• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2017 The Android Open Source Project
3  *
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
18 #define LOG_TAG "bpfloader"
19 #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 
54 using android::base::EndsWith;
55 using android::bpf::domain;
56 using std::string;
57 
exists(const char * const path)58 bool 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 
69 constexpr 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.
79 constexpr 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 
96 // Networking-related program types are limited to the Tethering Apex
97 // to prevent things from breaking due to conflicts on mainline updates
98 // (exception made for socket filters, ie. xt_bpf for potential use in iptables,
99 // or for attaching to sockets directly)
100 constexpr bpf_prog_type kPlatformAllowedProgTypes[] = {
101         BPF_PROG_TYPE_KPROBE,
102         BPF_PROG_TYPE_PERF_EVENT,
103         BPF_PROG_TYPE_SOCKET_FILTER,
104         BPF_PROG_TYPE_TRACEPOINT,
105         BPF_PROG_TYPE_UNSPEC,  // Will be replaced with fuse bpf program type
106 };
107 
108 // see b/162057235. For arbitrary program types, the concern is that due to the lack of
109 // SELinux access controls over BPF program attachpoints, we have no way to control the
110 // attachment of programs to shared resources (or to detect when a shared resource
111 // has one BPF program replace another that is attached there)
112 constexpr bpf_prog_type kVendorAllowedProgTypes[] = {
113         BPF_PROG_TYPE_SOCKET_FILTER,
114 };
115 
116 
117 const android::bpf::Location locations[] = {
118         // S+ Tethering mainline module (network_stack): tether offload
119         {
120                 .dir = "/apex/com.android.tethering/etc/bpf/",
121                 .prefix = "tethering/",
122                 .allowedDomainBitmask = kTetheringApexDomainBitmask,
123                 .allowedProgTypes = kTetheringApexAllowedProgTypes,
124                 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
125         },
126         // T+ Tethering mainline module (shared with netd & system server)
127         // netutils_wrapper (for iptables xt_bpf) has access to programs
128         {
129                 .dir = "/apex/com.android.tethering/etc/bpf/netd_shared/",
130                 .prefix = "netd_shared/",
131                 .allowedDomainBitmask = kTetheringApexDomainBitmask,
132                 .allowedProgTypes = kTetheringApexAllowedProgTypes,
133                 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
134         },
135         // T+ Tethering mainline module (shared with netd & system server)
136         // netutils_wrapper has no access, netd has read only access
137         {
138                 .dir = "/apex/com.android.tethering/etc/bpf/netd_readonly/",
139                 .prefix = "netd_readonly/",
140                 .allowedDomainBitmask = kTetheringApexDomainBitmask,
141                 .allowedProgTypes = kTetheringApexAllowedProgTypes,
142                 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
143         },
144         // T+ Tethering mainline module (shared with system server)
145         {
146                 .dir = "/apex/com.android.tethering/etc/bpf/net_shared/",
147                 .prefix = "net_shared/",
148                 .allowedDomainBitmask = kTetheringApexDomainBitmask,
149                 .allowedProgTypes = kTetheringApexAllowedProgTypes,
150                 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
151         },
152         // T+ Tethering mainline module (not shared, just network_stack)
153         {
154                 .dir = "/apex/com.android.tethering/etc/bpf/net_private/",
155                 .prefix = "net_private/",
156                 .allowedDomainBitmask = kTetheringApexDomainBitmask,
157                 .allowedProgTypes = kTetheringApexAllowedProgTypes,
158                 .allowedProgTypesLength = arraysize(kTetheringApexAllowedProgTypes),
159         },
160         // Core operating system
161         {
162                 .dir = "/system/etc/bpf/",
163                 .prefix = "",
164                 .allowedDomainBitmask = domainToBitmask(domain::platform),
165                 .allowedProgTypes = kPlatformAllowedProgTypes,
166                 .allowedProgTypesLength = arraysize(kPlatformAllowedProgTypes),
167         },
168         // Vendor operating system
169         {
170                 .dir = "/vendor/etc/bpf/",
171                 .prefix = "vendor/",
172                 .allowedDomainBitmask = domainToBitmask(domain::vendor),
173                 .allowedProgTypes = kVendorAllowedProgTypes,
174                 .allowedProgTypesLength = arraysize(kVendorAllowedProgTypes),
175         },
176 };
177 
loadAllElfObjects(const android::bpf::Location & location)178 int loadAllElfObjects(const android::bpf::Location& location) {
179     int retVal = 0;
180     DIR* dir;
181     struct dirent* ent;
182 
183     if ((dir = opendir(location.dir)) != NULL) {
184         while ((ent = readdir(dir)) != NULL) {
185             string s = ent->d_name;
186             if (!EndsWith(s, ".o")) continue;
187 
188             string progPath(location.dir);
189             progPath += s;
190 
191             bool critical;
192             int ret = android::bpf::loadProg(progPath.c_str(), &critical, location);
193             if (ret) {
194                 if (critical) retVal = ret;
195                 ALOGE("Failed to load object: %s, ret: %s", progPath.c_str(), std::strerror(-ret));
196             } else {
197                 ALOGI("Loaded object: %s", progPath.c_str());
198             }
199         }
200         closedir(dir);
201     }
202     return retVal;
203 }
204 
createSysFsBpfSubDir(const char * const prefix)205 int createSysFsBpfSubDir(const char* const prefix) {
206     if (*prefix) {
207         mode_t prevUmask = umask(0);
208 
209         string s = "/sys/fs/bpf/";
210         s += prefix;
211 
212         errno = 0;
213         int ret = mkdir(s.c_str(), S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO);
214         if (ret && errno != EEXIST) {
215             const int err = errno;
216             ALOGE("Failed to create directory: %s, ret: %s", s.c_str(), std::strerror(err));
217             return -err;
218         }
219 
220         umask(prevUmask);
221     }
222     return 0;
223 }
224 
225 // Technically 'value' doesn't need to be newline terminated, but it's best
226 // to include a newline to match 'echo "value" > /proc/sys/...foo' behaviour,
227 // which is usually how kernel devs test the actual sysctl interfaces.
