citadel/threads.c

   1 /*
   2  * $Id: sysdep.c 5882 2007-12-13 19:46:05Z davew $
   3  *
   4  * Citadel "system dependent" stuff.
   5  * See copyright.txt for copyright information.
   6  *
   7  * Here's where we have the Citadel thread implimentation
   8  *
   9  */
  10
  11 #include <sys/types.h>
  12 #include <errno.h>
  13 #include <sys/socket.h>
  14 #include <unistd.h>
  15 #include <fcntl.h>
  16 #include <signal.h>
  17
  18 #if TIME_WITH_SYS_TIME
  19 # include <sys/time.h>
  20 # include <time.h>
  21 #else
  22 # if HAVE_SYS_TIME_H
  23 #  include <sys/time.h>
  24 # else
  25 #  include <time.h>
  26 # endif
  27 #endif
  28
  29 #include "threads.h"
  30 #include "ctdl_module.h"
  31 #include "modules_init.h"
  32 #include "housekeeping.h"
  33 #include "config.h"
  34 #include "citserver.h"
  35 #include "sysdep_decls.h"
  36
  37 /*
  38  * define this to use the new worker_thread method of handling connections
  39  */
  40 //#define NEW_WORKER
  41
  42 /*
  43  * New thread interface.
  44  * To create a thread you must call one of the create thread functions.
  45  * You must pass it the address of (a pointer to a CtdlThreadNode initialised to NULL) like this
  46  * struct CtdlThreadNode *node = NULL;
  47  * pass in &node
  48  * If the thread is created *node will point to the thread control structure for the created thread.
  49  * If the thread creation fails *node remains NULL
  50  * Do not free the memory pointed to by *node, it doesn't belong to you.
  51  * This new interface duplicates much of the eCrash stuff. We should go for closer integration since that would
  52  * remove the need for the calls to eCrashRegisterThread and friends
  53  */
  54
  55 static int num_threads = 0;                     /* Current number of threads */
  56 static int num_workers = 0;                     /* Current number of worker threads */
  57
  58 CtdlThreadNode *CtdlThreadList = NULL;
  59 CtdlThreadNode *CtdlThreadSchedList = NULL;
  60
  61 static CtdlThreadNode *GC_thread = NULL;
  62 static char *CtdlThreadStates[CTDL_THREAD_LAST_STATE];
  63 double CtdlThreadLoadAvg = 0;
  64 double CtdlThreadWorkerAvg = 0;
  65 citthread_key_t ThreadKey;
  66
  67 citthread_mutex_t Critters[MAX_SEMAPHORES];     /* Things needing locking */
  68
  69
  70
  71 void InitialiseSemaphores(void)
  72 {
  73         int i;
  74
  75         /* Set up a bunch of semaphores to be used for critical sections */
  76         for (i=0; i<MAX_SEMAPHORES; ++i) {
  77                 citthread_mutex_init(&Critters[i], NULL);
  78         }
  79 }
  80
  81
  82
  83
  84 /*
  85  * Obtain a semaphore lock to begin a critical section.
  86  * but only if no one else has one
  87  */
  88 int try_critical_section(int which_one)
  89 {
  90         /* For all types of critical sections except those listed here,
  91          * ensure nobody ever tries to do a critical section within a
  92          * transaction; this could lead to deadlock.
  93          */
  94         if (    (which_one != S_FLOORCACHE)
  95 #ifdef DEBUG_MEMORY_LEAKS
  96                 && (which_one != S_DEBUGMEMLEAKS)
  97 #endif
  98                 && (which_one != S_RPLIST)
  99         ) {
 100                 cdb_check_handles();
 101         }
 102         return (citthread_mutex_trylock(&Critters[which_one]));
 103 }
 104
 105
 106 /*
 107  * Obtain a semaphore lock to begin a critical section.
 108  */
 109 void begin_critical_section(int which_one)
 110 {
 111         /* CtdlLogPrintf(CTDL_DEBUG, "begin_critical_section(%d)\n", which_one); */
 112
 113         /* For all types of critical sections except those listed here,
 114          * ensure nobody ever tries to do a critical section within a
 115          * transaction; this could lead to deadlock.
 116          */
 117         if (    (which_one != S_FLOORCACHE)
 118 #ifdef DEBUG_MEMORY_LEAKS
 119                 && (which_one != S_DEBUGMEMLEAKS)
 120 #endif
 121                 && (which_one != S_RPLIST)
 122         ) {
 123                 cdb_check_handles();
 124         }
 125         citthread_mutex_lock(&Critters[which_one]);
 126 }
 127
 128 /*
 129  * Release a semaphore lock to end a critical section.
 130  */
 131 void end_critical_section(int which_one)
 132 {
 133         citthread_mutex_unlock(&Critters[which_one]);
 134 }
 135
 136
 137 /*
 138  * A function to destroy the TSD
 139  */
 140 static void ctdl_thread_internal_dest_tsd(void *arg)
 141 {
 142         if (arg != NULL) {
 143                 check_handles(arg);
 144                 free(arg);
 145         }
 146 }
 147
 148
 149 /*
 150  * A function to initialise the thread TSD
 151  */
 152 void ctdl_thread_internal_init_tsd(void)
 153 {
 154         int ret;
 155
 156         if ((ret = citthread_key_create(&ThreadKey, ctdl_thread_internal_dest_tsd))) {
 157                 CtdlLogPrintf(CTDL_EMERG, "citthread_key_create: %s\n", strerror(ret));
 158                 exit(CTDLEXIT_DB);
 159         }
 160 }
 161
 162 /*
 163  * Ensure that we have a key for thread-specific data.
