Aurimas Liutikas | 88c7ff1 | 2023-08-10 12:42:26 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2013 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 | package com.android.server.am; |
| 18 | |
| 19 | import static android.system.OsConstants.AF_UNIX; |
| 20 | import static android.system.OsConstants.SOCK_STREAM; |
| 21 | import static android.system.OsConstants.SOL_SOCKET; |
| 22 | import static android.system.OsConstants.SO_RCVTIMEO; |
| 23 | import static android.system.OsConstants.SO_SNDTIMEO; |
| 24 | |
| 25 | import android.app.ApplicationErrorReport.CrashInfo; |
| 26 | import android.system.ErrnoException; |
| 27 | import android.system.Os; |
| 28 | import android.system.StructTimeval; |
| 29 | import android.system.UnixSocketAddress; |
| 30 | import android.util.Slog; |
| 31 | |
| 32 | import java.io.ByteArrayOutputStream; |
| 33 | import java.io.File; |
| 34 | import java.io.FileDescriptor; |
| 35 | import java.io.InterruptedIOException; |
| 36 | |
| 37 | /** |
| 38 | * Set up a Unix domain socket that debuggerd will connect() to in |
| 39 | * order to write a description of a native crash. The crash info is |
| 40 | * then parsed and forwarded to the ActivityManagerService's normal |
| 41 | * crash handling code. |
| 42 | * |
| 43 | * Note that this component runs in a separate thread. |
| 44 | */ |
| 45 | final class NativeCrashListener extends Thread { |
| 46 | static final String TAG = "NativeCrashListener"; |
| 47 | static final boolean DEBUG = false; |
| 48 | static final boolean MORE_DEBUG = DEBUG && false; |
| 49 | |
| 50 | // Must match the path defined in debuggerd.c. |
| 51 | static final String DEBUGGERD_SOCKET_PATH = "/data/system/ndebugsocket"; |
| 52 | |
| 53 | // Use a short timeout on socket operations and abandon the connection |
| 54 | // on hard errors, just in case debuggerd goes out to lunch. |
| 55 | static final long SOCKET_TIMEOUT_MILLIS = 10000; // 10 seconds |
| 56 | |
| 57 | final ActivityManagerService mAm; |
| 58 | |
| 59 | /* |
| 60 | * Spin the actual work of handling a debuggerd crash report into a |
| 61 | * separate thread so that the listener can go immediately back to |
| 62 | * accepting incoming connections. |
| 63 | */ |
| 64 | class NativeCrashReporter extends Thread { |
| 65 | ProcessRecord mApp; |
| 66 | int mSignal; |
| 67 | boolean mGwpAsanRecoverableCrash; |
| 68 | String mCrashReport; |
| 69 | |
| 70 | NativeCrashReporter(ProcessRecord app, int signal, boolean gwpAsanRecoverableCrash, |
| 71 | String report) { |
| 72 | super("NativeCrashReport"); |
| 73 | mApp = app; |
| 74 | mSignal = signal; |
| 75 | mGwpAsanRecoverableCrash = gwpAsanRecoverableCrash; |
| 76 | mCrashReport = report; |
| 77 | } |
| 78 | |
| 79 | @Override |
| 80 | public void run() { |
| 81 | try { |
| 82 | CrashInfo ci = new CrashInfo(); |
| 83 | ci.exceptionClassName = "Native crash"; |
| 84 | ci.exceptionMessage = Os.strsignal(mSignal); |
| 85 | ci.throwFileName = "unknown"; |
| 86 | ci.throwClassName = "unknown"; |
| 87 | ci.throwMethodName = "unknown"; |
| 88 | ci.stackTrace = mCrashReport; |
| 89 | |
| 90 | if (DEBUG) Slog.v(TAG, "Calling handleApplicationCrash()"); |
| 91 | mAm.handleApplicationCrashInner( |
| 92 | mGwpAsanRecoverableCrash ? "native_recoverable_crash" : "native_crash", |
| 93 | mApp, mApp.processName, ci); |
| 94 | if (DEBUG) Slog.v(TAG, "<-- handleApplicationCrash() returned"); |
| 95 | } catch (Exception e) { |
| 96 | Slog.e(TAG, "Unable to report native crash", e); |
| 97 | } |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | /* |
| 102 | * Daemon thread that accept()s incoming domain socket connections from debuggerd |
| 103 | * and processes the crash dump that is passed through. |
