1 /*
2 * Copyright 2006, 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 #include <arpa/inet.h>
18 #include <dirent.h>
19 #include <elf.h>
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <pthread.h>
23 #include <signal.h>
24 #include <stdarg.h>
25 #include <stdio.h>
26 #include <sys/poll.h>
27 #include <sys/prctl.h>
28 #include <sys/ptrace.h>
29 #include <sys/socket.h>
30 #include <sys/stat.h>
31 #include <sys/types.h>
32 #include <sys/wait.h>
33 #include <sys/un.h>
34 #include <time.h>
35
36 #include <memory>
37 #include <set>
38 #include <string>
39
40 #include <selinux/android.h>
41
42 #include <log/logger.h>
43
44 #include <android-base/file.h>
45 #include <android-base/unique_fd.h>
46 #include <cutils/debugger.h>
47 #include <cutils/properties.h>
48 #include <cutils/sockets.h>
49 #include <nativehelper/ScopedFd.h>
50
51 #include <linux/input.h>
52
53 #include <private/android_filesystem_config.h>
54
55 #include "backtrace.h"
56 #include "getevent.h"
57 #include "signal_sender.h"
58 #include "tombstone.h"
59 #include "utility.h"
60
61 // If the 32 bit executable is compiled on a 64 bit system,
62 // use the 32 bit socket name.
63 #if defined(TARGET_IS_64_BIT) && !defined(__LP64__)
64 #define SOCKET_NAME DEBUGGER32_SOCKET_NAME
65 #else
66 #define SOCKET_NAME DEBUGGER_SOCKET_NAME
67 #endif
68
69 struct debugger_request_t {
70 debugger_action_t action;
71 pid_t pid, tid;
72 uid_t uid, gid;
73 uintptr_t abort_msg_address;
74 int32_t original_si_code;
75 };
76
wait_for_user_action(const debugger_request_t & request)77 static void wait_for_user_action(const debugger_request_t& request) {
78 // Explain how to attach the debugger.
79 ALOGI("***********************************************************\n"
80 "* Process %d has been suspended while crashing.\n"
81 "* To attach gdbserver and start gdb, run this on the host:\n"
82 "*\n"
83 "* gdbclient.py -p %d\n"
84 "*\n"
85 "* Wait for gdb to start, then press the VOLUME DOWN key\n"
86 "* to let the process continue crashing.\n"
87 "***********************************************************",
88 request.pid, request.tid);
89
90 // Wait for VOLUME DOWN.
91 while (true) {
92 input_event e;
93 if (get_event(&e, -1) == 0) {
94 if (e.type == EV_KEY && e.code == KEY_VOLUMEDOWN && e.value == 0) {
95 break;
96 }
97 }
98 }
99
100 ALOGI("debuggerd resuming process %d", request.pid);
101 }
102
get_process_info(pid_t tid,pid_t * out_pid,uid_t * out_uid,uid_t * out_gid)103 static int get_process_info(pid_t tid, pid_t* out_pid, uid_t* out_uid, uid_t* out_gid) {
104 char path[64];
105 snprintf(path, sizeof(path), "/proc/%d/status", tid);
106
107 FILE* fp = fopen(path, "r");
108 if (!fp) {
109 return -1;
110 }
111
112 int fields = 0;
113 char line[1024];
114 while (fgets(line, sizeof(line), fp)) {
115 size_t len = strlen(line);
116 if (len > 6 && !memcmp(line, "Tgid:\t", 6)) {
117 *out_pid = atoi(line + 6);
118 fields |= 1;
119 } else if (len > 5 && !memcmp(line, "Uid:\t", 5)) {
120 *out_uid = atoi(line + 5);
121 fields |= 2;
122 } else if (len > 5 && !memcmp(line, "Gid:\t", 5)) {
123 *out_gid = atoi(line + 5);
124 fields |= 4;
125 }
126 }
127 fclose(fp);
128 return fields == 7 ? 0 : -1;
129 }
130
131 /*
132 * Corresponds with debugger_action_t enum type in
133 * include/cutils/debugger.h.