 164  *
 165  * This should be called immediately after startup by any thread
 166  *
 167  */
 168 void CtdlThreadAllocTSD(void)
 169 {
 170         ThreadTSD *tsd;
 171
 172         if (citthread_getspecific(ThreadKey) != NULL)
 173                 return;
 174
 175         tsd = malloc(sizeof(ThreadTSD));
 176
 177         tsd->tid = NULL;
 178
 179         memset(tsd->cursors, 0, sizeof tsd->cursors);
 180         tsd->self = NULL;
 181
 182         citthread_setspecific(ThreadKey, tsd);
 183 }
 184
 185
 186 void ctdl_thread_internal_free_tsd(void)
 187 {
 188         ctdl_thread_internal_dest_tsd(citthread_getspecific(ThreadKey));
 189         citthread_setspecific(ThreadKey, NULL);
 190 }
 191
 192
 193 void ctdl_thread_internal_cleanup(void)
 194 {
 195         int i;
 196         CtdlThreadNode *this_thread, *that_thread;
 197
 198         for (i=0; i<CTDL_THREAD_LAST_STATE; i++)
 199         {
 200                 free (CtdlThreadStates[i]);
 201         }
 202
 203         /* Clean up the scheduled thread list */
 204         this_thread = CtdlThreadSchedList;
 205         while (this_thread)
 206         {
 207                 that_thread = this_thread;
 208                 this_thread = this_thread->next;
 209                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 210                 citthread_cond_destroy(&that_thread->ThreadCond);
 211                 citthread_mutex_destroy(&that_thread->SleepMutex);
 212                 citthread_cond_destroy(&that_thread->SleepCond);
 213                 citthread_attr_destroy(&that_thread->attr);
 214                 free(that_thread);
 215         }
 216         ctdl_thread_internal_free_tsd();
 217 }
 218
 219 void ctdl_thread_internal_init(void)
 220 {
 221         CtdlThreadNode *this_thread;
 222         int ret = 0;
 223
 224         CtdlThreadStates[CTDL_THREAD_INVALID] = strdup ("Invalid Thread");
 225         CtdlThreadStates[CTDL_THREAD_VALID] = strdup("Valid Thread");
 226         CtdlThreadStates[CTDL_THREAD_CREATE] = strdup("Thread being Created");
 227         CtdlThreadStates[CTDL_THREAD_CANCELLED] = strdup("Thread Cancelled");
 228         CtdlThreadStates[CTDL_THREAD_EXITED] = strdup("Thread Exited");
 229         CtdlThreadStates[CTDL_THREAD_STOPPING] = strdup("Thread Stopping");
 230         CtdlThreadStates[CTDL_THREAD_STOP_REQ] = strdup("Thread Stop Requested");
 231         CtdlThreadStates[CTDL_THREAD_SLEEPING] = strdup("Thread Sleeping");
 232         CtdlThreadStates[CTDL_THREAD_RUNNING] = strdup("Thread Running");
 233         CtdlThreadStates[CTDL_THREAD_BLOCKED] = strdup("Thread Blocked");
 234
 235         /* Get ourself a thread entry */
 236         this_thread = malloc(sizeof(CtdlThreadNode));
 237         if (this_thread == NULL) {
 238                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 239                 return;
 240         }
 241         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 242         memset (this_thread, 0, sizeof(CtdlThreadNode));
 243
 244         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 245         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 246         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 247         citthread_cond_init (&(this_thread->SleepCond), NULL);
 248
 249         /* We are garbage collector so create us as running */
 250         this_thread->state = CTDL_THREAD_RUNNING;
 251
 252         if ((ret = citthread_attr_init(&this_thread->attr))) {
 253                 CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 254                 free(this_thread);
 255                 return;
 256         }
 257
 258         this_thread->name = "Garbage Collection Thread";
 259
 260         this_thread->tid = citthread_self();
 261         GC_thread = this_thread;
 262         CT = this_thread;
 263
 264         num_threads++;  // Increase the count of threads in the system.
 265
 266         this_thread->next = CtdlThreadList;
 267         CtdlThreadList = this_thread;
 268         if (this_thread->next)
 269                 this_thread->next->prev = this_thread;
 270         /* Set up start times */
 271         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 272         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 273 }
 274
 275
 276 /*
 277  * A function to update a threads load averages
 278  */
 279  void ctdl_thread_internal_update_avgs(CtdlThreadNode *this_thread)
 280  {
 281         struct timeval now, result;
 282         double last_duration;
 283
 284         gettimeofday(&now, NULL);
 285         timersub(&now, &(this_thread->last_state_change), &result);
 286         /* I don't think these mutex's are needed here */
 287         citthread_mutex_lock(&this_thread->ThreadMutex);
 288         // result now has a timeval for the time we spent in the last state since we last updated
 289         last_duration = (double)result.tv_sec + ((double)result.tv_usec / (double) 1000000);
 290         if (this_thread->state == CTDL_THREAD_SLEEPING)
 291                 this_thread->avg_sleeping += last_duration;
 292         if (this_thread->state == CTDL_THREAD_RUNNING)
 293                 this_thread->avg_running += last_duration;
 294         if (this_thread->state == CTDL_THREAD_BLOCKED)
 295                 this_thread->avg_blocked += last_duration;
 296         memcpy (&this_thread->last_state_change, &now, sizeof (struct timeval));
 297         citthread_mutex_unlock(&this_thread->ThreadMutex);
 298 }
 299
 300 /*
 301  * A function to chenge the state of a thread
 302  */
 303 void ctdl_thread_internal_change_state (CtdlThreadNode *this_thread, enum CtdlThreadState new_state)
 304 {
 305         /*
 306          * Wether we change state or not we need update the load values
 307          */
 308         ctdl_thread_internal_update_avgs(this_thread);
 309         /* This mutex not needed here? */
 310         citthread_mutex_lock(&this_thread->ThreadMutex); /* To prevent race condition of a sleeping thread */
 311         if ((new_state == CTDL_THREAD_STOP_REQ) && (this_thread->state > CTDL_THREAD_STOP_REQ))
 312                 this_thread->state = new_state;
 313         if (((new_state == CTDL_THREAD_SLEEPING) || (new_state == CTDL_THREAD_BLOCKED)) && (this_thread->state == CTDL_THREAD_RUNNING))
 314                 this_thread->state = new_state;
 315         if ((new_state == CTDL_THREAD_RUNNING) && ((this_thread->state == CTDL_THREAD_SLEEPING) || (this_thread->state == CTDL_THREAD_BLOCKED)))
 316                 this_thread->state = new_state;
 317         citthread_mutex_unlock(&this_thread->ThreadMutex);
 318 }
 319
 320
 321 /*
 322  * A function to tell all threads to exit
 323  */
 324 void CtdlThreadStopAll(void)
 325 {
 326         //FIXME: The signalling of the condition should not be in the critical_section
 327         // We need to build a list of threads we are going to signal and then signal them afterwards
 328
 329         CtdlThreadNode *this_thread;
 330
 331         begin_critical_section(S_THREAD_LIST);
 332         this_thread = CtdlThreadList;
 333         // Ask the GC thread to stop first so everything knows we are shutting down.
