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 long statcount = 0;             /* are we doing a stats check? */
  64 static long stats_done = 0;
  65
  66 CtdlThreadNode *CtdlThreadList = NULL;
  67 CtdlThreadNode *CtdlThreadSchedList = NULL;
  68
  69 static CtdlThreadNode *GC_thread = NULL;
  70 static char *CtdlThreadStates[CTDL_THREAD_LAST_STATE];
  71 double CtdlThreadLoadAvg = 0;
  72 double CtdlThreadWorkerAvg = 0;
  73 citthread_key_t ThreadKey;
  74
  75 citthread_mutex_t Critters[MAX_SEMAPHORES];     /* Things needing locking */
  76
  77
  78
  79 void InitialiseSemaphores(void)
  80 {
  81         int i;
  82
  83         /* Set up a bunch of semaphores to be used for critical sections */
  84         for (i=0; i<MAX_SEMAPHORES; ++i) {
  85                 citthread_mutex_init(&Critters[i], NULL);
  86         }
  87 }
  88
  89
  90
  91
  92 /*
  93  * Obtain a semaphore lock to begin a critical section.
  94  * but only if no one else has one
  95  */
  96 int try_critical_section(int which_one)
  97 {
  98         /* For all types of critical sections except those listed here,
  99          * ensure nobody ever tries to do a critical section within a
 100          * transaction; this could lead to deadlock.
 101          */
 102         if (    (which_one != S_FLOORCACHE)
 103 #ifdef DEBUG_MEMORY_LEAKS
 104                 && (which_one != S_DEBUGMEMLEAKS)
 105 #endif
 106                 && (which_one != S_RPLIST)
 107         ) {
 108                 cdb_check_handles();
 109         }
 110         return (citthread_mutex_trylock(&Critters[which_one]));
 111 }
 112
 113
 114 /*
 115  * Obtain a semaphore lock to begin a critical section.
 116  */
 117 void begin_critical_section(int which_one)
 118 {
 119         /* CtdlLogPrintf(CTDL_DEBUG, "begin_critical_section(%d)\n", which_one); */
 120
 121         /* For all types of critical sections except those listed here,
 122          * ensure nobody ever tries to do a critical section within a
 123          * transaction; this could lead to deadlock.
 124          */
 125         if (    (which_one != S_FLOORCACHE)
 126 #ifdef DEBUG_MEMORY_LEAKS
 127                 && (which_one != S_DEBUGMEMLEAKS)
 128 #endif
 129                 && (which_one != S_RPLIST)
 130         ) {
 131                 cdb_check_handles();
 132         }
 133         citthread_mutex_lock(&Critters[which_one]);
 134 }
 135
 136 /*
 137  * Release a semaphore lock to end a critical section.
 138  */
 139 void end_critical_section(int which_one)
 140 {
 141         citthread_mutex_unlock(&Critters[which_one]);
 142 }
 143
 144
 145 /*
 146  * A function to destroy the TSD
 147  */
 148 static void ctdl_thread_internal_dest_tsd(void *arg)
 149 {
 150         if (arg != NULL) {
 151                 check_handles(arg);
 152                 free(arg);
 153         }
 154 }
 155
 156
 157 /*
 158  * A function to initialise the thread TSD
 159  */
 160 void ctdl_thread_internal_init_tsd(void)
 161 {
 162         int ret;
 163
 164         if ((ret = citthread_key_create(&ThreadKey, ctdl_thread_internal_dest_tsd))) {
 165                 CtdlLogPrintf(CTDL_EMERG, "citthread_key_create: %s\n", strerror(ret));
 166                 exit(CTDLEXIT_DB);
 167         }
 168 }
 169
 170 /*
 171  * Ensure that we have a key for thread-specific data.
 172  *
 173  * This should be called immediately after startup by any thread
 174  *
 175  */
 176 void CtdlThreadAllocTSD(void)
 177 {
 178         ThreadTSD *tsd;
 179
 180         if (citthread_getspecific(ThreadKey) != NULL)
 181                 return;
 182
 183         tsd = malloc(sizeof(ThreadTSD));
 184
 185         tsd->tid = NULL;
 186
 187         memset(tsd->cursors, 0, sizeof tsd->cursors);
 188         tsd->self = NULL;
 189
 190         citthread_setspecific(ThreadKey, tsd);
 191 }
 192
 193
 194 void ctdl_thread_internal_free_tsd(void)
 195 {
 196         ctdl_thread_internal_dest_tsd(citthread_getspecific(ThreadKey));
 197         citthread_setspecific(ThreadKey, NULL);
 198 }
 199
 200
 201 void ctdl_thread_internal_cleanup(void)
 202 {
 203         int i;
 204         CtdlThreadNode *this_thread, *that_thread;
 205
 206         for (i=0; i<CTDL_THREAD_LAST_STATE; i++)
 207         {
 208                 free (CtdlThreadStates[i]);
 209         }
 210
 211         /* Clean up the scheduled thread list */
 212         this_thread = CtdlThreadSchedList;
 213         while (this_thread)
 214         {
 215                 that_thread = this_thread;
 216                 this_thread = this_thread->next;
 217                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 218                 citthread_cond_destroy(&that_thread->ThreadCond);
 219                 citthread_mutex_destroy(&that_thread->SleepMutex);
 220                 citthread_cond_destroy(&that_thread->SleepCond);
 221                 citthread_attr_destroy(&that_thread->attr);
 222                 free(that_thread);
 223         }
 224         ctdl_thread_internal_free_tsd();
 225 }
 226
 227 void ctdl_thread_internal_init(void)
 228 {
 229         CtdlThreadNode *this_thread;
 230         int ret = 0;
 231
 232         CtdlThreadStates[CTDL_THREAD_INVALID] = strdup ("Invalid Thread");
 233         CtdlThreadStates[CTDL_THREAD_VALID] = strdup("Valid Thread");
 234         CtdlThreadStates[CTDL_THREAD_CREATE] = strdup("Thread being Created");
 235         CtdlThreadStates[CTDL_THREAD_CANCELLED] = strdup("Thread Cancelled");
 236         CtdlThreadStates[CTDL_THREAD_EXITED] = strdup("Thread Exited");
 237         CtdlThreadStates[CTDL_THREAD_STOPPING] = strdup("Thread Stopping");
 238         CtdlThreadStates[CTDL_THREAD_STOP_REQ] = strdup("Thread Stop Requested");
 239         CtdlThreadStates[CTDL_THREAD_SLEEPING] = strdup("Thread Sleeping");
 240         CtdlThreadStates[CTDL_THREAD_RUNNING] = strdup("Thread Running");
 241         CtdlThreadStates[CTDL_THREAD_BLOCKED] = strdup("Thread Blocked");
 242
 243         /* Get ourself a thread entry */
 244         this_thread = malloc(sizeof(CtdlThreadNode));
 245         if (this_thread == NULL) {
 246                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 247                 return;
 248         }
 249         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 250         memset (this_thread, 0, sizeof(CtdlThreadNode));
 251
 252         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 253         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 254         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 255         citthread_cond_init (&(this_thread->SleepCond), NULL);
 256
 257         /* We are garbage collector so create us as running */
 258         this_thread->state = CTDL_THREAD_RUNNING;
 259
 260         if ((ret = citthread_attr_init(&this_thread->attr))) {
 261                 CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 262                 free(this_thread);
 263                 return;
 264         }
 265
 266         this_thread->name = "Garbage Collection Thread";
 267
 268         this_thread->tid = citthread_self();
 269         GC_thread = this_thread;
 270         CT = this_thread;
 271
 272         num_threads++;  // Increase the count of threads in the system.
