citadel/ecrash.c

   1 /*
   2  * author: David Frascone
   3  *
   4  * eCrash Implementation
   5  *
   6  * eCrash will allow you to capture stack traces in the
   7  * event of a crash, and write those traces to disk, stdout,
   8  * or any other file handle.
   9  *
  10  * modified to integrate closer into citadel by Wilfried Goesgens
  11  *
  12  * vim: ts=4
  13  *
  14  * This program is free software; you can redistribute it and/or modify
  15  * it under the terms of the GNU General Public License as published by
  16  * the Free Software Foundation; either version 3 of the License, or
  17  * (at your option) any later version.
  18  *
  19  * This program is distributed in the hope that it will be useful,
  20  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  * GNU General Public License for more details.
  23  *
  24  * You should have received a copy of the GNU General Public License
  25  * along with this program; if not, write to the Free Software
  26  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  27  */
  28
  29 #include "sysdep.h"
  30 #include <stdio.h>
  31 #include <unistd.h>
  32 #include <stdlib.h>
  33 #include <stdarg.h>
  34 #include <string.h>
  35 #include <fcntl.h>
  36 #include <syslog.h>
  37 #include <sys/types.h>
  38 #include <sys/stat.h>
  39 #include <pthread.h>
  40 #include <libcitadel.h>
  41 #include "server.h"
  42 #include "sysdep_decls.h"
  43 #include "support.h"
  44 #include "config.h"
  45 #include "citserver.h"
  46 #include "ecrash.h"
  47
  48 #define NIY()   printf("%s: Not Implemented Yet!\n", __FUNCTION__)
  49 #ifdef HAVE_BACKTRACE
  50 #include <execinfo.h>
  51 static eCrashParameters gbl_params;
  52
  53 static int    gbl_backtraceEntries;
  54 static void **gbl_backtraceBuffer;
  55 static char **gbl_backtraceSymbols;
  56 static int    gbl_backtraceDoneFlag = 0;
  57
  58 static void *stack_frames[50];
  59 static size_t size, NThread;
  60 static char **strings;
  61 static char StaticBuf[SIZ];
  62
  63 /*
  64  * Private structures for our thread list
  65  */
  66 typedef struct thread_list_node{
  67         char *threadName;
  68         pthread_t thread;
  69         int backtraceSignal;
  70         sighandler_t oldHandler;
  71         struct thread_list_node *Next;
  72 } ThreadListNode;
  73
  74 static pthread_mutex_t ThreadListMutex = PTHREAD_MUTEX_INITIALIZER;
  75 static ThreadListNode *ThreadList = NULL;
  76
  77 /*********************************************************************
  78  *********************************************************************
  79  **     P  R  I  V  A  T  E      F  U  N  C  T  I  O  N  S
  80  *********************************************************************
  81  ********************************************************************/
  82
  83
  84 /*!
  85  * Insert a node into our threadList
  86  *
  87  * @param name   Text string indicating our thread
  88  * @param thread Our Thread Id
  89  * @param signo  Signal to create backtrace with
  90  * @param old_handler Our old handler for signo
  91  *
  92  * @returns zero on success
  93  */
  94 static int addThreadToList(char *name, pthread_t thread,int signo,
  95                                            sighandler_t old_handler)
  96 {
  97         ThreadListNode *node;
  98
  99         node = malloc(sizeof(ThreadListNode));
 100         if (!node) return -1;
 101
 102         DPRINTF(ECRASH_DEBUG_VERBOSE,
 103                                         "Adding thread 0x%08x (%s)\n", (unsigned int)thread, name);
 104         node->threadName = strdup(name);
 105         node->thread = thread;
 106         node->backtraceSignal = signo;
 107         node->oldHandler = old_handler;
 108
 109         /* And, add it to the list */
 110         pthread_mutex_lock(&ThreadListMutex);
 111         node->Next = ThreadList;
 112         ThreadList = node;
 113         pthread_mutex_unlock(&ThreadListMutex);
 114
 115         return 0;
 116
 117 } // addThreadToList
 118
 119 /*!