 334         GC_thread->state = CTDL_THREAD_STOP_REQ;
 335         while(this_thread)
 336         {
 337 #ifdef THREADS_USESIGNALS
 338                 if (!citthread_equal(this_thread->tid, GC_thread->tid))
 339                         citthread_kill(this_thread->tid, SIGHUP);
 340 #endif
 341                 ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 342                 citthread_cond_signal(&this_thread->ThreadCond);
 343                 citthread_cond_signal(&this_thread->SleepCond);
 344                 this_thread->stop_ticker = time(NULL);
 345                 CtdlLogPrintf(CTDL_DEBUG, "Thread system stopping thread \"%s\" (0x%08lx).\n",
 346                         this_thread->name, this_thread->tid);
 347                 this_thread = this_thread->next;
 348         }
 349         end_critical_section(S_THREAD_LIST);
 350 }
 351
 352
 353 /*
 354  * A function to wake up all sleeping threads
 355  */
 356 void CtdlThreadWakeAll(void)
 357 {
 358         CtdlThreadNode *this_thread;
 359
 360         CtdlLogPrintf(CTDL_DEBUG, "Thread system waking all threads.\n");
 361
 362         begin_critical_section(S_THREAD_LIST);
 363         this_thread = CtdlThreadList;
 364         while(this_thread)
 365         {
 366                 if (!this_thread->thread_func)
 367                 {
 368                         citthread_cond_signal(&this_thread->ThreadCond);
 369                         citthread_cond_signal(&this_thread->SleepCond);
 370                 }
 371                 this_thread = this_thread->next;
 372         }
 373         end_critical_section(S_THREAD_LIST);
 374 }
 375
 376
 377 /*
 378  * A function to return the number of threads running in the system
 379  */
 380 int CtdlThreadGetCount(void)
 381 {
 382         return  num_threads;
 383 }
 384
 385 int CtdlThreadGetWorkers(void)
 386 {
 387         return  num_workers;
 388 }
 389
 390 double CtdlThreadGetWorkerAvg(void)
 391 {
 392         double ret;
 393
 394         begin_critical_section(S_THREAD_LIST);
 395         ret =  CtdlThreadWorkerAvg;
 396         end_critical_section(S_THREAD_LIST);
 397         return ret;
 398 }
 399
 400 double CtdlThreadGetLoadAvg(void)
 401 {
 402         double ret;
 403
 404         begin_critical_section(S_THREAD_LIST);
 405         ret =  CtdlThreadLoadAvg;
 406         end_critical_section(S_THREAD_LIST);
 407         return ret;
 408 }
 409
 410
 411
 412
 413 /*
 414  * A function to rename a thread
 415  * Returns a const char *
 416  */
 417 const char *CtdlThreadName(const char *name)
 418 {
 419         const char *old_name;
 420
 421         if (!CT)
 422         {
 423                 CtdlLogPrintf(CTDL_WARNING, "Thread system WARNING. Attempt to CtdlThreadRename() a non thread. %s\n", name);
 424                 return NULL;
 425         }
 426         old_name = CT->name;
 427         if (name)
 428                 CT->name = name;
 429         return (old_name);
 430 }
 431
 432
 433 /*
 434  * A function to force a thread to exit
 435  */
 436 void CtdlThreadCancel(CtdlThreadNode *thread)
 437 {
 438         CtdlThreadNode *this_thread;
 439
 440         if (!thread)
 441                 this_thread = CT;
 442         else
 443                 this_thread = thread;
 444         if (!this_thread)
 445         {
 446                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() a non thread.\n");
 447                 CtdlThreadStopAll();
 448                 return;
 449         }
 450
 451         if (!this_thread->thread_func)
 452         {
 453                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() the garbage collector.\n");
 454                 CtdlThreadStopAll();
 455                 return;
 456         }
 457
 458         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_CANCELLED);
 459         citthread_cancel(this_thread->tid);
 460 }
 461
 462
 463 /*
 464  * A function for a thread to check if it has been asked to stop
 465  */
 466 int CtdlThreadCheckStop(void)
 467 {
 468         int state;
 469
 470         if (!CT)
 471         {
 472                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, CtdlThreadCheckStop() called by a non thread.\n");
 473                 CtdlThreadStopAll();
 474                 return -1;
 475         }
 476
 477         state = CT->state;
 478
 479 #ifdef THREADS_USESIGNALS
 480         if (CT->signal)
 481         {
 482                 CtdlLogPrintf(CTDL_DEBUG, "Thread \"%s\" caught signal %d.\n", CT->name, CT->signal);
 483                 CT->signal = 0;
 484         }
 485 #endif
 486         if(state == CTDL_THREAD_STOP_REQ)
 487         {
 488                 CT->state = CTDL_THREAD_STOPPING;
 489                 return -1;
 490         }
 491         else if((state < CTDL_THREAD_STOP_REQ) && (state > CTDL_THREAD_CREATE))
 492         {
 493                 return -1;
 494         }
 495         return 0;
 496 }
 497
 498
 499 /*
 500  * A function to ask a thread to exit
 501  * The thread must call CtdlThreadCheckStop() periodically to determine if it should exit
 502  */
 503 void CtdlThreadStop(CtdlThreadNode *thread)
 504 {
 505         CtdlThreadNode *this_thread;
 506
 507         if (!thread)
 508                 this_thread = CT;
 509         else
 510                 this_thread = thread;
 511         if (!this_thread)
 512                 return;
 513         if (!(this_thread->thread_func))
 514                 return;         // Don't stop garbage collector
 515 #ifdef THREADS_USESIGNALS
 516         if (!citthread_equal(this_thread->tid, GC_thread->tid))
 517                 citthread_kill(this_thread->tid, SIGHUP);
 518 #endif
 519         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 520         citthread_cond_signal(&this_thread->ThreadCond);
 521         citthread_cond_signal(&this_thread->SleepCond);
 522         this_thread->stop_ticker = time(NULL);
 523 }
 524
 525 /*
 526  * So we now have a sleep command that works with threads but it is in seconds
 527  */
 528 void CtdlThreadSleep(int secs)
 529 {
 530         struct timespec wake_time;
 531         struct timeval time_now;
 532
 533
 534         if (!CT)
 535         {
 536                 CtdlLogPrintf(CTDL_WARNING, "CtdlThreadSleep() called by something that is not a thread. Should we die?\n");
 537                 return;
 538         }
 539
 540         memset (&wake_time, 0, sizeof(struct timespec));
 541         gettimeofday(&time_now, NULL);
 542         wake_time.tv_sec = time_now.tv_sec + secs;
 543         wake_time.tv_nsec = time_now.tv_usec * 10;
 544
 545         ctdl_thread_internal_change_state (CT, CTDL_THREAD_SLEEPING);
 546
 547         citthread_mutex_lock(&CT->ThreadMutex); /* Prevent something asking us to awaken before we've gone to sleep */
 548         citthread_cond_timedwait(&CT->SleepCond, &CT->ThreadMutex, &wake_time);
 549         citthread_mutex_unlock(&CT->ThreadMutex);
 550
 551         ctdl_thread_internal_change_state (CT, CTDL_THREAD_RUNNING);
 552 }
 553
 554
 555 /*
 556  * Routine to clean up our thread function on exit
 557  */
 558 static void ctdl_internal_thread_cleanup(void *arg)
 559 {
 560         /*
 561          * In here we were called by the current thread because it is exiting
 562          * NB. WE ARE THE CURRENT THREAD
 563          */
 564         CtdlLogPrintf(CTDL_NOTICE, "Thread \"%s\" (0x%08lx) exited.\n", CT->name, CT->tid);
 565
 566         #ifdef HAVE_BACKTRACE
 567         eCrash_UnregisterThread();
 568         #endif
 569
 570         citthread_mutex_lock(&CT->ThreadMutex);
 571         CT->state = CTDL_THREAD_EXITED; // needs to be last thing else house keeping will unlink us too early
 572         citthread_mutex_unlock(&CT->ThreadMutex);
 573 }
 574
 575 /*
 576  * A quick function to show the load averages
 577  */
 578 void ctdl_thread_internal_calc_loadavg(void)
 579 {
 580         CtdlThreadNode *that_thread;
 581         double load_avg, worker_avg;
 582         int workers = 0;
 583
 584         that_thread = CtdlThreadList;
 585         load_avg = 0;
 586         worker_avg = 0;
 587         while(that_thread)
 588         {
 589                 /* Update load averages */
 590                 ctdl_thread_internal_update_avgs(that_thread);
 591                 citthread_mutex_lock(&that_thread->ThreadMutex);
 592                 that_thread->load_avg = (that_thread->avg_sleeping + that_thread->avg_running) / (that_thread->avg_sleeping + that_thread->avg_running + that_thread->avg_blocked) * 100;
 593                 that_thread->avg_sleeping /= 2;
 594                 that_thread->avg_running /= 2;
 595                 that_thread->avg_blocked /= 2;
 596                 load_avg += that_thread->load_avg;
 597                 if (that_thread->flags & CTDLTHREAD_WORKER)
 598                 {
 599                         worker_avg += that_thread->load_avg;
 600                         workers++;
 601                 }
 602 #ifdef WITH_THREADLOG
 603                 CtdlLogPrintf(CTDL_DEBUG, "CtdlThread, \"%s\" (%lu) \"%s\" %.2f %.2f %.2f %.2f\n",
 604                         that_thread->name,
 605                         that_thread->tid,
 606                         CtdlThreadStates[that_thread->state],
 607                         that_thread->avg_sleeping,
 608                         that_thread->avg_running,
 609                         that_thread->avg_blocked,
 610                         that_thread->load_avg);
 611 #endif
 612                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 613                 that_thread = that_thread->next;
 614         }
 615         CtdlThreadLoadAvg = load_avg/num_threads;
 616         CtdlThreadWorkerAvg = worker_avg/workers;
 617 #ifdef WITH_THREADLOG
 618         CtdlLogPrintf(CTDL_INFO, "System load average %.2f, workers averag %.2f, threads %d, workers %d, sessions %d\n", CtdlThreadLoadAvg, CtdlThreadWorkerAvg, num_threads, num_workers, num_sessions);
 619 #endif
 620 }
 621
 622
 623 /*
 624  * Garbage collection routine.
