citadel/threads.c

   1 /*
   2  * Thread handling stuff for Citadel server
   3  *
   4  * Copyright (c) 1987-2011 by the citadel.org team
   5  *
   6  * This program is open source software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 3 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  */
  20
  21 #include <stdlib.h>
  22 #include <unistd.h>
  23 #include <stdio.h>
  24 #include <sys/types.h>
  25 #include <errno.h>
  26 #include <sys/socket.h>
  27 #include <unistd.h>
  28 #include <fcntl.h>
  29 #include <signal.h>
  30 #include <syslog.h>
  31
  32 #include "sysdep.h"
  33 #if TIME_WITH_SYS_TIME
  34 # include <sys/time.h>
  35 # include <time.h>
  36 #else
  37 # if HAVE_SYS_TIME_H
  38 #  include <sys/time.h>
  39 # else
  40 #  include <time.h>
  41 # endif
  42 #endif
  43
  44 #ifdef HAVE_SYSCALL_H
  45 # include <syscall.h>
  46 #else
  47 # if HAVE_SYS_SYSCALL_H
  48 #  include <sys/syscall.h>
  49 # endif
  50 #endif
  51
  52 #include <libcitadel.h>
  53
  54 #include "threads.h"
  55 #include "ctdl_module.h"
  56 #include "modules_init.h"
  57 #include "housekeeping.h"
  58 #include "config.h"
  59 #include "citserver.h"
  60 #include "sysdep_decls.h"
  61 #include "context.h"
  62 #include "event_client.h"
  63
  64
  65 /*
  66  * To create a thread you must call one of the create thread functions.
  67  * You must pass it the address of (a pointer to a CtdlThreadNode initialised to NULL) like this
  68  * struct CtdlThreadNode *node = NULL;
  69  * pass in &node
  70  * If the thread is created *node will point to the thread control structure for the created thread.
  71  * If the thread creation fails *node remains NULL
  72  * Do not free the memory pointed to by *node, it doesn't belong to you.
  73  * This new interface duplicates much of the eCrash stuff. We should go for closer integration since that would
  74  * remove the need for the calls to eCrashRegisterThread and friends
  75  */
  76
  77 static int num_threads = 0;                     /* Current number of threads */
  78 static int num_workers = 0;                     /* Current number of worker threads */
  79
  80 CtdlThreadNode *CtdlThreadList = NULL;
  81 CtdlThreadNode *CtdlThreadSchedList = NULL;
  82
  83 static CtdlThreadNode *GC_thread = NULL;
  84 static char *CtdlThreadStates[CTDL_THREAD_LAST_STATE];
  85 double CtdlThreadLoadAvg = 0;
  86 double CtdlThreadWorkerAvg = 0;
  87 citthread_key_t ThreadKey;
  88
  89 citthread_mutex_t Critters[MAX_SEMAPHORES];     /* Things needing locking */
  90
  91
  92
  93 void InitialiseSemaphores(void)
  94 {
  95         int i;
  96
  97         /* Set up a bunch of semaphores to be used for critical sections */
  98         for (i=0; i<MAX_SEMAPHORES; ++i) {
  99                 citthread_mutex_init(&Critters[i], NULL);
 100         }
 101 }
 102
 103
 104
 105
 106 /*
 107  * Obtain a semaphore lock to begin a critical section.
 108  * but only if no one else has one
 109  */
 110 int try_critical_section(int which_one)
 111 {
 112         /* For all types of critical sections except those listed here,
 113          * ensure nobody ever tries to do a critical section within a
 114          * transaction; this could lead to deadlock.
 115          */
 116         if (    (which_one != S_FLOORCACHE)
 117                 && (which_one != S_RPLIST)
 118         ) {
 119                 cdb_check_handles();
 120         }
 121         return (citthread_mutex_trylock(&Critters[which_one]));
 122 }
 123
 124
 125 /*
 126  * Obtain a semaphore lock to begin a critical section.