134 */
135 static const char *debuggerd_perms[] = {
136 NULL, /* crash is only used on self, no check applied */
137 "dump_tombstone",
138 "dump_backtrace"
139 };
140
audit_callback(void * data,security_class_t,char * buf,size_t len)141 static int audit_callback(void* data, security_class_t /* cls */, char* buf, size_t len)
142 {
143 struct debugger_request_t* req = reinterpret_cast<debugger_request_t*>(data);
144
145 if (!req) {
146 ALOGE("No debuggerd request audit data");
147 return 0;
148 }
149
150 snprintf(buf, len, "pid=%d uid=%d gid=%d", req->pid, req->uid, req->gid);
151 return 0;
152 }
153
selinux_action_allowed(int s,debugger_request_t * request)154 static bool selinux_action_allowed(int s, debugger_request_t* request)
155 {
156 char *scon = NULL, *tcon = NULL;
157 const char *tclass = "debuggerd";
158 const char *perm;
159 bool allowed = false;
160
161 if (request->action <= 0 || request->action >= (sizeof(debuggerd_perms)/sizeof(debuggerd_perms[0]))) {
162 ALOGE("SELinux: No permission defined for debugger action %d", request->action);
163 return false;
164 }
165
166 perm = debuggerd_perms[request->action];
167
168 if (getpeercon(s, &scon) < 0) {
169 ALOGE("Cannot get peer context from socket\n");
170 goto out;
171 }
172
173 if (getpidcon(request->tid, &tcon) < 0) {
174 ALOGE("Cannot get context for tid %d\n", request->tid);
175 goto out;
176 }
177
178 allowed = (selinux_check_access(scon, tcon, tclass, perm, reinterpret_cast<void*>(request)) == 0);
179
180 out:
181 freecon(scon);
182 freecon(tcon);
183 return allowed;
184 }
185
pid_contains_tid(pid_t pid,pid_t tid)186 static bool pid_contains_tid(pid_t pid, pid_t tid) {
187 char task_path[PATH_MAX];
188 if (snprintf(task_path, PATH_MAX, "/proc/%d/task/%d", pid, tid) >= PATH_MAX) {
189 ALOGE("debuggerd: task path overflow (pid = %d, tid = %d)\n", pid, tid);
190 exit(1);
191 }
192
193 return access(task_path, F_OK) == 0;
194 }
195
read_request(int fd,debugger_request_t * out_request)196 static int read_request(int fd, debugger_request_t* out_request) {
197 ucred cr;
198 socklen_t len = sizeof(cr);
199 int status = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &cr, &len);
200 if (status != 0) {
201 ALOGE("cannot get credentials");
202 return -1;
203 }
204
205 ALOGV("reading tid");
206 fcntl(fd, F_SETFL, O_NONBLOCK);
207
208 pollfd pollfds[1];
209 pollfds[0].fd = fd;
210 pollfds[0].events = POLLIN;
211 pollfds[0].revents = 0;
212 status = TEMP_FAILURE_RETRY(poll(pollfds, 1, 3000));
213 if (status != 1) {
214 ALOGE("timed out reading tid (from pid=%d uid=%d)\n", cr.pid, cr.uid);
215 return -1;
216 }
217
218 debugger_msg_t msg;
219 memset(&msg, 0, sizeof(msg));
220 status = TEMP_FAILURE_RETRY(read(fd, &msg, sizeof(msg)));
221 if (status < 0) {
222 ALOGE("read failure? %s (pid=%d uid=%d)\n", strerror(errno), cr.pid, cr.uid);
223 return -1;
224 }
225 if (status != sizeof(debugger_msg_t)) {
226 ALOGE("invalid crash request of size %d (from pid=%d uid=%d)\n", status, cr.pid, cr.uid);
227 return -1;
228 }
229
230 out_request->action = static_cast<debugger_action_t>(msg.action);
231 out_request->tid = msg.tid;
232 out_request->pid = cr.pid;
233 out_request->uid = cr.uid;
234 out_request->gid = cr.gid;
235 out_request->abort_msg_address = msg.abort_msg_address;
236 out_request->original_si_code = msg.original_si_code;
237
238 if (msg.action == DEBUGGER_ACTION_CRASH) {
239 // Ensure that the tid reported by the crashing process is valid.
240 // This check needs to happen again after ptracing the requested thread to prevent a race.
241 if (!pid_contains_tid(out_request->pid, out_request->tid)) {
242 ALOGE("tid %d does not exist in pid %d. ignoring debug request\n", out_request->tid,
243 out_request->pid);
244 return -1;
245 }
246 } else if (cr.uid == 0 || (cr.uid == AID_SYSTEM && msg.action == DEBUGGER_ACTION_DUMP_BACKTRACE)) {
247 // Only root or system can ask us to attach to any process and dump it explicitly.
248 // However, system is only allowed to collect backtraces but cannot dump tombstones.
249 status = get_process_info(out_request->tid, &out_request->pid,
250 &out_request->uid, &out_request->gid);
251 if (status < 0) {
252 ALOGE("tid %d does not exist. ignoring explicit dump request\n", out_request->tid);
253 return -1;
254 }
255
256 if (!selinux_action_allowed(fd, out_request))
257 return -1;
258 } else {
259 // No one else is allowed to dump arbitrary processes.