 273
 274         this_thread->next = CtdlThreadList;
 275         CtdlThreadList = this_thread;
 276         if (this_thread->next)
 277                 this_thread->next->prev = this_thread;
 278         /* Set up start times */
 279         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 280         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 281 }
 282
 283
 284 /*
 285  * A function to update a threads load averages
 286  */
 287  void ctdl_thread_internal_update_avgs(CtdlThreadNode *this_thread)
 288  {
 289         struct timeval now, result;
 290         double last_duration;
 291
 292         gettimeofday(&now, NULL);
 293         timersub(&now, &(this_thread->last_state_change), &result);
 294         /* I don't think these mutex's are needed here */
 295         citthread_mutex_lock(&this_thread->ThreadMutex);
 296         // result now has a timeval for the time we spent in the last state since we last updated
 297         last_duration = (double)result.tv_sec + ((double)result.tv_usec / (double) 1000000);
 298         if (this_thread->state == CTDL_THREAD_SLEEPING)
 299                 this_thread->avg_sleeping += last_duration;
 300         if (this_thread->state == CTDL_THREAD_RUNNING)
 301                 this_thread->avg_running += last_duration;
 302         if (this_thread->state == CTDL_THREAD_BLOCKED)
 303                 this_thread->avg_blocked += last_duration;
 304         memcpy (&this_thread->last_state_change, &now, sizeof (struct timeval));
 305         citthread_mutex_unlock(&this_thread->ThreadMutex);
 306 }
 307
 308 /*
 309  * A function to chenge the state of a thread
 310  */
 311 void ctdl_thread_internal_change_state (CtdlThreadNode *this_thread, enum CtdlThreadState new_state)
 312 {
 313         /*
 314          * Wether we change state or not we need update the load values
 315          */
 316         ctdl_thread_internal_update_avgs(this_thread);
 317         /* This mutex not needed here? */
 318         citthread_mutex_lock(&this_thread->ThreadMutex); /* To prevent race condition of a sleeping thread */
 319         if ((new_state == CTDL_THREAD_STOP_REQ) && (this_thread->state > CTDL_THREAD_STOP_REQ))
 320                 this_thread->state = new_state;
 321         if (((new_state == CTDL_THREAD_SLEEPING) || (new_state == CTDL_THREAD_BLOCKED)) && (this_thread->state == CTDL_THREAD_RUNNING))
 322                 this_thread->state = new_state;
 323         if ((new_state == CTDL_THREAD_RUNNING) && ((this_thread->state == CTDL_THREAD_SLEEPING) || (this_thread->state == CTDL_THREAD_BLOCKED)))
 324                 this_thread->state = new_state;
 325         citthread_mutex_unlock(&this_thread->ThreadMutex);
 326 }
 327
 328
 329 /*
 330  * A function to tell all threads to exit
 331  */
 332 void CtdlThreadStopAll(void)
 333 {
 334         /* First run any registered shutdown hooks.  This probably doesn't belong here. */
 335         PerformSessionHooks(EVT_SHUTDOWN);
 336
 337         //FIXME: The signalling of the condition should not be in the critical_section
 338         // We need to build a list of threads we are going to signal and then signal them afterwards
 339
 340         CtdlThreadNode *this_thread;
 341
 342         begin_critical_section(S_THREAD_LIST);
 343         this_thread = CtdlThreadList;
 344         // Ask the GC thread to stop first so everything knows we are shutting down.
 345         GC_thread->state = CTDL_THREAD_STOP_REQ;
 346         while(this_thread)
 347         {
 348 #ifdef THREADS_USESIGNALS
 349                 if (!citthread_equal(this_thread->tid, GC_thread->tid))
 350                         citthread_kill(this_thread->tid, SIGHUP);
 351 #endif
 352                 ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 353                 citthread_cond_signal(&this_thread->ThreadCond);
 354                 citthread_cond_signal(&this_thread->SleepCond);
 355                 this_thread->stop_ticker = time(NULL);
 356                 CtdlLogPrintf(CTDL_DEBUG, "Thread system stopping thread \"%s\" (0x%08lx).\n",
 357                         this_thread->name, this_thread->tid);
 358                 this_thread = this_thread->next;
 359         }
 360         end_critical_section(S_THREAD_LIST);
 361 }
 362
 363
 364 /*
 365  * A function to wake up all sleeping threads
 366  */
 367 void CtdlThreadWakeAll(void)
 368 {
 369         CtdlThreadNode *this_thread;
 370
 371         CtdlLogPrintf(CTDL_DEBUG, "Thread system waking all threads.\n");
 372
 373         begin_critical_section(S_THREAD_LIST);
 374         this_thread = CtdlThreadList;
 375         while(this_thread)
 376         {
 377                 if (!this_thread->thread_func)
 378                 {
 379                         citthread_cond_signal(&this_thread->ThreadCond);
 380                         citthread_cond_signal(&this_thread->SleepCond);
 381                 }
 382                 this_thread = this_thread->next;
 383         }
 384         end_critical_section(S_THREAD_LIST);
 385 }
 386
 387
 388 /*
 389  * A function to return the number of threads running in the system
 390  */
 391 int CtdlThreadGetCount(void)
 392 {
 393         return  num_threads;
 394 }
 395
 396 int CtdlThreadGetWorkers(void)
 397 {
 398         return  num_workers;
 399 }
 400
 401 double CtdlThreadGetWorkerAvg(void)
 402 {
 403         double ret;
 404
 405         begin_critical_section(S_THREAD_LIST);
 406         ret =  CtdlThreadWorkerAvg;
 407         end_critical_section(S_THREAD_LIST);
 408         return ret;
 409 }
 410
 411 double CtdlThreadGetLoadAvg(void)
 412 {
 413         double load_avg[3] ;
 414
 415         int ret;
 416         int smp_num_cpus;
 417
 418         /* Borrowed this straight from procps */
 419         smp_num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 420         if(smp_num_cpus<1) smp_num_cpus=1; /* SPARC glibc is buggy */
 421
 422         ret = getloadavg(load_avg, 3);
 423         if (ret < 0)
 424                 return 0;
 425         return load_avg[0] / smp_num_cpus;
 426 /*
 427  * This old chunk of code return a value that indicated the load on citserver
 428  * This value could easily reach 100 % even when citserver was doing very little and
 429  * hence the machine has much more spare capacity.
 430  * Because this value was used to determine if the machine was under heavy load conditions
 431  * from other processes in the system then citserver could be strangled un-necesarily
 432  * What we are actually trying to achieve is to strangle citserver if the machine is heavily loaded.
 433  * So we have changed this.
