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 <stdio.h> 18#include <errno.h> 19#include <signal.h> 20#include <pthread.h> 21#include <stdarg.h> 22#include <fcntl.h> 23#include <sys/types.h> 24#include <dirent.h> 25#include <time.h> 26 27#include <sys/ptrace.h> 28#include <sys/wait.h> 29#include <elf.h> 30#include <sys/stat.h> 31#include <sys/poll.h> 32 33#include <selinux/android.h> 34 35#include <log/logger.h> 36 37#include <cutils/sockets.h> 38#include <cutils/properties.h> 39#include <cutils/debugger.h> 40 41#include <linux/input.h> 42 43#include <private/android_filesystem_config.h> 44 45#include "backtrace.h" 46#include "getevent.h" 47#include "tombstone.h" 48#include "utility.h" 49 50// If the 32 bit executable is compiled on a 64 bit system, 51// use the 32 bit socket name. 52#if defined(TARGET_IS_64_BIT) && !defined(__LP64__) 53#define SOCKET_NAME DEBUGGER32_SOCKET_NAME 54#else 55#define SOCKET_NAME DEBUGGER_SOCKET_NAME 56#endif 57 58struct debugger_request_t { 59 debugger_action_t action; 60 pid_t pid, tid; 61 uid_t uid, gid; 62 uintptr_t abort_msg_address; 63 int32_t original_si_code; 64}; 65 66static void wait_for_user_action(const debugger_request_t &request) { 67 // Find out the name of the process that crashed. 68 char path[64]; 69 snprintf(path, sizeof(path), "/proc/%d/exe", request.pid); 70 71 char exe[PATH_MAX]; 72 int count; 73 if ((count = readlink(path, exe, sizeof(exe) - 1)) == -1) { 74 ALOGE("readlink('%s') failed: %s", path, strerror(errno)); 75 strlcpy(exe, "unknown", sizeof(exe)); 76 } else { 77 exe[count] = '\0'; 78 } 79 80 // Explain how to attach the debugger. 81 ALOGI("********************************************************\n" 82 "* Process %d has been suspended while crashing.\n" 83 "* To attach gdbserver for a gdb connection on port 5039\n" 84 "* and start gdbclient:\n" 85 "*\n" 86 "* gdbclient %s :5039 %d\n" 87 "*\n" 88 "* Wait for gdb to start, then press the VOLUME DOWN key\n" 89 "* to let the process continue crashing.\n" 90 "********************************************************", 91 request.pid, exe, request.tid); 92 93 // Wait for VOLUME DOWN. 94 if (init_getevent() == 0) { 95 while (true) { 96 input_event e; 97 if (get_event(&e, -1) == 0) { 98 if (e.type == EV_KEY && e.code == KEY_VOLUMEDOWN && e.value == 0) { 99 break; 100 } 101 } 102 } 103 uninit_getevent(); 104 } 105 106 ALOGI("debuggerd resuming process %d", request.pid); 107} 108 109static int get_process_info(pid_t tid, pid_t* out_pid, uid_t* out_uid, uid_t* out_gid) { 110 char path[64]; 111 snprintf(path, sizeof(path), "/proc/%d/status", tid); 112 113 FILE* fp = fopen(path, "r"); 114 if (!fp) { 115 return -1; 116 } 117 118 int fields = 0; 119 char line[1024]; 120 while (fgets(line, sizeof(line), fp)) { 121 size_t len = strlen(line); 122 if (len > 6 && !memcmp(line, "Tgid:\t", 6)) { 123 *out_pid = atoi(line + 6); 124 fields |= 1; 125 } else if (len > 5 && !memcmp(line, "Uid:\t", 5)) { 126 *out_uid = atoi(line + 5); 127 fields |= 2; 128 } else if (len > 5 && !memcmp(line, "Gid:\t", 5)) { 129 *out_gid = atoi(line + 5); 130 fields |= 4; 131 } 132 } 133 fclose(fp); 134 return fields == 7 ? 