260 return -1;
261 }
262 return 0;
263 }
264
activity_manager_connect()265 static int activity_manager_connect() {
266 android::base::unique_fd amfd(socket(PF_UNIX, SOCK_STREAM, 0));
267 if (amfd.get() < -1) {
268 ALOGE("debuggerd: Unable to connect to activity manager (socket failed: %s)", strerror(errno));
269 return -1;
270 }
271
272 struct sockaddr_un address;
273 memset(&address, 0, sizeof(address));
274 address.sun_family = AF_UNIX;
275 // The path used here must match the value defined in NativeCrashListener.java.
276 strncpy(address.sun_path, "/data/system/ndebugsocket", sizeof(address.sun_path));
277 if (TEMP_FAILURE_RETRY(connect(amfd.get(), reinterpret_cast<struct sockaddr*>(&address),
278 sizeof(address))) == -1) {
279 ALOGE("debuggerd: Unable to connect to activity manager (connect failed: %s)", strerror(errno));
280 return -1;
281 }
282
283 struct timeval tv;
284 memset(&tv, 0, sizeof(tv));
285 tv.tv_sec = 1; // tight leash
286 if (setsockopt(amfd.get(), SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)) == -1) {
287 ALOGE("debuggerd: Unable to connect to activity manager (setsockopt SO_SNDTIMEO failed: %s)",
288 strerror(errno));
289 return -1;
290 }
291
292 tv.tv_sec = 3; // 3 seconds on handshake read
293 if (setsockopt(amfd.get(), SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)) == -1) {
294 ALOGE("debuggerd: Unable to connect to activity manager (setsockopt SO_RCVTIMEO failed: %s)",
295 strerror(errno));
296 return -1;
297 }
298
299 return amfd.release();
300 }
301
activity_manager_write(int pid,int signal,int amfd,const std::string & amfd_data)302 static void activity_manager_write(int pid, int signal, int amfd, const std::string& amfd_data) {
303 if (amfd == -1) {
304 return;
305 }
306
307 // Activity Manager protocol: binary 32-bit network-byte-order ints for the
308 // pid and signal number, followed by the raw text of the dump, culminating
309 // in a zero byte that marks end-of-data.
310 uint32_t datum = htonl(pid);
311 if (!android::base::WriteFully(amfd, &datum, 4)) {
312 ALOGE("AM pid write failed: %s\n", strerror(errno));
313 return;
314 }
315 datum = htonl(signal);
316 if (!android::base::WriteFully(amfd, &datum, 4)) {
317 ALOGE("AM signal write failed: %s\n", strerror(errno));
318 return;
319 }
320
321 if (!android::base::WriteFully(amfd, amfd_data.c_str(), amfd_data.size())) {
322 ALOGE("AM data write failed: %s\n", strerror(errno));
323 return;
324 }
325
326 // Send EOD to the Activity Manager, then wait for its ack to avoid racing
327 // ahead and killing the target out from under it.
328 uint8_t eodMarker = 0;
329 if (!android::base::WriteFully(amfd, &eodMarker, 1)) {
330 ALOGE("AM eod write failed: %s\n", strerror(errno));
331 return;
332 }
333 // 3 sec timeout reading the ack; we're fine if the read fails.
334 android::base::ReadFully(amfd, &eodMarker, 1);
335 }
336
should_attach_gdb(const debugger_request_t & request)337 static bool should_attach_gdb(const debugger_request_t& request) {
338 if (request.action == DEBUGGER_ACTION_CRASH) {
339 return property_get_bool("debug.debuggerd.wait_for_gdb", false);
340 }
341 return false;
342 }
343
344 #if defined(__LP64__)
is32bit(pid_t tid)345 static bool is32bit(pid_t tid) {
346 char* exeline;
347 if (asprintf(&exeline, "/proc/%d/exe", tid) == -1) {
348 return false;
349 }
350 int fd = TEMP_FAILURE_RETRY(open(exeline, O_RDONLY | O_CLOEXEC));
