citadel/ecrash.c

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