 127  */
 128 void begin_critical_section(int which_one)
 129 {
 130         /* syslog(LOG_DEBUG, "begin_critical_section(%d)\n", which_one); */
 131
 132         /* For all types of critical sections except those listed here,
 133          * ensure nobody ever tries to do a critical section within a
 134          * transaction; this could lead to deadlock.
 135          */
 136         if (    (which_one != S_FLOORCACHE)
 137                 && (which_one != S_RPLIST)
 138         ) {
 139                 cdb_check_handles();
 140         }
 141         citthread_mutex_lock(&Critters[which_one]);
 142 }
 143
 144 /*
 145  * Release a semaphore lock to end a critical section.
 146  */
 147 void end_critical_section(int which_one)
 148 {
 149         citthread_mutex_unlock(&Critters[which_one]);
 150 }
 151
 152
 153 /*
 154  * A function to destroy the TSD
 155  */
 156 static void ctdl_thread_internal_dest_tsd(void *arg)
 157 {
 158         if (arg != NULL) {
 159                 check_handles(arg);
 160                 free(arg);
 161         }
 162 }
 163
 164
 165 /*
 166  * A function to initialise the thread TSD
 167  */
 168 void ctdl_thread_internal_init_tsd(void)
 169 {
 170         int ret;
 171
 172         if ((ret = citthread_key_create(&ThreadKey, ctdl_thread_internal_dest_tsd))) {
 173                 syslog(LOG_EMERG, "citthread_key_create: %s\n", strerror(ret));
 174                 exit(CTDLEXIT_DB);
 175         }
 176 }
 177
 178 /*
 179  * Ensure that we have a key for thread-specific data.
 180  *
 181  * This should be called immediately after startup by any thread
 182  *
 183  */
 184 void CtdlThreadAllocTSD(void)
 185 {
 186         ThreadTSD *tsd;
 187
 188         if (citthread_getspecific(ThreadKey) != NULL)
 189                 return;
 190
 191         tsd = malloc(sizeof(ThreadTSD));
 192
 193         tsd->tid = NULL;
 194
 195         memset(tsd->cursors, 0, sizeof tsd->cursors);
 196         tsd->self = NULL;
 197
 198         citthread_setspecific(ThreadKey, tsd);
 199 }
 200
 201
 202 void ctdl_thread_internal_free_tsd(void)
 203 {
 204         ctdl_thread_internal_dest_tsd(citthread_getspecific(ThreadKey));
 205         citthread_setspecific(ThreadKey, NULL);
 206 }
 207
 208
 209 void ctdl_thread_internal_cleanup(void)
 210 {
 211         int i;
 212         CtdlThreadNode *this_thread, *that_thread;
 213
 214         for (i=0; i<CTDL_THREAD_LAST_STATE; i++)
 215         {
 216                 free (CtdlThreadStates[i]);
 217         }
 218
 219         /* Clean up the scheduled thread list */
 220         this_thread = CtdlThreadSchedList;
 221         while (this_thread)
 222         {
 223                 that_thread = this_thread;
 224                 this_thread = this_thread->next;
 225                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 226                 citthread_cond_destroy(&that_thread->ThreadCond);
 227                 citthread_mutex_destroy(&that_thread->SleepMutex);
 228                 citthread_cond_destroy(&that_thread->SleepCond);
 229                 citthread_attr_destroy(&that_thread->attr);
 230                 free(that_thread);
 231         }
 232         ctdl_thread_internal_free_tsd();
 233 }
 234
 235 void ctdl_thread_internal_init(void)
 236 {
 237         CtdlThreadNode *this_thread;
 238         int ret = 0;
 239
 240         CtdlThreadStates[CTDL_THREAD_INVALID] = strdup ("Invalid Thread");
 241         CtdlThreadStates[CTDL_THREAD_VALID] = strdup("Valid Thread");
 242         CtdlThreadStates[CTDL_THREAD_CREATE] = strdup("Thread being Created");
 243         CtdlThreadStates[CTDL_THREAD_CANCELLED] = strdup("Thread Cancelled");
 244         CtdlThreadStates[CTDL_THREAD_EXITED] = strdup("Thread Exited");
 245         CtdlThreadStates[CTDL_THREAD_STOPPING] = strdup("Thread Stopping");
 246         CtdlThreadStates[CTDL_THREAD_STOP_REQ] = strdup("Thread Stop Requested");