351 int saved_errno = errno;
352 free(exeline);
353 if (fd == -1) {
354 ALOGW("Failed to open /proc/%d/exe %s", tid, strerror(saved_errno));
355 return false;
356 }
357
358 char ehdr[EI_NIDENT];
359 ssize_t bytes = TEMP_FAILURE_RETRY(read(fd, &ehdr, sizeof(ehdr)));
360 close(fd);
361 if (bytes != (ssize_t) sizeof(ehdr) || memcmp(ELFMAG, ehdr, SELFMAG) != 0) {
362 return false;
363 }
364 if (ehdr[EI_CLASS] == ELFCLASS32) {
365 return true;
366 }
367 return false;
368 }
369
redirect_to_32(int fd,debugger_request_t * request)370 static void redirect_to_32(int fd, debugger_request_t* request) {
371 debugger_msg_t msg;
372 memset(&msg, 0, sizeof(msg));
373 msg.tid = request->tid;
374 msg.action = request->action;
375
376 int sock_fd = socket_local_client(DEBUGGER32_SOCKET_NAME, ANDROID_SOCKET_NAMESPACE_ABSTRACT,
377 SOCK_STREAM | SOCK_CLOEXEC);
378 if (sock_fd < 0) {
379 ALOGE("Failed to connect to debuggerd32: %s", strerror(errno));
380 return;
381 }
382
383 if (TEMP_FAILURE_RETRY(write(sock_fd, &msg, sizeof(msg))) != (ssize_t) sizeof(msg)) {
384 ALOGE("Failed to write request to debuggerd32 socket: %s", strerror(errno));
385 close(sock_fd);
386 return;
387 }
388
389 char ack;
390 if (TEMP_FAILURE_RETRY(read(sock_fd, &ack, 1)) == -1) {
391 ALOGE("Failed to read ack from debuggerd32 socket: %s", strerror(errno));
392 close(sock_fd);
393 return;
394 }
395
396 char buffer[1024];
397 ssize_t bytes_read;
398 while ((bytes_read = TEMP_FAILURE_RETRY(read(sock_fd, buffer, sizeof(buffer)))) > 0) {
399 ssize_t bytes_to_send = bytes_read;
400 ssize_t bytes_written;
401 do {
402 bytes_written = TEMP_FAILURE_RETRY(write(fd, buffer + bytes_read - bytes_to_send,
403 bytes_to_send));
404 if (bytes_written == -1) {
405 if (errno == EAGAIN) {
406 // Retry the write.
407 continue;
408 }
409 ALOGE("Error while writing data to fd: %s", strerror(errno));
410 break;
411 }
412 bytes_to_send -= bytes_written;
413 } while (bytes_written != 0 && bytes_to_send > 0);
414 if (bytes_to_send != 0) {
415 ALOGE("Failed to write all data to fd: read %zd, sent %zd", bytes_read, bytes_to_send);
416 break;
417 }
418 }
419 close(sock_fd);
420 }
421 #endif
422
423 // Attach to a thread, and verify that it's still a member of the given process
ptrace_attach_thread(pid_t pid,pid_t tid)424 static bool ptrace_attach_thread(pid_t pid, pid_t tid) {
425 if (ptrace(PTRACE_ATTACH, tid, 0, 0) != 0) {
426 return false;
427 }
428
429 // Make sure that the task we attached to is actually part of the pid we're dumping.
430 if (!pid_contains_tid(pid, tid)) {
431 if (ptrace(PTRACE_DETACH, tid, 0, 0) != 0) {
432 ALOGE("debuggerd: failed to detach from thread '%d'", tid);
433 exit(1);
434 }
435 return false;
436 }
437
438 return true;
439 }
440
ptrace_siblings(pid_t pid,pid_t main_tid,std::set<pid_t> & tids)441 static void ptrace_siblings(pid_t pid, pid_t main_tid, std::set<pid_t>& tids) {
442 char task_path[PATH_MAX];
443
444 if (snprintf(task_path, PATH_MAX, "/proc/%d/task", pid) >= PATH_MAX) {
445 ALOGE("debuggerd: task path overflow (pid = %d)\n", pid);
446 abort();
447 }
448
449 std::unique_ptr<DIR, int (*)(DIR*)> d(opendir(task_path), closedir);
450
451 // Bail early if the task directory cannot be opened.
452 if (!d) {
453 ALOGE("debuggerd: failed to open /proc/%d/task: %s", pid, strerror(errno));
454 return;
455 }
456
457 struct dirent* de;
458 while ((de = readdir(d.get())) != NULL) {
459 // Ignore "." and "..".
