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         vsyslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, format, ap);
 189 } // outputPrintf
 190
 191
 192
 193 /*!
 194  * Dump our backtrace into a global location
 195  *
 196  * This function will dump out our backtrace into our
 197  * global holding area.
 198  *
 199  */
 200 static void createGlobalBacktrace( void )
 201 {
 202
 203         size = backtrace(stack_frames, sizeof(stack_frames) / sizeof(void*));
 204         for (NThread = 0; NThread < size; NThread++)
 205         {
 206                 syslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, "RAW: %p  ", stack_frames[NThread]);
 207         }
 208         strings = backtrace_symbols(stack_frames, size);
 209         for (NThread = 0; NThread < size; NThread++) {
 210                 if (strings != NULL) {
 211                         syslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, "RAW: %p  ", strings[NThread]);
 212                 }
 213         }
 214 } /* createGlobalBacktrace */
 215 static void outputRawtrace( void )
 216 {
 217
 218         size = backtrace(stack_frames, sizeof(stack_frames) / sizeof(void*));
 219         for (NThread = 0; NThread < size; NThread++)
 220         {
 221                 syslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, "RAW: %p  ", stack_frames[NThread]);
 222         }
 223 } /* createGlobalBacktrace */
 224
 225 /*!
 226  * Print out (to all the fds, etc), or global backtrace
 227  */
 228 static void outputGlobalBacktrace ( void )
 229 {
 230         int i;
 231
 232         for (i=0; i < gbl_backtraceEntries; i++) {
 233                 if (gbl_backtraceSymbols != FALSE) {
 234                         outputPrintf("*      Frame %02x: %s\n",
 235                                      i, gbl_backtraceSymbols[i]);
 236                 } else {
 237                         outputPrintf("*      Frame %02x: %p\n", i,
 238                                      gbl_backtraceBuffer[i]);
 239                 }
 240         }
 241 } // outputGlobalBacktrace
 242
 243 /*!
 244  * Output our current stack's backtrace
 245  */
 246 static void outputBacktrace( void )
 247 {
 248         createGlobalBacktrace();
 249         outputGlobalBacktrace();
 250 } /* outputBacktrace */
 251
 252 static void outputBacktraceThreads( void )
 253 {
 254         ThreadListNode *probe;
 255         int i;
 256
 257         // When we're backtracing, don't worry about the mutex . . hopefully
 258         // we're in a safe place.
 259
 260         for (probe=ThreadList; probe; probe=probe->Next) {
 261                 gbl_backtraceDoneFlag = 0;
 262                 pthread_kill(probe->thread, probe->backtraceSignal);
 263                 for (i=0; i < gbl_params.threadWaitTime; i++) {
 264                         if (gbl_backtraceDoneFlag)
 265                                 break;
 266                         sleep(1);
 267                 }
 268                 if (gbl_backtraceDoneFlag) {
 269                         outputPrintf("*  Backtrace of \"%s\" (0x%08x)\n",
 270                                                  probe->threadName, (unsigned int)probe->thread);
 271                         outputGlobalBacktrace();
 272                 } else {
 273                         outputPrintf("*  Error: unable to get backtrace of \"%s\" (0x%08x)\n",
 274                                                  probe->threadName, (unsigned int)probe->thread);
 275                 }
 276                 outputPrintf("*\n");
 277         }
 278 } // outputBacktraceThreads
 279
 280
 281 /*!
 282  * Handle signals (crash signals)
 283  *
 284  * This function will catch all crash signals, and will output the
 285  * crash dump.
 286  *
 287  * It will physically write (and sync) the current thread's information
 288  * before it attempts to send signals to other threads.
 289  *
 290  * @param signum Signal received.
 291  */
 292 static void crash_handler(int signo)
 293 {
 294         outputRawtrace();
 295         outputPrintf("*********************************************************\n");
 296         outputPrintf("*               eCrash Crash Handler\n");
 297         outputPrintf("*********************************************************\n");
 298         outputPrintf("*\n");
 299         outputPrintf("*  Got a crash! signo=%d\n", signo);
 300         outputPrintf("*\n");
 301         outputPrintf("*  Offending Thread's Backtrace:\n");
 302         outputPrintf("*\n");
 303         outputBacktrace();
 304         outputPrintf("*\n");
 305
 306         if (gbl_params.dumpAllThreads != FALSE) {
 307                 outputBacktraceThreads();
 308         }
 309
 310         outputPrintf("*\n");
 311         outputPrintf("*********************************************************\n");
 312         outputPrintf("*               eCrash Crash Handler\n");
 313         outputPrintf("*********************************************************\n");
 314
 315         exit(signo);
 316 } // crash_handler
 317
 318 /*!