 434
 435         begin_critical_section(S_THREAD_LIST);
 436         ret =  CtdlThreadLoadAvg;
 437         end_critical_section(S_THREAD_LIST);
 438         return ret;
 439 */
 440 }
 441
 442
 443
 444
 445 /*
 446  * A function to rename a thread
 447  * Returns a const char *
 448  */
 449 const char *CtdlThreadName(const char *name)
 450 {
 451         const char *old_name;
 452
 453         if (!CT)
 454         {
 455                 CtdlLogPrintf(CTDL_WARNING, "Thread system WARNING. Attempt to CtdlThreadRename() a non thread. %s\n", name);
 456                 return NULL;
 457         }
 458         old_name = CT->name;
 459         if (name)
 460                 CT->name = name;
 461         return (old_name);
 462 }
 463
 464
 465 /*
 466  * A function to force a thread to exit
 467  */
 468 void CtdlThreadCancel(CtdlThreadNode *thread)
 469 {
 470         CtdlThreadNode *this_thread;
 471
 472         if (!thread)
 473                 this_thread = CT;
 474         else
 475                 this_thread = thread;
 476         if (!this_thread)
 477         {
 478                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() a non thread.\n");
 479                 CtdlThreadStopAll();
 480                 return;
 481         }
 482
 483         if (!this_thread->thread_func)
 484         {
 485                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() the garbage collector.\n");
 486                 CtdlThreadStopAll();
 487                 return;
 488         }
 489
 490         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_CANCELLED);
 491         citthread_cancel(this_thread->tid);
 492 }
 493
 494
 495 /*
 496  * A function for a thread to check if it has been asked to stop
 497  */
 498 int CtdlThreadCheckStop(void)
 499 {
 500         int state;
 501
 502         if (!CT)
 503         {
 504                 CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, CtdlThreadCheckStop() called by a non thread.\n");
 505                 CtdlThreadStopAll();
 506                 return -1;
 507         }
 508
 509         state = CT->state;
 510
 511 #ifdef THREADS_USESIGNALS
 512         if (CT->signal)
 513         {
 514                 CtdlLogPrintf(CTDL_DEBUG, "Thread \"%s\" caught signal %d.\n", CT->name, CT->signal);
 515                 CT->signal = 0;
 516         }
 517 #endif
 518         if(state == CTDL_THREAD_STOP_REQ)
 519         {
 520                 CT->state = CTDL_THREAD_STOPPING;
 521                 return -1;
 522         }
 523         else if((state < CTDL_THREAD_STOP_REQ) && (state > CTDL_THREAD_CREATE))
 524         {
 525                 return -1;
 526         }
 527         return 0;
 528 }
 529
 530
 531 /*
 532  * A function to ask a thread to exit
 533  * The thread must call CtdlThreadCheckStop() periodically to determine if it should exit
 534  */
 535 void CtdlThreadStop(CtdlThreadNode *thread)
 536 {
 537         CtdlThreadNode *this_thread;
 538
 539         if (!thread)
 540                 this_thread = CT;
 541         else
 542                 this_thread = thread;
 543         if (!this_thread)
 544                 return;
 545         if (!(this_thread->thread_func))
 546                 return;         // Don't stop garbage collector
 547 #ifdef THREADS_USESIGNALS
 548         if (!citthread_equal(this_thread->tid, GC_thread->tid))
 549                 citthread_kill(this_thread->tid, SIGHUP);
 550 #endif
 551         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 552         citthread_cond_signal(&this_thread->ThreadCond);
 553         citthread_cond_signal(&this_thread->SleepCond);
 554         this_thread->stop_ticker = time(NULL);
 555 }
 556
 557 /*
 558  * So we now have a sleep command that works with threads but it is in seconds
 559  */
 560 void CtdlThreadSleep(int secs)
 561 {
 562         struct timespec wake_time;
 563         struct timeval time_now;
 564
 565
 566         if (!CT)
 567         {
 568                 CtdlLogPrintf(CTDL_WARNING, "CtdlThreadSleep() called by something that is not a thread. Should we die?\n");
 569                 return;
 570         }
 571
 572         memset (&wake_time, 0, sizeof(struct timespec));
 573         gettimeofday(&time_now, NULL);
 574         wake_time.tv_sec = time_now.tv_sec + secs;
 575         wake_time.tv_nsec = time_now.tv_usec * 10;
 576
 577         ctdl_thread_internal_change_state (CT, CTDL_THREAD_SLEEPING);
 578
 579         citthread_mutex_lock(&CT->ThreadMutex); /* Prevent something asking us to awaken before we've gone to sleep */
 580         citthread_cond_timedwait(&CT->SleepCond, &CT->ThreadMutex, &wake_time);
 581         citthread_mutex_unlock(&CT->ThreadMutex);
 582
 583         ctdl_thread_internal_change_state (CT, CTDL_THREAD_RUNNING);
 584 }
 585
 586
 587 /*
 588  * Routine to clean up our thread function on exit
 589  */
 590 static void ctdl_internal_thread_cleanup(void *arg)
 591 {
 592         /*
 593          * In here we were called by the current thread because it is exiting
 594          * NB. WE ARE THE CURRENT THREAD
 595          */
 596         CtdlLogPrintf(CTDL_NOTICE, "Thread \"%s\" (0x%08lx) exited.\n", CT->name, CT->tid);
 597
 598         #ifdef HAVE_BACKTRACE
 599         eCrash_UnregisterThread();
 600         #endif
 601
 602         citthread_mutex_lock(&CT->ThreadMutex);
 603         CT->state = CTDL_THREAD_EXITED; // needs to be last thing else house keeping will unlink us too early
 604         citthread_mutex_unlock(&CT->ThreadMutex);
 605 }
 606
 607 /*
 608  * A quick function to show the load averages
 609  */
 610 void ctdl_thread_internal_calc_loadavg(void)
 611 {
 612         CtdlThreadNode *that_thread;
 613         double load_avg, worker_avg;
 614         int workers = 0;
 615
 616         that_thread = CtdlThreadList;
 617         load_avg = 0;
 618         worker_avg = 0;
 619         while(that_thread)
 620         {
 621                 /* Update load averages */
 622                 ctdl_thread_internal_update_avgs(that_thread);
 623                 citthread_mutex_lock(&that_thread->ThreadMutex);
 624                 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;
 625                 that_thread->avg_sleeping /= 2;
 626                 that_thread->avg_running /= 2;
 627                 that_thread->avg_blocked /= 2;
 628                 load_avg += that_thread->load_avg;
 629                 if (that_thread->flags & CTDLTHREAD_WORKER)
 630                 {
 631                         worker_avg += that_thread->load_avg;
 632                         workers++;
 633                 }
 634 #ifdef WITH_THREADLOG
 635                 CtdlLogPrintf(CTDL_DEBUG, "CtdlThread, \"%s\" (%lu) \"%s\" %.2f %.2f %.2f %.2f\n",
 636                         that_thread->name,
 637                         that_thread->tid,
 638                         CtdlThreadStates[that_thread->state],
 639                         that_thread->avg_sleeping,
 640                         that_thread->avg_running,
 641                         that_thread->avg_blocked,
 642                         that_thread->load_avg);
 643 #endif
 644                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 645                 that_thread = that_thread->next;
 646         }
 647         CtdlThreadLoadAvg = load_avg/num_threads;
 648         CtdlThreadWorkerAvg = worker_avg/workers;
 649 #ifdef WITH_THREADLOG
 650         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);
 651 #endif
 652 }
 653
 654
 655 /*
 656  * Garbage collection routine.
 657  * Gets called by main() in a loop to clean up the thread list periodically.
