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                 vsyslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, format, ap);
 191
 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                         syslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, "RAW: %p  ", stack_frames[NThread]);
 215                 }
 216         else
 217                 for (NThread = 0; NThread < size; NThread++)
 218                         CtdlLogPrintf(1, "RAW: %p\n", stack_frames[NThread]);
 219         strings = backtrace_symbols(stack_frames, size);
 220         for (NThread = 0; NThread < size; NThread++) {
 221                 if (strings != NULL) {
 222                         if (enable_syslog)
 223                         {// vsyslogs printf compliance sucks.
 224                                 syslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, "RAW: %p  ", strings[NThread]);
 225                         }
 226                         else
 227                                 CtdlLogPrintf(1, "%s\n", strings[NThread]);
 228                 }
 229         }
 230 } /* createGlobalBacktrace */
 231 static void outputRawtrace( void )
 232 {
 233
 234         size = backtrace(stack_frames, sizeof(stack_frames) / sizeof(void*));
 235         if (enable_syslog)
 236                 for (NThread = 0; NThread < size; NThread++)
 237                 {
 238                         syslog(LOG_CRIT|LOG_NDELAY|LOG_MAIL, "RAW: %p  ", stack_frames[NThread]);
 239                 }
 240         else
 241                 for (NThread = 0; NThread < size; NThread++)
 242                         CtdlLogPrintf(1, "RAW: %p\n", stack_frames[NThread]);
 243 } /* createGlobalBacktrace */
 244
 245 /*!
 246  * Print out (to all the fds, etc), or global backtrace
 247  */
 248 static void outputGlobalBacktrace ( void )
 249 {
 250         int i;
 251
 252         for (i=0; i < gbl_backtraceEntries; i++) {
 253                 if (gbl_backtraceSymbols != FALSE) {
 254                         outputPrintf("*      Frame %02x: %s\n",
 255                                      i, gbl_backtraceSymbols[i]);
 256                 } else {
 257                         outputPrintf("*      Frame %02x: %p\n", i,
 258                                      gbl_backtraceBuffer[i]);
 259                 }
 260         }
 261 } // outputGlobalBacktrace
 262
 263 /*!
 264  * Output our current stack's backtrace
 265  */
 266 static void outputBacktrace( void )
 267 {
 268         createGlobalBacktrace();
 269         outputGlobalBacktrace();
 270 } /* outputBacktrace */
 271
 272 static void outputBacktraceThreads( void )
 273 {
 274         ThreadListNode *probe;
 275         int i;
 276
 277         // When we're backtracing, don't worry about the mutex . . hopefully
 278         // we're in a safe place.
 279
 280         for (probe=ThreadList; probe; probe=probe->Next) {
 281                 gbl_backtraceDoneFlag = 0;
 282                 pthread_kill(probe->thread, probe->backtraceSignal);
 283                 for (i=0; i < gbl_params.threadWaitTime; i++) {
 284                         if (gbl_backtraceDoneFlag)
 285                                 break;
 286                         sleep(1);
 287                 }
 288                 if (gbl_backtraceDoneFlag) {
 289                         outputPrintf("*  Backtrace of \"%s\" (0x%08x)\n",
 290                                                  probe->threadName, (unsigned int)probe->thread);
 291                         outputGlobalBacktrace();
 292                 } else {
 293                         outputPrintf("*  Error: unable to get backtrace of \"%s\" (0x%08x)\n",
 294                                                  probe->threadName, (unsigned int)probe->thread);
 295                 }
 296                 outputPrintf("*\n");
 297         }
 298 } // outputBacktraceThreads
 299
 300
 301 /*!
 302  * Handle signals (crash signals)
 303  *
 304  * This function will catch all crash signals, and will output the
 305  * crash dump.
 306  *
 307  * It will physically write (and sync) the current thread's information
 308  * before it attempts to send signals to other threads.
 309  *
 310  * @param signum Signal received.
