citadel/threads.c

   1 /*
   2  * $Id$
   3  *
   4  * Citadel "system dependent" stuff.
   5  * See COPYING for copyright information.
   6  *
   7  * Here's where we have the Citadel thread implimentation
   8  *
   9  */
  10
  11 #include <stdlib.h>
  12 #include <unistd.h>
  13 #include <stdio.h>
  14 #include <sys/types.h>
  15 #include <errno.h>
  16 #include <sys/socket.h>
  17 #include <unistd.h>
  18 #include <fcntl.h>
  19 #include <signal.h>
  20
  21 #if TIME_WITH_SYS_TIME
  22 # include <sys/time.h>
  23 # include <time.h>
  24 #else
  25 # if HAVE_SYS_TIME_H
  26 #  include <sys/time.h>
  27 # else
  28 #  include <time.h>
  29 # endif
  30 #endif
  31
  32 #include <libcitadel.h>
  33
  34 #include "threads.h"
  35 #include "ctdl_module.h"
  36 #include "modules_init.h"
  37 #include "housekeeping.h"
  38 #include "config.h"
  39 #include "citserver.h"
  40 #include "sysdep_decls.h"
  41 #include "context.h"
  42
  43 /*
  44  * define this to use the new worker_thread method of handling connections
  45  */
  46 //#define NEW_WORKER
  47
  48 /*
  49  * New thread interface.
  50  * To create a thread you must call one of the create thread functions.
  51  * You must pass it the address of (a pointer to a CtdlThreadNode initialised to NULL) like this
  52  * struct CtdlThreadNode *node = NULL;
  53  * pass in &node
  54  * If the thread is created *node will point to the thread control structure for the created thread.
  55  * If the thread creation fails *node remains NULL
  56  * Do not free the memory pointed to by *node, it doesn't belong to you.
  57  * This new interface duplicates much of the eCrash stuff. We should go for closer integration since that would
  58  * remove the need for the calls to eCrashRegisterThread and friends
  59  */
  60
  61 static int num_threads = 0;                     /* Current number of threads */
  62 static int num_workers = 0;                     /* Current number of worker threads */
  63
  64 CtdlThreadNode *CtdlThreadList = NULL;
  65 CtdlThreadNode *CtdlThreadSchedList = NULL;
  66
  67 static CtdlThreadNode *GC_thread = NULL;
  68 static char *CtdlThreadStates[CTDL_THREAD_LAST_STATE];
  69 double CtdlThreadLoadAvg = 0;
  70 double CtdlThreadWorkerAvg = 0;
  71 citthread_key_t ThreadKey;
  72
  73 citthread_mutex_t Critters[MAX_SEMAPHORES];     /* Things needing locking */
  74
  75
  76
  77 void InitialiseSemaphores(void)
  78 {
  79         int i;
  80
  81         /* Set up a bunch of semaphores to be used for critical sections */
  82         for (i=0; i<MAX_SEMAPHORES; ++i) {
  83                 citthread_mutex_init(&Critters[i], NULL);
  84         }
  85 }
  86
  87
  88
  89
  90 /*
  91  * Obtain a semaphore lock to begin a critical section.
  92  * but only if no one else has one
  93  */
  94 int try_critical_section(int which_one)
  95 {
  96         /* For all types of critical sections except those listed here,
  97          * ensure nobody ever tries to do a critical section within a
  98          * transaction; this could lead to deadlock.
  99          */
 100         if (    (which_one != S_FLOORCACHE)
 101 #ifdef DEBUG_MEMORY_LEAKS
 102                 && (which_one != S_DEBUGMEMLEAKS)
 103 #endif
 104                 && (which_one != S_RPLIST)
 105         ) {
 106                 cdb_check_handles();
 107         }
 108         return (citthread_mutex_trylock(&Critters[which_one]));
 109 }
 110
 111
 112 /*
 113  * Obtain a semaphore lock to begin a critical section.
 114  */
 115 void begin_critical_section(int which_one)
 116 {
 117         /* CtdlLogPrintf(CTDL_DEBUG, "begin_critical_section(%d)\n", which_one); */
 118
 119         /* For all types of critical sections except those listed here,
 120          * ensure nobody ever tries to do a critical section within a
 121          * transaction; this could lead to deadlock.
 122          */
 123         if (    (which_one != S_FLOORCACHE)
 124 #ifdef DEBUG_MEMORY_LEAKS
 125                 && (which_one != S_DEBUGMEMLEAKS)
 126 #endif
 127                 && (which_one != S_RPLIST)
 128         ) {
 129                 cdb_check_handles();
 130         }
 131         citthread_mutex_lock(&Critters[which_one]);
 132 }
 133
 134 /*
 135  * Release a semaphore lock to end a critical section.
 136  */
 137 void end_critical_section(int which_one)
 138 {
 139         citthread_mutex_unlock(&Critters[which_one]);
 140 }
 141
 142
 143 /*
 144  * A function to destroy the TSD
 145  */
 146 static void ctdl_thread_internal_dest_tsd(void *arg)
 147 {
 148         if (arg != NULL) {
 149                 check_handles(arg);
 150                 free(arg);
 151         }
 152 }
 153
 154
 155 /*
 156  * A function to initialise the thread TSD
 157  */
 158 void ctdl_thread_internal_init_tsd(void)
 159 {
 160         int ret;
 161
 162         if ((ret = citthread_key_create(&ThreadKey, ctdl_thread_internal_dest_tsd))) {
 163                 CtdlLogPrintf(CTDL_EMERG, "citthread_key_create: %s\n", strerror(ret));
 164                 exit(CTDLEXIT_DB);
 165         }
 166 }
 167
 168 /*
 169  * Ensure that we have a key for thread-specific data.