460 if (!strcmp(de->d_name, ".") || !strcmp(de->d_name, "..")) {
461 continue;
462 }
463
464 char* end;
465 pid_t tid = strtoul(de->d_name, &end, 10);
466 if (*end) {
467 continue;
468 }
469
470 if (tid == main_tid) {
471 continue;
472 }
473
474 if (!ptrace_attach_thread(pid, tid)) {
475 ALOGE("debuggerd: ptrace attach to %d failed: %s", tid, strerror(errno));
476 continue;
477 }
478
479 tids.insert(tid);
480 }
481 }
482
perform_dump(const debugger_request_t & request,int fd,int tombstone_fd,BacktraceMap * backtrace_map,const std::set<pid_t> & siblings,int * crash_signal,std::string * amfd_data)483 static bool perform_dump(const debugger_request_t& request, int fd, int tombstone_fd,
484 BacktraceMap* backtrace_map, const std::set<pid_t>& siblings,
485 int* crash_signal, std::string* amfd_data) {
486 if (TEMP_FAILURE_RETRY(write(fd, "\0", 1)) != 1) {
487 ALOGE("debuggerd: failed to respond to client: %s\n", strerror(errno));
488 return false;
489 }
490
491 int total_sleep_time_usec = 0;
492 while (true) {
493 int signal = wait_for_signal(request.tid, &total_sleep_time_usec);
494 switch (signal) {
495 case -1:
496 ALOGE("debuggerd: timed out waiting for signal");
497 return false;
498
499 case SIGSTOP:
500 if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) {
501 ALOGV("debuggerd: stopped -- dumping to tombstone");
502 engrave_tombstone(tombstone_fd, backtrace_map, request.pid, request.tid, siblings, signal,
503 request.original_si_code, request.abort_msg_address, amfd_data);
504 } else if (request.action == DEBUGGER_ACTION_DUMP_BACKTRACE) {
505 ALOGV("debuggerd: stopped -- dumping to fd");
506 dump_backtrace(fd, backtrace_map, request.pid, request.tid, siblings, nullptr);
507 } else {
508 ALOGV("debuggerd: stopped -- continuing");
509 if (ptrace(PTRACE_CONT, request.tid, 0, 0) != 0) {
510 ALOGE("debuggerd: ptrace continue failed: %s", strerror(errno));
511 return false;
512 }
513 continue; // loop again
514 }
515 break;
516
517 case SIGABRT:
518 case SIGBUS:
519 case SIGFPE:
520 case SIGILL:
521 case SIGSEGV:
522 #ifdef SIGSTKFLT
523 case SIGSTKFLT:
524 #endif
525 case SIGSYS:
526 case SIGTRAP:
527 ALOGV("stopped -- fatal signal\n");
528 *crash_signal = signal;
529 engrave_tombstone(tombstone_fd, backtrace_map, request.pid, request.tid, siblings, signal,
530 request.original_si_code, request.abort_msg_address, amfd_data);
531 break;
532
533 default:
534 ALOGE("debuggerd: process stopped due to unexpected signal %d\n", signal);
535 break;
536 }
537 break;
538 }
539
540 return true;
541 }
542
drop_privileges()543 static bool drop_privileges() {
544 // AID_LOG: for reading the logs data associated with the crashing process.
545 // AID_READPROC: for reading /proc/<PID>/{comm,cmdline}.
546 gid_t groups[] = { AID_DEBUGGERD, AID_LOG, AID_READPROC };
547 if (setgroups(sizeof(groups)/sizeof(groups[0]), groups) != 0) {
548 ALOGE("debuggerd: failed to setgroups: %s", strerror(errno));
549 return false;
550 }
551
552 if (setresgid(AID_DEBUGGERD, AID_DEBUGGERD, AID_DEBUGGERD) != 0) {
553 ALOGE("debuggerd: failed to setresgid: %s", strerror(errno));
554 return false;
555 }
556
557 if (setresuid(AID_DEBUGGERD, AID_DEBUGGERD, AID_DEBUGGERD) != 0) {
558 ALOGE("debuggerd: failed to setresuid: %s", strerror(errno));
559 return false;
560 }
561
562 return true;
563 }
564
worker_process(int fd,debugger_request_t & request)565 static void worker_process(int fd, debugger_request_t& request) {
566 // Open the tombstone file if we need it.
567 std::string tombstone_path;
568 int tombstone_fd = -1;
569 switch (request.action) {
570 case DEBUGGER_ACTION_DUMP_TOMBSTONE:
571 case DEBUGGER_ACTION_CRASH:
572 tombstone_fd = open_tombstone(&tombstone_path);
573 if (tombstone_fd == -1) {
574 ALOGE("debuggerd: failed to open tombstone file: %s\n", strerror(errno));
575 exit(1);
576 }
577 break;
578
579 case DEBUGGER_ACTION_DUMP_BACKTRACE:
580 break;
581
582 default:
583 ALOGE("debuggerd: unexpected request action: %d", request.action);
584 exit(1);
585 }
586
587 // At this point, the thread that made the request is blocked in
588 // a read() call. If the thread has crashed, then this gives us
589 // time to PTRACE_ATTACH to it before it has a chance to really fault.
590 //
591 // The PTRACE_ATTACH sends a SIGSTOP to the target process, but it
592 // won't necessarily have stopped by the time ptrace() returns. (We
593 // currently assume it does.) We write to the file descriptor to
594 // ensure that it can run as soon as we call PTRACE_CONT below.