 658  */
 659 void CtdlThreadGC (void)
 660 {
 661         CtdlThreadNode *this_thread, *that_thread;
 662         int workers = 0, sys_workers;
 663         int ret=0;
 664
 665         begin_critical_section(S_THREAD_LIST);
 666
 667         /* Handle exiting of garbage collector thread */
 668         if(num_threads == 1)
 669                 CtdlThreadList->state = CTDL_THREAD_EXITED;
 670
 671 #ifdef WITH_THREADLOG
 672         CtdlLogPrintf(CTDL_DEBUG, "Thread system running garbage collection.\n");
 673 #endif
 674         /*
 675          * Woke up to do garbage collection
 676          */
 677         this_thread = CtdlThreadList;
 678         while(this_thread)
 679         {
 680                 that_thread = this_thread;
 681                 this_thread = this_thread->next;
 682
 683                 if ((that_thread->state == CTDL_THREAD_STOP_REQ || that_thread->state == CTDL_THREAD_STOPPING)
 684                         && (!citthread_equal(that_thread->tid, citthread_self())))
 685                                 CtdlLogPrintf(CTDL_DEBUG, "Waiting for thread %s (0x%08lx) to exit.\n", that_thread->name, that_thread->tid);
 686                 else
 687                 {
 688                         /**
 689                          * Catch the situation where a worker was asked to stop but couldn't and we are not
 690                          * shutting down.
 691                          */
 692                         that_thread->stop_ticker = 0;
 693                 }
 694
 695                 if (that_thread->stop_ticker + 5 == time(NULL))
 696                 {
 697                         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);
 698                         if ((that_thread->flags & CTDLTHREAD_WORKER) == 0)
 699                                 CtdlLogPrintf(CTDL_INFO, "Thread System: A non worker thread would have been canceled this may cause message loss.\n");
 700 //                      that_thread->state = CTDL_THREAD_CANCELLED;
 701                         that_thread->stop_ticker++;
 702 //                      citthread_cancel(that_thread->tid);
 703 //                      continue;
 704                 }
 705
 706                 /* Do we need to clean up this thread? */
 707                 if ((that_thread->state != CTDL_THREAD_EXITED) && (that_thread->state != CTDL_THREAD_CANCELLED))
 708                 {
 709                         if(that_thread->flags & CTDLTHREAD_WORKER)
 710                                 workers++;      /* Sanity check on number of worker threads */
 711                         continue;
 712                 }
 713
 714                 if (citthread_equal(that_thread->tid, citthread_self()) && that_thread->thread_func)
 715                 {       /* Sanity check */
 716                         end_critical_section(S_THREAD_LIST);
 717                         CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, a thread is trying to clean up after itself.\n");
 718                         abort();
 719                         return;
 720                 }
 721
 722                 if (num_threads <= 0)
 723                 {       /* Sanity check */
 724                         end_critical_section(S_THREAD_LIST);
 725                         CtdlLogPrintf(CTDL_EMERG, "Thread system PANIC, num_threads <= 0 and trying to do Garbage Collection.\n");
 726                         abort();
 727                         return;
 728                 }
 729
 730                 if(that_thread->flags & CTDLTHREAD_WORKER)
 731                         num_workers--;  /* This is a wroker thread so reduce the count. */
 732                 num_threads--;
 733                 /* If we are unlinking the list head then the next becomes the list head */
 734                 if(that_thread->prev)
 735                         that_thread->prev->next = that_thread->next;
 736                 else
 737                         CtdlThreadList = that_thread->next;
 738                 if(that_thread->next)
 739                         that_thread->next->prev = that_thread->prev;
 740
 741                 citthread_cond_signal(&that_thread->ThreadCond);
 742                 citthread_cond_signal(&that_thread->SleepCond); // Make sure this thread is awake
 743                 citthread_mutex_lock(&that_thread->ThreadMutex);        // Make sure it has done what its doing
 744                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 745                 /*
 746                  * Join on the thread to do clean up and prevent memory leaks
 747                  * Also makes sure the thread has cleaned up after itself before we remove it from the list
 748                  * We can join on the garbage collector thread the join should just return EDEADLCK
 749                  */
 750                 ret = citthread_join (that_thread->tid, NULL);
 751                 if (ret == EDEADLK)
 752                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection on own thread.\n");
 753                 else if (ret == EINVAL)
 754                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, that thread already joined on.\n");
 755                 else if (ret == ESRCH)
 756                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, no thread to join on.\n");
 757                 else if (ret != 0)
 758                         CtdlLogPrintf(CTDL_DEBUG, "Garbage collection, citthread_join returned an unknown error(%d).\n", ret);
 759                 /*
 760                  * Now we own that thread entry
 761                  */
 762                 CtdlLogPrintf(CTDL_INFO, "Garbage Collection for thread \"%s\" (0x%08lx).\n",
 763                         that_thread->name, that_thread->tid);
 764                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 765                 citthread_cond_destroy(&that_thread->ThreadCond);
 766                 citthread_mutex_destroy(&that_thread->SleepMutex);
 767                 citthread_cond_destroy(&that_thread->SleepCond);
 768                 citthread_attr_destroy(&that_thread->attr);
 769                 free(that_thread);
 770         }
 771         sys_workers = num_workers;
 772         end_critical_section(S_THREAD_LIST);
 773
 774         /* Sanity check number of worker threads */
 775         if (workers != sys_workers)
 776         {
 777                 CtdlLogPrintf(CTDL_EMERG,
 778                         "Thread system PANIC, discrepancy in number of worker threads. Counted %d, should be %d.\n",
 779                         workers, sys_workers
 780                         );
 781                 abort();
 782         }
 783 }
 784
 785
 786
 787
 788 /*
 789  * Runtime function for a Citadel Thread.
 790  * This initialises the threads environment and then calls the user supplied thread function
 791  * Note that this is the REAL thread function and wraps the users thread function.
 792  */
 793 static void *ctdl_internal_thread_func (void *arg)
 794 {
 795         CtdlThreadNode *this_thread;
 796         void *ret = NULL;
 797
 798         /* lock and unlock the thread list.
 799          * This causes this thread to wait until all its creation stuff has finished before it
 800          * can continue its execution.
 801          */
 802         begin_critical_section(S_THREAD_LIST);
 803         this_thread = (CtdlThreadNode *) arg;
 804         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 805 //      citthread_mutex_lock(&this_thread->ThreadMutex);
 806
 807         // Register the cleanup function to take care of when we exit.
 808         citthread_cleanup_push(ctdl_internal_thread_cleanup, NULL);
 809         // Get our thread data structure
 810         CtdlThreadAllocTSD();
 811         CT = this_thread;
 812         this_thread->pid = getpid();
 813         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 814         /* Only change to running state if we weren't asked to stop during the create cycle
 815          * Other wise there is a window to allow this threads creation to continue to full grown and
 816          * therby prevent a shutdown of the server.
 817          */
 818 //      citthread_mutex_unlock(&this_thread->ThreadMutex);
 819
 820         if (!CtdlThreadCheckStop())
 821         {
 822                 citthread_mutex_lock(&this_thread->ThreadMutex);
 823                 this_thread->state = CTDL_THREAD_RUNNING;
 824                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 825         }
 826         end_critical_section(S_THREAD_LIST);
 827
 828         // Register for tracing
 829         #ifdef HAVE_BACKTRACE
 830         eCrash_RegisterThread(this_thread->name, 0);
 831         #endif
 832
 833         // Tell the world we are here
 834         CtdlLogPrintf(CTDL_NOTICE, "Created a new thread \"%s\" (0x%08lx).\n",
 835                 this_thread->name, this_thread->tid);
 836
 837         /*
 838          * run the thread to do the work but only if we haven't been asked to stop
 839          */
 840         if (!CtdlThreadCheckStop())
 841                 ret = (this_thread->thread_func)(this_thread->user_args);
 842
 843         /*
 844          * Our thread is exiting either because it wanted to end or because the server is stopping
 845          * We need to clean up
 846          */
 847         citthread_cleanup_pop(1);       // Execute our cleanup routine and remove it
 848
 849         return(ret);
 850 }
 851
 852
 853
 854
 855 /*
 856  * Function to initialise an empty thread structure
 857  */
 858 CtdlThreadNode *ctdl_internal_init_thread_struct(CtdlThreadNode *this_thread, long flags)
 859 {
 860         int ret = 0;
 861
 862         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 863         memset (this_thread, 0, sizeof(CtdlThreadNode));
 864
 865         /* Create the mutex's early so we can use them */
 866         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 867         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 868         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 869         citthread_cond_init (&(this_thread->SleepCond), NULL);
 870
 871         this_thread->state = CTDL_THREAD_CREATE;
 872
 873         if ((ret = citthread_attr_init(&this_thread->attr))) {
 874                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 875                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 876                 citthread_cond_destroy(&(this_thread->ThreadCond));
 877                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 878                 citthread_cond_destroy(&(this_thread->SleepCond));
 879                 CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 880                 free(this_thread);
 881                 return NULL;
 882         }
 883
 884         /* Our per-thread stacks need to be bigger than the default size,
 885          * otherwise the MIME parser crashes on FreeBSD, and the IMAP service
 886          * crashes on 64-bit Linux.