 170  *
 171  * This should be called immediately after startup by any thread
 172  *
 173  */
 174 void CtdlThreadAllocTSD(void)
 175 {
 176         ThreadTSD *tsd;
 177
 178         if (citthread_getspecific(ThreadKey) != NULL)
 179                 return;
 180
 181         tsd = malloc(sizeof(ThreadTSD));
 182
 183         tsd->tid = NULL;
 184
 185         memset(tsd->cursors, 0, sizeof tsd->cursors);
 186         tsd->self = NULL;
 187
 188         citthread_setspecific(ThreadKey, tsd);
 189 }
 190
 191
 192 void ctdl_thread_internal_free_tsd(void)
 193 {
 194         ctdl_thread_internal_dest_tsd(citthread_getspecific(ThreadKey));
 195         citthread_setspecific(ThreadKey, NULL);
 196 }
 197
 198
 199 void ctdl_thread_internal_cleanup(void)
 200 {
 201         int i;
 202         CtdlThreadNode *this_thread, *that_thread;
 203
 204         for (i=0; i<CTDL_THREAD_LAST_STATE; i++)
 205         {
 206                 free (CtdlThreadStates[i]);
 207         }
 208
 209         /* Clean up the scheduled thread list */
 210         this_thread = CtdlThreadSchedList;
 211         while (this_thread)
 212         {
 213                 that_thread = this_thread;
 214                 this_thread = this_thread->next;
 215                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 216                 citthread_cond_destroy(&that_thread->ThreadCond);
 217                 citthread_mutex_destroy(&that_thread->SleepMutex);
 218                 citthread_cond_destroy(&that_thread->SleepCond);
 219                 citthread_attr_destroy(&that_thread->attr);
 220                 free(that_thread);
 221         }
 222         ctdl_thread_internal_free_tsd();
 223 }
 224
 225 void ctdl_thread_internal_init(void)
 226 {
 227         CtdlThreadNode *this_thread;
 228         int ret = 0;
 229
 230         CtdlThreadStates[CTDL_THREAD_INVALID] = strdup ("Invalid Thread");
 231         CtdlThreadStates[CTDL_THREAD_VALID] = strdup("Valid Thread");
 232         CtdlThreadStates[CTDL_THREAD_CREATE] = strdup("Thread being Created");
 233         CtdlThreadStates[CTDL_THREAD_CANCELLED] = strdup("Thread Cancelled");
 234         CtdlThreadStates[CTDL_THREAD_EXITED] = strdup("Thread Exited");
 235         CtdlThreadStates[CTDL_THREAD_STOPPING] = strdup("Thread Stopping");
 236         CtdlThreadStates[CTDL_THREAD_STOP_REQ] = strdup("Thread Stop Requested");
 237         CtdlThreadStates[CTDL_THREAD_SLEEPING] = strdup("Thread Sleeping");
 238         CtdlThreadStates[CTDL_THREAD_RUNNING] = strdup("Thread Running");
 239         CtdlThreadStates[CTDL_THREAD_BLOCKED] = strdup("Thread Blocked");
 240
 241         /* Get ourself a thread entry */
 242         this_thread = malloc(sizeof(CtdlThreadNode));
 243         if (this_thread == NULL) {
 244                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 245                 return;
 246         }
 247         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 248         memset (this_thread, 0, sizeof(CtdlThreadNode));
 249
 250         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 251         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 252         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 253         citthread_cond_init (&(this_thread->SleepCond), NULL);
 254
 255         /* We are garbage collector so create us as running */
 256         this_thread->state = CTDL_THREAD_RUNNING;
 257
 258         if ((ret = citthread_attr_init(&this_thread->attr))) {
 259                 CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 260                 free(this_thread);
 261                 return;
 262         }
 263
 264         this_thread->name = "Garbage Collection Thread";
 265
 266         this_thread->tid = citthread_self();
 267         GC_thread = this_thread;
 268         CT = this_thread;
 269
 270         num_threads++;  // Increase the count of threads in the system.
 271
 272         this_thread->next = CtdlThreadList;
 273         CtdlThreadList = this_thread;
 274         if (this_thread->next)
 275                 this_thread->next->prev = this_thread;
 276         /* Set up start times */
 277         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 278         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 279 }
 280
 281
 282 /*
 283  * A function to update a threads load averages
 284  */
 285  void ctdl_thread_internal_update_avgs(CtdlThreadNode *this_thread)
 286  {
 287         struct timeval now, result;
 288         double last_duration;
 289
 290         gettimeofday(&now, NULL);
 291         timersub(&now, &(this_thread->last_state_change), &result);
 292         /* I don't think these mutex's are needed here */
 293         citthread_mutex_lock(&this_thread->ThreadMutex);
 294         // result now has a timeval for the time we spent in the last state since we last updated
 295         last_duration = (double)result.tv_sec + ((double)result.tv_usec / (double) 1000000);
 296         if (this_thread->state == CTDL_THREAD_SLEEPING)
 297                 this_thread->avg_sleeping += last_duration;
 298         if (this_thread->state == CTDL_THREAD_RUNNING)
 299                 this_thread->avg_running += last_duration;
 300         if (this_thread->state == CTDL_THREAD_BLOCKED)
 301                 this_thread->avg_blocked += last_duration;
 302         memcpy (&this_thread->last_state_change, &now, sizeof (struct timeval));
 303         citthread_mutex_unlock(&this_thread->ThreadMutex);
 304 }
 305
 306 /*
 307  * A function to chenge the state of a thread
 308  */
 309 void ctdl_thread_internal_change_state (CtdlThreadNode *this_thread, enum CtdlThreadState new_state)
 310 {
 311         /*
 312          * Wether we change state or not we need update the load values
 313          */
 314         ctdl_thread_internal_update_avgs(this_thread);
 315         /* This mutex not needed here? */
 316         citthread_mutex_lock(&this_thread->ThreadMutex); /* To prevent race condition of a sleeping thread */
 317         if ((new_state == CTDL_THREAD_STOP_REQ) && (this_thread->state > CTDL_THREAD_STOP_REQ))
 318                 this_thread->state = new_state;
 319         if (((new_state == CTDL_THREAD_SLEEPING) || (new_state == CTDL_THREAD_BLOCKED)) && (this_thread->state == CTDL_THREAD_RUNNING))
 320                 this_thread->state = new_state;
 321         if ((new_state == CTDL_THREAD_RUNNING) && ((this_thread->state == CTDL_THREAD_SLEEPING) || (this_thread->state == CTDL_THREAD_BLOCKED)))
 322                 this_thread->state = new_state;
 323         citthread_mutex_unlock(&this_thread->ThreadMutex);
 324 }
 325
 326
 327 /*
 328  * A function to tell all threads to exit
 329  */
 330 void CtdlThreadStopAll(void)
 331 {
 332         /* First run any registered shutdown hooks.  This probably doesn't belong here. */
 333         PerformSessionHooks(EVT_SHUTDOWN);
 334
 335         //FIXME: The signalling of the condition should not be in the critical_section
 336         // We need to build a list of threads we are going to signal and then signal them afterwards
 337
 338         CtdlThreadNode *this_thread;
 339
 340         begin_critical_section(S_THREAD_LIST);
 341         this_thread = CtdlThreadList;
 342         // Ask the GC thread to stop first so everything knows we are shutting down.
 343         GC_thread->state = CTDL_THREAD_STOP_REQ;
 344         while(this_thread)
 345         {
 346 #ifdef THREADS_USESIGNALS
 347                 if (!citthread_equal(this_thread->tid, GC_thread->tid))
 348                         citthread_kill(this_thread->tid, SIGHUP);
 349 #endif
 350                 ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 351                 citthread_cond_signal(&this_thread->ThreadCond);
 352                 citthread_cond_signal(&this_thread->SleepCond);
 353                 this_thread->stop_ticker = time(NULL);
 354                 CtdlLogPrintf(CTDL_DEBUG, "Thread system stopping thread \"%s\" (0x%08lx).\n",
 355                         this_thread->name, this_thread->tid);
 356                 this_thread = this_thread->next;
 357         }
 358         end_critical_section(S_THREAD_LIST);
 359 }
 360
 361
 362 /*
 363  * A function to wake up all sleeping threads
 364  */
 365 void CtdlThreadWakeAll(void)
 366 {
 367         CtdlThreadNode *this_thread;
 368
 369         CtdlLogPrintf(CTDL_DEBUG, "Thread system waking all threads.\n");
 370
 371         begin_critical_section(S_THREAD_LIST);
 372         this_thread = CtdlThreadList;
 373         while(this_thread)
 374         {
 375                 if (!this_thread->thread_func)
 376                 {
 377                         citthread_cond_signal(&this_thread->ThreadCond);
 378                         citthread_cond_signal(&this_thread->SleepCond);
 379                 }
 380                 this_thread = this_thread->next;
 381         }
 382         end_critical_section(S_THREAD_LIST);
 383 }
 384
 385
 386 /*
 387  * A function to return the number of threads running in the system
 388  */
 389 int CtdlThreadGetCount(void)
 390 {
 391         return  num_threads;
 392 }
 393
 394 int CtdlThreadGetWorkers(void)
 395 {
 396         return  num_workers;
 397 }
 398
 399 double CtdlThreadGetWorkerAvg(void)
 400 {
 401         double ret;
 402
 403         begin_critical_section(S_THREAD_LIST);
 404         ret =  CtdlThreadWorkerAvg;
 405         end_critical_section(S_THREAD_LIST);
 406         return ret;
 407 }
 408
 409 double CtdlThreadGetLoadAvg(void)
 410 {
 411         double load_avg[3] ;
 412
 413         int ret;
 414         int smp_num_cpus;
 415
 416         /* Borrowed this straight from procps */
 417         smp_num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 418         if(smp_num_cpus<1) smp_num_cpus=1; /* SPARC glibc is buggy */
 419
 420         ret = getloadavg(load_avg, 3);
 421         if (ret < 0)
 422                 return 0;
 423         return load_avg[0] / smp_num_cpus;
 424 /*
 425  * This old chunk of code return a value that indicated the load on citserver
 426  * This value could easily reach 100 % even when citserver was doing very little and
 427  * hence the machine has much more spare capacity.