595 // See details in bionic/libc/linker/debugger.c, in function
596 // debugger_signal_handler().
597
598 // Attach to the target process.
599 if (!ptrace_attach_thread(request.pid, request.tid)) {
600 ALOGE("debuggerd: ptrace attach failed: %s", strerror(errno));
601 exit(1);
602 }
603
604 // DEBUGGER_ACTION_CRASH requests can come from arbitrary processes and the tid field in the
605 // request is sent from the other side. If an attacker can cause a process to be spawned with the
606 // pid of their process, they could trick debuggerd into dumping that process by exiting after
607 // sending the request. Validate the trusted request.uid/gid to defend against this.
608 if (request.action == DEBUGGER_ACTION_CRASH) {
609 pid_t pid;
610 uid_t uid;
611 gid_t gid;
612 if (get_process_info(request.tid, &pid, &uid, &gid) != 0) {
613 ALOGE("debuggerd: failed to get process info for tid '%d'", request.tid);
614 exit(1);
615 }
616
617 if (pid != request.pid || uid != request.uid || gid != request.gid) {
618 ALOGE(
619 "debuggerd: attached task %d does not match request: "
620 "expected pid=%d,uid=%d,gid=%d, actual pid=%d,uid=%d,gid=%d",
621 request.tid, request.pid, request.uid, request.gid, pid, uid, gid);
622 exit(1);
623 }
624 }
625
626 // Don't attach to the sibling threads if we want to attach gdb.
627 // Supposedly, it makes the process less reliable.
628 bool attach_gdb = should_attach_gdb(request);
629 if (attach_gdb) {
630 // Open all of the input devices we need to listen for VOLUMEDOWN before dropping privileges.
631 if (init_getevent() != 0) {
632 ALOGE("debuggerd: failed to initialize input device, not waiting for gdb");
633 attach_gdb = false;
634 }
635
636 }
637
638 std::set<pid_t> siblings;
639 if (!attach_gdb) {
640 ptrace_siblings(request.pid, request.tid, siblings);
641 }
642
643 // Generate the backtrace map before dropping privileges.
644 std::unique_ptr<BacktraceMap> backtrace_map(BacktraceMap::Create(request.pid));
645
646 int amfd = -1;
647 std::unique_ptr<std::string> amfd_data;
648 if (request.action == DEBUGGER_ACTION_CRASH) {
649 // Connect to the activity manager before dropping privileges.
650 amfd = activity_manager_connect();
651 amfd_data.reset(new std::string);
652 }
653
654 bool succeeded = false;
655
656 // Now that we've done everything that requires privileges, we can drop them.
657 if (!drop_privileges()) {
658 ALOGE("debuggerd: failed to drop privileges, exiting");
659 _exit(1);
660 }
661
662 int crash_signal = SIGKILL;
663 succeeded = perform_dump(request, fd, tombstone_fd, backtrace_map.get(), siblings,
664 &crash_signal, amfd_data.get());
665 if (succeeded) {
666 if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) {
667 if (!tombstone_path.empty()) {
668 android::base::WriteFully(fd, tombstone_path.c_str(), tombstone_path.length());
669 }
670 }
671 }
672
673 if (attach_gdb) {
674 // Tell the signal process to send SIGSTOP to the target.
675 if (!send_signal(request.pid, 0, SIGSTOP)) {
676 ALOGE("debuggerd: failed to stop process for gdb attach: %s", strerror(errno));
677 attach_gdb = false;
678 }
679 }
680
681 if (!attach_gdb) {
682 // Tell the Activity Manager about the crashing process. If we are
683 // waiting for gdb to attach, do not send this or Activity Manager
684 // might kill the process before anyone can attach.
685 activity_manager_write(request.pid, crash_signal, amfd, *amfd_data.get());
686 }
687
688 if (ptrace(PTRACE_DETACH, request.tid, 0, 0) != 0) {
689 ALOGE("debuggerd: ptrace detach from %d failed: %s", request.tid, strerror(errno));
690 }
691
692 for (pid_t sibling : siblings) {
693 ptrace(PTRACE_DETACH, sibling, 0, 0);
694 }
695
696 // Send the signal back to the process if it crashed and we're not waiting for gdb.
697 if (!attach_gdb && request.action == DEBUGGER_ACTION_CRASH) {
698 if (!send_signal(request.pid, request.tid, crash_signal)) {
699 ALOGE("debuggerd: failed to kill process %d: %s", request.pid, strerror(errno));
700 }
701 }
702
703 // Wait for gdb, if requested.
704 if (attach_gdb) {
705 wait_for_user_action(request);
706
707 // Now tell the activity manager about this process.
708 activity_manager_write(request.pid, crash_signal, amfd, *amfd_data.get());
709
710 // Tell the signal process to send SIGCONT to the target.