 887          */
 888         if (flags & CTDLTHREAD_BIGSTACK)
 889         {
 890 #ifdef WITH_THREADLOG
 891                 CtdlLogPrintf(CTDL_INFO, "Thread system. Creating BIG STACK thread.\n");
 892 #endif
 893                 if ((ret = citthread_attr_setstacksize(&this_thread->attr, THREADSTACKSIZE))) {
 894                         citthread_mutex_unlock(&this_thread->ThreadMutex);
 895                         citthread_mutex_destroy(&(this_thread->ThreadMutex));
 896                         citthread_cond_destroy(&(this_thread->ThreadCond));
 897                         citthread_mutex_destroy(&(this_thread->SleepMutex));
 898                         citthread_cond_destroy(&(this_thread->SleepCond));
 899                         citthread_attr_destroy(&this_thread->attr);
 900                         CtdlLogPrintf(CTDL_EMERG, "Thread system, citthread_attr_setstacksize: %s\n",
 901                                 strerror(ret));
 902                         free(this_thread);
 903                         return NULL;
 904                 }
 905         }
 906
 907         /* Set this new thread with an avg_blocked of 2. We do this so that its creation affects the
 908          * load average for the system. If we don't do this then we create a mass of threads at the same time
 909          * because the creation didn't affect the load average.
 910          */
 911         this_thread->avg_blocked = 2;
 912
 913         return (this_thread);
 914 }
 915
 916
 917
 918
 919 /*
 920  * Internal function to create a thread.
 921  */
 922 CtdlThreadNode *ctdl_internal_create_thread(char *name, long flags, void *(*thread_func) (void *arg), void *args)
 923 {
 924         int ret = 0;
 925         CtdlThreadNode *this_thread;
 926
 927         if (num_threads >= 32767)
 928         {
 929                 CtdlLogPrintf(CTDL_EMERG, "Thread system. Thread list full.\n");
 930                 return NULL;
 931         }
 932
 933         this_thread = malloc(sizeof(CtdlThreadNode));
 934         if (this_thread == NULL) {
 935                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 936                 return NULL;
 937         }
 938
 939         /* Initialise the thread structure */
 940         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
 941         {
 942                 free(this_thread);
 943                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
 944                 return NULL;
 945         }
 946         /*
 947          * If we got here we are going to create the thread so we must initilise the structure
 948          * first because most implimentations of threading can't create it in a stopped state
 949          * and it might want to do things with its structure that aren't initialised otherwise.
 950          */
 951         if(name)
 952         {
 953                 this_thread->name = name;
 954         }
 955         else
 956         {
 957                 this_thread->name = "Un-named Thread";
 958         }
 959
 960         this_thread->flags = flags;
 961         this_thread->thread_func = thread_func;
 962         this_thread->user_args = args;
 963
 964 //      citthread_mutex_lock(&this_thread->ThreadMutex);
 965
 966         begin_critical_section(S_THREAD_LIST);
 967         /*
 968          * We pass this_thread into the thread as its args so that it can find out information
 969          * about itself and it has a bit of storage space for itself, not to mention that the REAL
 970          * thread function needs to finish off the setup of the structure
 971          */
 972         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
 973         {
 974                 end_critical_section(S_THREAD_LIST);
 975                 CtdlLogPrintf(CTDL_ALERT, "Thread system, Can't create thread: %s\n",
 976                         strerror(ret));
 977                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 978                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 979                 citthread_cond_destroy(&(this_thread->ThreadCond));
 980                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 981                 citthread_cond_destroy(&(this_thread->SleepCond));
 982                 citthread_attr_destroy(&this_thread->attr);
 983                 free(this_thread);
 984                 return NULL;
 985         }
 986
 987         num_threads++;  // Increase the count of threads in the system.
 988         if(this_thread->flags & CTDLTHREAD_WORKER)
 989                 num_workers++;
 990
 991         this_thread->next = CtdlThreadList;
 992         CtdlThreadList = this_thread;
 993         if (this_thread->next)
 994                 this_thread->next->prev = this_thread;
 995         ctdl_thread_internal_calc_loadavg();
 996
 997 //      citthread_mutex_unlock(&this_thread->ThreadMutex);
 998         end_critical_section(S_THREAD_LIST);
 999
1000         return this_thread;
1001 }
1002
1003 /*
1004  * Wrapper function to create a thread
1005  * ensures the critical section and other protections are in place.
1006  * char *name = name to give to thread, if NULL, use generic name
1007  * int flags = flags to determine type of thread and standard facilities
1008  */
1009 CtdlThreadNode *CtdlThreadCreate(char *name, long flags, void *(*thread_func) (void *arg), void *args)
1010 {
1011         CtdlThreadNode *ret = NULL;
1012
1013         ret = ctdl_internal_create_thread(name, flags, thread_func, args);
1014         return ret;
1015 }
1016
1017
1018
1019 /*
1020  * Internal function to schedule a thread.
1021  * Must be called from within a S_THREAD_LIST critical section
1022  */
1023 CtdlThreadNode *CtdlThreadSchedule(char *name, long flags, void *(*thread_func) (void *arg), void *args, time_t when)
1024 {
1025         CtdlThreadNode *this_thread;
1026
1027         if (num_threads >= 32767)
1028         {
1029                 CtdlLogPrintf(CTDL_EMERG, "Thread system. Thread list full.\n");
1030                 return NULL;
1031         }
1032
1033         this_thread = malloc(sizeof(CtdlThreadNode));
1034         if (this_thread == NULL) {
1035                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
1036                 return NULL;
1037         }
1038         /* Initialise the thread structure */
1039         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
1040         {
1041                 free(this_thread);
1042                 CtdlLogPrintf(CTDL_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
1043                 return NULL;
1044         }
1045
1046         /*
1047          * If we got here we are going to create the thread so we must initilise the structure
1048          * first because most implimentations of threading can't create it in a stopped state
1049          * and it might want to do things with its structure that aren't initialised otherwise.