 428  * Because this value was used to determine if the machine was under heavy load conditions
 429  * from other processes in the system then citserver could be strangled un-necesarily
 430  * What we are actually trying to achieve is to strangle citserver if the machine is heavily loaded.
 431  * So we have changed this.
 432
 433         begin_critical_section(S_THREAD_LIST);
 434         ret =  CtdlThreadLoadAvg;
 435         end_critical_section(S_THREAD_LIST);
 436         return ret;
 437 */
 438 }
 439
 440
 441
 442
 443 /*
 444  * A function to rename a thread
 445  * Returns a const char *
 446  */
 447 const char *CtdlThreadName(const char *name)
 448 {
 449         const char *old_name;
 450
 451         if (!CT)
 452         {
 453                 CtdlLogPrintf(CTDL_WARNING, "Thread system WARNING. Attempt to CtdlThreadRename() a non thread. %s\n", name);
 454                 return NULL;
 455         }
 456         old_name = CT->name;
 457         if (name)
 458                 CT->name = name;
 459         return (old_name);
 460 }
 461
 462
 463 /*
 464  * A function to force a thread to exit
 465  */
 466 void CtdlThreadCancel(CtdlThreadNode *thread)
 467 {
 468         CtdlThreadNode *this_thread;
 469
 470         if (!thread)
 471                 this_thread = CT;
 472         else
 473                 this_thread = thread;
 474         if (!this_thread)
 475         {
 476                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() a non thread.\n");
 477                 CtdlThreadStopAll();
 478                 return;
 479         }
 480
 481         if (!this_thread->thread_func)
 482         {
 483                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() the garbage collector.\n");
 484                 CtdlThreadStopAll();
 485                 return;
 486         }
 487
 488         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_CANCELLED);
 489         citthread_cancel(this_thread->tid);
 490 }
 491
 492
 493 /*
 494  * A function for a thread to check if it has been asked to stop
 495  */
 496 int CtdlThreadCheckStop(void)
 497 {
 498         int state;
 499
 500         if (!CT)
 501         {
 502                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, CtdlThreadCheckStop() called by a non thread.\n");
 503                 CtdlThreadStopAll();
 504                 return -1;
 505         }
 506
 507         state = CT->state;
 508
 509 #ifdef THREADS_USESIGNALS
 510         if (CT->signal)
 511         {
 512                 CtdlLogPrintf(CTDL_DEBUG, "Thread \"%s\" caught signal %d.\n", CT->name, CT->signal);
 513                 CT->signal = 0;
 514         }
 515 #endif
 516         if(state == CTDL_THREAD_STOP_REQ)
 517         {
 518                 CT->state = CTDL_THREAD_STOPPING;
 519                 return -1;
 520         }
 521         else if((state < CTDL_THREAD_STOP_REQ) && (state > CTDL_THREAD_CREATE))
 522         {
 523                 return -1;
 524         }
 525         return 0;
 526 }
 527
 528
 529 /*
 530  * A function to ask a thread to exit
 531  * The thread must call CtdlThreadCheckStop() periodically to determine if it should exit
 532  */
 533 void CtdlThreadStop(CtdlThreadNode *thread)
 534 {
 535         CtdlThreadNode *this_thread;
 536
 537         if (!thread)
 538                 this_thread = CT;
 539         else
 540                 this_thread = thread;
 541         if (!this_thread)
 542                 return;
 543         if (!(this_thread->thread_func))
 544                 return;         // Don't stop garbage collector
 545 #ifdef THREADS_USESIGNALS
 546         if (!citthread_equal(this_thread->tid, GC_thread->tid))
 547                 citthread_kill(this_thread->tid, SIGHUP);
 548 #endif
 549         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 550         citthread_cond_signal(&this_thread->ThreadCond);
 551         citthread_cond_signal(&this_thread->SleepCond);
 552         this_thread->stop_ticker = time(NULL);
 553 }
 554
 555 /*
 556  * So we now have a sleep command that works with threads but it is in seconds
 557  */
 558 void CtdlThreadSleep(int secs)
 559 {
 560         struct timespec wake_time;
 561         struct timeval time_now;
 562
 563
 564         if (!CT)
 565         {
 566                 CtdlLogPrintf(CTDL_WARNING, "CtdlThreadSleep() called by something that is not a thread. Should we die?\n");
 567                 return;
 568         }
 569
 570         memset (&wake_time, 0, sizeof(struct timespec));
 571         gettimeofday(&time_now, NULL);
 572         wake_time.tv_sec = time_now.tv_sec + secs;
 573         wake_time.tv_nsec = time_now.tv_usec * 10;
 574
 575         ctdl_thread_internal_change_state (CT, CTDL_THREAD_SLEEPING);
 576
 577         citthread_mutex_lock(&CT->ThreadMutex); /* Prevent something asking us to awaken before we've gone to sleep */
 578         citthread_cond_timedwait(&CT->SleepCond, &CT->ThreadMutex, &wake_time);
 579         citthread_mutex_unlock(&CT->ThreadMutex);
 580
 581         ctdl_thread_internal_change_state (CT, CTDL_THREAD_RUNNING);
 582 }
 583
 584
 585 /*
 586  * Routine to clean up our thread function on exit
 587  */
 588 static void ctdl_internal_thread_cleanup(void *arg)
 589 {
 590         /*
 591          * In here we were called by the current thread because it is exiting
 592          * NB. WE ARE THE CURRENT THREAD
 593          */
 594         CtdlLogPrintf(CTDL_NOTICE, "Thread \"%s\" (0x%08lx) exited.\n", CT->name, CT->tid);
 595
 596         #ifdef HAVE_BACKTRACE
 597         eCrash_UnregisterThread();
 598         #endif
 599
 600         citthread_mutex_lock(&CT->ThreadMutex);
 601         CT->state = CTDL_THREAD_EXITED; // needs to be last thing else house keeping will unlink us too early
 602         citthread_mutex_unlock(&CT->ThreadMutex);
 603 }
 604
 605 /*
 606  * A quick function to show the load averages
 607  */
 608 void ctdl_thread_internal_calc_loadavg(void)
 609 {
 610         CtdlThreadNode *that_thread;
 611         double load_avg, worker_avg;
 612         int workers = 0;
 613
 614         that_thread = CtdlThreadList;
 615         load_avg = 0;
 616         worker_avg = 0;
 617         while(that_thread)
 618         {
 619                 /* Update load averages */
 620                 ctdl_thread_internal_update_avgs(that_thread);
 621                 citthread_mutex_lock(&that_thread->ThreadMutex);
 622                 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;
 623                 that_thread->avg_sleeping /= 2;
 624                 that_thread->avg_running /= 2;
 625                 that_thread->avg_blocked /= 2;
 626                 load_avg += that_thread->load_avg;
 627                 if (that_thread->flags & CTDLTHREAD_WORKER)
 628                 {
 629                         worker_avg += that_thread->load_avg;
 630                         workers++;
 631                 }
 632 #ifdef WITH_THREADLOG
 633                 CtdlLogPrintf(CTDL_DEBUG, "CtdlThread, \"%s\" (%lu) \"%s\" %.2f %.2f %.2f %.2f\n",
 634                         that_thread->name,
 635                         that_thread->tid,
 636                         CtdlThreadStates[that_thread->state],
 637                         that_thread->avg_sleeping,
 638                         that_thread->avg_running,
 639                         that_thread->avg_blocked,
 640                         that_thread->load_avg);
 641 #endif
 642                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 643                 that_thread = that_thread->next;
 644         }
 645         CtdlThreadLoadAvg = load_avg/num_threads;
 646         CtdlThreadWorkerAvg = worker_avg/workers;
 647 #ifdef WITH_THREADLOG
 648         CtdlLogPrintf(CTDL_INFO, "System load average %.2f, workers averag %.2f, threads %d, workers %d, sessions %d\n", CtdlThreadGetLoadAvg(), CtdlThreadWorkerAvg, num_threads, num_workers, num_sessions);
 649 #endif
 650 }
 651
 652
 653 /*
 654  * Garbage collection routine.