| 104 | */ |
| 105 | NativeCrashListener(ActivityManagerService am) { |
| 106 | mAm = am; |
| 107 | } |
| 108 | |
| 109 | @Override |
| 110 | public void run() { |
| 111 | final byte[] ackSignal = new byte[1]; |
| 112 | |
| 113 | if (DEBUG) Slog.i(TAG, "Starting up"); |
| 114 | |
| 115 | // The file system entity for this socket is created with 0777 perms, owned |
| 116 | // by system:system. selinux restricts things so that only crash_dump can |
| 117 | // access it. |
| 118 | { |
| 119 | File socketFile = new File(DEBUGGERD_SOCKET_PATH); |
| 120 | if (socketFile.exists()) { |
| 121 | socketFile.delete(); |
| 122 | } |
| 123 | } |
| 124 | |
| 125 | try { |
| 126 | FileDescriptor serverFd = Os.socket(AF_UNIX, SOCK_STREAM, 0); |
| 127 | final UnixSocketAddress sockAddr = UnixSocketAddress.createFileSystem( |
| 128 | DEBUGGERD_SOCKET_PATH); |
| 129 | Os.bind(serverFd, sockAddr); |
| 130 | Os.listen(serverFd, 1); |
| 131 | Os.chmod(DEBUGGERD_SOCKET_PATH, 0777); |
| 132 | |
| 133 | while (true) { |
| 134 | FileDescriptor peerFd = null; |
| 135 | try { |
| 136 | if (MORE_DEBUG) Slog.v(TAG, "Waiting for debuggerd connection"); |
| 137 | peerFd = Os.accept(serverFd, null /* peerAddress */); |
| 138 | if (MORE_DEBUG) Slog.v(TAG, "Got debuggerd socket " + peerFd); |
| 139 | if (peerFd != null) { |
| 140 | // the reporting thread may take responsibility for |
| 141 | // acking the debugger; make sure we play along. |
| 142 | consumeNativeCrashData(peerFd); |
| 143 | } |
| 144 | } catch (Exception e) { |
| 145 | Slog.w(TAG, "Error handling connection", e); |
| 146 | } finally { |
| 147 | // Always ack crash_dump's connection to us. The actual |
| 148 | // byte written is irrelevant. |
| 149 | if (peerFd != null) { |
| 150 | try { |
| 151 | Os.write(peerFd, ackSignal, 0, 1); |
| 152 | } catch (Exception e) { |
| 153 | /* we don't care about failures here */ |
| 154 | if (MORE_DEBUG) { |
| 155 | Slog.d(TAG, "Exception writing ack: " + e.getMessage()); |
| 156 | } |
| 157 | } |
| 158 | try { |
| 159 | Os.close(peerFd); |
| 160 | } catch (ErrnoException e) { |
| 161 | if (MORE_DEBUG) { |
| 162 | Slog.d(TAG, "Exception closing socket: " + e.getMessage()); |
| 163 | } |
| 164 | } |
| 165 | } |
| 166 | } |
| 167 | } |
| 168 | } catch (Exception e) { |
| 169 | Slog.e(TAG, "Unable to init native debug socket!", e); |
| 170 | } |
| 171 | } |
| 172 | |
| 173 | static int unpackInt(byte[] buf, int offset) { |
| 174 | int b0, b1, b2, b3; |
| 175 | |
| 176 | b0 = ((int) buf[offset]) & 0xFF; // mask against sign extension |
| 177 | b1 = ((int) buf[offset+1]) & 0xFF; |
| 178 | b2 = ((int) buf[offset+2]) & 0xFF; |
| 179 | b3 = ((int) buf[offset+3]) & 0xFF; |
| 180 | return (b0 << 24) | (b1 << 16) | (b2 << 8) | b3; |
| 181 | } |
| 182 | |
| 183 | static int readExactly(FileDescriptor fd, byte[] buffer, int offset, int numBytes) |
| 184 | throws ErrnoException, InterruptedIOException { |
| 185 | int totalRead = 0; |
| 186 | while (numBytes > 0) { |
| 187 | int n = Os.read(fd, buffer, offset + totalRead, numBytes); |
| 188 | if (n <= 0) { |
| 189 | if (DEBUG) { |
| 190 | Slog.w(TAG, "Needed " + numBytes + " but saw " + n); |
| 191 | } |
| 192 | return -1; // premature EOF or timeout |
| 193 | } |
| 194 | numBytes -= n; |
| 195 | totalRead += n; |
| 196 | } |
| 197 | return totalRead; |
| 198 | } |
| 199 | |
| 200 | // Read a crash report from the connection |
| 201 | void consumeNativeCrashData(FileDescriptor fd) { |
| 202 | if (MORE_DEBUG) Slog.i(TAG, "debuggerd connected"); |
| 203 | final byte[] buf = new byte[4096]; |
| 204 | final ByteArrayOutputStream os = new ByteArrayOutputStream(4096); |
| 205 | |
| 206 | try { |
| 207 | StructTimeval timeout = StructTimeval.fromMillis(SOCKET_TIMEOUT_MILLIS); |