writeProcSysFile(const char * filename,const char * value)228 int writeProcSysFile(const char *filename, const char *value) {
229     android::base::unique_fd fd(open(filename, O_WRONLY | O_CLOEXEC));
230     if (fd < 0) {
231         const int err = errno;
232         ALOGE("open('%s', O_WRONLY | O_CLOEXEC) -> %s", filename, strerror(err));
233         return -err;
234     }
235     int len = strlen(value);
236     int v = write(fd, value, len);
237     if (v < 0) {
238         const int err = errno;
239         ALOGE("write('%s', '%s', %d) -> %s", filename, value, len, strerror(err));
240         return -err;
241     }
242     if (v != len) {
243         // In practice, due to us only using this for /proc/sys/... files, this can't happen.
244         ALOGE("write('%s', '%s', %d) -> short write [%d]", filename, value, len, v);
245         return -EINVAL;
246     }
247     return 0;
248 }
249 
main(int argc,char ** argv)250 int main(int argc, char** argv) {
251     (void)argc;
252     android::base::InitLogging(argv, &android::base::KernelLogger);
253 
254     // Ensure we can determine the Android build type.
255     if (!android::bpf::isEng() && !android::bpf::isUser() && !android::bpf::isUserdebug()) {
256         ALOGE("Failed to determine the build type: got %s, want 'eng', 'user', or 'userdebug'",
257               android::bpf::getBuildType().c_str());
258         return 1;
259     }
260 
261     // Linux 5.16-rc1 changed the default to 2 (disabled but changeable), but we need 0 (enabled)
262     // (this writeFile is known to fail on at least 4.19, but always defaults to 0 on pre-5.13,
263     // on 5.13+ it depends on CONFIG_BPF_UNPRIV_DEFAULT_OFF)
264     if (writeProcSysFile("/proc/sys/kernel/unprivileged_bpf_disabled", "0\n") &&
265         android::bpf::isAtLeastKernelVersion(5, 13, 0)) return 1;
266 
267     // Enable the eBPF JIT -- but do note that on 64-bit kernels it is likely
268     // already force enabled by the kernel config option BPF_JIT_ALWAYS_ON.
269     // (Note: this (open) will fail with ENOENT 'No such file or directory' if
270     //  kernel does not have CONFIG_BPF_JIT=y)
271     // BPF_JIT is required by R VINTF (which means 4.14/4.19/5.4 kernels),
272     // but 4.14/4.19 were released with P & Q, and only 5.4 is new in R+.
273     if (writeProcSysFile("/proc/sys/net/core/bpf_jit_enable", "1\n") &&
274         android::bpf::isAtLeastKernelVersion(5, 4, 0)) return 1;
275 
276     // Enable JIT kallsyms export for privileged users only
277     // (Note: this (open) will fail with ENOENT 'No such file or directory' if
278     //  kernel does not have CONFIG_HAVE_EBPF_JIT=y)
279     if (writeProcSysFile("/proc/sys/net/core/bpf_jit_kallsyms", "1\n") &&
280         android::bpf::isAtLeastKernelVersion(5, 4, 0)) return 1;
281 
282     // Create all the pin subdirectories
283     // (this must be done first to allow selinux_context and pin_subdir functionality,
284     //  which could otherwise fail with ENOENT during object pinning or renaming,
285     //  due to ordering issues)
286     for (const auto& location : locations) {
287         if (createSysFsBpfSubDir(location.prefix)) return 1;
288     }
289 
290     // Note: there's no actual src dir for fs_bpf_loader .o's,
291     // so it is not listed in 'locations[].prefix'.
292     // This is because this is primarily meant for triggering genfscon rules,
293     // and as such this will likely always be the case.
294     // Thus we need to manually create the /sys/fs/bpf/loader subdirectory.
295     if (createSysFsBpfSubDir("loader")) return 1;
296 
297     // Load all ELF objects, create programs and maps, and pin them
298     for (const auto& location : locations) {
299         if (loadAllElfObjects(location) != 0) {
300             ALOGE("=== CRITICAL FAILURE LOADING BPF PROGRAMS FROM %s ===", location.dir);
301             ALOGE("If this triggers reliably, you're probably missing kernel options or patches.");
302             ALOGE("If this triggers randomly, you might be hitting some memory allocation "
303                   "problems or startup script race.");
304             ALOGE("--- DO NOT EXPECT SYSTEM TO BOOT SUCCESSFULLY ---");
305             sleep(20);
306             return 2;
307         }
308     }
309 
310     int key = 1;
311     int value = 123;
312     android::base::unique_fd map(
313             android::bpf::createMap(BPF_MAP_TYPE_ARRAY, sizeof(key), sizeof(value), 2, 0));
314     if (android::bpf::writeToMapEntry(map, &key, &value, BPF_ANY)) {
315         ALOGE("Critical kernel bug - failure to write into index 1 of 2 element bpf map array.");
316         return 1;
317     }
318 
319     if (android::base::SetProperty("bpf.progs_loaded", "1") == false) {
320         ALOGE("Failed to set bpf.progs_loaded property");
321         return 1;
322     }
323 
324     return 0;
325 }
326