711 if (!send_signal(request.pid, 0, SIGCONT)) {
712 ALOGE("debuggerd: failed to resume process %d: %s", request.pid, strerror(errno));
713 }
714
715 uninit_getevent();
716 }
717
718 close(amfd);
719
720 exit(!succeeded);
721 }
722
monitor_worker_process(int child_pid,const debugger_request_t & request)723 static void monitor_worker_process(int child_pid, const debugger_request_t& request) {
724 struct timespec timeout = {.tv_sec = 10, .tv_nsec = 0 };
725 if (should_attach_gdb(request)) {
726 // If wait_for_gdb is enabled, set the timeout to something large.
727 timeout.tv_sec = INT_MAX;
728 }
729
730 sigset_t signal_set;
731 sigemptyset(&signal_set);
732 sigaddset(&signal_set, SIGCHLD);
733
734 bool kill_worker = false;
735 bool kill_target = false;
736 bool kill_self = false;
737
738 int status;
739 siginfo_t siginfo;
740 int signal = TEMP_FAILURE_RETRY(sigtimedwait(&signal_set, &siginfo, &timeout));
741 if (signal == SIGCHLD) {
742 pid_t rc = waitpid(-1, &status, WNOHANG | WUNTRACED);
743 if (rc != child_pid) {
744 ALOGE("debuggerd: waitpid returned unexpected pid (%d), committing murder-suicide", rc);
745
746 if (WIFEXITED(status)) {
747 ALOGW("debuggerd: pid %d exited with status %d", rc, WEXITSTATUS(status));
748 } else if (WIFSIGNALED(status)) {
749 ALOGW("debuggerd: pid %d received signal %d", rc, WTERMSIG(status));
750 } else if (WIFSTOPPED(status)) {
751 ALOGW("debuggerd: pid %d stopped by signal %d", rc, WSTOPSIG(status));
752 } else if (WIFCONTINUED(status)) {
753 ALOGW("debuggerd: pid %d continued", rc);
754 }
755
756 kill_worker = true;
757 kill_target = true;
758 kill_self = true;
759 } else if (WIFSIGNALED(status)) {
760 ALOGE("debuggerd: worker process %d terminated due to signal %d", child_pid, WTERMSIG(status));
761 kill_worker = false;
762 kill_target = true;
763 } else if (WIFSTOPPED(status)) {
764 ALOGE("debuggerd: worker process %d stopped due to signal %d", child_pid, WSTOPSIG(status));
765 kill_worker = true;
766 kill_target = true;
767 }
768 } else {
769 ALOGE("debuggerd: worker process %d timed out", child_pid);
770 kill_worker = true;
771 kill_target = true;
772 }
773
774 if (kill_worker) {
775 // Something bad happened, kill the worker.
776 if (kill(child_pid, SIGKILL) != 0) {
777 ALOGE("debuggerd: failed to kill worker process %d: %s", child_pid, strerror(errno));
778 } else {
779 waitpid(child_pid, &status, 0);
780 }
781 }
782
783 int exit_signal = SIGCONT;
784 if (kill_target && request.action == DEBUGGER_ACTION_CRASH) {
785 ALOGE("debuggerd: killing target %d", request.pid);
786 exit_signal = SIGKILL;
787 } else {
788 ALOGW("debuggerd: resuming target %d", request.pid);
789 }
790
791 if (kill(request.pid, exit_signal) != 0) {
792 ALOGE("debuggerd: failed to send signal %d to target: %s", exit_signal, strerror(errno));
793 }
794
795 if (kill_self) {
796 stop_signal_sender();
797 _exit(1);
798 }
799 }
800
handle_request(int fd)801 static void handle_request(int fd) {
802 ALOGV("handle_request(%d)\n", fd);
803
804 ScopedFd closer(fd);
805 debugger_request_t request;
806 memset(&request, 0, sizeof(request));
807 int status = read_request(fd, &request);
808 if (status != 0) {
809 return;
810 }
811
812 ALOGW("debuggerd: handling request: pid=%d uid=%d gid=%d tid=%d\n", request.pid, request.uid,
813 request.gid, request.tid);
814
815 #if defined(__LP64__)
816 // On 64 bit systems, requests to dump 32 bit and 64 bit tids come
817 // to the 64 bit debuggerd. If the process is a 32 bit executable,
818 // redirect the request to the 32 bit debuggerd.
819 if (is32bit(request.tid)) {
820 // Only dump backtrace and dump tombstone requests can be redirected.
