1/*
2 * Copyright (C) 2008 The Android Open Source Project
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *  * Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 *  * Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in
12 *    the documentation and/or other materials provided with the
13 *    distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include "linker.h"
30#include "linker_gdb_support.h"
31
32#include <errno.h>
33#include <inttypes.h>
34#include <pthread.h>
35#include <signal.h>
36#include <stdio.h>
37#include <stdlib.h>
38#include <string.h>
39#include <sys/mman.h>
40#include <sys/prctl.h>
41#include <sys/socket.h>
42#include <sys/syscall.h>
43#include <sys/un.h>
44#include <unistd.h>
45
46extern "C" int tgkill(int tgid, int tid, int sig);
47
48// Crash actions have to be sent to the proper debuggerd.
49// On 64 bit systems, the 32 bit debuggerd is named differently.
50#if defined(TARGET_IS_64_BIT) && !defined(__LP64__)
51#define DEBUGGER_SOCKET_NAME "android:debuggerd32"
52#else
53#define DEBUGGER_SOCKET_NAME "android:debuggerd"
54#endif
55
56enum debugger_action_t {
57  // dump a crash
58  DEBUGGER_ACTION_CRASH,
59  // dump a tombstone file
60  DEBUGGER_ACTION_DUMP_TOMBSTONE,
61  // dump a backtrace only back to the socket
62  DEBUGGER_ACTION_DUMP_BACKTRACE,
63};
64
65// Message sent over the socket.
66// NOTE: Any changes to this structure must also be reflected in
67//       system/core/include/cutils/debugger.h.
68struct __attribute__((packed)) debugger_msg_t {
69  int32_t action;
70  pid_t tid;
71  uint64_t abort_msg_address;
72  int32_t original_si_code;
73};
74
75// see man(2) prctl, specifically the section about PR_GET_NAME
76#define MAX_TASK_NAME_LEN (16)
77
78static int socket_abstract_client(const char* name, int type) {
79  sockaddr_un addr;
80
81  // Test with length +1 for the *initial* '\0'.
82  size_t namelen = strlen(name);
83  if ((namelen + 1) > sizeof(addr.sun_path)) {
84    errno = EINVAL;
85    return -1;
86  }
87
88  // This is used for abstract socket namespace, we need
89  // an initial '\0' at the start of the Unix socket path.
90  //
91  // Note: The path in this case is *not* supposed to be
92  // '\0'-terminated. ("man 7 unix" for the gory details.)
93  memset(&addr, 0, sizeof(addr));
94  addr.sun_family = AF_LOCAL;
95  addr.sun_path[0] = 0;
96  memcpy(addr.sun_path + 1, name, namelen);
97
98  socklen_t alen = namelen + offsetof(sockaddr_un, sun_path) + 1;
99
100  int s = socket(AF_LOCAL, type, 0);
101  if (s == -1) {
102    return -1;
103  }
104
105  int rc = TEMP_FAILURE_RETRY(connect(s, reinterpret_cast<sockaddr*>(&addr), alen));
106  if (rc == -1) {
107    close(s);
108    return -1;
109  }
110
111  return s;
112}
113
114/*
115 * Writes a summary of the signal to the log file.  We do this so that, if
116 * for some reason we're not able to contact debuggerd, there is still some
117 * indication of the failure in the log.
118 *
119 * We could be here as a result of native heap corruption, or while a
120 * mutex is being held, so we don't want to use any libc functions that
121 * could allocate memory or hold a lock.