 655  * Gets called by main() in a loop to clean up the thread list periodically.
 656  */
 657 void CtdlThreadGC (void)
 658 {
 659         CtdlThreadNode *this_thread, *that_thread;
 660         int workers = 0, sys_workers;
 661         int ret=0;
 662
 663         begin_critical_section(S_THREAD_LIST);
 664
 665         /* Handle exiting of garbage collector thread */
 666         if(num_threads == 1)
 667                 CtdlThreadList->state = CTDL_THREAD_EXITED;
 668
 669 #ifdef WITH_THREADLOG
 670         CtdlLogPrintf(CTDL_DEBUG, "Thread system running garbage collection.\n");
 671 #endif
 672         /*
 673          * Woke up to do garbage collection
 674          */
 675         this_thread = CtdlThreadList;
 676         while(this_thread)
 677         {
 678                 that_thread = this_thread;
 679                 this_thread = this_thread->next;
 680
 681                 if ((that_thread->state == CTDL_THREAD_STOP_REQ || that_thread->state == CTDL_THREAD_STOPPING)
 682                         && (!citthread_equal(that_thread->tid, citthread_self())))
 683                                 CtdlLogPrintf(CTDL_DEBUG, "Waiting for thread %s (0x%08lx) to exit.\n", that_thread->name, that_thread->tid);
 684                 else
 685                 {
 686                         /**
 687                          * Catch the situation where a worker was asked to stop but couldn't and we are not
 688                          * shutting down.
 689                          */
 690                         that_thread->stop_ticker = 0;
 691                 }
 692
 693                 if (that_thread->stop_ticker + 5 == time(NULL))
 694                 {
 695                         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);
 696                         if ((that_thread->flags & CTDLTHREAD_WORKER) == 0)
 697                                 CtdlLogPrintf(CTDL_INFO, "Thread System: A non worker thread would have been canceled this may cause message loss.\n");
 698 //                      that_thread->state = CTDL_THREAD_CANCELLED;
 699                         that_thread->stop_ticker++;
 700 //                      citthread_cancel(that_thread->tid);
 701 //                      continue;
 702                 }
 703
 704                 /* Do we need to clean up this thread? */
 705                 if ((that_thread->state != CTDL_THREAD_EXITED) && (that_thread->state != CTDL_THREAD_CANCELLED))
 706                 {
 707                         if(that_thread->flags & CTDLTHREAD_WORKER)
 708                                 workers++;      /* Sanity check on number of worker threads */
 709                         continue;
 710                 }
 711
 712                 if (citthread_equal(that_thread->tid, citthread_self()) && that_thread->thread_func)
 713                 {       /* Sanity check */
 714                         end_critical_section(S_THREAD_LIST);
 715                         CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, a thread is trying to clean up after itself.\n");
 716                         abort();
 717                         return;
 718                 }
 719
 720                 if (num_threads <= 0)
 721                 {       /* Sanity check */
 722                         end_critical_section(S_THREAD_LIST);
 723                         CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, num_threads <= 0 and trying to do Garbage Collection.\n");
 724                         abort();
 725                         return;
 726                 }
 727
 728                 if(that_thread->flags & CTDLTHREAD_WORKER)
 729                         num_workers--;  /* This is a wroker thread so reduce the count. */
 730                 num_threads--;
 731                 /* If we are unlinking the list head then the next becomes the list head */
 732                 if(that_thread->prev)
 733                         that_thread->prev->next = that_thread->next;
 734                 else
 735                         CtdlThreadList = that_thread->next;
 736                 if(that_thread->next)
 737                         that_thread->next->prev = that_thread->prev;
 738
 739                 citthread_cond_signal(&that_thread->ThreadCond);
 740                 citthread_cond_signal(&that_thread->SleepCond); // Make sure this thread is awake
 741                 citthread_mutex_lock(&that_thread->ThreadMutex);        // Make sure it has done what its doing
 742                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 743                 /*
 744                  * Join on the thread to do clean up and prevent memory leaks
 745                  * Also makes sure the thread has cleaned up after itself before we remove it from the list
 746                  * We can join on the garbage collector thread the join should just return EDEADLCK
 747                  */
 748                 ret = citthread_join (that_thread->tid, NULL);
 749                 if (ret == EDEADLK)
 750                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection on own thread.\n");
 751                 else if (ret == EINVAL)
 752                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, that thread already joined on.\n");
 753                 else if (ret == ESRCH)
 754                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, no thread to join on.\n");
 755                 else if (ret != 0)
 756                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, citthread_join returned an unknown error(%d).\n", ret);
 757                 /*
 758                  * Now we own that thread entry
 759                  */
 760                 CtdlLogPrintf(CTDL_INFO, "Garbage Collection for thread \"%s\" (0x%08lx).\n",
 761                         that_thread->name, that_thread->tid);
 762                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 763                 citthread_cond_destroy(&that_thread->ThreadCond);
 764                 citthread_mutex_destroy(&that_thread->SleepMutex);
 765                 citthread_cond_destroy(&that_thread->SleepCond);
 766                 citthread_attr_destroy(&that_thread->attr);
 767                 free(that_thread);
 768         }
 769         sys_workers = num_workers;
 770         end_critical_section(S_THREAD_LIST);
 771
 772         /* Sanity check number of worker threads */
 773         if (workers != sys_workers)
 774         {
 775                 CtdlLogPrintf(CTDL_EMERG,
 776                         "Thread system PANIC, discrepancy in number of worker threads. Counted %d, should be %d.\n",
 777                         workers, sys_workers
 778                         );
 779                 abort();
 780         }
 781 }
 782
 783
 784
 785
 786 /*
 787  * Runtime function for a Citadel Thread.
 788  * This initialises the threads environment and then calls the user supplied thread function
 789  * Note that this is the REAL thread function and wraps the users thread function.
 790  */
 791 static void *ctdl_internal_thread_func (void *arg)
 792 {
 793         CtdlThreadNode *this_thread;
 794         void *ret = NULL;
 795
 796         /* lock and unlock the thread list.
 797          * This causes this thread to wait until all its creation stuff has finished before it
 798          * can continue its execution.