1050          */
1051         if(name)
1052         {
1053                 this_thread->name = name;
1054         }
1055         else
1056         {
1057                 this_thread->name = "Un-named Thread";
1058         }
1059
1060         this_thread->flags = flags;
1061         this_thread->thread_func = thread_func;
1062         this_thread->user_args = args;
1063
1064         /*
1065          * When to start this thread
1066          */
1067         this_thread->when = when;
1068
1069         begin_critical_section(S_SCHEDULE_LIST);
1070         this_thread->next = CtdlThreadSchedList;
1071         CtdlThreadSchedList = this_thread;
1072         if (this_thread->next)
1073                 this_thread->next->prev = this_thread;
1074         end_critical_section(S_SCHEDULE_LIST);
1075
1076         return this_thread;
1077 }
1078
1079
1080
1081 CtdlThreadNode *ctdl_thread_internal_start_scheduled (CtdlThreadNode *this_thread)
1082 {
1083         int ret = 0;
1084
1085 //      citthread_mutex_lock(&that_thread->ThreadMutex);
1086         begin_critical_section(S_THREAD_LIST);
1087         /*
1088          * We pass this_thread into the thread as its args so that it can find out information
1089          * about itself and it has a bit of storage space for itself, not to mention that the REAL
1090          * thread function needs to finish off the setup of the structure
1091          */
1092         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
1093         {
1094                 end_critical_section(S_THREAD_LIST);
1095                 CtdlLogPrintf(CTDL_DEBUG, "Failed to start scheduled thread \"%s\": %s\n", this_thread->name, strerror(ret));
1096 //              citthread_mutex_unlock(&this_thread->ThreadMutex);
1097                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
1098                 citthread_cond_destroy(&(this_thread->ThreadCond));
1099                 citthread_mutex_destroy(&(this_thread->SleepMutex));
1100                 citthread_cond_destroy(&(this_thread->SleepCond));
1101                 citthread_attr_destroy(&this_thread->attr);
1102                 free(this_thread);
1103                 return NULL;
1104         }
1105
1106
1107         num_threads++;  // Increase the count of threads in the system.
1108         if(this_thread->flags & CTDLTHREAD_WORKER)
1109                 num_workers++;
1110
1111         this_thread->next = CtdlThreadList;
1112         CtdlThreadList = this_thread;
1113         if (this_thread->next)
1114                 this_thread->next->prev = this_thread;
1115 //      citthread_mutex_unlock(&that_thread->ThreadMutex);
1116
1117         ctdl_thread_internal_calc_loadavg();
1118         end_critical_section(S_THREAD_LIST);
1119
1120
1121         return this_thread;
1122 }
1123
1124
1125
1126 void ctdl_thread_internal_check_scheduled(void)
1127 {
1128         CtdlThreadNode *this_thread, *that_thread;
1129         time_t now;
1130
1131         /* Don't start scheduled threads if the system wants single user mode */
1132         if (CtdlWantSingleUser())
1133                 return;
1134
1135         if (try_critical_section(S_SCHEDULE_LIST))
1136                 return; /* If this list is locked we wait till the next chance */
1137
1138         now = time(NULL);
1139
1140 #ifdef WITH_THREADLOG
1141         CtdlLogPrintf(CTDL_DEBUG, "Checking for scheduled threads to start.\n");
1142 #endif
1143
1144         this_thread = CtdlThreadSchedList;
1145         while(this_thread)
1146         {
1147                 that_thread = this_thread;
1148                 this_thread = this_thread->next;
1149
1150                 if (now > that_thread->when)
1151                 {
1152                         /* Unlink from schedule list */
1153                         if (that_thread->prev)
1154                                 that_thread->prev->next = that_thread->next;
1155                         else
1156                                 CtdlThreadSchedList = that_thread->next;
1157                         if (that_thread->next)
1158                                 that_thread->next->prev = that_thread->prev;
1159
1160                         that_thread->next = that_thread->prev = NULL;
1161 #ifdef WITH_THREADLOG
1162                         CtdlLogPrintf(CTDL_DEBUG, "About to start scheduled thread \"%s\".\n", that_thread->name);
1163 #endif
1164                         if (CT->state > CTDL_THREAD_STOP_REQ)
1165                         {       /* Only start it if the system is not stopping */
1166                                 if (ctdl_thread_internal_start_scheduled (that_thread))
1167                                 {
1168 #ifdef WITH_THREADLOG
1169                                         CtdlLogPrintf(CTDL_INFO, "Thread system, Started a scheduled thread \"%s\" (0x%08lx).\n",
1170                                                 that_thread->name, that_thread->tid);
1171 #endif
1172                                 }
1173                         }
1174                 }
1175 #ifdef WITH_THREADLOG
1176                 else
1177                 {
1178                         CtdlLogPrintf(CTDL_DEBUG, "Thread \"%s\" will start in %ld seconds.\n",
1179                                 that_thread->name, that_thread->when - time(NULL));
1180                 }
1181 #endif
1182         }
1183         end_critical_section(S_SCHEDULE_LIST);
1184 }
1185
1186
1187 /*
1188  * A warapper function for select so we can show a thread as blocked
1189  */
1190 int CtdlThreadSelect(int n, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout)
1191 {
1192         int ret = 0;
1193
1194         ctdl_thread_internal_change_state(CT, CTDL_THREAD_BLOCKED);
1195         if (!CtdlThreadCheckStop())
1196                 ret = select(n, readfds, writefds, exceptfds, timeout);
1197         /**
1198          * If the select returned <= 0 then it failed due to an error
1199          * or timeout so this thread could stop if asked to do so.
1200          * Anything else means it needs to continue unless the system is shutting down
1201          */
1202         if (ret > 0)
1203         {
1204                 /**
1205                  * The select says this thread needs to do something useful.
1206                  * This thread was in an idle state so it may have been asked to stop
1207                  * but if the system isn't shutting down this thread is no longer
1208                  * idle and select has given it a task to do so it must not stop
1209                  * In this condition we need to force it into the running state.
1210                  * CtdlThreadGC will clear its ticker for us.
1211                  *
1212                  * FIXME: there is still a small hole here. It is possible for the sequence of locking
1213                  * to allow the state to get changed to STOP_REQ just after this code if the other thread
1214                  * has decided to change the state before this lock, it there fore has to wait till the lock
1215                  * completes but it will continue to change the state. We need something a bit better here.
1216                  */
1217                 citthread_mutex_lock(&CT->ThreadMutex); /* To prevent race condition of a sleeping thread */
1218                 if (GC_thread->state > CTDL_THREAD_STOP_REQ && CT->state <= CTDL_THREAD_STOP_REQ)
1219                 {
1220                         CtdlLogPrintf(CTDL_DEBUG, "Thread %s (0x%08lx) refused stop request.\n", CT->name, CT->tid);
1221                         CT->state = CTDL_THREAD_RUNNING;
1222                 }
1223                 citthread_mutex_unlock(&CT->ThreadMutex);
1224         }
1225
1226         ctdl_thread_internal_change_state(CT, CTDL_THREAD_RUNNING);
1227
1228         return ret;
1229 }
1230
1231
1232
1233 void *new_worker_thread(void *arg);
1234 extern void close_masters (void);
1235
1236
1237 void *simulation_worker (void*arg) {
1238         struct CitContext *this;
1239
1240         this = CreateNewContext();
1241         CtdlThreadSleep(1);
1242         this->kill_me = 1;
1243         this->state = CON_IDLE;
1244         dead_session_purge(1);
1245         begin_critical_section(S_SESSION_TABLE);
1246         stats_done++;
1247         end_critical_section(S_SESSION_TABLE);
1248         return NULL;
1249 }
1250
1251
1252 void *simulation_thread (void *arg)
1253 {
1254         long stats = statcount;
1255
1256         while(stats && !CtdlThreadCheckStop()) {
1257                 CtdlThreadCreate("Connection simulation worker", CTDLTHREAD_BIGSTACK, simulation_worker, NULL);
1258                 stats--;
1259         }
1260         CtdlThreadStopAll();
1261         return NULL;
1262 }
1263
1264 void go_threading(void)
1265 {
1266         int i;
1267         CtdlThreadNode *last_worker;
1268         struct timeval start, now, result;
1269         double last_duration;
1270
1271         /*
1272          * Initialise the thread system
1273          */
1274         ctdl_thread_internal_init();
1275
1276         /* Second call to module init functions now that threading is up */
1277         if (!statcount)
1278                 initialise_modules(1);
1279         else {
1280                 CtdlLogPrintf(CTDL_EMERG, "Running connection simulation stats\n");
1281                 gettimeofday(&start, NULL);
1282                 CtdlThreadCreate("Connection simulation master", CTDLTHREAD_BIGSTACK, simulation_thread, NULL);
1283         }
1284
1285
1286         /*
1287          * This thread is now used for garbage collection of other threads in the thread list
1288          */
1289         CtdlLogPrintf(CTDL_INFO, "Startup thread %d becoming garbage collector,\n", citthread_self());
1290
1291         /*
1292          * We do a lot of locking and unlocking of the thread list in here.