 319  * Handle signals (bt signals)
 320  *
 321  * This function shoudl be called to generate a crashdump into our
 322  * global area.  Once the dump has been completed, this function will
 323  * return after tickling a global.  Since mutexes are not async
 324  * signal safe, the main thread, after signaling us to generate our
 325  * own backtrace, will sleep for a few seconds waiting for us to complete.
 326  *
 327  * @param signum Signal received.
 328  */
 329 static void bt_handler(int signo)
 330 {
 331         createGlobalBacktrace();
 332         gbl_backtraceDoneFlag=1;
 333 } // bt_handler
 334
 335 /*!
 336  * Validate a passed-in symbol table
 337  *
 338  * For now, just print it out (if verbose), and make sure it's
 339  * sorted and none of the pointers are zero.
 340  */
 341 static int ValidateSymbolTable( void )
 342 {
 343         int i;
 344         int rc=0;
 345         unsigned long lastAddress =0;
 346
 347         // Get out of here if the table is empty
 348         if (!gbl_params.symbolTable) return 0;
 349
 350         // Dump it in verbose mode
 351         DPRINTF(ECRASH_DEBUG_VERBOSE,
 352                                         "Symbol Table Provided with %d symbols\n",
 353                                         gbl_params.symbolTable->numSymbols);
 354         for (i=0; i < gbl_params.symbolTable->numSymbols; i++){
 355                 // Dump it in verbose mode
 356                 DPRINTF(ECRASH_DEBUG_VERBOSE,
 357                                 "%-30s %p\n",
 358                                 gbl_params.symbolTable->symbols[i].function,
 359                                 gbl_params.symbolTable->symbols[i].address);
 360                 if (lastAddress >
 361                     (unsigned long)gbl_params.symbolTable->symbols[i].address) {
 362                         DPRINTF(ECRASH_DEBUG_ERROR,
 363                                         "Error: symbol table is not sorted (last=%p, current=%p)\n",
 364                                         (void *)lastAddress,
 365                                         gbl_params.symbolTable->symbols[i].address);
 366                         rc = -1;
 367                 }
 368
 369         } // for
 370
 371         return rc;
 372
 373 } // ValidateSymbolTable
 374
 375 /*********************************************************************
 376  *********************************************************************
 377  **      P  U  B  L  I  C      F  U  N  C  T  I  O  N  S
 378  *********************************************************************
 379  ********************************************************************/
 380
 381 /*!
 382  * Initialize eCrash.
 383  *
 384  * This function must be called before calling any other eCrash
 385  * functions.  It sets up the global behavior of the system, and
 386  * registers the calling thread for crash dumps.
 387  *
 388  * @param params Our input parameters.  The passed in structure will be copied.
 389  *
 390  * @return Zero on success.
 391  */
 392 int eCrash_Init(eCrashParameters *params)
 393 {
 394         int sigIndex;
 395         int ret = 0;
 396 #ifdef DO_SIGNALS_RIGHT
 397         sigset_t blocked;
 398         struct sigaction act;
 399 #endif
 400
 401         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE,"Init Starting params = %p\n", params);
 402
 403         // Allocate our backtrace area
 404         gbl_backtraceBuffer = malloc(sizeof(void *) * (params->maxStackDepth+5));
 405
 406 #ifdef DO_SIGNALS_RIGHT
 407         sigemptyset(&blocked);
 408         act.sa_sigaction = crash_handler;
 409         act.sa_mask = blocked;
 410         act.sa_flags = SA_SIGINFO;
 411 #endif
 412
 413         if (params != NULL) {
 414                 // Make ourselves a global copy of params.
