blob: 8e47ea87a14716959b40057aa3bd1602a325b66e [file] [log] [blame]
Maciej Żenczykowski60c159f2023-10-02 14:54:48 -07001/*
Maciej Żenczykowski283c25a2023-10-02 19:43:30 -07002 * Copyright (C) 2017-2023 The Android Open Source Project
Maciej Żenczykowski60c159f2023-10-02 14:54:48 -07003 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#ifndef LOG_TAG
Maciej Żenczykowski283c25a2023-10-02 19:43:30 -070018#define LOG_TAG "NetBpfLoad"
Maciej Żenczykowski60c159f2023-10-02 14:54:48 -070019#endif
20
21#include <arpa/inet.h>
22#include <dirent.h>
23#include <elf.h>
24#include <error.h>
25#include <fcntl.h>
26#include <inttypes.h>
27#include <linux/bpf.h>
28#include <linux/unistd.h>
29#include <net/if.h>
30#include <stdint.h>
31#include <stdio.h>
32#include <stdlib.h>
33#include <string.h>
34#include <unistd.h>
35
36#include <sys/mman.h>
37#include <sys/socket.h>
38#include <sys/stat.h>
39#include <sys/types.h>
40
41#include <android-base/logging.h>
42#include <android-base/macros.h>
43#include <android-base/properties.h>
44#include <android-base/stringprintf.h>
45#include <android-base/strings.h>
46#include <android-base/unique_fd.h>
47#include <libbpf_android.h>
48#include <log/log.h>
49#include <netdutils/Misc.h>
50#include <netdutils/Slice.h>
51#include "BpfSyscallWrappers.h"
52#include "bpf/BpfUtils.h"
53
54using android::base::EndsWith;
55using android::bpf::domain;
56using std::string;
57
58bool exists(const char* const path) {
59 int v = access(path, F_OK);
60 if (!v) {
61 ALOGI("%s exists.", path);
62 return true;
63 }
64 if (errno == ENOENT) return false;
65 ALOGE("FATAL: access(%s, F_OK) -> %d [%d:%s]", path, v, errno, strerror(errno));
66 abort(); // can only hit this if permissions (likely selinux) are screwed up
67}
68
69constexpr unsigned long long kTetheringApexDomainBitmask =
70 domainToBitmask(domain::tethering) |
71 domainToBitmask(domain::net_private) |
72 domainToBitmask(domain::net_shared) |
73 domainToBitmask(domain::netd_readonly) |
74 domainToBitmask(domain::netd_shared);
75
76// Programs shipped inside the tethering apex should be limited to networking stuff,
77// as KPROBE, PERF_EVENT, TRACEPOINT are dangerous to use from mainline updatable code,
78// since they are less stable abi/api and may conflict with platform uses of bpf.
79constexpr bpf_prog_type kTetheringApexAllowedProgTypes[] = {
80 BPF_PROG_TYPE_CGROUP_SKB,
81 BPF_PROG_TYPE_CGROUP_SOCK,
82 BPF_PROG_TYPE_CGROUP_SOCKOPT,
83 BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
84 BPF_PROG_TYPE_CGROUP_SYSCTL,
85 BPF_PROG_TYPE_LWT_IN,
86 BPF_PROG_TYPE_LWT_OUT,
87 BPF_PROG_TYPE_LWT_SEG6LOCAL,
88 BPF_PROG_TYPE_LWT_XMIT,
89 BPF_PROG_TYPE_SCHED_ACT,
90 BPF_PROG_TYPE_SCHED_CLS,
91 BPF_PROG_TYPE_SOCKET_FILTER,
92 BPF_PROG_TYPE_SOCK_OPS,
93 BPF_PROG_TYPE_XDP,
94};
95
Maciej Żenczykowski60c159f2023-10-02 14:54:48 -070096
97const android::bpf::Location locations[] = {
98 // S+ Tethering mainline module (network_stack): tether offload
99 {
100 .dir = "/apex/com.android.tethering/etc/bpf/",
101 .prefix = "tethering/",
102 .allowedDomainBitmask = kTetheringApexDomainBitmask,
103 .allowedProgTypes = kTetheringApexAllowedProgTypes,
104 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
105 },
106 // T+ Tethering mainline module (shared with netd & system server)
107 // netutils_wrapper (for iptables xt_bpf) has access to programs
108 {
109 .dir = "/apex/com.android.tethering/etc/bpf/netd_shared/",
110 .prefix = "netd_shared/",
111 .allowedDomainBitmask = kTetheringApexDomainBitmask,
112 .allowedProgTypes = kTetheringApexAllowedProgTypes,
113 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
114 },
115 // T+ Tethering mainline module (shared with netd & system server)
116 // netutils_wrapper has no access, netd has read only access
117 {