 247         CtdlThreadStates[CTDL_THREAD_SLEEPING] = strdup("Thread Sleeping");
 248         CtdlThreadStates[CTDL_THREAD_RUNNING] = strdup("Thread Running");
 249         CtdlThreadStates[CTDL_THREAD_BLOCKED] = strdup("Thread Blocked");
 250
 251         /* Get ourself a thread entry */
 252         this_thread = malloc(sizeof(CtdlThreadNode));
 253         if (this_thread == NULL) {
 254                 syslog(LOG_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 255                 return;
 256         }
 257         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 258         memset (this_thread, 0, sizeof(CtdlThreadNode));
 259
 260         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 261         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 262         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 263         citthread_cond_init (&(this_thread->SleepCond), NULL);
 264
 265         /* We are garbage collector so create us as running */
 266         this_thread->state = CTDL_THREAD_RUNNING;
 267
 268         if ((ret = citthread_attr_init(&this_thread->attr))) {
 269                 syslog(LOG_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 270                 free(this_thread);
 271                 return;
 272         }
 273
 274         this_thread->name = "Garbage Collection Thread";
 275
 276         this_thread->tid = citthread_self();
 277         GC_thread = this_thread;
 278         CT = this_thread;
 279
 280         num_threads++;  // Increase the count of threads in the system.
 281
 282         this_thread->next = CtdlThreadList;
 283         CtdlThreadList = this_thread;
 284         if (this_thread->next)
 285                 this_thread->next->prev = this_thread;
 286         /* Set up start times */
 287         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 288         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 289 }
 290
 291
 292 /*
 293  * A function to chenge the state of a thread
 294  */
 295 void ctdl_thread_internal_change_state (CtdlThreadNode *this_thread, enum CtdlThreadState new_state)
 296 {
 297         /*
 298          * Wether we change state or not we need update the load values
 299          */
 300         /* This mutex not needed here? */
 301         citthread_mutex_lock(&this_thread->ThreadMutex); /* To prevent race condition of a sleeping thread */
 302         if ((new_state == CTDL_THREAD_STOP_REQ) && (this_thread->state > CTDL_THREAD_STOP_REQ))
 303                 this_thread->state = new_state;
 304         if (((new_state == CTDL_THREAD_SLEEPING) || (new_state == CTDL_THREAD_BLOCKED)) && (this_thread->state == CTDL_THREAD_RUNNING))
 305                 this_thread->state = new_state;
 306         if ((new_state == CTDL_THREAD_RUNNING) && ((this_thread->state == CTDL_THREAD_SLEEPING) || (this_thread->state == CTDL_THREAD_BLOCKED)))
 307                 this_thread->state = new_state;
 308         citthread_mutex_unlock(&this_thread->ThreadMutex);
 309 }
 310
 311
 312 /*
 313  * A function to tell all threads to exit
 314  */
 315 void CtdlThreadStopAll(void)
 316 {
 317         /* First run any registered shutdown hooks.  This probably doesn't belong here. */
 318         PerformSessionHooks(EVT_SHUTDOWN);
 319
 320         /* then close all tcp ports so nobody else can talk to us anymore. */
 321         CtdlShutdownServiceHooks();
 322         //FIXME: The signalling of the condition should not be in the critical_section
 323         // We need to build a list of threads we are going to signal and then signal them afterwards
 324
 325         ShutDownEventQueue();
 326
 327         CtdlThreadNode *this_thread;
 328
 329         begin_critical_section(S_THREAD_LIST);
 330         this_thread = CtdlThreadList;
 331         // Ask the GC thread to stop first so everything knows we are shutting down.