 311  */
 312 static void crash_handler(int signo)
 313 {
 314         outputRawtrace();
 315         outputPrintf("*********************************************************\n");
 316         outputPrintf("*               eCrash Crash Handler\n");
 317         outputPrintf("*********************************************************\n");
 318         outputPrintf("*\n");
 319         outputPrintf("*  Got a crash! signo=%d\n", signo);
 320         outputPrintf("*\n");
 321         outputPrintf("*  Offending Thread's Backtrace:\n");
 322         outputPrintf("*\n");
 323         outputBacktrace();
 324         outputPrintf("*\n");
 325
 326         if (gbl_params.dumpAllThreads != FALSE) {
 327                 outputBacktraceThreads();
 328         }
 329
 330         outputPrintf("*\n");
 331         outputPrintf("*********************************************************\n");
 332         outputPrintf("*               eCrash Crash Handler\n");
 333         outputPrintf("*********************************************************\n");
 334
 335         exit(signo);
 336 } // crash_handler
 337
 338 /*!
 339  * Handle signals (bt signals)
 340  *
 341  * This function shoudl be called to generate a crashdump into our
 342  * global area.  Once the dump has been completed, this function will
 343  * return after tickling a global.  Since mutexes are not async
 344  * signal safe, the main thread, after signaling us to generate our
 345  * own backtrace, will sleep for a few seconds waiting for us to complete.
 346  *
 347  * @param signum Signal received.
 348  */
 349 static void bt_handler(int signo)
 350 {
 351         createGlobalBacktrace();
 352         gbl_backtraceDoneFlag=1;
 353 } // bt_handler
 354
 355 /*!
 356  * Validate a passed-in symbol table
 357  *
 358  * For now, just print it out (if verbose), and make sure it's
 359  * sorted and none of the pointers are zero.
 360  */
 361 static int ValidateSymbolTable( void )
 362 {
 363         int i;
 364         int rc=0;
 365         unsigned long lastAddress =0;
 366
 367         // Get out of here if the table is empty
 368         if (!gbl_params.symbolTable) return 0;
 369
 370         // Dump it in verbose mode
 371         DPRINTF(ECRASH_DEBUG_VERBOSE,
 372                                         "Symbol Table Provided with %d symbols\n",
 373                                         gbl_params.symbolTable->numSymbols);
 374         for (i=0; i < gbl_params.symbolTable->numSymbols; i++){
 375                 // Dump it in verbose mode
 376                 DPRINTF(ECRASH_DEBUG_VERBOSE,
 377                                 "%-30s %p\n",
 378                                 gbl_params.symbolTable->symbols[i].function,
 379                                 gbl_params.symbolTable->symbols[i].address);
 380                 if (lastAddress >
 381                     (unsigned long)gbl_params.symbolTable->symbols[i].address) {
 382                         DPRINTF(ECRASH_DEBUG_ERROR,
 383                                         "Error: symbol table is not sorted (last=%p, current=%p)\n",
 384                                         (void *)lastAddress,
 385                                         gbl_params.symbolTable->symbols[i].address);
 386                         rc = -1;
 387                 }
 388
 389         } // for
 390
 391         return rc;
 392
 393 } // ValidateSymbolTable
 394
 395 /*********************************************************************
 396  *********************************************************************
 397  **      P  U  B  L  I  C      F  U  N  C  T  I  O  N  S
 398  *********************************************************************
 399  ********************************************************************/
 400
 401 /*!
 402  * Initialize eCrash.
 403  *
 404  * This function must be called before calling any other eCrash
 405  * functions.  It sets up the global behavior of the system, and
 406  * registers the calling thread for crash dumps.
 407  *
 408  * @param params Our input parameters.  The passed in structure will be copied.
 409  *
 410  * @return Zero on success.
 411  */
 412 int eCrash_Init(eCrashParameters *params)
 413 {
 414         int sigIndex;
 415         int ret = 0;
 416 #ifdef DO_SIGNALS_RIGHT
 417         sigset_t blocked;
 418         struct sigaction act;
 419 #endif
 420
 421         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE,"Init Starting params = %p\n", params);
 422
 423         // Allocate our backtrace area
 424         gbl_backtraceBuffer = malloc(sizeof(void *) * (params->maxStackDepth+5));
 425
 426 #ifdef DO_SIGNALS_RIGHT
 427         sigemptyset(&blocked);
 428         act.sa_sigaction = crash_handler;
 429         act.sa_mask = blocked;
 430         act.sa_flags = SA_SIGINFO;
 431 #endif
 432
 433         if (params != NULL) {
 434                 // Make ourselves a global copy of params.