 799          */
 800         begin_critical_section(S_THREAD_LIST);
 801         this_thread = (CtdlThreadNode *) arg;
 802         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 803 //      citthread_mutex_lock(&this_thread->ThreadMutex);
 804
 805         // Register the cleanup function to take care of when we exit.
 806         citthread_cleanup_push(ctdl_internal_thread_cleanup, NULL);
 807         // Get our thread data structure
 808         CtdlThreadAllocTSD();
 809         CT = this_thread;
 810         this_thread->pid = getpid();
 811         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 812         /* Only change to running state if we weren't asked to stop during the create cycle
 813          * Other wise there is a window to allow this threads creation to continue to full grown and
 814          * therby prevent a shutdown of the server.
 815          */
 816 //      citthread_mutex_unlock(&this_thread->ThreadMutex);
 817
 818         if (!CtdlThreadCheckStop())
 819         {
 820                 citthread_mutex_lock(&this_thread->ThreadMutex);
 821                 this_thread->state = CTDL_THREAD_RUNNING;
 822                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 823         }
 824         end_critical_section(S_THREAD_LIST);
 825
 826         // Register for tracing
 827         #ifdef HAVE_BACKTRACE
 828         eCrash_RegisterThread(this_thread->name, 0);
 829         #endif
 830
 831         // Tell the world we are here
 832         CtdlLogPrintf(CTDL_NOTICE, "Created a new thread \"%s\" (0x%08lx).\n",
 833                 this_thread->name, this_thread->tid);
 834
 835         /*
 836          * run the thread to do the work but only if we haven't been asked to stop
 837          */
 838         if (!CtdlThreadCheckStop())
 839                 ret = (this_thread->thread_func)(this_thread->user_args);
 840
 841         /*
 842          * Our thread is exiting either because it wanted to end or because the server is stopping
 843          * We need to clean up
 844          */
 845         citthread_cleanup_pop(1);       // Execute our cleanup routine and remove it
 846
 847         return(ret);
 848 }
 849
 850
 851
 852
 853 /*
 854  * Function to initialise an empty thread structure
 855  */
 856 CtdlThreadNode *ctdl_internal_init_thread_struct(CtdlThreadNode *this_thread, long flags)
 857 {
 858         int ret = 0;
 859
 860         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 861         memset (this_thread, 0, sizeof(CtdlThreadNode));
 862
 863         /* Create the mutex's early so we can use them */
 864         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 865         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 866         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 867         citthread_cond_init (&(this_thread->SleepCond), NULL);
 868
 869         this_thread->state = CTDL_THREAD_CREATE;
 870
 871         if ((ret = citthread_attr_init(&this_thread->attr))) {
 872                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 873                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 874                 citthread_cond_destroy(&(this_thread->ThreadCond));
 875                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 876                 citthread_cond_destroy(&(this_thread->SleepCond));
 877                 CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 878                 free(this_thread);
 879                 return NULL;
 880         }
 881
 882         /* Our per-thread stacks need to be bigger than the default size,
 883          * otherwise the MIME parser crashes on FreeBSD, and the IMAP service
 884          * crashes on 64-bit Linux.
 885          */
 886         if (flags & CTDLTHREAD_BIGSTACK)
 887         {
 888 #ifdef WITH_THREADLOG
 889                 CtdlLogPrintf(CTDL_INFO, "Thread system. Creating BIG STACK thread.\n");
 890 #endif
 891                 if ((ret = citthread_attr_setstacksize(&this_thread->attr, THREADSTACKSIZE))) {
 892                         citthread_mutex_unlock(&this_thread->ThreadMutex);
 893                         citthread_mutex_destroy(&(this_thread->ThreadMutex));
 894                         citthread_cond_destroy(&(this_thread->ThreadCond));
 895                         citthread_mutex_destroy(&(this_thread->SleepMutex));
 896                         citthread_cond_destroy(&(this_thread->SleepCond));
 897                         citthread_attr_destroy(&this_thread->attr);
 898                         CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_setstacksize: %s\n",
 899                                 strerror(ret));
 900                         free(this_thread);
 901                         return NULL;
 902                 }
 903         }
 904
 905         /* Set this new thread with an avg_blocked of 2. We do this so that its creation affects the
 906          * load average for the system. If we don't do this then we create a mass of threads at the same time
 907          * because the creation didn't affect the load average.
 908          */
 909         this_thread->avg_blocked = 2;
 910
 911         return (this_thread);
 912 }
 913
 914
 915
 916
 917 /*
 918  * Internal function to create a thread.
 919  */
 920 CtdlThreadNode *ctdl_internal_create_thread(char *name, long flags, void *(*thread_func) (void *arg), void *args)
 921 {
 922         int ret = 0;
 923         CtdlThreadNode *this_thread;
 924
 925         if (num_threads >= 32767)
 926         {
 927                 CtdlLogPrintf(CTDL_EMERG, "Thread system. Thread list full.\n");
 928                 return NULL;
 929         }
 930
 931         this_thread = malloc(sizeof(CtdlThreadNode));
 932         if (this_thread == NULL) {
 933                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 934                 return NULL;
 935         }
 936
 937         /* Initialise the thread structure */
 938         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
 939         {
 940                 free(this_thread);
 941                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
 942                 return NULL;
 943         }
 944         /*
 945          * If we got here we are going to create the thread so we must initilise the structure
 946          * first because most implimentations of threading can't create it in a stopped state
 947          * and it might want to do things with its structure that aren't initialised otherwise.
 948          */
 949         if(name)
 950         {
 951                 this_thread->name = name;
 952         }
 953         else
 954         {
 955                 this_thread->name = "Un-named Thread";
 956         }
 957
 958         this_thread->flags = flags;
 959         this_thread->thread_func = thread_func;
 960         this_thread->user_args = args;
 961
 962 //      citthread_mutex_lock(&this_thread->ThreadMutex);
 963
 964         begin_critical_section(S_THREAD_LIST);
 965         /*
 966          * We pass this_thread into the thread as its args so that it can find out information
 967          * about itself and it has a bit of storage space for itself, not to mention that the REAL
 968          * thread function needs to finish off the setup of the structure
 969          */
 970         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
 971         {
 972                 end_critical_section(S_THREAD_LIST);
 973                 CtdlLogPrintf(CTDL_ALERT, "Thread system, Can't create thread: %s\n",
 974                         strerror(ret));
 975                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 976                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 977                 citthread_cond_destroy(&(this_thread->ThreadCond));
 978                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 979                 citthread_cond_destroy(&(this_thread->SleepCond));
 980                 citthread_attr_destroy(&this_thread->attr);
 981                 free(this_thread);
 982                 return NULL;
 983         }
 984
 985         num_threads++;  // Increase the count of threads in the system.
 986         if(this_thread->flags & CTDLTHREAD_WORKER)
 987                 num_workers++;
 988
 989         this_thread->next = CtdlThreadList;
 990         CtdlThreadList = this_thread;
 991         if (this_thread->next)
 992                 this_thread->next->prev = this_thread;
 993         ctdl_thread_internal_calc_loadavg();
 994
 995 //      citthread_mutex_unlock(&this_thread->ThreadMutex);
 996         end_critical_section(S_THREAD_LIST);
 997
 998         return this_thread;
 999 }
1000
1001 /*
1002  * Wrapper function to create a thread
1003  * ensures the critical section and other protections are in place.