 332         GC_thread->state = CTDL_THREAD_STOP_REQ;
 333         while(this_thread)
 334         {
 335                 if (!citthread_equal(this_thread->tid, GC_thread->tid))
 336                         citthread_kill(this_thread->tid, SIGHUP);
 337
 338                 ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 339                 citthread_cond_signal(&this_thread->ThreadCond);
 340                 citthread_cond_signal(&this_thread->SleepCond);
 341                 this_thread->stop_ticker = time(NULL);
 342                 syslog(LOG_DEBUG, "Thread system stopping thread \"%s\" (0x%08lx).\n",
 343                         this_thread->name, this_thread->tid);
 344                 this_thread = this_thread->next;
 345         }
 346         end_critical_section(S_THREAD_LIST);
 347 }
 348
 349
 350 /*
 351  * A function to wake up all sleeping threads
 352  */
 353 void CtdlThreadWakeAll(void)
 354 {
 355         CtdlThreadNode *this_thread;
 356
 357         syslog(LOG_DEBUG, "Thread system waking all threads.\n");
 358
 359         begin_critical_section(S_THREAD_LIST);
 360         this_thread = CtdlThreadList;
 361         while(this_thread)
 362         {
 363                 if (!this_thread->thread_func)
 364                 {
 365                         citthread_cond_signal(&this_thread->ThreadCond);
 366                         citthread_cond_signal(&this_thread->SleepCond);
 367                 }
 368                 this_thread = this_thread->next;
 369         }
 370         end_critical_section(S_THREAD_LIST);
 371 }
 372
 373
 374 /*
 375  * A function to return the number of threads running in the system
 376  */
 377 int CtdlThreadGetCount(void)
 378 {
 379         return  num_threads;
 380 }
 381
 382 int CtdlThreadGetWorkers(void)
 383 {
 384         return  num_workers;
 385 }
 386
 387 double CtdlThreadGetWorkerAvg(void)
 388 {
 389         double ret;
 390
 391         begin_critical_section(S_THREAD_LIST);
 392         ret =  CtdlThreadWorkerAvg;
 393         end_critical_section(S_THREAD_LIST);
 394         return ret;
 395 }
 396
 397 double CtdlThreadGetLoadAvg(void)
 398 {
 399         double load_avg[3] = {0.0, 0.0, 0.0};
 400
 401         int ret = 0;
 402         int smp_num_cpus;
 403
 404         /* Borrowed this straight from procps */
 405         smp_num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 406         if(smp_num_cpus<1) smp_num_cpus=1; /* SPARC glibc is buggy */
 407
 408 #ifdef HAVE_GETLOADAVG
 409         ret = getloadavg(load_avg, 3);
 410 #endif
 411         if (ret < 0)
 412                 return 0;
 413         return load_avg[0] / smp_num_cpus;
 414 /*
 415  * This old chunk of code return a value that indicated the load on citserver
 416  * This value could easily reach 100 % even when citserver was doing very little and
 417  * hence the machine has much more spare capacity.
 418  * Because this value was used to determine if the machine was under heavy load conditions
 419  * from other processes in the system then citserver could be strangled un-necesarily
 420  * What we are actually trying to achieve is to strangle citserver if the machine is heavily loaded.
 421  * So we have changed this.