 625  * Gets called by main() in a loop to clean up the thread list periodically.
 626  */
 627 void CtdlThreadGC (void)
 628 {
 629         CtdlThreadNode *this_thread, *that_thread;
 630         int workers = 0, sys_workers;
 631         int ret=0;
 632
 633         begin_critical_section(S_THREAD_LIST);
 634
 635         /* Handle exiting of garbage collector thread */
 636         if(num_threads == 1)
 637                 CtdlThreadList->state = CTDL_THREAD_EXITED;
 638
 639 #ifdef WITH_THREADLOG
 640         CtdlLogPrintf(CTDL_DEBUG, "Thread system running garbage collection.\n");
 641 #endif
 642         /*
 643          * Woke up to do garbage collection
 644          */
 645         this_thread = CtdlThreadList;
 646         while(this_thread)
 647         {
 648                 that_thread = this_thread;
 649                 this_thread = this_thread->next;
 650
 651                 if ((that_thread->state == CTDL_THREAD_STOP_REQ || that_thread->state == CTDL_THREAD_STOPPING)
 652                         && (!citthread_equal(that_thread->tid, citthread_self())))
 653                                 CtdlLogPrintf(CTDL_DEBUG, "Waiting for thread %s (0x%08lx) to exit.\n", that_thread->name, that_thread->tid);
 654                 else
 655                 {
 656                         /**
 657                          * Catch the situation where a worker was asked to stop but couldn't and we are not
 658                          * shutting down.
 659                          */
 660                         that_thread->stop_ticker = 0;
 661                 }
 662
 663                 if (that_thread->stop_ticker + 5 == time(NULL))
 664                 {
 665                         CtdlLogPrintf(CTDL_DEBUG, "Thread System: The thread \"%s\" (0x%08lx) failed to self terminate within 5 ticks. It would be cancelled now.\n", that_thread->name, that_thread->tid);
 666                         if ((that_thread->flags & CTDLTHREAD_WORKER) == 0)
 667                                 CtdlLogPrintf(CTDL_INFO, "Thread System: A non worker thread would have been canceled this may cause message loss.\n");
 668 //                      that_thread->state = CTDL_THREAD_CANCELLED;
 669                         that_thread->stop_ticker++;
 670 //                      citthread_cancel(that_thread->tid);
 671 //                      continue;
 672                 }
 673
 674                 /* Do we need to clean up this thread? */
 675                 if ((that_thread->state != CTDL_THREAD_EXITED) && (that_thread->state != CTDL_THREAD_CANCELLED))
 676                 {
 677                         if(that_thread->flags & CTDLTHREAD_WORKER)
 678                                 workers++;      /* Sanity check on number of worker threads */
 679                         continue;
 680                 }
 681
 682                 if (citthread_equal(that_thread->tid, citthread_self()) && that_thread->thread_func)
 683                 {       /* Sanity check */
 684                         end_critical_section(S_THREAD_LIST);
 685                         CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, a thread is trying to clean up after itself.\n");
 686                         abort();
 687                         return;
 688                 }
 689
 690                 if (num_threads <= 0)
 691                 {       /* Sanity check */
 692                         end_critical_section(S_THREAD_LIST);
 693                         CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, num_threads <= 0 and trying to do Garbage Collection.\n");
 694                         abort();
 695                         return;
 696                 }
 697
 698                 if(that_thread->flags & CTDLTHREAD_WORKER)
 699                         num_workers--;  /* This is a wroker thread so reduce the count. */
 700                 num_threads--;
 701                 /* If we are unlinking the list head then the next becomes the list head */
 702                 if(that_thread->prev)
 703                         that_thread->prev->next = that_thread->next;
 704                 else
 705                         CtdlThreadList = that_thread->next;
 706                 if(that_thread->next)
 707                         that_thread->next->prev = that_thread->prev;
 708
 709                 citthread_cond_signal(&that_thread->ThreadCond);
 710                 citthread_cond_signal(&that_thread->SleepCond); // Make sure this thread is awake
 711                 citthread_mutex_lock(&that_thread->ThreadMutex);        // Make sure it has done what its doing
 712                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 713                 /*
 714                  * Join on the thread to do clean up and prevent memory leaks
 715                  * Also makes sure the thread has cleaned up after itself before we remove it from the list
 716                  * We can join on the garbage collector thread the join should just return EDEADLCK
 717                  */
 718                 ret = citthread_join (that_thread->tid, NULL);
 719                 if (ret == EDEADLK)
 720                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection on own thread.\n");
 721                 else if (ret == EINVAL)
 722                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, that thread already joined on.\n");
 723                 else if (ret == ESRCH)
 724                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, no thread to join on.\n");
 725                 else if (ret != 0)
 726                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, citthread_join returned an unknown error(%d).\n", ret);
 727                 /*
 728                  * Now we own that thread entry
 729                  */
 730                 CtdlLogPrintf(CTDL_INFO, "Garbage Collection for thread \"%s\" (0x%08lx).\n",
 731                         that_thread->name, that_thread->tid);
 732                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 733                 citthread_cond_destroy(&that_thread->ThreadCond);
 734                 citthread_mutex_destroy(&that_thread->SleepMutex);
 735                 citthread_cond_destroy(&that_thread->SleepCond);
 736                 citthread_attr_destroy(&that_thread->attr);
 737                 free(that_thread);
 738         }
 739         sys_workers = num_workers;
 740         end_critical_section(S_THREAD_LIST);
 741
 742         /* Sanity check number of worker threads */
 743         if (workers != sys_workers)
 744         {
 745                 CtdlLogPrintf(CTDL_EMERG,
 746                         "Thread system PANIC, discrepancy in number of worker threads. Counted %d, should be %d.\n",
 747                         workers, sys_workers
 748                         );
 749                 abort();
 750         }
 751 }
 752
 753
 754
 755
 756 /*
 757  * Runtime function for a Citadel Thread.
 758  * This initialises the threads environment and then calls the user supplied thread function
 759  * Note that this is the REAL thread function and wraps the users thread function.
 760  */
 761 static void *ctdl_internal_thread_func (void *arg)
 762 {
 763         CtdlThreadNode *this_thread;
 764         void *ret = NULL;
 765
 766         /* lock and unlock the thread list.
 767          * This causes this thread to wait until all its creation stuff has finished before it
 768          * can continue its execution.
 769          */
 770         begin_critical_section(S_THREAD_LIST);
 771         this_thread = (CtdlThreadNode *) arg;
 772         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 773 //      citthread_mutex_lock(&this_thread->ThreadMutex);
 774
 775         // Register the cleanup function to take care of when we exit.