1004  * char *name = name to give to thread, if NULL, use generic name
1005  * int flags = flags to determine type of thread and standard facilities
1006  */
1007 CtdlThreadNode *CtdlThreadCreate(char *name, long flags, void *(*thread_func) (void *arg), void *args)
1008 {
1009         CtdlThreadNode *ret = NULL;
1010
1011         ret = ctdl_internal_create_thread(name, flags, thread_func, args);
1012         return ret;
1013 }
1014
1015
1016
1017 /*
1018  * Internal function to schedule a thread.
1019  * Must be called from within a S_THREAD_LIST critical section
1020  */
1021 CtdlThreadNode *CtdlThreadSchedule(char *name, long flags, void *(*thread_func) (void *arg), void *args, time_t when)
1022 {
1023         CtdlThreadNode *this_thread;
1024
1025         if (num_threads >= 32767)
1026         {
1027                 CtdlLogPrintf(CTDL_EMERG, "Thread system. Thread list full.\n");
1028                 return NULL;
1029         }
1030
1031         this_thread = malloc(sizeof(CtdlThreadNode));
1032         if (this_thread == NULL) {
1033                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
1034                 return NULL;
1035         }
1036         /* Initialise the thread structure */
1037         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
1038         {
1039                 free(this_thread);
1040                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
1041                 return NULL;
1042         }
1043
1044         /*
1045          * If we got here we are going to create the thread so we must initilise the structure
1046          * first because most implimentations of threading can't create it in a stopped state
1047          * and it might want to do things with its structure that aren't initialised otherwise.
1048          */
1049         if(name)
1050         {
1051                 this_thread->name = name;
1052         }
1053         else
1054         {
1055                 this_thread->name = "Un-named Thread";
1056         }
1057
1058         this_thread->flags = flags;
1059         this_thread->thread_func = thread_func;
1060         this_thread->user_args = args;
1061
1062         /*
1063          * When to start this thread
1064          */
1065         this_thread->when = when;
1066
1067         begin_critical_section(S_SCHEDULE_LIST);
1068         this_thread->next = CtdlThreadSchedList;
1069         CtdlThreadSchedList = this_thread;
1070         if (this_thread->next)
1071                 this_thread->next->prev = this_thread;
1072         end_critical_section(S_SCHEDULE_LIST);
1073
1074         return this_thread;
1075 }
1076
1077
1078
1079 CtdlThreadNode *ctdl_thread_internal_start_scheduled (CtdlThreadNode *this_thread)
1080 {
1081         int ret = 0;
1082
1083 //      citthread_mutex_lock(&that_thread->ThreadMutex);
1084         begin_critical_section(S_THREAD_LIST);
1085         /*
1086          * We pass this_thread into the thread as its args so that it can find out information
1087          * about itself and it has a bit of storage space for itself, not to mention that the REAL
1088          * thread function needs to finish off the setup of the structure
1089          */
1090         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
1091         {
1092                 end_critical_section(S_THREAD_LIST);
1093                 CtdlLogPrintf(CTDL_DEBUG, "Failed to start scheduled thread \"%s\": %s\n", this_thread->name, strerror(ret));
1094 //              citthread_mutex_unlock(&this_thread->ThreadMutex);
1095                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
1096                 citthread_cond_destroy(&(this_thread->ThreadCond));
1097                 citthread_mutex_destroy(&(this_thread->SleepMutex));
1098                 citthread_cond_destroy(&(this_thread->SleepCond));
1099                 citthread_attr_destroy(&this_thread->attr);
1100                 free(this_thread);
1101                 return NULL;
1102         }
1103
1104
1105         num_threads++;  // Increase the count of threads in the system.
1106         if(this_thread->flags & CTDLTHREAD_WORKER)
1107                 num_workers++;
1108
1109         this_thread->next = CtdlThreadList;
1110         CtdlThreadList = this_thread;
1111         if (this_thread->next)
1112                 this_thread->next->prev = this_thread;
1113 //      citthread_mutex_unlock(&that_thread->ThreadMutex);
1114
1115         ctdl_thread_internal_calc_loadavg();
1116         end_critical_section(S_THREAD_LIST);
1117
1118
1119         return this_thread;
1120 }
1121
1122
1123
1124 void ctdl_thread_internal_check_scheduled(void)
1125 {
1126         CtdlThreadNode *this_thread, *that_thread;
1127         time_t now;
1128
1129         /* Don't start scheduled threads if the system wants single user mode */
1130         if (CtdlWantSingleUser())
1131                 return;
1132
1133         if (try_critical_section(S_SCHEDULE_LIST))
1134                 return; /* If this list is locked we wait till the next chance */
1135
1136         now = time(NULL);
1137
1138 #ifdef WITH_THREADLOG
1139         CtdlLogPrintf(CTDL_DEBUG, "Checking for scheduled threads to start.\n");
1140 #endif
1141
1142         this_thread = CtdlThreadSchedList;
1143         while(this_thread)
1144         {
1145                 that_thread = this_thread;
1146                 this_thread = this_thread->next;
1147
1148                 if (now > that_thread->when)
1149                 {
1150                         /* Unlink from schedule list */
1151                         if (that_thread->prev)
1152                                 that_thread->prev->next = that_thread->next;
1153                         else
1154                                 CtdlThreadSchedList = that_thread->next;
1155                         if (that_thread->next)
1156                                 that_thread->next->prev = that_thread->prev;
1157
1158                         that_thread->next = that_thread->prev = NULL;
1159 #ifdef WITH_THREADLOG
1160                         CtdlLogPrintf(CTDL_DEBUG, "About to start scheduled thread \"%s\".\n", that_thread->name);
1161 #endif
1162                         if (CT->state > CTDL_THREAD_STOP_REQ)
1163                         {       /* Only start it if the system is not stopping */
1164                                 if (ctdl_thread_internal_start_scheduled (that_thread))
1165                                 {
1166 #ifdef WITH_THREADLOG
1167                                         CtdlLogPrintf(CTDL_INFO, "Thread system, Started a scheduled thread \"%s\" (0x%08lx).\n",
1168                                                 that_thread->name, that_thread->tid);
1169 #endif
1170                                 }
1171                         }
1172                 }
1173 #ifdef WITH_THREADLOG
1174                 else
1175                 {
1176                         CtdlLogPrintf(CTDL_DEBUG, "Thread \"%s\" will start in %ld seconds.\n",
1177                                 that_thread->name, that_thread->when - time(NULL));
1178                 }
1179 #endif
1180         }
1181         end_critical_section(S_SCHEDULE_LIST);
1182 }
1183
1184
1185 /*
1186  * A warapper function for select so we can show a thread as blocked
1187  */
1188 int CtdlThreadSelect(int n, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout)
1189 {
1190         int ret = 0;
1191
1192         ctdl_thread_internal_change_state(CT, CTDL_THREAD_BLOCKED);
1193         if (!CtdlThreadCheckStop())
1194                 ret = select(n, readfds, writefds, exceptfds, timeout);
1195         /**
1196          * If the select returned <= 0 then it failed due to an error
1197          * or timeout so this thread could stop if asked to do so.
1198          * Anything else means it needs to continue unless the system is shutting down
1199          */
1200         if (ret > 0)
1201         {
1202                 /**
1203                  * The select says this thread needs to do something useful.
1204                  * This thread was in an idle state so it may have been asked to stop
1205                  * but if the system isn't shutting down this thread is no longer
1206                  * idle and select has given it a task to do so it must not stop
1207                  * In this condition we need to force it into the running state.
1208                  * CtdlThreadGC will clear its ticker for us.