 422
 423         begin_critical_section(S_THREAD_LIST);
 424         ret =  CtdlThreadLoadAvg;
 425         end_critical_section(S_THREAD_LIST);
 426         return ret;
 427 */
 428 }
 429
 430
 431
 432
 433 /*
 434  * A function to rename a thread
 435  * Returns a const char *
 436  */
 437 const char *CtdlThreadName(const char *name)
 438 {
 439         const char *old_name;
 440
 441         if (!CT)
 442         {
 443                 syslog(LOG_WARNING, "Thread system WARNING. Attempt to CtdlThreadRename() a non thread. %s\n", name);
 444                 return NULL;
 445         }
 446         old_name = CT->name;
 447         if (name)
 448                 CT->name = name;
 449         return (old_name);
 450 }
 451
 452
 453 /*
 454  * A function to force a thread to exit
 455  */
 456 void CtdlThreadCancel(CtdlThreadNode *thread)
 457 {
 458         CtdlThreadNode *this_thread;
 459
 460         if (!thread)
 461                 this_thread = CT;
 462         else
 463                 this_thread = thread;
 464         if (!this_thread)
 465         {
 466                 syslog(LOG_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() a non thread.\n");
 467                 CtdlThreadStopAll();
 468                 return;
 469         }
 470
 471         if (!this_thread->thread_func)
 472         {
 473                 syslog(LOG_EMERG, "Thread system PANIC. Attempt to CtdlThreadCancel() the garbage collector.\n");
 474                 CtdlThreadStopAll();
 475                 return;
 476         }
 477
 478         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_CANCELLED);
 479         citthread_cancel(this_thread->tid);
 480 }
 481
 482
 483 /*
 484  * A function for a thread to check if it has been asked to stop
 485  */
 486 int CtdlThreadCheckStop(void)
 487 {
 488         int state;
 489
 490         if (!CT)
 491         {
 492                 syslog(LOG_EMERG, "Thread system PANIC, CtdlThreadCheckStop() called by a non thread.\n");
 493                 CtdlThreadStopAll();
 494                 return -1;
 495         }
 496
 497         state = CT->state;
 498
 499         if (CT->signal)
 500         {
 501                 syslog(LOG_DEBUG, "Thread \"%s\" caught signal %d.\n", CT->name, CT->signal);
 502                 if (CT->signal == SIGHUP)
 503                         CT->state = CTDL_THREAD_STOP_REQ;
 504                 CT->signal = 0;
 505         }
 506         if(state == CTDL_THREAD_STOP_REQ)
 507         {
 508                 CT->state = CTDL_THREAD_STOPPING;
 509                 return -1;
 510         }
 511         else if((state < CTDL_THREAD_STOP_REQ) && (state > CTDL_THREAD_CREATE))
 512         {
 513                 return -1;
 514         }
 515         return 0;
 516 }
 517
 518
 519 /*
 520  * A function to ask a thread to exit
 521  * The thread must call CtdlThreadCheckStop() periodically to determine if it should exit
 522  */
 523 void CtdlThreadStop(CtdlThreadNode *thread)
 524 {
 525         CtdlThreadNode *this_thread;
 526
 527         if (!thread)
 528                 this_thread = CT;
 529         else
 530                 this_thread = thread;
 531         if (!this_thread)
 532                 return;
 533         if (!(this_thread->thread_func))
 534                 return;         // Don't stop garbage collector
 535
 536         if (!citthread_equal(this_thread->tid, GC_thread->tid))
 537                 citthread_kill(this_thread->tid, SIGHUP);
 538
 539         ctdl_thread_internal_change_state (this_thread, CTDL_THREAD_STOP_REQ);
 540         citthread_cond_signal(&this_thread->ThreadCond);
 541         citthread_cond_signal(&this_thread->SleepCond);
 542         this_thread->stop_ticker = time(NULL);
 543 }
 544
 545 /*
 546  * So we now have a sleep command that works with threads but it is in seconds
 547  */
 548 void CtdlThreadSleep(int secs)
 549 {
 550         struct timespec wake_time;
 551         struct timeval time_now;
 552
 553
 554         if (!CT)
 555         {
 556                 syslog(LOG_WARNING, "CtdlThreadSleep() called by something that is not a thread. Should we die?\n");
 557                 return;
 558         }
 559
 560         memset (&wake_time, 0, sizeof(struct timespec));
 561         gettimeofday(&time_now, NULL);
 562         wake_time.tv_sec = time_now.tv_sec + secs;
 563         wake_time.tv_nsec = time_now.tv_usec * 10;
 564
 565         ctdl_thread_internal_change_state (CT, CTDL_THREAD_SLEEPING);
 566
 567         citthread_mutex_lock(&CT->ThreadMutex); /* Prevent something asking us to awaken before we've gone to sleep */
 568         citthread_cond_timedwait(&CT->SleepCond, &CT->ThreadMutex, &wake_time);
 569         citthread_mutex_unlock(&CT->ThreadMutex);
 570
 571         ctdl_thread_internal_change_state (CT, CTDL_THREAD_RUNNING);
 572 }
 573
 574
 575 /*
 576  * Routine to clean up our thread function on exit
 577  */
 578 static void ctdl_internal_thread_cleanup(void *arg)
 579 {
 580         /*
 581          * In here we were called by the current thread because it is exiting
 582          * NB. WE ARE THE CURRENT THREAD
 583          */
 584         if (CT)
 585         {
 586                 const char *name = CT->name;
 587                 const pid_t tid = CT->tid;
 588
 589                 syslog(LOG_NOTICE, "Thread \"%s\" (0x%08lx) exited.\n", name, (unsigned long) tid);
 590         }
 591         else
 592         {
 593                 syslog(LOG_NOTICE, "some ((unknown ? ? ?) Thread exited.\n");
 594         }
 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  * Garbage collection routine.