 776         citthread_cleanup_push(ctdl_internal_thread_cleanup, NULL);
 777         // Get our thread data structure
 778         CtdlThreadAllocTSD();
 779         CT = this_thread;
 780         this_thread->pid = getpid();
 781         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 782         /* Only change to running state if we weren't asked to stop during the create cycle
 783          * Other wise there is a window to allow this threads creation to continue to full grown and
 784          * therby prevent a shutdown of the server.
 785          */
 786 //      citthread_mutex_unlock(&this_thread->ThreadMutex);
 787
 788         if (!CtdlThreadCheckStop())
 789         {
 790                 citthread_mutex_lock(&this_thread->ThreadMutex);
 791                 this_thread->state = CTDL_THREAD_RUNNING;
 792                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 793         }
 794         end_critical_section(S_THREAD_LIST);
 795
 796         // Register for tracing
 797         #ifdef HAVE_BACKTRACE
 798         eCrash_RegisterThread(this_thread->name, 0);
 799         #endif
 800
 801         // Tell the world we are here
 802         CtdlLogPrintf(CTDL_NOTICE, "Created a new thread \"%s\" (0x%08lx).\n",
 803                 this_thread->name, this_thread->tid);
 804
 805         /*
 806          * run the thread to do the work but only if we haven't been asked to stop
 807          */
 808         if (!CtdlThreadCheckStop())
 809                 ret = (this_thread->thread_func)(this_thread->user_args);
 810
 811         /*
 812          * Our thread is exiting either because it wanted to end or because the server is stopping
 813          * We need to clean up
 814          */
 815         citthread_cleanup_pop(1);       // Execute our cleanup routine and remove it
 816
 817         return(ret);
 818 }
 819
 820
 821
 822
 823 /*
 824  * Function to initialise an empty thread structure
 825  */
 826 CtdlThreadNode *ctdl_internal_init_thread_struct(CtdlThreadNode *this_thread, long flags)
 827 {
 828         int ret = 0;
 829
 830         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 831         memset (this_thread, 0, sizeof(CtdlThreadNode));
 832
 833         /* Create the mutex's early so we can use them */
 834         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 835         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 836         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 837         citthread_cond_init (&(this_thread->SleepCond), NULL);
 838
 839         this_thread->state = CTDL_THREAD_CREATE;
 840
 841         if ((ret = citthread_attr_init(&this_thread->attr))) {
 842                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 843                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 844                 citthread_cond_destroy(&(this_thread->ThreadCond));
 845                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 846                 citthread_cond_destroy(&(this_thread->SleepCond));
 847                 CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 848                 free(this_thread);
 849                 return NULL;
 850         }
 851
 852         /* Our per-thread stacks need to be bigger than the default size,
 853          * otherwise the MIME parser crashes on FreeBSD, and the IMAP service
 854          * crashes on 64-bit Linux.
 855          */
 856         if (flags & CTDLTHREAD_BIGSTACK)
 857         {
 858 #ifdef WITH_THREADLOG
 859                 CtdlLogPrintf(CTDL_INFO, "Thread system. Creating BIG STACK thread.\n");
 860 #endif
 861                 if ((ret = citthread_attr_setstacksize(&this_thread->attr, THREADSTACKSIZE))) {
 862                         citthread_mutex_unlock(&this_thread->ThreadMutex);
 863                         citthread_mutex_destroy(&(this_thread->ThreadMutex));
 864                         citthread_cond_destroy(&(this_thread->ThreadCond));
 865                         citthread_mutex_destroy(&(this_thread->SleepMutex));
 866                         citthread_cond_destroy(&(this_thread->SleepCond));
 867                         citthread_attr_destroy(&this_thread->attr);
 868                         CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_setstacksize: %s\n",
 869                                 strerror(ret));
 870                         free(this_thread);
 871                         return NULL;
 872                 }
 873         }
 874
 875         /* Set this new thread with an avg_blocked of 2. We do this so that its creation affects the
 876          * load average for the system. If we don't do this then we create a mass of threads at the same time
 877          * because the creation didn't affect the load average.
 878          */
 879         this_thread->avg_blocked = 2;
 880
 881         return (this_thread);
 882 }
 883
 884
 885
 886
 887 /*
 888  * Internal function to create a thread.
 889  */
 890 CtdlThreadNode *ctdl_internal_create_thread(char *name, long flags, void *(*thread_func) (void *arg), void *args)
 891 {
 892         int ret = 0;
 893         CtdlThreadNode *this_thread;
 894
 895         if (num_threads >= 32767)
 896         {
 897                 CtdlLogPrintf(CTDL_EMERG, "Thread system. Thread list full.\n");
 898                 return NULL;
 899         }
 900
 901         this_thread = malloc(sizeof(CtdlThreadNode));
 902         if (this_thread == NULL) {
 903                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 904                 return NULL;
 905         }
 906
 907         /* Initialise the thread structure */
 908         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
 909         {
 910                 free(this_thread);
 911                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
 912                 return NULL;
 913         }
 914         /*
 915          * If we got here we are going to create the thread so we must initilise the structure
 916          * first because most implimentations of threading can't create it in a stopped state
 917          * and it might want to do things with its structure that aren't initialised otherwise.
 918          */
 919         if(name)
 920         {
 921                 this_thread->name = name;
 922         }
 923         else
 924         {
 925                 this_thread->name = "Un-named Thread";
 926         }
 927
 928         this_thread->flags = flags;
 929         this_thread->thread_func = thread_func;
 930         this_thread->user_args = args;
 931
 932 //      citthread_mutex_lock(&this_thread->ThreadMutex);
 933
 934         begin_critical_section(S_THREAD_LIST);
 935         /*
 936          * We pass this_thread into the thread as its args so that it can find out information
 937          * about itself and it has a bit of storage space for itself, not to mention that the REAL
 938          * thread function needs to finish off the setup of the structure
 939          */
 940         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
 941         {
 942                 end_critical_section(S_THREAD_LIST);
 943                 CtdlLogPrintf(CTDL_ALERT, "Thread system, Can't create thread: %s\n",
 944                         strerror(ret));
 945                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 946                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 947                 citthread_cond_destroy(&(this_thread->ThreadCond));
 948                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 949                 citthread_cond_destroy(&(this_thread->SleepCond));
 950                 citthread_attr_destroy(&this_thread->attr);
 951                 free(this_thread);
 952                 return NULL;
 953         }
 954
 955         num_threads++;  // Increase the count of threads in the system.
 956         if(this_thread->flags & CTDLTHREAD_WORKER)
 957                 num_workers++;
 958
 959         this_thread->next = CtdlThreadList;
 960         CtdlThreadList = this_thread;
 961         if (this_thread->next)
 962                 this_thread->next->prev = this_thread;
 963         ctdl_thread_internal_calc_loadavg();
 964
 965 //      citthread_mutex_unlock(&this_thread->ThreadMutex);
 966         end_critical_section(S_THREAD_LIST);
 967
 968         return this_thread;
 969 }
 970
 971 /*
 972  * Wrapper function to create a thread
 973  * ensures the critical section and other protections are in place.
 974  * char *name = name to give to thread, if NULL, use generic name
 975  * int flags = flags to determine type of thread and standard facilities
 976  */
 977 CtdlThreadNode *CtdlThreadCreate(char *name, long flags, void *(*thread_func) (void *arg), void *args)
 978 {
 979         CtdlThreadNode *ret = NULL;
 980
 981         ret = ctdl_internal_create_thread(name, flags, thread_func, args);
 982         return ret;
 983 }
 984
 985
 986
 987 /*
 988  * Internal function to schedule a thread.