 435                 gbl_params = *params;
 436                 gbl_params.filename = strdup(params->filename);
 437
 438                 // Set our defaults, if they weren't specified
 439                 if (gbl_params.maxStackDepth == 0 )
 440                         gbl_params.maxStackDepth = ECRASH_DEFAULT_STACK_DEPTH;
 441
 442                 if (gbl_params.defaultBacktraceSignal == 0 )
 443                         gbl_params.defaultBacktraceSignal = ECRASH_DEFAULT_BACKTRACE_SIGNAL;
 444
 445                 if (gbl_params.threadWaitTime == 0 )
 446                         gbl_params.threadWaitTime = ECRASH_DEFAULT_THREAD_WAIT_TIME;
 447
 448                 if (gbl_params.debugLevel == 0 )
 449                         gbl_params.debugLevel = ECRASH_DEBUG_DEFAULT;
 450
 451                 // Copy our symbol table
 452                 if (gbl_params.symbolTable) {
 453                     DPRINTF(ECRASH_DEBUG_VERBOSE,
 454                                                         "symbolTable @ %p -- %d symbols\n", gbl_params.symbolTable,
 455                                                 gbl_params.symbolTable->numSymbols);
 456                         // Make a copy of our symbol table
 457                         gbl_params.symbolTable = malloc(sizeof(eCrashSymbolTable));
 458                         memcpy(gbl_params.symbolTable, params->symbolTable,
 459                                    sizeof(eCrashSymbolTable));
 460
 461                         // Now allocate / copy the actual table.
 462                         gbl_params.symbolTable->symbols = malloc(sizeof(eCrashSymbol) *
 463                                                                      gbl_params.symbolTable->numSymbols);
 464                         memcpy(gbl_params.symbolTable->symbols,
 465                                    params->symbolTable->symbols,
 466                                    sizeof(eCrashSymbol) * gbl_params.symbolTable->numSymbols);
 467
 468                         ValidateSymbolTable();
 469                 }
 470
 471                 // And, finally, register for our signals
 472                 for (sigIndex=0; gbl_params.signals[sigIndex] != 0; sigIndex++) {
 473                         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE,
 474                                                         "   Catching signal[%d] %d\n", sigIndex,
 475                                         gbl_params.signals[sigIndex]);
 476
 477                         // I know there's a better way to catch signals with pthreads.
 478                         // I'll do it later TODO
 479                         signal(gbl_params.signals[sigIndex], crash_handler);
 480                 }
 481         } else {
 482                 DPRINTF(ECRASH_DEBUG_ERROR, "   Error:  Null Params!\n");
 483                 ret = -1;
 484         }
 485         DPRINTF(ECRASH_DEBUG_VERY_VERBOSE, "Init Complete ret=%d\n", ret);
 486         return ret;
 487 } /* eCrash_Init */
 488
 489 /*!
 490  * UnInitialize eCrash.
 491  *
 492  * This function may be called to de-activate eCrash, release the
 493  * signal handlers, and free any memory allocated by eCrash.
 494  *
 495  * @return Zero on success.
 496  */
 497 int eCrash_Uninit( void )
 498 {
 499         NIY();
 500
 501         return 0;
 502 } /* eCrash_Uninit */
 503
 504 /*!
 505  * Register a thread for backtracing on crash.
 506  *
 507  * This function must be called by any thread wanting it's stack
 508  * dumped in the event of a crash.  The thread my specify what
 509  * signal should be used, or the default, SIGUSR1 will be used.
 510  *
 511  * @param signo Signal to use to generate dump (default: SIGUSR1)
 512  *
 513  * @return Zero on success.
 514  */
 515 int eCrash_RegisterThread(char *name, int signo)
 516 {
 517         sighandler_t old_handler;
 518
 519         // Register for our signal
 520         if (signo == 0) {
 521                 signo = gbl_params.defaultBacktraceSignal;
 522         }
 523
 524         old_handler = signal(signo, bt_handler);
 525         return addThreadToList(name, pthread_self(), signo, old_handler);
 526
 527 } /* eCrash_RegisterThread */
 528
 529 /*!
 530  * Un-register a thread for stack dumps.
 531  *
 532  * This function may be called to un-register any previously
 533  * registered thread.
 534  *
 535  * @return Zero on success.
 536  */
 537 int eCrash_UnregisterThread( void )
 538 {
 539         return removeThreadFromList(pthread_self());
 540 } /* eCrash_UnregisterThread */
 541
 542 #endif