118 .dir = "/apex/com.android.tethering/etc/bpf/netd_readonly/",
119 .prefix = "netd_readonly/",
120 .allowedDomainBitmask = kTetheringApexDomainBitmask,
121 .allowedProgTypes = kTetheringApexAllowedProgTypes,
122 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
123 },
124 // T+ Tethering mainline module (shared with system server)
125 {
126 .dir = "/apex/com.android.tethering/etc/bpf/net_shared/",
127 .prefix = "net_shared/",
128 .allowedDomainBitmask = kTetheringApexDomainBitmask,
129 .allowedProgTypes = kTetheringApexAllowedProgTypes,
130 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
131 },
132 // T+ Tethering mainline module (not shared, just network_stack)
133 {
134 .dir = "/apex/com.android.tethering/etc/bpf/net_private/",
135 .prefix = "net_private/",
136 .allowedDomainBitmask = kTetheringApexDomainBitmask,
137 .allowedProgTypes = kTetheringApexAllowedProgTypes,
138 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
139 },
Maciej Żenczykowski60c159f2023-10-02 14:54:48 -0700140};
141
142int loadAllElfObjects(const android::bpf::Location& location) {
143 int retVal = 0;
144 DIR* dir;
145 struct dirent* ent;
146
147 if ((dir = opendir(location.dir)) != NULL) {
148 while ((ent = readdir(dir)) != NULL) {
149 string s = ent->d_name;
150 if (!EndsWith(s, ".o")) continue;
151
152 string progPath(location.dir);
153 progPath += s;
154
155 bool critical;
156 int ret = android::bpf::loadProg(progPath.c_str(), &critical, location);
157 if (ret) {
158 if (critical) retVal = ret;
159 ALOGE("Failed to load object: %s, ret: %s", progPath.c_str(), std::strerror(-ret));
160 } else {
161 ALOGI("Loaded object: %s", progPath.c_str());
162 }
163 }
164 closedir(dir);
165 }
166 return retVal;
167}
168
169int createSysFsBpfSubDir(const char* const prefix) {
170 if (*prefix) {
171 mode_t prevUmask = umask(0);
172
173 string s = "/sys/fs/bpf/";
174 s += prefix;
175
176 errno = 0;
177 int ret = mkdir(s.c_str(), S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO);
178 if (ret && errno != EEXIST) {
179 const int err = errno;
180 ALOGE("Failed to create directory: %s, ret: %s", s.c_str(), std::strerror(err));
181 return -err;
182 }
183
184 umask(prevUmask);
185 }
186 return 0;
187}
188
189// Technically 'value' doesn't need to be newline terminated, but it's best
190// to include a newline to match 'echo "value" > /proc/sys/...foo' behaviour,
191// which is usually how kernel devs test the actual sysctl interfaces.
192int writeProcSysFile(const char *filename, const char *value) {
193 android::base::unique_fd fd(open(filename, O_WRONLY | O_CLOEXEC));
194 if (fd < 0) {
195 const int err = errno;
196 ALOGE("open('%s', O_WRONLY | O_CLOEXEC) -> %s", filename, strerror(err));
197 return -err;
198 }
199 int len = strlen(value);
200 int v = write(fd, value, len);
201 if (v < 0) {
202 const int err = errno;
203 ALOGE("write('%s', '%s', %d) -> %s", filename, value, len, strerror(err));
204 return -err;
205 }
206 if (v != len) {
207 // In practice, due to us only using this for /proc/sys/... files, this can't happen.
208 ALOGE("write('%s', '%s', %d) -> short write [%d]", filename, value, len, v);
209 return -EINVAL;
210 }
211 return 0;
212}
213
214int main(int argc, char** argv) {
215 (void)argc;
216 android::base::InitLogging(argv, &android::base::KernelLogger);
217
218 if (!android::bpf::isAtLeastKernelVersion(4, 19, 0)) {
219 ALOGE("Android U QPR2 requires kernel 4.19.");
220 return 1;
221 }
222
223 if (android::bpf::isUserspace32bit() && android::bpf::isAtLeastKernelVersion(6, 2, 0)) {
224 /* Android 14/U should only launch on 64-bit kernels
225 * T launches on 5.10/5.15
226 * U launches on 5.15/6.1
227 * So >=5.16 implies isKernel64Bit()
228 *
229 * We thus added a test to V VTS which requires 5.16+ devices to use 64-bit kernels.