| 208 | Os.setsockoptTimeval(fd, SOL_SOCKET, SO_RCVTIMEO, timeout); |
| 209 | Os.setsockoptTimeval(fd, SOL_SOCKET, SO_SNDTIMEO, timeout); |
| 210 | |
| 211 | // The socket is guarded by an selinux neverallow rule that only |
| 212 | // permits crash_dump to connect to it. This allows us to trust the |
| 213 | // received values. |
| 214 | |
| 215 | // Activity Manager protocol: |
| 216 | // - 32-bit network-byte-order: pid |
| 217 | // - 32-bit network-byte-order: signal number |
| 218 | // - byte: gwpAsanRecoverableCrash |
| 219 | // - bytes: raw text of the dump |
| 220 | // - null terminator |
| 221 | int headerBytes = readExactly(fd, buf, 0, 9); |
| 222 | if (headerBytes != 9) { |
| 223 | // protocol failure; give up |
| 224 | Slog.e(TAG, "Unable to read from debuggerd"); |
| 225 | return; |
| 226 | } |
| 227 | |
| 228 | int pid = unpackInt(buf, 0); |
| 229 | int signal = unpackInt(buf, 4); |
| 230 | boolean gwpAsanRecoverableCrash = buf[8] != 0; |
| 231 | if (DEBUG) { |
| 232 | Slog.v(TAG, "Read pid=" + pid + " signal=" + signal |
| 233 | + " recoverable=" + gwpAsanRecoverableCrash); |
| 234 | } |
| 235 | if (pid < 0) { |
| 236 | Slog.e(TAG, "Bogus pid!"); |
| 237 | return; |
| 238 | } |
| 239 | |
| 240 | // now the text of the dump |
| 241 | final ProcessRecord pr; |
| 242 | synchronized (mAm.mPidsSelfLocked) { |
| 243 | pr = mAm.mPidsSelfLocked.get(pid); |
| 244 | } |
| 245 | if (pr == null) { |
| 246 | Slog.w(TAG, "Couldn't find ProcessRecord for pid " + pid); |
| 247 | return; |
| 248 | } |
| 249 | |
| 250 | // Don't attempt crash reporting for persistent apps |
| 251 | if (pr.isPersistent()) { |
| 252 | if (DEBUG) { |
| 253 | Slog.v(TAG, "Skipping report for persistent app " + pr); |
| 254 | } |
| 255 | return; |
| 256 | } |
| 257 | |
| 258 | int bytes; |
| 259 | do { |
| 260 | // get some data |
| 261 | bytes = Os.read(fd, buf, 0, buf.length); |
| 262 | if (bytes > 0) { |
| 263 | if (MORE_DEBUG) { |
| 264 | String s = new String(buf, 0, bytes, "UTF-8"); |
| 265 | Slog.v(TAG, "READ=" + bytes + "> " + s); |
| 266 | } |
| 267 | // did we just get the EOD null byte? |
| 268 | if (buf[bytes - 1] == 0) { |
| 269 | os.write(buf, 0, bytes - 1); // exclude the EOD token |
| 270 | break; |
| 271 | } |
| 272 | // no EOD, so collect it and read more |
| 273 | os.write(buf, 0, bytes); |
| 274 | } |
| 275 | } while (bytes > 0); |
| 276 | |
| 277 | // Okay, we've got the report. |
| 278 | if (DEBUG) Slog.v(TAG, "processing"); |
| 279 | |
| 280 | // Mark the process record as being a native crash so that the |
| 281 | // cleanup mechanism knows we're still submitting the report even |
| 282 | // though the process will vanish as soon as we let debuggerd |
| 283 | // proceed. This isn't relevant for recoverable crashes, as we don't |
| 284 | // show the user an "app crashed" dialogue because the app (by |
| 285 | // design) didn't crash. |
| 286 | if (!gwpAsanRecoverableCrash) { |
| 287 | synchronized (mAm) { |
| 288 | synchronized (mAm.mProcLock) { |
| 289 | pr.mErrorState.setCrashing(true); |
| 290 | pr.mErrorState.setForceCrashReport(true); |
| 291 | } |
| 292 | } |
| 293 | } |
| 294 | |
| 295 | // Crash reporting is synchronous but we want to let debuggerd |
| 296 | // go about it business right away, so we spin off the actual |
| 297 | // reporting logic on a thread and let it take it's time. |
| 298 | final String reportString = new String(os.toByteArray(), "UTF-8"); |
| 299 | (new NativeCrashReporter(pr, signal, gwpAsanRecoverableCrash, reportString)).start(); |
| 300 | } catch (Exception e) { |
| 301 | Slog.e(TAG, "Exception dealing with report", e); |
| 302 | // ugh, fail. |
| 303 | } |
| 304 | } |
| 305 | |
| 306 | } |