0 : -1; 135} 136 137static int selinux_enabled; 138 139/* 140 * Corresponds with debugger_action_t enum type in 141 * include/cutils/debugger.h. 142 */ 143static const char *debuggerd_perms[] = { 144 NULL, /* crash is only used on self, no check applied */ 145 "dump_tombstone", 146 "dump_backtrace" 147}; 148 149static bool selinux_action_allowed(int s, pid_t tid, debugger_action_t action) 150{ 151 char *scon = NULL, *tcon = NULL; 152 const char *tclass = "debuggerd"; 153 const char *perm; 154 bool allowed = false; 155 156 if (selinux_enabled <= 0) 157 return true; 158 159 if (action <= 0 || action >= (sizeof(debuggerd_perms)/sizeof(debuggerd_perms[0]))) { 160 ALOGE("SELinux: No permission defined for debugger action %d", action); 161 return false; 162 } 163 164 perm = debuggerd_perms[action]; 165 166 if (getpeercon(s, &scon) < 0) { 167 ALOGE("Cannot get peer context from socket\n"); 168 goto out; 169 } 170 171 if (getpidcon(tid, &tcon) < 0) { 172 ALOGE("Cannot get context for tid %d\n", tid); 173 goto out; 174 } 175 176 allowed = (selinux_check_access(scon, tcon, tclass, perm, NULL) == 0); 177 178out: 179 freecon(scon); 180 freecon(tcon); 181 return allowed; 182} 183 184static int read_request(int fd, debugger_request_t* out_request) { 185 ucred cr; 186 socklen_t len = sizeof(cr); 187 int status = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &cr, &len); 188 if (status != 0) { 189 ALOGE("cannot get credentials"); 190 return -1; 191 } 192 193 ALOGV("reading tid"); 194 fcntl(fd, F_SETFL, O_NONBLOCK); 195 196 pollfd pollfds[1]; 197 pollfds[0].fd = fd; 198 pollfds[0].events = POLLIN; 199 pollfds[0].revents = 0; 200 status = TEMP_FAILURE_RETRY(poll(pollfds, 1, 3000)); 201 if (status != 1) { 202 ALOGE("timed out reading tid (from pid=%d uid=%d)\n", cr.pid, cr.uid); 203 return -1; 204 } 205 206 debugger_msg_t msg; 207 memset(&msg, 0, sizeof(msg)); 208 status = TEMP_FAILURE_RETRY(read(fd, &msg, sizeof(msg))); 209 if (status < 0) { 210 ALOGE("read failure? %s (pid=%d uid=%d)\n", strerror(errno), cr.pid, cr.uid); 211 return -1; 212 } 213 if (status != sizeof(debugger_msg_t)) { 214 ALOGE("invalid crash request of size %d (from pid=%d uid=%d)\n", status, cr.pid, cr.uid); 215 return -1; 216 } 217 218 out_request->action = static_cast<debugger_action_t>(msg.action); 219 out_request->tid = msg.tid; 220 out_request->pid = cr.pid; 221 out_request->uid = cr.uid; 222 out_request->gid = cr.gid; 223 out_request->abort_msg_address = msg.abort_msg_address; 224 out_request->original_si_code = msg.original_si_code; 225 226 if (msg.action == DEBUGGER_ACTION_CRASH) { 227 // Ensure that the tid reported by the crashing process is valid. 228 char buf[64]; 229 struct stat s; 230 snprintf(buf, sizeof buf, "/proc/%d/task/%d", out_request->pid, out_request->tid); 231 if (stat(buf, &s)) { 232 ALOGE("tid %d does not exist in pid %d. ignoring debug request\n", 233 out_request->tid, out_request->pid); 234 return -1; 235 } 236 } else if (cr.uid == 0 237 || (cr.uid == AID_SYSTEM && msg.action == DEBUGGER_ACTION_DUMP_BACKTRACE)) { 238 // Only root or system can ask us to attach to any process and dump it explicitly. 239 // However, system is only allowed to collect backtraces but cannot dump tombstones. 240 status = get_process_info(out_request->tid, &out_request->pid, 241 &out_request->uid, &out_request->gid); 242 if (status < 0) { 243 ALOGE("tid %d does not exist. ignoring explicit dump request\n", out_request->tid); 244 return -1; 245 } 246 247 if (!selinux_action_allowed(fd, out_request->tid, out_request->action)) 248 return -1; 249 } else { 250 // No one else is allowed to dump arbitrary processes. 