230 *
231 * Starting with Android V, which is the first to support a post 6.1 Linux Kernel,
232 * we also require 64-bit userspace.
233 *
234 * There are various known issues with 32-bit userspace talking to various
235 * kernel interfaces (especially CAP_NET_ADMIN ones) on a 64-bit kernel.
236 * Some of these have userspace or kernel workarounds/hacks.
237 * Some of them don't...
238 * We're going to be removing the hacks.
239 *
240 * Additionally the 32-bit kernel jit support is poor,
241 * and 32-bit userspace on 64-bit kernel bpf ringbuffer compatibility is broken.
242 */
243 ALOGE("64-bit userspace required on 6.2+ kernels.");
244 return 1;
245 }
246
247 // Ensure we can determine the Android build type.
248 if (!android::bpf::isEng() && !android::bpf::isUser() && !android::bpf::isUserdebug()) {
249 ALOGE("Failed to determine the build type: got %s, want 'eng', 'user', or 'userdebug'",
250 android::bpf::getBuildType().c_str());
251 return 1;
252 }
253
254 // Linux 5.16-rc1 changed the default to 2 (disabled but changeable), but we need 0 (enabled)
255 // (this writeFile is known to fail on at least 4.19, but always defaults to 0 on pre-5.13,
256 // on 5.13+ it depends on CONFIG_BPF_UNPRIV_DEFAULT_OFF)
257 if (writeProcSysFile("/proc/sys/kernel/unprivileged_bpf_disabled", "0\n") &&
258 android::bpf::isAtLeastKernelVersion(5, 13, 0)) return 1;
259
260 // Enable the eBPF JIT -- but do note that on 64-bit kernels it is likely
261 // already force enabled by the kernel config option BPF_JIT_ALWAYS_ON.
262 // (Note: this (open) will fail with ENOENT 'No such file or directory' if
263 // kernel does not have CONFIG_BPF_JIT=y)
264 // BPF_JIT is required by R VINTF (which means 4.14/4.19/5.4 kernels),
265 // but 4.14/4.19 were released with P & Q, and only 5.4 is new in R+.
266 if (writeProcSysFile("/proc/sys/net/core/bpf_jit_enable", "1\n")) return 1;
267
268 // Enable JIT kallsyms export for privileged users only
269 // (Note: this (open) will fail with ENOENT 'No such file or directory' if
270 // kernel does not have CONFIG_HAVE_EBPF_JIT=y)
271 if (writeProcSysFile("/proc/sys/net/core/bpf_jit_kallsyms", "1\n")) return 1;
272
273 // Create all the pin subdirectories
274 // (this must be done first to allow selinux_context and pin_subdir functionality,
275 // which could otherwise fail with ENOENT during object pinning or renaming,
276 // due to ordering issues)
277 for (const auto& location : locations) {
278 if (createSysFsBpfSubDir(location.prefix)) return 1;
279 }
280
281 // Note: there's no actual src dir for fs_bpf_loader .o's,
282 // so it is not listed in 'locations[].prefix'.
283 // This is because this is primarily meant for triggering genfscon rules,
284 // and as such this will likely always be the case.
285 // Thus we need to manually create the /sys/fs/bpf/loader subdirectory.
286 if (createSysFsBpfSubDir("loader")) return 1;
287
288 // Load all ELF objects, create programs and maps, and pin them
289 for (const auto& location : locations) {
290 if (loadAllElfObjects(location) != 0) {
291 ALOGE("=== CRITICAL FAILURE LOADING BPF PROGRAMS FROM %s ===", location.dir);
292 ALOGE("If this triggers reliably, you're probably missing kernel options or patches.");
293 ALOGE("If this triggers randomly, you might be hitting some memory allocation "
294 "problems or startup script race.");
295 ALOGE("--- DO NOT EXPECT SYSTEM TO BOOT SUCCESSFULLY ---");
296 sleep(20);
297 return 2;
298 }
299 }
300
301 int key = 1;
302 int value = 123;
303 android::base::unique_fd map(
304 android::bpf::createMap(BPF_MAP_TYPE_ARRAY, sizeof(key), sizeof(value), 2, 0));
305 if (android::bpf::writeToMapEntry(map, &key, &value, BPF_ANY)) {
306 ALOGE("Critical kernel bug - failure to write into index 1 of 2 element bpf map array.");
307 return 1;
308 }
309
310 if (android::base::SetProperty("bpf.progs_loaded", "1") == false) {
311 ALOGE("Failed to set bpf.progs_loaded property");
312 return 1;
313 }
314
315 return 0;
316}