 120  * Remove a node from our threadList
 121  *
 122  * @param thread Our Thread Id
 123  *
 124  * @returns zero on success
 125  */
 126 static int removeThreadFromList(pthread_t thread)
 127 {
 128         ThreadListNode *Probe, *Prev=NULL;
 129         ThreadListNode *Removed = NULL;
 130
 131         DPRINTF(ECRASH_DEBUG_VERBOSE,
 132                                         "Removing thread 0x%08x from list . . .\n", (unsigned int)thread);
 133         pthread_mutex_lock(&ThreadListMutex);
 134         for (Probe=ThreadList;Probe != NULL; Probe = Probe->Next) {
 135                 if (Probe->thread == thread) {
 136                         // We found it!  Unlink it and move on!
 137                         Removed = Probe;
 138                         if (Prev == NULL) { // head of list
 139                                 ThreadList = Probe->Next;
 140                         } else {
 141                                 // Prev != null, so we need to link around ourselves.
 142                                 Prev->Next = Probe->Next;
 143                         }
 144                         Removed->Next = NULL;
 145                         break;
 146                 }
 147
 148                 Prev = Probe;
 149         }
 150         pthread_mutex_unlock(&ThreadListMutex);
 151
 152         // Now, if something is in Removed, free it, and return success
 153         if (Removed) {
 154             DPRINTF(ECRASH_DEBUG_VERBOSE,
 155                                                 "   Found %s -- removing\n", Removed->threadName);
 156                 // Reset the signal handler
 157                 signal(Removed->backtraceSignal, Removed->oldHandler);
 158
 159                 // And free the allocated memory
 160                 free (Removed->threadName);
 161                 free (Removed);
 162
 163                 return 0;
 164         } else {
 165             DPRINTF(ECRASH_DEBUG_VERBOSE,
 166                                                 "   Not Found\n");
 167                 return -1; // Not Found
 168         }
 169 } // removeThreadFromList
 170
 171 /*!
 172  * Print out a line of output to all our destinations
 173  *
 174  * One by one, output a line of text to all of our output destinations.
 175  *
 176  * Return failure if we fail to output to any of them.
 177  *
 178  * @param format   Normal printf style vararg format
 179  *
 180  * @returns nothing// bytes written, or error on failure.
 181  */
 182 static void outputPrintf(char *format, ...)
 183 {
 184         va_list ap;
 185
 186         va_start(ap, format);
 187
 188         if (enable_syslog)
 189         {
 190                 snprintf (StaticBuf, SIZ, format, ap);
 191                 syslog( LOG_CRIT|LOG_NDELAY|LOG_MAIL, StaticBuf);
 192         }
 193         else
 194                 CtdlLogPrintf(CTDL_EMERG, format, ap);
 195
 196 } // outputPrintf
 197
 198
 199
 200 /*!
 201  * Dump our backtrace into a global location
 202  *
 203  * This function will dump out our backtrace into our
 204  * global holding area.
 205  *
 206  */
 207 static void createGlobalBacktrace( void )
 208 {
 209
 210         size = backtrace(stack_frames, sizeof(stack_frames) / sizeof(void*));
 211         if (enable_syslog)
 212                 for (NThread = 0; NThread < size; NThread++)
 213                 {
 214                         snprintf (StaticBuf, SIZ, "RAW: %p  ", stack_frames[NThread]);
 215                         syslog( LOG_CRIT|LOG_NDELAY|LOG_MAIL, StaticBuf);
 216                 }
 217         else
 218                 for (NThread = 0; NThread < size; NThread++)
 219                         CtdlLogPrintf(1, "RAW: %p\n", stack_frames[NThread]);
 220         strings = backtrace_symbols(stack_frames, size);
 221         for (NThread = 0; NThread < size; NThread++) {
 222                 if (strings != NULL) {
 223                         if (enable_syslog)
 224                         {// vsyslogs printf compliance sucks.