821 if (request.action == DEBUGGER_ACTION_DUMP_BACKTRACE ||
822 request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) {
823 redirect_to_32(fd, &request);
824 } else {
825 ALOGE("debuggerd: Not allowed to redirect action %d to 32 bit debuggerd\n", request.action);
826 }
827 return;
828 }
829 #endif
830
831 // Fork a child to handle the rest of the request.
832 pid_t fork_pid = fork();
833 if (fork_pid == -1) {
834 ALOGE("debuggerd: failed to fork: %s\n", strerror(errno));
835 } else if (fork_pid == 0) {
836 worker_process(fd, request);
837 } else {
838 monitor_worker_process(fork_pid, request);
839 }
840 }
841
do_server()842 static int do_server() {
843 // debuggerd crashes can't be reported to debuggerd.
844 // Reset all of the crash handlers.
845 signal(SIGABRT, SIG_DFL);
846 signal(SIGBUS, SIG_DFL);
847 signal(SIGFPE, SIG_DFL);
848 signal(SIGILL, SIG_DFL);
849 signal(SIGSEGV, SIG_DFL);
850 #ifdef SIGSTKFLT
851 signal(SIGSTKFLT, SIG_DFL);
852 #endif
853 signal(SIGTRAP, SIG_DFL);
854
855 // Ignore failed writes to closed sockets
856 signal(SIGPIPE, SIG_IGN);
857
858 // Block SIGCHLD so we can sigtimedwait for it.
859 sigset_t sigchld;
860 sigemptyset(&sigchld);
861 sigaddset(&sigchld, SIGCHLD);
862 sigprocmask(SIG_SETMASK, &sigchld, nullptr);
863
864 int s = socket_local_server(SOCKET_NAME, ANDROID_SOCKET_NAMESPACE_ABSTRACT,
865 SOCK_STREAM | SOCK_CLOEXEC);
866 if (s == -1) return 1;
867
868 // Fork a process that stays root, and listens on a pipe to pause and resume the target.
869 if (!start_signal_sender()) {
870 ALOGE("debuggerd: failed to fork signal sender");
871 return 1;
872 }
873
874 ALOGI("debuggerd: starting\n");
875
876 for (;;) {
877 sockaddr_storage ss;
878 sockaddr* addrp = reinterpret_cast<sockaddr*>(&ss);
879 socklen_t alen = sizeof(ss);
880
881 ALOGV("waiting for connection\n");
882 int fd = accept4(s, addrp, &alen, SOCK_CLOEXEC);
883 if (fd == -1) {
884 ALOGE("accept failed: %s\n", strerror(errno));
885 continue;
886 }
887
888 handle_request(fd);
889 }
890 return 0;
891 }
892
do_explicit_dump(pid_t tid,bool dump_backtrace)893 static int do_explicit_dump(pid_t tid, bool dump_backtrace) {
894 fprintf(stdout, "Sending request to dump task %d.\n", tid);
895
896 if (dump_backtrace) {
897 fflush(stdout);
898 if (dump_backtrace_to_file(tid, fileno(stdout)) < 0) {
899 fputs("Error dumping backtrace.\n", stderr);
900 return 1;
901 }
902 } else {
903 char tombstone_path[PATH_MAX];
904 if (dump_tombstone(tid, tombstone_path, sizeof(tombstone_path)) < 0) {
905 fputs("Error dumping tombstone.\n", stderr);
906 return 1;
907 }
908 fprintf(stderr, "Tombstone written to: %s\n", tombstone_path);
909 }
910 return 0;
911 }
912
usage()913 static void usage() {
914 fputs("Usage: -b [<tid>]\n"
915 " -b dump backtrace to console, otherwise dump full tombstone file\n"
916 "\n"
917 "If tid specified, sends a request to debuggerd to dump that task.\n"
918 "Otherwise, starts the debuggerd server.\n", stderr);
919 }
920
main(int argc,char ** argv)921 int main(int argc, char** argv) {
922 union selinux_callback cb;
923 if (argc == 1) {
924 cb.func_audit = audit_callback;
925 selinux_set_callback(SELINUX_CB_AUDIT, cb);
926 cb.func_log = selinux_log_callback;
927 selinux_set_callback(SELINUX_CB_LOG, cb);
928 return do_server();
929 }
930
931 bool dump_backtrace = false;
932 bool have_tid = false;
933 pid_t tid = 0;
934 for (int i = 1; i < argc; i++) {
935 if (!strcmp(argv[i], "-b")) {
936 dump_backtrace = true;
937 } else if (!have_tid) {
938 tid = atoi(argv[i]);
939 have_tid = true;
940 } else {
941 usage();
942 return 1;
943 }
944 }
945 if (!have_tid) {
946 usage();
947 return 1;
948 }
949 return do_explicit_dump(tid, dump_backtrace);
950 }
951