 989  * Must be called from within a S_THREAD_LIST critical section
 990  */
 991 CtdlThreadNode *CtdlThreadSchedule(char *name, long flags, void *(*thread_func) (void *arg), void *args, time_t when)
 992 {
 993         CtdlThreadNode *this_thread;
 994
 995         if (num_threads >= 32767)
 996         {
 997                 CtdlLogPrintf(CTDL_EMERG, "Thread system. Thread list full.\n");
 998                 return NULL;
 999         }
1000
1001         this_thread = malloc(sizeof(CtdlThreadNode));
1002         if (this_thread == NULL) {
1003                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
1004                 return NULL;
1005         }
1006         /* Initialise the thread structure */
1007         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
1008         {
1009                 free(this_thread);
1010                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
1011                 return NULL;
1012         }
1013
1014         /*
1015          * If we got here we are going to create the thread so we must initilise the structure
1016          * first because most implimentations of threading can't create it in a stopped state
1017          * and it might want to do things with its structure that aren't initialised otherwise.
1018          */
1019         if(name)
1020         {
1021                 this_thread->name = name;
1022         }
1023         else
1024         {
1025                 this_thread->name = "Un-named Thread";
1026         }
1027
1028         this_thread->flags = flags;
1029         this_thread->thread_func = thread_func;
1030         this_thread->user_args = args;
1031
1032         /*
1033          * When to start this thread
1034          */
1035         this_thread->when = when;
1036
1037         begin_critical_section(S_SCHEDULE_LIST);
1038         this_thread->next = CtdlThreadSchedList;
1039         CtdlThreadSchedList = this_thread;
1040         if (this_thread->next)
1041                 this_thread->next->prev = this_thread;
1042         end_critical_section(S_SCHEDULE_LIST);
1043
1044         return this_thread;
1045 }
1046
1047
1048
1049 CtdlThreadNode *ctdl_thread_internal_start_scheduled (CtdlThreadNode *this_thread)
1050 {
1051         int ret = 0;
1052
1053 //      citthread_mutex_lock(&that_thread->ThreadMutex);
1054         begin_critical_section(S_THREAD_LIST);
1055         /*
1056          * We pass this_thread into the thread as its args so that it can find out information
1057          * about itself and it has a bit of storage space for itself, not to mention that the REAL
1058          * thread function needs to finish off the setup of the structure
1059          */
1060         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
1061         {
1062                 end_critical_section(S_THREAD_LIST);
1063                 CtdlLogPrintf(CTDL_DEBUG, "Failed to start scheduled thread \"%s\": %s\n", this_thread->name, strerror(ret));
1064 //              citthread_mutex_unlock(&this_thread->ThreadMutex);
1065                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
1066                 citthread_cond_destroy(&(this_thread->ThreadCond));
1067                 citthread_mutex_destroy(&(this_thread->SleepMutex));
1068                 citthread_cond_destroy(&(this_thread->SleepCond));
1069                 citthread_attr_destroy(&this_thread->attr);
1070                 free(this_thread);
1071                 return NULL;
1072         }
1073
1074
1075         num_threads++;  // Increase the count of threads in the system.
1076         if(this_thread->flags & CTDLTHREAD_WORKER)
1077                 num_workers++;
1078
1079         this_thread->next = CtdlThreadList;
1080         CtdlThreadList = this_thread;
1081         if (this_thread->next)
1082                 this_thread->next->prev = this_thread;
1083 //      citthread_mutex_unlock(&that_thread->ThreadMutex);
1084
1085         ctdl_thread_internal_calc_loadavg();
1086         end_critical_section(S_THREAD_LIST);
1087
1088
1089         return this_thread;
1090 }
1091
1092
1093
1094 void ctdl_thread_internal_check_scheduled(void)
1095 {
1096         CtdlThreadNode *this_thread, *that_thread;
1097         time_t now;
1098
1099         if (try_critical_section(S_SCHEDULE_LIST))
1100                 return; /* If this list is locked we wait till the next chance */
1101
1102         now = time(NULL);
1103
1104 #ifdef WITH_THREADLOG
1105         CtdlLogPrintf(CTDL_DEBUG, "Checking for scheduled threads to start.\n");
1106 #endif
1107
1108         this_thread = CtdlThreadSchedList;
1109         while(this_thread)
1110         {
1111                 that_thread = this_thread;
1112                 this_thread = this_thread->next;
1113
1114                 if (now > that_thread->when)
1115                 {
1116                         /* Unlink from schedule list */
1117                         if (that_thread->prev)
1118                                 that_thread->prev->next = that_thread->next;
1119                         else
1120                                 CtdlThreadSchedList = that_thread->next;
1121                         if (that_thread->next)
1122                                 that_thread->next->prev = that_thread->prev;
1123
1124                         that_thread->next = that_thread->prev = NULL;
1125 #ifdef WITH_THREADLOG
1126                         CtdlLogPrintf(CTDL_DEBUG, "About to start scheduled thread \"%s\".\n", that_thread->name);
1127 #endif
1128                         if (CT->state > CTDL_THREAD_STOP_REQ)
1129                         {       /* Only start it if the system is not stopping */
1130                                 if (ctdl_thread_internal_start_scheduled (that_thread))
1131                                 {
1132 #ifdef WITH_THREADLOG
1133                                         CtdlLogPrintf(CTDL_INFO, "Thread system, Started a scheduled thread \"%s\" (0x%08lx).\n",
1134                                                 that_thread->name, that_thread->tid);
1135 #endif
1136                                 }
1137                         }
1138                 }
1139 #ifdef WITH_THREADLOG
1140                 else
1141                 {
1142                         CtdlLogPrintf(CTDL_DEBUG, "Thread \"%s\" will start in %ld seconds.\n",
1143                                 that_thread->name, that_thread->when - time(NULL));
1144                 }
1145 #endif
1146         }
1147         end_critical_section(S_SCHEDULE_LIST);
1148 }
1149
1150
1151 /*
1152  * A warapper function for select so we can show a thread as blocked
1153  */
1154 int CtdlThreadSelect(int n, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout)
1155 {
1156         int ret = 0;
1157
1158         ctdl_thread_internal_change_state(CT, CTDL_THREAD_BLOCKED);
1159         if (!CtdlThreadCheckStop())
1160                 ret = select(n, readfds, writefds, exceptfds, timeout);
1161         /**
1162          * If the select returned <= 0 then it failed due to an error
1163          * or timeout so this thread could stop if asked to do so.
1164          * Anything else means it needs to continue unless the system is shutting down
1165          */
1166         if (ret <= 0)
1167         {
1168                 /**
1169                  * select says nothing to do so we can change to running if we haven't been asked to stop.
1170                  */
1171                 ctdl_thread_internal_change_state(CT, CTDL_THREAD_RUNNING);
1172         }
1173         else
1174         {
1175                 /**
1176                  * The select says this thread needs to do something useful.
1177                  * This thread was in an idle state so it may have been asked to stop
1178                  * but if the system isn't shutting down this thread is no longer
1179                  * idle and select has given it a task to do so it must not stop
1180                  * In this condition we need to force it into the running state.
1181                  * CtdlThreadGC will clear its ticker for us.