 225                                 snprintf (StaticBuf, SIZ, "RAW: %p  ", strings[NThread]);
 226                                 syslog( LOG_CRIT|LOG_NDELAY|LOG_MAIL, StaticBuf);
 227                         }
 228                         else
 229                                 CtdlLogPrintf(1, "%s\n", strings[NThread]);
 230                 }
 231         }
 232 } /* createGlobalBacktrace */
 233 static void outputRawtrace( void )
 234 {
 235
 236         size = backtrace(stack_frames, sizeof(stack_frames) / sizeof(void*));
 237         if (enable_syslog)
 238                 for (NThread = 0; NThread < size; NThread++)
 239                 {
 240                         snprintf (StaticBuf, SIZ, "RAW: %p  ", stack_frames[NThread]);
 241                         syslog( LOG_CRIT|LOG_NDELAY|LOG_MAIL, StaticBuf);
 242                 }
 243         else
 244                 for (NThread = 0; NThread < size; NThread++)
 245                         CtdlLogPrintf(1, "RAW: %p\n", stack_frames[NThread]);
 246 } /* createGlobalBacktrace */
 247
 248 /*!
 249  * Print out (to all the fds, etc), or global backtrace
 250  */
 251 static void outputGlobalBacktrace ( void )
 252 {
 253         int i;
 254
 255         for (i=0; i < gbl_backtraceEntries; i++) {
 256                 if (gbl_backtraceSymbols != FALSE) {
 257                         outputPrintf("*      Frame %02x: %s\n",
 258                                      i, gbl_backtraceSymbols[i]);
 259                 } else {
 260                         outputPrintf("*      Frame %02x: %p\n", i,
 261                                      gbl_backtraceBuffer[i]);
 262                 }
 263         }
 264 } // outputGlobalBacktrace
 265
 266 /*!
 267  * Output our current stack's backtrace
 268  */
 269 static void outputBacktrace( void )
 270 {
 271         createGlobalBacktrace();
 272         outputGlobalBacktrace();
 273 } /* outputBacktrace */
 274
 275 static void outputBacktraceThreads( void )
 276 {
 277         ThreadListNode *probe;
 278         int i;
 279
 280         // When we're backtracing, don't worry about the mutex . . hopefully
 281         // we're in a safe place.
 282
 283         for (probe=ThreadList; probe; probe=probe->Next) {
 284                 gbl_backtraceDoneFlag = 0;
 285                 pthread_kill(probe->thread, probe->backtraceSignal);
 286                 for (i=0; i < gbl_params.threadWaitTime; i++) {
 287                         if (gbl_backtraceDoneFlag)
 288                                 break;
 289                         sleep(1);
 290                 }
 291                 if (gbl_backtraceDoneFlag) {
 292                         outputPrintf("*  Backtrace of \"%s\" (0x%08x)\n",
 293                                                  probe->threadName, (unsigned int)probe->thread);
 294                         outputGlobalBacktrace();
 295                 } else {
 296                         outputPrintf("*  Error: unable to get backtrace of \"%s\" (0x%08x)\n",
 297                                                  probe->threadName, (unsigned int)probe->thread);
 298                 }
 299                 outputPrintf("*\n");
 300         }
 301 } // outputBacktraceThreads
 302
 303
 304 /*!
 305  * Handle signals (crash signals)
 306  *
 307  * This function will catch all crash signals, and will output the
 308  * crash dump.
 309  *
 310  * It will physically write (and sync) the current thread's information
 311  * before it attempts to send signals to other threads.
 312  *
 313  * @param signum Signal received.
 314  */
 315 static void crash_handler(int signo)
 316 {
 317         outputRawtrace();
 318         outputPrintf("*********************************************************\n");
 319         outputPrintf("*               eCrash Crash Handler\n");
 320         outputPrintf("*********************************************************\n");
 321         outputPrintf("*\n");
 322         outputPrintf("*  Got a crash! signo=%d\n", signo);
 323         outputPrintf("*\n");
 324         outputPrintf("*  Offending Thread's Backtrace:\n");
 325         outputPrintf("*\n");
 326         outputBacktrace();
 327         outputPrintf("*\n");
 328
 329         if (gbl_params.dumpAllThreads != FALSE) {
 330                 outputBacktraceThreads();
 331         }
 332
 333         outputPrintf("*\n");
 334         outputPrintf("*********************************************************\n");
 335         outputPrintf("*               eCrash Crash Handler\n");
 336         outputPrintf("*********************************************************\n");
 337
 338         exit(signo);
 339 } // crash_handler
 340
 341 /*!