1209                  *
1210                  * FIXME: there is still a small hole here. It is possible for the sequence of locking
1211                  * to allow the state to get changed to STOP_REQ just after this code if the other thread
1212                  * has decided to change the state before this lock, it there fore has to wait till the lock
1213                  * completes but it will continue to change the state. We need something a bit better here.
1214                  */
1215                 citthread_mutex_lock(&CT->ThreadMutex); /* To prevent race condition of a sleeping thread */
1216                 if (GC_thread->state > CTDL_THREAD_STOP_REQ && CT->state <= CTDL_THREAD_STOP_REQ)
1217                 {
1218                         CtdlLogPrintf(CTDL_DEBUG, "Thread %s (0x%08lx) refused stop request.\n", CT->name, CT->tid);
1219                         CT->state = CTDL_THREAD_RUNNING;
1220                 }
1221                 citthread_mutex_unlock(&CT->ThreadMutex);
1222         }
1223
1224         ctdl_thread_internal_change_state(CT, CTDL_THREAD_RUNNING);
1225
1226         return ret;
1227 }
1228
1229
1230
1231 void *new_worker_thread(void *arg);
1232 extern void close_masters (void);
1233
1234
1235
1236 void go_threading(void)
1237 {
1238         int i;
1239         CtdlThreadNode *last_worker;
1240
1241         /*
1242          * Initialise the thread system
1243          */
1244         ctdl_thread_internal_init();
1245
1246         /* Second call to module init functions now that threading is up */
1247         initialise_modules(1);
1248
1249         /*
1250          * This thread is now used for garbage collection of other threads in the thread list
1251          */
1252         CtdlLogPrintf(CTDL_INFO, "Startup thread %d becoming garbage collector,\n", citthread_self());
1253
1254         /*
1255          * We do a lot of locking and unlocking of the thread list in here.
1256          * We do this so that we can repeatedly release time for other threads
1257          * that may be waiting on the thread list.
1258          * We are a low priority thread so we can afford to do this
1259          */
1260
1261         while (CtdlThreadGetCount())
1262         {
1263                 if (CT->signal)
1264                         exit_signal = CT->signal;
1265                 if (exit_signal)
1266                 {
1267                         CtdlThreadStopAll();
1268 //                      close_masters();
1269                 }
1270                 check_sched_shutdown();
1271                 if (CT->state > CTDL_THREAD_STOP_REQ)
1272                 {
1273                         begin_critical_section(S_THREAD_LIST);
1274                         ctdl_thread_internal_calc_loadavg();
1275                         end_critical_section(S_THREAD_LIST);
1276
1277                         ctdl_thread_internal_check_scheduled(); /* start scheduled threads */
1278                 }
1279
1280                 /* Reduce the size of the worker thread pool if necessary. */
1281                 if ((CtdlThreadGetWorkers() > config.c_min_workers + 1) && (CtdlThreadWorkerAvg < 20) && (CT->state > CTDL_THREAD_STOP_REQ))
1282                 {
1283                         /* Ask a worker thread to stop as we no longer need it */
1284                         begin_critical_section(S_THREAD_LIST);
1285                         last_worker = CtdlThreadList;
1286                         while (last_worker)
1287                         {
1288                                 citthread_mutex_lock(&last_worker->ThreadMutex);
1289                                 if (last_worker->flags & CTDLTHREAD_WORKER && (last_worker->state > CTDL_THREAD_STOPPING) && (last_worker->Context == NULL))
1290                                 {
1291                                         citthread_mutex_unlock(&last_worker->ThreadMutex);
1292                                         break;
1293                                 }
1294                                 citthread_mutex_unlock(&last_worker->ThreadMutex);
1295                                 last_worker = last_worker->next;
1296                         }
1297                         end_critical_section(S_THREAD_LIST);
1298                         if (last_worker)
1299                         {
1300 #ifdef WITH_THREADLOG
1301                                 CtdlLogPrintf(CTDL_DEBUG, "Thread system, stopping excess worker thread \"%s\" (0x%08lx).\n",
1302                                         last_worker->name,
1303                                         last_worker->tid
1304                                         );
1305 #endif
1306                                 CtdlThreadStop(last_worker);
1307                         }
1308                 }
1309
1310                 /*
1311                  * If all our workers are working hard, start some more to help out
1312                  * with things
1313                  */
1314                 /* FIXME: come up with a better way to dynamically alter the number of threads
1315                  * based on the system load
1316                  */
1317 #ifdef NEW_WORKER
1318                 if ((((CtdlThreadGetWorkers() < config.c_max_workers) && (CtdlThreadGetWorkers() <= num_sessions) ) || CtdlThreadGetWorkers() < config.c_min_workers) && (CT->state > CTDL_THREAD_STOP_REQ))
1319 #else
1320                 if ((((CtdlThreadGetWorkers() < config.c_max_workers) && (CtdlThreadGetWorkerAvg() > 60)) || CtdlThreadGetWorkers() < config.c_min_workers) && (CT->state > CTDL_THREAD_STOP_REQ))
1321 #endif /* NEW_WORKER */
1322                 {
1323                         /* Only start new threads if we are not going to overload the machine */
1324                         if (CtdlThreadGetLoadAvg() < ((double)1.00)) {
1325                                 for (i=0; i<5 ; i++) {
1326 #ifdef NEW_WORKER
1327                                         CtdlThreadCreate("Worker Thread (new)",
1328                                                 CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1329                                                 new_worker_thread,
1330                                                 NULL
1331                                                 );
1332 #else
1333                                         CtdlThreadCreate("Worker Thread",
1334                                                 CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1335                                                 worker_thread,
1336                                                 NULL
1337                                                 );
1338 #endif /* NEW_WORKER */
1339                                 }
1340                         }
1341                         else
1342                                 CtdlLogPrintf (CTDL_WARNING, "Server strangled due to machine load average too high.\n");
1343                 }
1344
1345                 CtdlThreadGC();
1346
1347                 if (CtdlThreadGetCount() <= 1) // Shutting down clean up the garbage collector
1348                 {
1349                         CtdlThreadGC();
1350                 }
1351
1352 #ifdef THREADS_USESIGNALS
1353                 if (CtdlThreadGetCount() && CT->state > CTDL_THREAD_STOP_REQ)
1354 #else
1355                 if (CtdlThreadGetCount())
1356 #endif
1357                         CtdlThreadSleep(1);
1358         }
1359         /*
1360          * If the above loop exits we must be shutting down since we obviously have no threads
1361          */
1362         ctdl_thread_internal_cleanup();
1363 }
1364
1365
1366
1367
1368 /*
1369  * Starting a new implimentation of a worker thread.
1370  * This new implimentation will be faster and do more work per thread.
1371  */
1372
1373 /*
1374  * Select on master socket.
1375  * First worker thread in here acquires the lock and builds an FDSET of master sockets.
1376  * then it goes into a loop selecting on the master sockets timing out every few milliseconds.
1377  * If it times out it rebiulds its list and loops.
1378  * If the select succeeds it creates a new context and returns.
1379  * During this time the other workers are selecting on existing contexts or sleeping.