122 */
123static void log_signal_summary(int signum, const siginfo_t* info) {
124  const char* signal_name = "???";
125  bool has_address = false;
126  switch (signum) {
127    case SIGABRT:
128      signal_name = "SIGABRT";
129      break;
130    case SIGBUS:
131      signal_name = "SIGBUS";
132      has_address = true;
133      break;
134    case SIGFPE:
135      signal_name = "SIGFPE";
136      has_address = true;
137      break;
138    case SIGILL:
139      signal_name = "SIGILL";
140      has_address = true;
141      break;
142    case SIGSEGV:
143      signal_name = "SIGSEGV";
144      has_address = true;
145      break;
146#if defined(SIGSTKFLT)
147    case SIGSTKFLT:
148      signal_name = "SIGSTKFLT";
149      break;
150#endif
151    case SIGTRAP:
152      signal_name = "SIGTRAP";
153      break;
154  }
155
156  char thread_name[MAX_TASK_NAME_LEN + 1]; // one more for termination
157  if (prctl(PR_GET_NAME, reinterpret_cast<unsigned long>(thread_name), 0, 0, 0) != 0) {
158    strcpy(thread_name, "<name unknown>");
159  } else {
160    // short names are null terminated by prctl, but the man page
161    // implies that 16 byte names are not.
162    thread_name[MAX_TASK_NAME_LEN] = 0;
163  }
164
165  // "info" will be null if the siginfo_t information was not available.
166  // Many signals don't have an address or a code.
167  char code_desc[32]; // ", code -6"
168  char addr_desc[32]; // ", fault addr 0x1234"
169  addr_desc[0] = code_desc[0] = 0;
170  if (info != nullptr) {
171    // For a rethrown signal, this si_code will be right and the one debuggerd shows will
172    // always be SI_TKILL.
173    __libc_format_buffer(code_desc, sizeof(code_desc), ", code %d", info->si_code);
174    if (has_address) {
175      __libc_format_buffer(addr_desc, sizeof(addr_desc), ", fault addr %p", info->si_addr);
176    }
177  }
178  __libc_format_log(ANDROID_LOG_FATAL, "libc",
179                    "Fatal signal %d (%s)%s%s in tid %d (%s)",
180                    signum, signal_name, code_desc, addr_desc, gettid(), thread_name);
181}
182
183/*
184 * Returns true if the handler for signal "signum" has SA_SIGINFO set.
185 */
186static bool have_siginfo(int signum) {
187  struct sigaction old_action, new_action;
188
189  memset(&new_action, 0, sizeof(new_action));
190  new_action.sa_handler = SIG_DFL;
191  new_action.sa_flags = SA_RESTART;
192  sigemptyset(&new_action.sa_mask);
193
194  if (sigaction(signum, &new_action, &old_action) < 0) {
195    __libc_format_log(ANDROID_LOG_WARN, "libc", "Failed testing for SA_SIGINFO: %s",
196                      strerror(errno));
197    return false;
198  }
199  bool result = (old_action.sa_flags & SA_SIGINFO) != 0;
200
201  if (sigaction(signum, &old_action, nullptr) == -1) {
202    __libc_format_log(ANDROID_LOG_WARN, "libc", "Restore failed in test for SA_SIGINFO: %s",
203                      strerror(errno));
204  }
205  return result;
206}
207
208static void send_debuggerd_packet(siginfo_t* info) {
209  // Mutex to prevent multiple crashing threads from trying to talk
210  // to debuggerd at the same time.
211  static pthread_mutex_t crash_mutex = PTHREAD_MUTEX_INITIALIZER;
212  int ret = pthread_mutex_trylock(&crash_mutex);
213  if (ret != 0) {
214    if (ret == EBUSY) {
215      __libc_format_log(ANDROID_LOG_INFO, "libc",
216          "Another thread contacted debuggerd first; not contacting debuggerd.");
217      // This will never complete since the lock is never released.
218      pthread_mutex_lock(&crash_mutex);
219    } else {
220      __libc_format_log(ANDROID_LOG_INFO, "libc",
221                        "pthread_mutex_trylock failed: %s", strerror(ret));
222    }
223    return;
224  }
225
226  int s = socket_abstract_client(DEBUGGER_SOCKET_NAME, SOCK_STREAM | SOCK_CLOEXEC);
227  if (s == -1) {
228    __libc_format_log(ANDROID_LOG_FATAL, "libc", "Unable to open connection to debuggerd: %s",
229                      strerror(errno));
230    return;
231  }
232
233  // debuggerd knows our pid from the credentials on the
234  // local socket but we need to tell it the tid of the crashing thread.