1293          * We do this so that we can repeatedly release time for other threads
1294          * that may be waiting on the thread list.
1295          * We are a low priority thread so we can afford to do this
1296          */
1297
1298         while (CtdlThreadGetCount())
1299         {
1300                 if (CT->signal)
1301                         exit_signal = CT->signal;
1302                 if (exit_signal)
1303                 {
1304                         CtdlThreadStopAll();
1305 //                      close_masters();
1306                 }
1307                 check_sched_shutdown();
1308                 if (CT->state > CTDL_THREAD_STOP_REQ)
1309                 {
1310                         begin_critical_section(S_THREAD_LIST);
1311                         ctdl_thread_internal_calc_loadavg();
1312                         end_critical_section(S_THREAD_LIST);
1313
1314                         ctdl_thread_internal_check_scheduled(); /* start scheduled threads */
1315                 }
1316
1317                 /* Reduce the size of the worker thread pool if necessary. */
1318                 if ((CtdlThreadGetWorkers() > config.c_min_workers + 1) && (CtdlThreadWorkerAvg < 20) && (CT->state > CTDL_THREAD_STOP_REQ))
1319                 {
1320                         /* Ask a worker thread to stop as we no longer need it */
1321                         begin_critical_section(S_THREAD_LIST);
1322                         last_worker = CtdlThreadList;
1323                         while (last_worker)
1324                         {
1325                                 citthread_mutex_lock(&last_worker->ThreadMutex);
1326                                 if (last_worker->flags & CTDLTHREAD_WORKER && (last_worker->state > CTDL_THREAD_STOPPING) && (last_worker->Context == NULL))
1327                                 {
1328                                         citthread_mutex_unlock(&last_worker->ThreadMutex);
1329                                         break;
1330                                 }
1331                                 citthread_mutex_unlock(&last_worker->ThreadMutex);
1332                                 last_worker = last_worker->next;
1333                         }
1334                         end_critical_section(S_THREAD_LIST);
1335                         if (last_worker)
1336                         {
1337 #ifdef WITH_THREADLOG
1338                                 CtdlLogPrintf(CTDL_DEBUG, "Thread system, stopping excess worker thread \"%s\" (0x%08lx).\n",
1339                                         last_worker->name,
1340                                         last_worker->tid
1341                                         );
1342 #endif
1343                                 CtdlThreadStop(last_worker);
1344                         }
1345                 }
1346
1347                 /*
1348                  * If all our workers are working hard, start some more to help out
1349                  * with things
1350                  */
1351                 /* FIXME: come up with a better way to dynamically alter the number of threads
1352                  * based on the system load
1353                  */
1354                 if (!statcount) {
1355 #ifdef NEW_WORKER
1356                 if ((((CtdlThreadGetWorkers() < config.c_max_workers) && (CtdlThreadGetWorkers() <= num_sessions) ) || CtdlThreadGetWorkers() < config.c_min_workers) && (CT->state > CTDL_THREAD_STOP_REQ))
1357 #else
1358                 if ((((CtdlThreadGetWorkers() < config.c_max_workers) && (CtdlThreadGetWorkerAvg() > 60)) || CtdlThreadGetWorkers() < config.c_min_workers) && (CT->state > CTDL_THREAD_STOP_REQ))
1359 #endif /* NEW_WORKER */
1360                 {
1361                         /* Only start new threads if we are not going to overload the machine */
1362                         /* Temporarily set to 10 should be enough to make sure we don't stranglew the server
1363                          * at least until we make this a config option */
1364                         if (CtdlThreadGetLoadAvg() < ((double)10.00)) {
1365                                 for (i=0; i<5 ; i++) {
1366 #ifdef NEW_WORKER
1367                                         CtdlThreadCreate("Worker Thread (new)",
1368                                                 CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1369                                                 new_worker_thread,
1370                                                 NULL
1371                                                 );
1372 #else
1373                                         CtdlThreadCreate("Worker Thread",
1374                                                 CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1375                                                 worker_thread,
1376                                                 NULL
1377                                                 );
1378 #endif /* NEW_WORKER */
1379                                 }
1380                         }
1381                         else
1382                                 CtdlLogPrintf (CTDL_WARNING, "Server strangled due to machine load average too high.\n");
1383                 }
1384                 }
1385
1386                 CtdlThreadGC();
1387
1388                 if (CtdlThreadGetCount() <= 1) // Shutting down clean up the garbage collector
1389                 {
1390                         CtdlThreadGC();
1391                 }
1392
1393 #ifdef THREADS_USESIGNALS
1394                 if (CtdlThreadGetCount() && CT->state > CTDL_THREAD_STOP_REQ)
1395 #else
1396                 if (CtdlThreadGetCount())
1397 #endif
1398                         CtdlThreadSleep(1);
1399         }
1400         /*
1401          * If the above loop exits we must be shutting down since we obviously have no threads
1402          */
1403         ctdl_thread_internal_cleanup();
1404
1405         if (statcount) {
1406                 gettimeofday(&now, NULL);
1407                 timersub(&now, &start, &result);
1408                 last_duration = (double)result.tv_sec + ((double)result.tv_usec / (double) 1000000);
1409                 CtdlLogPrintf(CTDL_EMERG, "Simulated %ld connections in %f seconds\n", stats_done, last_duration);
1410         }
1411 }
1412
1413
1414
1415
1416 /*
1417  * Starting a new implimentation of a worker thread.
1418  * This new implimentation will be faster and do more work per thread.
1419  */
1420
1421 /*
1422  * Select on master socket.
1423  * First worker thread in here acquires the lock and builds an FDSET of master sockets.
1424  * then it goes into a loop selecting on the master sockets timing out every few milliseconds.
1425  * If it times out it rebiulds its list and loops.
1426  * If the select succeeds it creates a new context and returns.
1427  * During this time the other workers are selecting on existing contexts or sleeping.