1182                  *
1183                  * FIXME: there is still a small hole here. It is possible for the sequence of locking
1184                  * to allow the state to get changed to STOP_REQ just after this code if the other thread
1185                  * has decided to change the state before this lock, it there fore has to wait till the lock
1186                  * completes but it will continue to change the state. We need something a bit better here.
1187                  */
1188                 citthread_mutex_lock(&CT->ThreadMutex); /* To prevent race condition of a sleeping thread */
1189                 if (GC_thread->state > CTDL_THREAD_STOP_REQ && CT->state <= CTDL_THREAD_STOP_REQ)
1190                 {
1191                         CtdlLogPrintf(CTDL_DEBUG, "Thread %s (0x%08lx) refused stop request.\n", CT->name, CT->tid);
1192                         CT->state = CTDL_THREAD_RUNNING;
1193                 }
1194                 citthread_mutex_unlock(&CT->ThreadMutex);
1195         }
1196
1197         return ret;
1198 }
1199
1200
1201
1202 void *new_worker_thread(void *arg);
1203 extern void close_masters (void);
1204
1205
1206
1207 void go_threading(void)
1208 {
1209         int i;
1210         CtdlThreadNode *last_worker;
1211
1212         /*
1213          * Initialise the thread system
1214          */
1215         ctdl_thread_internal_init();
1216
1217         /* Second call to module init functions now that threading is up */
1218         initialise_modules(1);
1219
1220         /*
1221          * This thread is now used for garbage collection of other threads in the thread list
1222          */
1223         CtdlLogPrintf(CTDL_INFO, "Startup thread %d becoming garbage collector,\n", citthread_self());
1224
1225         /*
1226          * We do a lot of locking and unlocking of the thread list in here.
1227          * We do this so that we can repeatedly release time for other threads
1228          * that may be waiting on the thread list.
1229          * We are a low priority thread so we can afford to do this
1230          */
1231
1232         while (CtdlThreadGetCount())
1233         {
1234                 if (CT->signal)
1235                         exit_signal = CT->signal;
1236                 if (exit_signal)
1237                 {
1238                         CtdlThreadStopAll();
1239 //                      close_masters();
1240                 }
1241                 check_sched_shutdown();
1242                 if (CT->state > CTDL_THREAD_STOP_REQ)
1243                 {
1244                         begin_critical_section(S_THREAD_LIST);
1245                         ctdl_thread_internal_calc_loadavg();
1246                         end_critical_section(S_THREAD_LIST);
1247
1248                         ctdl_thread_internal_check_scheduled(); /* start scheduled threads */
1249                 }
1250
1251                 /* Reduce the size of the worker thread pool if necessary. */
1252                 if ((CtdlThreadGetWorkers() > config.c_min_workers + 1) && (CtdlThreadWorkerAvg < 20) && (CT->state > CTDL_THREAD_STOP_REQ))
1253                 {
1254                         /* Ask a worker thread to stop as we no longer need it */
1255                         begin_critical_section(S_THREAD_LIST);
1256                         last_worker = CtdlThreadList;
1257                         while (last_worker)
1258                         {
1259                                 citthread_mutex_lock(&last_worker->ThreadMutex);
1260                                 if (last_worker->flags & CTDLTHREAD_WORKER && (last_worker->state > CTDL_THREAD_STOPPING) && (last_worker->Context == NULL))
1261                                 {
1262                                         citthread_mutex_unlock(&last_worker->ThreadMutex);
1263                                         break;
1264                                 }
1265                                 citthread_mutex_unlock(&last_worker->ThreadMutex);
1266                                 last_worker = last_worker->next;
1267                         }
1268                         end_critical_section(S_THREAD_LIST);
1269                         if (last_worker)
1270                         {
1271 #ifdef WITH_THREADLOG
1272                                 CtdlLogPrintf(CTDL_DEBUG, "Thread system, stopping excess worker thread \"%s\" (0x%08lx).\n",
1273                                         last_worker->name,
1274                                         last_worker->tid
1275                                         );
1276 #endif
1277                                 CtdlThreadStop(last_worker);
1278                         }
1279                 }
1280
1281                 /*
1282                  * If all our workers are working hard, start some more to help out
1283                  * with things
1284                  */
1285                 /* FIXME: come up with a better way to dynamically alter the number of threads
1286                  * based on the system load
1287                  */
1288 #ifdef NEW_WORKER
1289                 if ((((CtdlThreadGetWorkers() < config.c_max_workers) && (CtdlThreadGetWorkers() <= num_sessions) ) || CtdlThreadGetWorkers() < config.c_min_workers) && (CT->state > CTDL_THREAD_STOP_REQ))
1290 #else
1291                 if ((((CtdlThreadGetWorkers() < config.c_max_workers) && (CtdlThreadGetWorkerAvg() > 60) && (CtdlThreadGetLoadAvg() < 90) ) || CtdlThreadGetWorkers() < config.c_min_workers) && (CT->state > CTDL_THREAD_STOP_REQ))
1292 #endif /* NEW_WORKER */
1293                 {
1294                         for (i=0; i<5 ; i++)
1295                         {
1296 #ifdef NEW_WORKER
1297                                 CtdlThreadCreate("Worker Thread (new)",
1298                                         CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1299                                         new_worker_thread,
1300                                         NULL
1301                                         );
1302 #else
1303                                 CtdlThreadCreate("Worker Thread",
1304                                         CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1305                                         worker_thread,
1306                                         NULL
1307                                         );
1308 #endif /* NEW_WORKER */
1309                         }
1310                 }
1311
1312                 CtdlThreadGC();
1313
1314                 if (CtdlThreadGetCount() <= 1) // Shutting down clean up the garbage collector
1315                 {
1316                         CtdlThreadGC();
1317                 }
1318
1319 #ifdef THREADS_USESIGNALS
1320                 if (CtdlThreadGetCount() && CT->state > CTDL_THREAD_STOP_REQ)
1321 #else
1322                 if (CtdlThreadGetCount())
1323 #endif
1324                         CtdlThreadSleep(1);
1325         }
1326         /*
1327          * If the above loop exits we must be shutting down since we obviously have no threads
1328          */
1329         ctdl_thread_internal_cleanup();
1330 }
1331
1332
1333
1334
1335 /*
1336  * Starting a new implimentation of a worker thread.
1337  * This new implimentation will be faster and do more work per thread.
1338  */
1339
1340 /*
1341  * Select on master socket.
1342  * First worker thread in here acquires the lock and builds an FDSET of master sockets.
1343  * then it goes into a loop selecting on the master sockets timing out every few milliseconds.
1344  * If it times out it rebiulds its list and loops.
1345  * If the select succeeds it creates a new context and returns.
1346  * During this time the other workers are selecting on existing contexts or sleeping.