 342  * Handle signals (bt signals)
 343  *
 344  * This function shoudl be called to generate a crashdump into our
 345  * global area.  Once the dump has been completed, this function will
 346  * return after tickling a global.  Since mutexes are not async
 347  * signal safe, the main thread, after signaling us to generate our
 348  * own backtrace, will sleep for a few seconds waiting for us to complete.
 349  *
 350  * @param signum Signal received.
 351  */
 352 static void bt_handler(int signo)
 353 {
 354         createGlobalBacktrace();
 355         gbl_backtraceDoneFlag=1;
 356 } // bt_handler
 357
 358 /*!
 359  * Validate a passed-in symbol table
 360  *
 361  * For now, just print it out (if verbose), and make sure it's
 362  * sorted and none of the pointers are zero.
 363  */
 364 static int ValidateSymbolTable( void )
 365 {
 366         int i;
 367         int rc=0;
 368         unsigned long lastAddress =0;
 369
 370         // Get out of here if the table is empty
 371         if (!gbl_params.symbolTable) return 0;
 372
 373         // Dump it in verbose mode
 374         DPRINTF(ECRASH_DEBUG_VERBOSE,
 375                                         "Symbol Table Provided with %d symbols\n",
 376                                         gbl_params.symbolTable->numSymbols);
 377         for (i=0; i < gbl_params.symbolTable->numSymbols; i++){
 378                 // Dump it in verbose mode
 379                 DPRINTF(ECRASH_DEBUG_VERBOSE,
 380                                 "%-30s %p\n",
 381                                 gbl_params.symbolTable->symbols[i].function,
 382                                 gbl_params.symbolTable->symbols[i].address);
 383                 if (lastAddress >
 384                     (unsigned long)gbl_params.symbolTable->symbols[i].address) {
 385                         DPRINTF(ECRASH_DEBUG_ERROR,
 386                                         "Error: symbol table is not sorted (last=%p, current=%p)\n",
 387                                         (void *)lastAddress,
 388                                         gbl_params.symbolTable->symbols[i].address);
 389                         rc = -1;
 390                 }
 391
 392         } // for
 393
 394         return rc;
 395
 396 } // ValidateSymbolTable
 397
 398 /*********************************************************************
 399  *********************************************************************
 400  **      P  U  B  L  I  C      F  U  N  C  T  I  O  N  S
 401  *********************************************************************
 402  ********************************************************************/
 403
 404 /*!
 405  * Initialize eCrash.
 406  *
 407  * This function must be called before calling any other eCrash
 408  * functions.  It sets up the global behavior of the system, and
 409  * registers the calling thread for crash dumps.
 410  *
 411  * @param params Our input parameters.  The passed in structure will be copied.
 412  *
 413  * @return Zero on success.
 414  */
 415 int eCrash_Init(eCrashParameters *params)
 416 {
 417         int sigIndex;
 418         int ret = 0;
 419 #ifdef DO_SIGNALS_RIGHT
 420         sigset_t blocked;
 421         struct sigaction act;
 422 #endif
 423
 424         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE,"Init Starting params = %p\n", params);
 425
 426         // Allocate our backtrace area
 427         gbl_backtraceBuffer = malloc(sizeof(void *) * (params->maxStackDepth+5));
 428
 429 #ifdef DO_SIGNALS_RIGHT
 430         sigemptyset(&blocked);
 431         act.sa_sigaction = crash_handler;
 432         act.sa_mask = blocked;
 433         act.sa_flags = SA_SIGINFO;
 434 #endif
 435
 436         if (params != NULL) {
 437                 // Make ourselves a global copy of params.