 415                 gbl_params = *params;
 416                 gbl_params.filename = strdup(params->filename);
 417
 418                 // Set our defaults, if they weren't specified
 419                 if (gbl_params.maxStackDepth == 0 )
 420                         gbl_params.maxStackDepth = ECRASH_DEFAULT_STACK_DEPTH;
 421
 422                 if (gbl_params.defaultBacktraceSignal == 0 )
 423                         gbl_params.defaultBacktraceSignal = ECRASH_DEFAULT_BACKTRACE_SIGNAL;
 424
 425                 if (gbl_params.threadWaitTime == 0 )
 426                         gbl_params.threadWaitTime = ECRASH_DEFAULT_THREAD_WAIT_TIME;
 427
 428                 if (gbl_params.debugLevel == 0 )
 429                         gbl_params.debugLevel = ECRASH_DEBUG_DEFAULT;
 430
 431                 // Copy our symbol table
 432                 if (gbl_params.symbolTable) {
 433                     DPRINTF(ECRASH_DEBUG_VERBOSE,
 434                                                         "symbolTable @ %p -- %d symbols\n", gbl_params.symbolTable,
 435                                                 gbl_params.symbolTable->numSymbols);
 436                         // Make a copy of our symbol table
 437                         gbl_params.symbolTable = malloc(sizeof(eCrashSymbolTable));
 438                         memcpy(gbl_params.symbolTable, params->symbolTable,
 439                                    sizeof(eCrashSymbolTable));
 440
 441                         // Now allocate / copy the actual table.
 442                         gbl_params.symbolTable->symbols = malloc(sizeof(eCrashSymbol) *
 443                                                                      gbl_params.symbolTable->numSymbols);
 444                         memcpy(gbl_params.symbolTable->symbols,
 445                                    params->symbolTable->symbols,
 446                                    sizeof(eCrashSymbol) * gbl_params.symbolTable->numSymbols);
 447
 448                         ValidateSymbolTable();
 449                 }
 450
 451                 // And, finally, register for our signals
 452                 for (sigIndex=0; gbl_params.signals[sigIndex] != 0; sigIndex++) {
 453                         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE,
 454                                                         "   Catching signal[%d] %d\n", sigIndex,
 455                                         gbl_params.signals[sigIndex]);
 456
 457                         // I know there's a better way to catch signals with pthreads.
 458                         // I'll do it later TODO
 459                         signal(gbl_params.signals[sigIndex], crash_handler);
 460                 }
 461         } else {
 462                 DPRINTF(ECRASH_DEBUG_ERROR, "   Error:  Null Params!\n");
 463                 ret = -1;
 464         }
 465         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE, "Init Complete ret=%d\n", ret);
 466         return ret;
 467 } /* eCrash_Init */
 468
 469 /*!
 470  * UnInitialize eCrash.
 471  *
 472  * This function may be called to de-activate eCrash, release the
 473  * signal handlers, and free any memory allocated by eCrash.
 474  *
 475  * @return Zero on success.
 476  */
 477 int eCrash_Uninit( void )
 478 {
 479         NIY();
 480
 481         return 0;
 482 } /* eCrash_Uninit */
 483
 484 /*!
 485  * Register a thread for backtracing on crash.
 486  *
 487  * This function must be called by any thread wanting it's stack
 488  * dumped in the event of a crash.  The thread my specify what
 489  * signal should be used, or the default, SIGUSR1 will be used.
 490  *
 491  * @param signo Signal to use to generate dump (default: SIGUSR1)
 492  *
 493  * @return Zero on success.
 494  */
 495 int eCrash_RegisterThread(char *name, int signo)
 496 {
 497         sighandler_t old_handler;
 498
 499         // Register for our signal
 500         if (signo == 0) {
 501                 signo = gbl_params.defaultBacktraceSignal;
 502         }
 503
 504         old_handler = signal(signo, bt_handler);
 505         return addThreadToList(name, pthread_self(), signo, old_handler);
 506
 507 } /* eCrash_RegisterThread */
 508
 509 /*!
 510  * Un-register a thread for stack dumps.
 511  *
 512  * This function may be called to un-register any previously
 513  * registered thread.
 514  *
 515  * @return Zero on success.
 516  */
 517 int eCrash_UnregisterThread( void )
 518 {
 519         return removeThreadFromList(pthread_self());
 520 } /* eCrash_UnregisterThread */
 521
 522 #endif