, 0x00, 0x00, \
(x)[4], (x)[5], (x)[6], (x)[7] \
}
#define MAP_CMD_SIZE 16
namespace android {
namespace bpf {
struct UidTag {
uint32_t uid;
uint32_t tag;
};
struct StatsKey {
uint32_t uid;
uint32_t tag;
uint32_t counterSet;
uint32_t ifaceIndex;
};
struct StatsValue {
uint64_t rxPackets;
uint64_t rxBytes;
uint64_t txPackets;
uint64_t txBytes;
};
struct Stats {
uint64_t rxBytes;
uint64_t rxPackets;
uint64_t txBytes;
uint64_t txPackets;
uint64_t tcpRxPackets;
uint64_t tcpTxPackets;
};
struct IfaceValue {
char name[IFNAMSIZ];
};
struct BpfProgInfo {
bpf_attach_type attachType;
const char* path;
const char* name;
bpf_prog_type loadType;
base::unique_fd fd;
};
int mapRetrieve(const char* pathname, uint32_t flags);
struct BpfMapInfo {
std::array<uint8_t, MAP_CMD_SIZE> search;
std::array<uint8_t, MAP_CMD_SIZE> replace;
const int fd;
std::string path;
BpfMapInfo(uint64_t dummyFd, const char* mapPath)
: BpfMapInfo(dummyFd, android::bpf::mapRetrieve(mapPath, 0)) {}
BpfMapInfo(uint64_t dummyFd, int realFd, const char* mapPath = "") : fd(realFd), path(mapPath) {
search = BPF_MAP_SEARCH_PATTERN((uint8_t*) &dummyFd);
replace = BPF_MAP_REPLACE_PATTERN((uint8_t*) &realFd);
}
};
#ifndef DEFAULT_OVERFLOWUID
#define DEFAULT_OVERFLOWUID 65534
#endif
constexpr const char* CGROUP_ROOT_PATH = "/dev/cg2_bpf";
constexpr const int OVERFLOW_COUNTERSET = 2;
constexpr const uint64_t NONEXISTENT_COOKIE = 0;
constexpr const int MINIMUM_API_REQUIRED = 28;
int createMap(bpf_map_type map_type, uint32_t key_size, uint32_t value_size,
uint32_t max_entries, uint32_t map_flags);
int writeToMapEntry(const base::unique_fd& map_fd, void* key, void* value, uint64_t flags);
int findMapEntry(const base::unique_fd& map_fd, void* key, void* value);
int deleteMapEntry(const base::unique_fd& map_fd, void* key);
int getNextMapKey(const base::unique_fd& map_fd, void* key, void* next_key);
int getFirstMapKey(const base::unique_fd& map_fd, void* firstKey);
int bpfProgLoad(bpf_prog_type prog_type, netdutils::Slice bpf_insns, const char* license,
uint32_t kern_version, netdutils::Slice bpf_log);
int bpfFdPin(const base::unique_fd& map_fd, const char* pathname);
int attachProgram(bpf_attach_type type, uint32_t prog_fd, uint32_t cg_fd);
int detachProgram(bpf_attach_type type, uint32_t cg_fd);
uint64_t getSocketCookie(int sockFd);
bool hasBpfSupport();
int parseProgramsFromFile(const char* path, BpfProgInfo* programs, size_t size,
const std::vector<BpfMapInfo>& mapPatterns);
#define SKIP_IF_BPF_NOT_SUPPORTED \
do { \
if (!hasBpfSupport()) return; \
} while (0)
constexpr int BPF_CONTINUE = 0;
constexpr int BPF_DELETED = 1;
bool operator==(const StatsValue& lhs, const StatsValue& rhs);
bool operator==(const UidTag& lhs, const UidTag& rhs);
bool operator==(const StatsKey& lhs, const StatsKey& rhs);
} // namespace bpf
} // namespace android
#endif