 607  * Gets called by main() in a loop to clean up the thread list periodically.
 608  */
 609 void CtdlThreadGC (void)
 610 {
 611
 612
 613         return;
 614         /* FIXME this is a big deal, but I think it's causing corruption */
 615
 616
 617         CtdlThreadNode *this_thread, *that_thread;
 618         int workers = 0, sys_workers;
 619         int ret=0;
 620
 621         begin_critical_section(S_THREAD_LIST);
 622
 623         /* Handle exiting of garbage collector thread */
 624         if(num_threads == 1)
 625                 CtdlThreadList->state = CTDL_THREAD_EXITED;
 626
 627 #ifdef WITH_THREADLOG
 628         syslog(LOG_DEBUG, "Thread system running garbage collection.\n");
 629 #endif
 630         /*
 631          * Woke up to do garbage collection
 632          */
 633         this_thread = CtdlThreadList;
 634         while(this_thread)
 635         {
 636                 that_thread = this_thread;
 637                 this_thread = this_thread->next;
 638
 639                 if ((that_thread->state == CTDL_THREAD_STOP_REQ || that_thread->state == CTDL_THREAD_STOPPING)
 640                         && (!citthread_equal(that_thread->tid, citthread_self())))
 641                 {
 642                         syslog(LOG_DEBUG, "Waiting for thread %s (0x%08lx) to exit.\n", that_thread->name, that_thread->tid);
 643                         terminate_stuck_sessions();
 644                 }
 645                 else
 646                 {
 647                         /**
 648                          * Catch the situation where a worker was asked to stop but couldn't and we are not
 649                          * shutting down.
 650                          */
 651                         that_thread->stop_ticker = 0;
 652                 }
 653
 654                 if (that_thread->stop_ticker + 5 == time(NULL))
 655                 {
 656                         syslog(LOG_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);
 657                         if ((that_thread->flags & CTDLTHREAD_WORKER) == 0)
 658                                 syslog(LOG_INFO, "Thread System: A non worker thread would have been canceled this may cause message loss.\n");
 659 //                      that_thread->state = CTDL_THREAD_CANCELLED;
 660                         that_thread->stop_ticker++;
 661 //                      citthread_cancel(that_thread->tid);
 662 //                      continue;
 663                 }
 664
 665                 /* Do we need to clean up this thread? */
 666                 if ((that_thread->state != CTDL_THREAD_EXITED) && (that_thread->state != CTDL_THREAD_CANCELLED))
 667                 {
 668                         if(that_thread->flags & CTDLTHREAD_WORKER)
 669                                 workers++;      /* Sanity check on number of worker threads */
 670                         continue;
 671                 }
 672
 673                 if (citthread_equal(that_thread->tid, citthread_self()) && that_thread->thread_func)
 674                 {       /* Sanity check */
 675                         end_critical_section(S_THREAD_LIST);
 676                         syslog(LOG_EMERG, "Thread system PANIC, a thread is trying to clean up after itself.\n");
 677                         abort();
 678                         return;
 679                 }
 680
 681                 if (num_threads <= 0)
 682                 {       /* Sanity check */
 683                         end_critical_section(S_THREAD_LIST);
 684                         syslog(LOG_EMERG, "Thread system PANIC, num_threads <= 0 and trying to do Garbage Collection.\n");
 685                         abort();
 686                         return;
 687                 }
 688
 689                 if(that_thread->flags & CTDLTHREAD_WORKER)