251 return -1; 252 } 253 return 0; 254} 255 256static bool should_attach_gdb(debugger_request_t* request) { 257 if (request->action == DEBUGGER_ACTION_CRASH) { 258 char value[PROPERTY_VALUE_MAX]; 259 property_get("debug.db.uid", value, "-1"); 260 int debug_uid = atoi(value); 261 return debug_uid >= 0 && request->uid <= (uid_t)debug_uid; 262 } 263 return false; 264} 265 266#if defined(__LP64__) 267static bool is32bit(pid_t tid) { 268 char* exeline; 269 if (asprintf(&exeline, "/proc/%d/exe", tid) == -1) { 270 return false; 271 } 272 int fd = TEMP_FAILURE_RETRY(open(exeline, O_RDONLY | O_CLOEXEC)); 273 int saved_errno = errno; 274 free(exeline); 275 if (fd == -1) { 276 ALOGW("Failed to open /proc/%d/exe %s", tid, strerror(saved_errno)); 277 return false; 278 } 279 280 char ehdr[EI_NIDENT]; 281 ssize_t bytes = TEMP_FAILURE_RETRY(read(fd, &ehdr, sizeof(ehdr))); 282 close(fd); 283 if (bytes != (ssize_t) sizeof(ehdr) || memcmp(ELFMAG, ehdr, SELFMAG) != 0) { 284 return false; 285 } 286 if (ehdr[EI_CLASS] == ELFCLASS32) { 287 return true; 288 } 289 return false; 290} 291 292static void redirect_to_32(int fd, debugger_request_t* request) { 293 debugger_msg_t msg; 294 memset(&msg, 0, sizeof(msg)); 295 msg.tid = request->tid; 296 msg.action = request->action; 297 298 int sock_fd = socket_local_client(DEBUGGER32_SOCKET_NAME, ANDROID_SOCKET_NAMESPACE_ABSTRACT, 299 SOCK_STREAM | SOCK_CLOEXEC); 300 if (sock_fd < 0) { 301 ALOGE("Failed to connect to debuggerd32: %s", strerror(errno)); 302 return; 303 } 304 305 if (TEMP_FAILURE_RETRY(write(sock_fd, &msg, sizeof(msg))) != (ssize_t) sizeof(msg)) { 306 ALOGE("Failed to write request to debuggerd32 socket: %s", strerror(errno)); 307 close(sock_fd); 308 return; 309 } 310 311 char ack; 312 if (TEMP_FAILURE_RETRY(read(sock_fd, &ack, 1)) == -1) { 313 ALOGE("Failed to read ack from debuggerd32 socket: %s", strerror(errno)); 314 close(sock_fd); 315 return; 316 } 317 318 char buffer[1024]; 319 ssize_t bytes_read; 320 while ((bytes_read = TEMP_FAILURE_RETRY(read(sock_fd, buffer, sizeof(buffer)))) > 0) { 321 ssize_t bytes_to_send = bytes_read; 322 ssize_t bytes_written; 323 do { 324 bytes_written = TEMP_FAILURE_RETRY(write(fd, buffer + bytes_read - bytes_to_send, 325 bytes_to_send)); 326 if (bytes_written == -1) { 327 if (errno == EAGAIN) { 328 // Retry the write. 329 continue; 330 } 331 ALOGE("Error while writing data to fd: %s", strerror(errno)); 332 break; 333 } 334 bytes_to_send -= bytes_written; 335 } while (bytes_written != 0 && bytes_to_send > 0); 336 if (bytes_to_send != 0) { 337 ALOGE("Failed to write all data to fd: read %zd, sent %zd", bytes_read, bytes_to_send); 338 break; 339 } 340 } 341 close(sock_fd); 342} 343#endif 344 345static void handle_request(int fd) { 346 ALOGV("handle_request(%d)\n", fd); 347 348 debugger_request_t request; 349 memset(&request, 0, sizeof(request)); 350 int status = read_request(fd, &request); 351 if (!status) { 352 ALOGV("BOOM: pid=%d uid=%d gid=%d tid=%d\n", 353 request.pid, request.uid, request.gid, request.tid); 354 355#if defined(__LP64__) 356 // On 64 bit systems, requests to dump 32 bit and 64 bit tids come 357 // to the 64 bit debuggerd. If the process is a 32 bit executable, 358 // redirect the request to the 32 bit debuggerd. 359 if (is32bit(request.tid)) { 360 // Only dump backtrace and dump tombstone requests can be redirected. 361 if (request.action == DEBUGGER_ACTION_DUMP_BACKTRACE 362 || request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) { 363 redirect_to_32(fd, &request); 364 } else { 365 ALOGE("debuggerd: Not allowed to redirect action %d to 32 bit debuggerd\n", 366 request.action); 367 } 368 close(fd); 369 return; 370 } 371#endif 372 373 // At this point, the thread that made the request is blocked in 374 // a read() call. If the thread has crashed, then this gives us 375 // time to PTRACE_ATTACH to it before it has a chance to really fault. 