 438                 gbl_params = *params;
 439                 gbl_params.filename = strdup(params->filename);
 440
 441                 // Set our defaults, if they weren't specified
 442                 if (gbl_params.maxStackDepth == 0 )
 443                         gbl_params.maxStackDepth = ECRASH_DEFAULT_STACK_DEPTH;
 444
 445                 if (gbl_params.defaultBacktraceSignal == 0 )
 446                         gbl_params.defaultBacktraceSignal = ECRASH_DEFAULT_BACKTRACE_SIGNAL;
 447
 448                 if (gbl_params.threadWaitTime == 0 )
 449                         gbl_params.threadWaitTime = ECRASH_DEFAULT_THREAD_WAIT_TIME;
 450
 451                 if (gbl_params.debugLevel == 0 )
 452                         gbl_params.debugLevel = ECRASH_DEBUG_DEFAULT;
 453
 454                 // Copy our symbol table
 455                 if (gbl_params.symbolTable) {
 456                     DPRINTF(ECRASH_DEBUG_VERBOSE,
 457                                                         "symbolTable @ %p -- %d symbols\n", gbl_params.symbolTable,
 458                                                 gbl_params.symbolTable->numSymbols);
 459                         // Make a copy of our symbol table
 460                         gbl_params.symbolTable = malloc(sizeof(eCrashSymbolTable));
 461                         memcpy(gbl_params.symbolTable, params->symbolTable,
 462                                    sizeof(eCrashSymbolTable));
 463
 464                         // Now allocate / copy the actual table.
 465                         gbl_params.symbolTable->symbols = malloc(sizeof(eCrashSymbol) *
 466                                                                      gbl_params.symbolTable->numSymbols);
 467                         memcpy(gbl_params.symbolTable->symbols,
 468                                    params->symbolTable->symbols,
 469                                    sizeof(eCrashSymbol) * gbl_params.symbolTable->numSymbols);
 470
 471                         ValidateSymbolTable();
 472                 }
 473
 474                 // And, finally, register for our signals
 475                 for (sigIndex=0; gbl_params.signals[sigIndex] != 0; sigIndex++) {
 476                         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE,
 477                                                         "   Catching signal[%d] %d\n", sigIndex,
 478                                         gbl_params.signals[sigIndex]);
 479
 480                         // I know there's a better way to catch signals with pthreads.
 481                         // I'll do it later TODO
 482                         signal(gbl_params.signals[sigIndex], crash_handler);
 483                 }
 484         } else {
 485                 DPRINTF(ECRASH_DEBUG_ERROR, "   Error:  Null Params!\n");
 486                 ret = -1;
 487         }
 488         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE, "Init Complete ret=%d\n", ret);
 489         return ret;
 490 } /* eCrash_Init */
 491
 492 /*!
 493  * UnInitialize eCrash.
 494  *
 495  * This function may be called to de-activate eCrash, release the
 496  * signal handlers, and free any memory allocated by eCrash.
 497  *
 498  * @return Zero on success.
 499  */
 500 int eCrash_Uninit( void )
 501 {
 502         NIY();
 503
 504         return 0;
 505 } /* eCrash_Uninit */
 506
 507 /*!
 508  * Register a thread for backtracing on crash.
 509  *
 510  * This function must be called by any thread wanting it's stack
 511  * dumped in the event of a crash.  The thread my specify what
 512  * signal should be used, or the default, SIGUSR1 will be used.
 513  *
 514  * @param signo Signal to use to generate dump (default: SIGUSR1)
 515  *
 516  * @return Zero on success.
 517  */
 518 int eCrash_RegisterThread(char *name, int signo)
 519 {
 520         sighandler_t old_handler;
 521
 522         // Register for our signal
 523         if (signo == 0) {
 524                 signo = gbl_params.defaultBacktraceSignal;
 525         }
 526
 527         old_handler = signal(signo, bt_handler);
 528         return addThreadToList(name, pthread_self(), signo, old_handler);
 529
 530 } /* eCrash_RegisterThread */
 531
 532 /*!
 533  * Un-register a thread for stack dumps.
 534  *
 535  * This function may be called to un-register any previously
 536  * registered thread.
 537  *
 538  * @return Zero on success.
 539  */
 540 int eCrash_UnregisterThread( void )
 541 {
 542         return removeThreadFromList(pthread_self());
 543 } /* eCrash_UnregisterThread */
 544
 545 #endif