1428  */
1429 void select_on_master(void)
1430 {
1431         fd_set readfds;
1432         struct ServiceFunctionHook *serviceptr;
1433         int ssock;                      /* Descriptor for client socket */
1434         int highest;
1435         int m, i;
1436         int retval = 0;
1437         struct timeval tv;
1438         CitContext *con;
1439         const char *old_name;
1440
1441
1442
1443         old_name = CtdlThreadName("select_on_master");
1444
1445         /* Initialize the fdset. */
1446         FD_ZERO(&readfds);
1447         highest = 0;
1448
1449         /* First, add the various master sockets to the fdset. */
1450         for (serviceptr = ServiceHookTable; serviceptr != NULL; serviceptr = serviceptr->next ) {
1451                 m = serviceptr->msock;
1452                 FD_SET(m, &readfds);
1453                 if (m > highest) {
1454                         highest = m;
1455                 }
1456         }
1457
1458         tv.tv_sec = 1;          /* wake up every 1 sec if no input */
1459         tv.tv_usec = 0;
1460         retval = CtdlThreadSelect(highest + 1, &readfds, NULL, NULL, &tv);
1461
1462         /* Select got an error or we are shutting down so get out */
1463         if (retval == 0 || CtdlThreadCheckStop()) {
1464                 CtdlThreadName(old_name);
1465                 return;
1466         }
1467
1468         /* Select says something happened on one of our master sockets so now we handle it */
1469         for (serviceptr = ServiceHookTable; serviceptr != NULL; serviceptr = serviceptr->next ) {
1470                 if (FD_ISSET(serviceptr->msock, &readfds)) {
1471                         ssock = accept(serviceptr->msock, NULL, 0);
1472                         if (ssock >= 0) {
1473                                 CtdlLogPrintf(CTDL_DEBUG, "New client socket %d\n", ssock);
1474                                 /* The master socket is non-blocking but the client
1475                                  * sockets need to be blocking, otherwise certain
1476                                  * operations barf on FreeBSD.  Not a fatal error.
1477                                  */
1478                                 if (fcntl(ssock, F_SETFL, 0) < 0) {
1479                                         CtdlLogPrintf(CTDL_EMERG,
1480                                                       "citserver: Can't set socket to blocking: %s\n",
1481                                                       strerror(errno));
1482                                 }
1483
1484                                 /* New context will be created already
1485                                  * set up in the CON_EXECUTING state.
1486                                  */
1487                                 con = CreateNewContext();
1488                                 CT->Context = con;
1489
1490                                 /* Assign our new socket number to it. */
1491                                 con->client_socket = ssock;
1492                                 con->h_command_function = serviceptr->h_command_function;
1493                                 con->h_async_function = serviceptr->h_async_function;
1494                                 con->ServiceName = serviceptr->ServiceName;
1495                                 /* Determine whether it's a local socket */
1496                                 if (serviceptr->sockpath != NULL)
1497                                         con->is_local_socket = 1;
1498
1499                                 /* Set the SO_REUSEADDR socket option */
1500                                 i = 1;
1501                                 setsockopt(ssock, SOL_SOCKET, SO_REUSEADDR, &i, sizeof(i));
1502
1503                                 become_session(con);
1504                                 begin_session(con);
1505                                 serviceptr->h_greeting_function();
1506                                 become_session(NULL);
1507                                 con->state = CON_IDLE;
1508                                 break;
1509                         }
1510                 }
1511         }
1512
1513         CtdlThreadName(old_name);
1514 }
1515
1516 /*
1517  * Select on client socket.
1518  * First worker thread in here acquires the lock and builds an FDSET of client sockets.
1519  * then it selects on the client sockets timing out after 1 second.
1520  * If it times out the thread goes off to check on housekeeping etc.
1521  * If the select succeeds the thread goes off to handle the client request.
1522  * If the list of client connections is empty the threads all sleep for one second
1523  */
1524 CitContext *select_on_client(void)
1525 {
1526         fd_set readfds;
1527         struct timeval tv;
1528         int retval = 0;
1529         int highest=0;
1530         const char *old_name;
1531
1532
1533         old_name = CtdlThreadName("select_on_client");
1534
1535         /* Initialise the fdset */
1536         FD_ZERO(&readfds);
1537         FD_SET(CT->Context->client_socket, &readfds);
1538         highest = CT->Context->client_socket;
1539         /* Now we can select on any connections that are waiting */
1540
1541         if (!CtdlThreadCheckStop())
1542         {
1543                 tv.tv_sec = config.c_sleeping;          /* wake up every second if no input */
1544                 tv.tv_usec = 0;
1545                 retval = select(highest + 1, &readfds, NULL, NULL, &tv);
1546         }
1547         else    /* Shutting down? */
1548         {
1549                 CtdlThreadName(old_name);
1550                 return(NULL);
1551         }
1552
1553
1554         /* Now figure out who made this select() unblock.
1555          * First, check for an error or exit condition.
1556          */
1557         if (retval < 0) {
1558                 if (errno == EBADF) {
1559                         CtdlLogPrintf(CTDL_NOTICE, "select() failed: (%s)\n",
1560                                 strerror(errno));
1561                 }
1562                 if (errno != EINTR) {
1563                         CtdlLogPrintf(CTDL_EMERG, "Exiting (%s)\n", strerror(errno));
1564                         CtdlThreadStopAll();
1565                 } else if (!CtdlThreadCheckStop()) {
1566                         CtdlLogPrintf(CTDL_DEBUG, "Un handled select failure.\n");
1567                 }
1568                 CtdlThreadName(old_name);
1569                 return NULL;
1570         }
1571         else if(retval == 0)
1572         {
1573                 CtdlThreadName(old_name);
1574                 CT->Context->kill_me = 1;
1575                 CT->Context = NULL;
1576                 return CT->Context;
1577         }
1578
1579         CT->Context->state = CON_EXECUTING;
1580         CT->Context->input_waiting = 1;
1581
1582         CtdlThreadName(old_name);
1583         return (CT->Context);
1584 }
1585
1586
1587
1588 /*
1589  * Do the worker threads work when needed
1590  */
1591 int execute_session(CitContext *bind_me)
1592 {
1593         int force_purge;
1594
1595         become_session(bind_me);
1596
1597         /* If the client has sent a command, execute it. */
1598         if (CC->input_waiting) {
1599                 CC->h_command_function();
1600                 CC->input_waiting = 0;
1601         }
1602
1603         /* If there are asynchronous messages waiting and the
1604          * client supports it, do those now */
1605         if ((CC->is_async) && (CC->async_waiting)
1606            && (CC->h_async_function != NULL)) {
1607                 CC->h_async_function();
1608                 CC->async_waiting = 0;
1609         }
1610
1611         force_purge = CC->kill_me;
1612         if (force_purge)
1613                 CT->Context = NULL;
1614         become_session(NULL);
1615         bind_me->state = CON_IDLE;
1616         return force_purge;
1617 }
1618
1619
1620
1621
1622 /*
1623  * A new worker_thread loop.
1624  */
1625
1626 void *new_worker_thread(void *arg)
1627 {
1628         CitContext *bind_me;
1629         int force_purge;
1630
1631         while (!CtdlThreadCheckStop()) {
1632
1633                 /* make doubly sure we're not holding any stale db handles
1634                  * which might cause a deadlock.
1635                  */
1636                 cdb_check_handles();
1637                 force_purge = 0;
1638                 bind_me = NULL;         /* Which session shall we handle? */
1639
1640                 if (CT->Context == NULL)
1641                         select_on_master();
1642                 if (CtdlThreadCheckStop())
1643                         break;
1644
1645                 if (CT->Context)
1646                         bind_me = select_on_client();
1647                 if (CtdlThreadCheckStop())
1648                         break;
1649
1650                 if (bind_me)
1651                         force_purge = execute_session(bind_me);
1652
1653                 dead_session_purge(force_purge);
1654                 if (CtdlThreadCheckStop())
1655                         break;
1656
1657                 do_housekeeping();
1658         }
1659         return NULL;
1660 }