376 // 377 // The PTRACE_ATTACH sends a SIGSTOP to the target process, but it 378 // won't necessarily have stopped by the time ptrace() returns. (We 379 // currently assume it does.) We write to the file descriptor to 380 // ensure that it can run as soon as we call PTRACE_CONT below. 381 // See details in bionic/libc/linker/debugger.c, in function 382 // debugger_signal_handler(). 383 if (ptrace(PTRACE_ATTACH, request.tid, 0, 0)) { 384 ALOGE("ptrace attach failed: %s\n", strerror(errno)); 385 } else { 386 bool detach_failed = false; 387 bool tid_unresponsive = false; 388 bool attach_gdb = should_attach_gdb(&request); 389 if (TEMP_FAILURE_RETRY(write(fd, "\0", 1)) != 1) { 390 ALOGE("failed responding to client: %s\n", strerror(errno)); 391 } else { 392 char* tombstone_path = NULL; 393 394 if (request.action == DEBUGGER_ACTION_CRASH) { 395 close(fd); 396 fd = -1; 397 } 398 399 int total_sleep_time_usec = 0; 400 for (;;) { 401 int signal = wait_for_sigstop(request.tid, &total_sleep_time_usec, &detach_failed); 402 if (signal == -1) { 403 tid_unresponsive = true; 404 break; 405 } 406 407 switch (signal) { 408 case SIGSTOP: 409 if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) { 410 ALOGV("stopped -- dumping to tombstone\n"); 411 tombstone_path = engrave_tombstone(request.pid, request.tid, 412 signal, request.original_si_code, 413 request.abort_msg_address, true, 414 &detach_failed, &total_sleep_time_usec); 415 } else if (request.action == DEBUGGER_ACTION_DUMP_BACKTRACE) { 416 ALOGV("stopped -- dumping to fd\n"); 417 dump_backtrace(fd, -1, request.pid, request.tid, &detach_failed, 418 &total_sleep_time_usec); 419 } else { 420 ALOGV("stopped -- continuing\n"); 421 status = ptrace(PTRACE_CONT, request.tid, 0, 0); 422 if (status) { 423 ALOGE("ptrace continue failed: %s\n", strerror(errno)); 424 } 425 continue; // loop again 426 } 427 break; 428 429 case SIGABRT: 430 case SIGBUS: 431 case SIGFPE: 432 case SIGILL: 433 case SIGPIPE: 434 case SIGSEGV: 435#ifdef SIGSTKFLT 436 case SIGSTKFLT: 437#endif 438 case SIGTRAP: 439 ALOGV("stopped -- fatal signal\n"); 440 // Send a SIGSTOP to the process to make all of 441 // the non-signaled threads stop moving. Without 442 // this we get a lot of "ptrace detach failed: 443 // No such process". 444 kill(request.pid, SIGSTOP); 445 // don't dump sibling threads when attaching to GDB because it 446 // makes the process less reliable, apparently... 447 tombstone_path = engrave_tombstone(request.pid, request.tid, 448 signal, request.original_si_code, 449 request.abort_msg_address, !attach_gdb, 450 &detach_failed, &total_sleep_time_usec); 451 break; 452 453 default: 454 ALOGE("process stopped due to unexpected signal %d\n", signal); 455 break; 456 } 457 break; 458 } 459 460 if (request.action == DEBUGGER_ACTION_DUMP_TOMBSTONE) { 461 if (tombstone_path) { 462 write(fd, tombstone_path, strlen(tombstone_path)); 463 } 464 close(fd); 465 fd = -1; 466 } 467 free(tombstone_path); 468 } 469 470 if (!tid_unresponsive) { 471 ALOGV("detaching"); 472 if (attach_gdb) { 473 // stop the process so we can debug 474 kill(request.pid, SIGSTOP); 475 } 476 if (ptrace(PTRACE_DETACH, request.tid, 0, 0)) { 477 ALOGE("ptrace detach from %d failed: %s", request.tid, strerror(errno)); 478 detach_failed = true; 479 } else if (attach_gdb) { 480 // if debug.db.uid is set, its value indicates if we should wait 481 // for user action for the crashing process. 