1347  */
1348 void select_on_master(void)
1349 {
1350         fd_set readfds;
1351         struct ServiceFunctionHook *serviceptr;
1352         int ssock;                      /* Descriptor for client socket */
1353         int highest;
1354         int m, i;
1355         int retval = 0;
1356         struct timeval tv;
1357         struct CitContext *con;
1358         const char *old_name;
1359
1360
1361
1362         old_name = CtdlThreadName("select_on_master");
1363
1364         /* Initialize the fdset. */
1365         FD_ZERO(&readfds);
1366         highest = 0;
1367
1368         /* First, add the various master sockets to the fdset. */
1369         for (serviceptr = ServiceHookTable; serviceptr != NULL; serviceptr = serviceptr->next ) {
1370                 m = serviceptr->msock;
1371                 FD_SET(m, &readfds);
1372                 if (m > highest) {
1373                         highest = m;
1374                 }
1375         }
1376
1377         tv.tv_sec = 1;          /* wake up every 1 sec if no input */
1378         tv.tv_usec = 0;
1379         retval = CtdlThreadSelect(highest + 1, &readfds, NULL, NULL, &tv);
1380
1381         /* Select got an error or we are shutting down so get out */
1382         if (retval == 0 || CtdlThreadCheckStop()) {
1383                 CtdlThreadName(old_name);
1384                 return;
1385         }
1386
1387         /* Select says something happened on one of our master sockets so now we handle it */
1388         for (serviceptr = ServiceHookTable; serviceptr != NULL; serviceptr = serviceptr->next ) {
1389                 if (FD_ISSET(serviceptr->msock, &readfds)) {
1390                         ssock = accept(serviceptr->msock, NULL, 0);
1391                         if (ssock >= 0) {
1392                                 CtdlLogPrintf(CTDL_DEBUG, "New client socket %d\n", ssock);
1393                                 /* The master socket is non-blocking but the client
1394                                  * sockets need to be blocking, otherwise certain
1395                                  * operations barf on FreeBSD.  Not a fatal error.
1396                                  */
1397                                 if (fcntl(ssock, F_SETFL, 0) < 0) {
1398                                         CtdlLogPrintf(CTDL_EMERG,
1399                                                       "citserver: Can't set socket to blocking: %s\n",
1400                                                       strerror(errno));
1401                                 }
1402
1403                                 /* New context will be created already
1404                                  * set up in the CON_EXECUTING state.
1405                                  */
1406                                 con = CreateNewContext();
1407                                 CT->Context = con;
1408
1409                                 /* Assign our new socket number to it. */
1410                                 con->client_socket = ssock;
1411                                 con->h_command_function = serviceptr->h_command_function;
1412                                 con->h_async_function = serviceptr->h_async_function;
1413                                 con->ServiceName = serviceptr->ServiceName;
1414                                 /* Determine whether it's a local socket */
1415                                 if (serviceptr->sockpath != NULL)
1416                                         con->is_local_socket = 1;
1417
1418                                 /* Set the SO_REUSEADDR socket option */
1419                                 i = 1;
1420                                 setsockopt(ssock, SOL_SOCKET, SO_REUSEADDR, &i, sizeof(i));
1421
1422                                 become_session(con);
1423                                 begin_session(con);
1424                                 serviceptr->h_greeting_function();
1425                                 become_session(NULL);
1426                                 con->state = CON_IDLE;
1427                                 break;
1428                         }
1429                 }
1430         }
1431
1432         CtdlThreadName(old_name);
1433 }
1434
1435 /*
1436  * Select on client socket.
1437  * First worker thread in here acquires the lock and builds an FDSET of client sockets.
1438  * then it selects on the client sockets timing out after 1 second.
1439  * If it times out the thread goes off to check on housekeeping etc.
1440  * If the select succeeds the thread goes off to handle the client request.
1441  * If the list of client connections is empty the threads all sleep for one second
1442  */
1443 struct CitContext *select_on_client(void)
1444 {
1445         fd_set readfds;
1446         struct timeval tv;
1447         int retval = 0;
1448         int highest=0;
1449         const char *old_name;
1450
1451
1452         old_name = CtdlThreadName("select_on_client");
1453
1454         /* Initialise the fdset */
1455         FD_ZERO(&readfds);
1456         FD_SET(CT->Context->client_socket, &readfds);
1457         highest = CT->Context->client_socket;
1458         /* Now we can select on any connections that are waiting */
1459
1460         if (!CtdlThreadCheckStop())
1461         {
1462                 tv.tv_sec = config.c_sleeping;          /* wake up every second if no input */
1463                 tv.tv_usec = 0;
1464                 retval = select(highest + 1, &readfds, NULL, NULL, &tv);
1465         }
1466         else    /* Shutting down? */
1467         {
1468                 CtdlThreadName(old_name);
1469                 return(NULL);
1470         }
1471
1472
1473         /* Now figure out who made this select() unblock.
1474          * First, check for an error or exit condition.
1475          */
1476         if (retval < 0) {
1477                 if (errno == EBADF) {
1478                         CtdlLogPrintf(CTDL_NOTICE, "select() failed: (%s)\n",
1479                                 strerror(errno));
1480                 }
1481                 if (errno != EINTR) {
1482                         CtdlLogPrintf(CTDL_EMERG, "Exiting (%s)\n", strerror(errno));
1483                         CtdlThreadStopAll();
1484                 } else if (!CtdlThreadCheckStop()) {
1485                         CtdlLogPrintf(CTDL_DEBUG, "Un handled select failure.\n");
1486                 }
1487                 CtdlThreadName(old_name);
1488                 return NULL;
1489         }
1490         else if(retval == 0)
1491         {
1492                 CtdlThreadName(old_name);
1493                 CT->Context->kill_me = 1;
1494                 CT->Context = NULL;
1495                 return CT->Context;
1496         }
1497
1498         CT->Context->state = CON_EXECUTING;
1499         CT->Context->input_waiting = 1;
1500
1501         CtdlThreadName(old_name);
1502         return (CT->Context);
1503 }
1504
1505
1506
1507 /*
1508  * Do the worker threads work when needed
1509  */
1510 int execute_session(struct CitContext *bind_me)
1511 {
1512         int force_purge;
1513
1514         become_session(bind_me);
1515
1516         /* If the client has sent a command, execute it. */
1517         if (CC->input_waiting) {
1518                 CC->h_command_function();
1519                 CC->input_waiting = 0;
1520         }
1521
1522         /* If there are asynchronous messages waiting and the
1523          * client supports it, do those now */
1524         if ((CC->is_async) && (CC->async_waiting)
1525            && (CC->h_async_function != NULL)) {
1526                 CC->h_async_function();
1527                 CC->async_waiting = 0;
1528         }
1529
1530         force_purge = CC->kill_me;
1531         if (force_purge)
1532                 CT->Context = NULL;
1533         become_session(NULL);
1534         bind_me->state = CON_IDLE;
1535         return force_purge;
1536 }
1537
1538
1539
1540 extern void dead_session_purge(int force);
1541
1542 /*
1543  * A new worker_thread loop.
1544  */
1545
1546 void *new_worker_thread(void *arg)
1547 {
1548         struct CitContext *bind_me;
1549         int force_purge;
1550
1551         while (!CtdlThreadCheckStop()) {
1552
1553                 /* make doubly sure we're not holding any stale db handles
1554                  * which might cause a deadlock.
1555                  */
1556                 cdb_check_handles();
1557                 force_purge = 0;
1558                 bind_me = NULL;         /* Which session shall we handle? */
1559
1560                 if (CT->Context == NULL)
1561                         select_on_master();
1562                 if (CtdlThreadCheckStop())
1563                         break;
1564
1565                 if (CT->Context)
1566                         bind_me = select_on_client();
1567                 if (CtdlThreadCheckStop())
1568                         break;
1569
1570                 if (bind_me)
1571                         force_purge = execute_session(bind_me);
1572
1573                 dead_session_purge(force_purge);
1574                 if (CtdlThreadCheckStop())
1575                         break;
1576
1577                 do_housekeeping();
1578         }
1579         return NULL;
1580 }