235  // debuggerd will be paranoid and verify that we sent a tid
236  // that's actually in our process.
237  debugger_msg_t msg;
238  msg.action = DEBUGGER_ACTION_CRASH;
239  msg.tid = gettid();
240  msg.abort_msg_address = reinterpret_cast<uintptr_t>(g_abort_message);
241  msg.original_si_code = (info != nullptr) ? info->si_code : 0;
242  ret = TEMP_FAILURE_RETRY(write(s, &msg, sizeof(msg)));
243  if (ret == sizeof(msg)) {
244    char debuggerd_ack;
245    ret = TEMP_FAILURE_RETRY(read(s, &debuggerd_ack, 1));
246    int saved_errno = errno;
247    notify_gdb_of_libraries();
248    errno = saved_errno;
249  } else {
250    // read or write failed -- broken connection?
251    __libc_format_log(ANDROID_LOG_FATAL, "libc", "Failed while talking to debuggerd: %s",
252                      strerror(errno));
253  }
254
255  close(s);
256}
257
258/*
259 * Catches fatal signals so we can ask debuggerd to ptrace us before
260 * we crash.
261 */
262static void debuggerd_signal_handler(int signal_number, siginfo_t* info, void*) {
263  // It's possible somebody cleared the SA_SIGINFO flag, which would mean
264  // our "info" arg holds an undefined value.
265  if (!have_siginfo(signal_number)) {
266    info = nullptr;
267  }
268
269  log_signal_summary(signal_number, info);
270
271  send_debuggerd_packet(info);
272
273  // We need to return from the signal handler so that debuggerd can dump the
274  // thread that crashed, but returning here does not guarantee that the signal
275  // will be thrown again, even for SIGSEGV and friends, since the signal could
276  // have been sent manually. Resend the signal with rt_tgsigqueueinfo(2) to
277  // preserve the SA_SIGINFO contents.
278  signal(signal_number, SIG_DFL);
279
280  struct siginfo si;
281  if (!info) {
282    memset(&si, 0, sizeof(si));
283    si.si_code = SI_USER;
284    si.si_pid = getpid();
285    si.si_uid = getuid();
286    info = &si;
287  } else if (info->si_code >= 0 || info->si_code == SI_TKILL) {
288    // rt_tgsigqueueinfo(2)'s documentation appears to be incorrect on kernels
289    // that contain commit 66dd34a (3.9+). The manpage claims to only allow
290    // negative si_code values that are not SI_TKILL, but 66dd34a changed the
291    // check to allow all si_code values in calls coming from inside the house.
292  }
293
294  int rc = syscall(SYS_rt_tgsigqueueinfo, getpid(), gettid(), signal_number, info);
295  if (rc != 0) {
296    __libc_format_log(ANDROID_LOG_FATAL, "libc", "failed to resend signal during crash: %s",
297                      strerror(errno));
298    _exit(0);
299  }
300}
301
302__LIBC_HIDDEN__ void debuggerd_init() {
303  struct sigaction action;
304  memset(&action, 0, sizeof(action));
305  sigemptyset(&action.sa_mask);
306  action.sa_sigaction = debuggerd_signal_handler;
307  action.sa_flags = SA_RESTART | SA_SIGINFO;
308
309  // Use the alternate signal stack if available so we can catch stack overflows.
310  action.sa_flags |= SA_ONSTACK;
311
312  sigaction(SIGABRT, &action, nullptr);
313  sigaction(SIGBUS, &action, nullptr);
314  sigaction(SIGFPE, &action, nullptr);
315  sigaction(SIGILL, &action, nullptr);
316  sigaction(SIGSEGV, &action, nullptr);
317#if defined(SIGSTKFLT)
318  sigaction(SIGSTKFLT, &action, nullptr);
319#endif
320  sigaction(SIGTRAP, &action, nullptr);
321}
322