482 // in this case, we log a message and turn the debug LED on 483 // waiting for a gdb connection (for instance) 484 wait_for_user_action(request); 485 } 486 } 487 488 // resume stopped process (so it can crash in peace). 489 kill(request.pid, SIGCONT); 490 491 // If we didn't successfully detach, we're still the parent, and the 492 // actual parent won't receive a death notification via wait(2). At this point 493 // there's not much we can do about that. 494 if (detach_failed) { 495 ALOGE("debuggerd committing suicide to free the zombie!\n"); 496 kill(getpid(), SIGKILL); 497 } 498 } 499 500 } 501 if (fd >= 0) { 502 close(fd); 503 } 504} 505 506static int do_server() { 507 // debuggerd crashes can't be reported to debuggerd. 508 // Reset all of the crash handlers. 509 signal(SIGABRT, SIG_DFL); 510 signal(SIGBUS, SIG_DFL); 511 signal(SIGFPE, SIG_DFL); 512 signal(SIGILL, SIG_DFL); 513 signal(SIGSEGV, SIG_DFL); 514#ifdef SIGSTKFLT 515 signal(SIGSTKFLT, SIG_DFL); 516#endif 517 signal(SIGTRAP, SIG_DFL); 518 519 // Ignore failed writes to closed sockets 520 signal(SIGPIPE, SIG_IGN); 521 522 int logsocket = socket_local_client("logd", ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_DGRAM); 523 if (logsocket < 0) { 524 logsocket = -1; 525 } else { 526 fcntl(logsocket, F_SETFD, FD_CLOEXEC); 527 } 528 529 struct sigaction act; 530 act.sa_handler = SIG_DFL; 531 sigemptyset(&act.sa_mask); 532 sigaddset(&act.sa_mask,SIGCHLD); 533 act.sa_flags = SA_NOCLDWAIT; 534 sigaction(SIGCHLD, &act, 0); 535 536 int s = socket_local_server(SOCKET_NAME, ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_STREAM); 537 if (s < 0) 538 return 1; 539 fcntl(s, F_SETFD, FD_CLOEXEC); 540 541 ALOGI("debuggerd: " __DATE__ " " __TIME__ "\n"); 542 543 for (;;) { 544 sockaddr addr; 545 socklen_t alen = sizeof(addr); 546 547 ALOGV("waiting for connection\n"); 548 int fd = accept(s, &addr, &alen); 549 if (fd < 0) { 550 ALOGV("accept failed: %s\n", strerror(errno)); 551 continue; 552 } 553 554 fcntl(fd, F_SETFD, FD_CLOEXEC); 555 556 handle_request(fd); 557 } 558 return 0; 559} 560 561static int do_explicit_dump(pid_t tid, bool dump_backtrace) { 562 fprintf(stdout, "Sending request to dump task %d.\n", tid); 563 564 if (dump_backtrace) { 565 fflush(stdout); 566 if (dump_backtrace_to_file(tid, fileno(stdout)) < 0) { 567 fputs("Error dumping backtrace.\n", stderr); 568 return 1; 569 } 570 } else { 571 char tombstone_path[PATH_MAX]; 572 if (dump_tombstone(tid, tombstone_path, sizeof(tombstone_path)) < 0) { 573 fputs("Error dumping tombstone.\n", stderr); 574 return 1; 575 } 576 fprintf(stderr, "Tombstone written to: %s\n", tombstone_path); 577 } 578 return 0; 579} 580 581static void usage() { 582 fputs("Usage: -b [<tid>]\n" 583 " -b dump backtrace to console, otherwise dump full tombstone file\n" 584 "\n" 585 "If tid specified, sends a request to debuggerd to dump that task.\n" 586 "Otherwise, starts the debuggerd server.\n", stderr); 587} 588 589int main(int argc, char** argv) { 590 union selinux_callback cb; 591 if (argc == 1) { 592 selinux_enabled = is_selinux_enabled(); 593 cb.func_log = selinux_log_callback; 594 selinux_set_callback(SELINUX_CB_LOG, cb); 595 return do_server(); 596 } 597 598 bool dump_backtrace = false; 599 bool have_tid = false; 600 pid_t tid = 0; 601 for (int i = 1; i < argc; i++) { 602 if (!strcmp(argv[i], "-b")) { 603 dump_backtrace = true; 604 } else if (!have_tid) { 605 tid = atoi(argv[i]); 606 have_tid = true; 607 } else { 608 usage(); 609 return 1; 610 } 611 } 612 if (!have_tid) { 613 usage(); 614 return 1; 615 } 616 return do_explicit_dump(tid, dump_backtrace); 617} 618