1380  */
1381 void select_on_master(void)
1382 {
1383         fd_set readfds;
1384         struct ServiceFunctionHook *serviceptr;
1385         int ssock;                      /* Descriptor for client socket */
1386         int highest;
1387         int m, i;
1388         int retval = 0;
1389         struct timeval tv;
1390         struct CitContext *con;
1391         const char *old_name;
1392
1393
1394
1395         old_name = CtdlThreadName("select_on_master");
1396
1397         /* Initialize the fdset. */
1398         FD_ZERO(&readfds);
1399         highest = 0;
1400
1401         /* First, add the various master sockets to the fdset. */
1402         for (serviceptr = ServiceHookTable; serviceptr != NULL; serviceptr = serviceptr->next ) {
1403                 m = serviceptr->msock;
1404                 FD_SET(m, &readfds);
1405                 if (m > highest) {
1406                         highest = m;
1407                 }
1408         }
1409
1410         tv.tv_sec = 1;          /* wake up every 1 sec if no input */
1411         tv.tv_usec = 0;
1412         retval = CtdlThreadSelect(highest + 1, &readfds, NULL, NULL, &tv);
1413
1414         /* Select got an error or we are shutting down so get out */
1415         if (retval == 0 || CtdlThreadCheckStop()) {
1416                 CtdlThreadName(old_name);
1417                 return;
1418         }
1419
1420         /* Select says something happened on one of our master sockets so now we handle it */
1421         for (serviceptr = ServiceHookTable; serviceptr != NULL; serviceptr = serviceptr->next ) {
1422                 if (FD_ISSET(serviceptr->msock, &readfds)) {
1423                         ssock = accept(serviceptr->msock, NULL, 0);
1424                         if (ssock >= 0) {
1425                                 CtdlLogPrintf(CTDL_DEBUG, "New client socket %d\n", ssock);
1426                                 /* The master socket is non-blocking but the client
1427                                  * sockets need to be blocking, otherwise certain
1428                                  * operations barf on FreeBSD.  Not a fatal error.
1429                                  */
1430                                 if (fcntl(ssock, F_SETFL, 0) < 0) {
1431                                         CtdlLogPrintf(CTDL_EMERG,
1432                                                       "citserver: Can't set socket to blocking: %s\n",
1433                                                       strerror(errno));
1434                                 }
1435
1436                                 /* New context will be created already
1437                                  * set up in the CON_EXECUTING state.
1438                                  */
1439                                 con = CreateNewContext();
1440                                 CT->Context = con;
1441
1442                                 /* Assign our new socket number to it. */
1443                                 con->client_socket = ssock;
1444                                 con->h_command_function = serviceptr->h_command_function;
1445                                 con->h_async_function = serviceptr->h_async_function;
1446                                 con->ServiceName = serviceptr->ServiceName;
1447                                 /* Determine whether it's a local socket */
1448                                 if (serviceptr->sockpath != NULL)
1449                                         con->is_local_socket = 1;
1450
1451                                 /* Set the SO_REUSEADDR socket option */
1452                                 i = 1;
1453                                 setsockopt(ssock, SOL_SOCKET, SO_REUSEADDR, &i, sizeof(i));
1454
1455                                 become_session(con);
1456                                 begin_session(con);
1457                                 serviceptr->h_greeting_function();
1458                                 become_session(NULL);
1459                                 con->state = CON_IDLE;
1460                                 break;
1461                         }
1462                 }
1463         }
1464
1465         CtdlThreadName(old_name);
1466 }
1467
1468 /*
1469  * Select on client socket.
1470  * First worker thread in here acquires the lock and builds an FDSET of client sockets.
1471  * then it selects on the client sockets timing out after 1 second.
1472  * If it times out the thread goes off to check on housekeeping etc.
1473  * If the select succeeds the thread goes off to handle the client request.
1474  * If the list of client connections is empty the threads all sleep for one second
1475  */
1476 struct CitContext *select_on_client(void)
1477 {
1478         fd_set readfds;
1479         struct timeval tv;
1480         int retval = 0;
1481         int highest=0;
1482         const char *old_name;
1483
1484
1485         old_name = CtdlThreadName("select_on_client");
1486
1487         /* Initialise the fdset */
1488         FD_ZERO(&readfds);
1489         FD_SET(CT->Context->client_socket, &readfds);
1490         highest = CT->Context->client_socket;
1491         /* Now we can select on any connections that are waiting */
1492
1493         if (!CtdlThreadCheckStop())
1494         {
1495                 tv.tv_sec = config.c_sleeping;          /* wake up every second if no input */
1496                 tv.tv_usec = 0;
1497                 retval = select(highest + 1, &readfds, NULL, NULL, &tv);
1498         }
1499         else    /* Shutting down? */
1500         {
1501                 CtdlThreadName(old_name);
1502                 return(NULL);
1503         }
1504
1505
1506         /* Now figure out who made this select() unblock.
1507          * First, check for an error or exit condition.
1508          */
1509         if (retval < 0) {
1510                 if (errno == EBADF) {
1511                         CtdlLogPrintf(CTDL_NOTICE, "select() failed: (%s)\n",
1512                                 strerror(errno));
1513                 }
1514                 if (errno != EINTR) {
1515                         CtdlLogPrintf(CTDL_EMERG, "Exiting (%s)\n", strerror(errno));
1516                         CtdlThreadStopAll();
1517                 } else if (!CtdlThreadCheckStop()) {
1518                         CtdlLogPrintf(CTDL_DEBUG, "Un handled select failure.\n");
1519                 }
1520                 CtdlThreadName(old_name);
1521                 return NULL;
1522         }
1523         else if(retval == 0)
1524         {
1525                 CtdlThreadName(old_name);
1526                 CT->Context->kill_me = 1;
1527                 CT->Context = NULL;
1528                 return CT->Context;
1529         }
1530
1531         CT->Context->state = CON_EXECUTING;
1532         CT->Context->input_waiting = 1;
1533
1534         CtdlThreadName(old_name);
1535         return (CT->Context);
1536 }
1537
1538
1539
1540 /*
1541  * Do the worker threads work when needed
1542  */
1543 int execute_session(struct CitContext *bind_me)
1544 {
1545         int force_purge;
1546
1547         become_session(bind_me);
1548
1549         /* If the client has sent a command, execute it. */
1550         if (CC->input_waiting) {
1551                 CC->h_command_function();
1552                 CC->input_waiting = 0;
1553         }
1554
1555         /* If there are asynchronous messages waiting and the
1556          * client supports it, do those now */
1557         if ((CC->is_async) && (CC->async_waiting)
1558            && (CC->h_async_function != NULL)) {
1559                 CC->h_async_function();
1560                 CC->async_waiting = 0;
1561         }
1562
1563         force_purge = CC->kill_me;
1564         if (force_purge)
1565                 CT->Context = NULL;
1566         become_session(NULL);
1567         bind_me->state = CON_IDLE;
1568         return force_purge;
1569 }
1570
1571
1572
1573
1574 /*
1575  * A new worker_thread loop.
1576  */
1577
1578 void *new_worker_thread(void *arg)
1579 {
1580         struct CitContext *bind_me;
1581         int force_purge;
1582
1583         while (!CtdlThreadCheckStop()) {
1584
1585                 /* make doubly sure we're not holding any stale db handles
1586                  * which might cause a deadlock.
1587                  */
1588                 cdb_check_handles();
1589                 force_purge = 0;
1590                 bind_me = NULL;         /* Which session shall we handle? */
1591
1592                 if (CT->Context == NULL)
1593                         select_on_master();
1594                 if (CtdlThreadCheckStop())
1595                         break;
1596
1597                 if (CT->Context)
1598                         bind_me = select_on_client();
1599                 if (CtdlThreadCheckStop())
1600                         break;
1601
1602                 if (bind_me)
1603                         force_purge = execute_session(bind_me);
1604
1605                 dead_session_purge(force_purge);
1606                 if (CtdlThreadCheckStop())
1607                         break;
1608
1609                 do_housekeeping();
1610         }
1611         return NULL;
1612 }