 690                         num_workers--;  /* This is a wroker thread so reduce the count. */
 691                 num_threads--;
 692                 /* If we are unlinking the list head then the next becomes the list head */
 693                 if(that_thread->prev)
 694                         that_thread->prev->next = that_thread->next;
 695                 else
 696                         CtdlThreadList = that_thread->next;
 697                 if(that_thread->next)
 698                         that_thread->next->prev = that_thread->prev;
 699
 700                 citthread_cond_signal(&that_thread->ThreadCond);
 701                 citthread_cond_signal(&that_thread->SleepCond); // Make sure this thread is awake
 702                 citthread_mutex_lock(&that_thread->ThreadMutex);        // Make sure it has done what its doing
 703                 citthread_mutex_unlock(&that_thread->ThreadMutex);
 704                 /*
 705                  * Join on the thread to do clean up and prevent memory leaks
 706                  * Also makes sure the thread has cleaned up after itself before we remove it from the list
 707                  * We can join on the garbage collector thread the join should just return EDEADLCK
 708                  */
 709                 ret = citthread_join (that_thread->tid, NULL);
 710                 if (ret == EDEADLK)
 711                         syslog(LOG_DEBUG, "Garbage collection on own thread.\n");
 712                 else if (ret == EINVAL)
 713                         syslog(LOG_DEBUG, "Garbage collection, that thread already joined on.\n");
 714                 else if (ret == ESRCH)
 715                         syslog(LOG_DEBUG, "Garbage collection, no thread to join on.\n");
 716                 else if (ret != 0)
 717                         syslog(LOG_DEBUG, "Garbage collection, citthread_join returned an unknown error(%d).\n", ret);
 718                 /*
 719                  * Now we own that thread entry
 720                  */
 721                 syslog(LOG_INFO, "Garbage Collection for thread \"%s\" (0x%08lx).\n",
 722                         that_thread->name, that_thread->tid);
 723                 citthread_mutex_destroy(&that_thread->ThreadMutex);
 724                 citthread_cond_destroy(&that_thread->ThreadCond);
 725                 citthread_mutex_destroy(&that_thread->SleepMutex);
 726                 citthread_cond_destroy(&that_thread->SleepCond);
 727                 citthread_attr_destroy(&that_thread->attr);
 728                 free(that_thread);
 729         }
 730         sys_workers = num_workers;
 731         end_critical_section(S_THREAD_LIST);
 732
 733         /* Sanity check number of worker threads */
 734         if (workers != sys_workers)
 735         {
 736                 syslog(LOG_EMERG,
 737                         "Thread system PANIC, discrepancy in number of worker threads. Counted %d, should be %d.\n",
 738                         workers, sys_workers
 739                         );
 740                 abort();
 741         }
 742 }
 743
 744
 745
 746
 747 /*
 748  * Runtime function for a Citadel Thread.
 749  * This initialises the threads environment and then calls the user supplied thread function
 750  * Note that this is the REAL thread function and wraps the users thread function.
 751  */
 752 static void *ctdl_internal_thread_func (void *arg)
 753 {
 754         CtdlThreadNode *this_thread;
 755         void *ret = NULL;
 756
 757         /* lock and unlock the thread list.
 758          * This causes this thread to wait until all its creation stuff has finished before it
 759          * can continue its execution.
 760          */
 761         begin_critical_section(S_THREAD_LIST);
 762         this_thread = (CtdlThreadNode *) arg;
 763         gettimeofday(&this_thread->start_time, NULL);           /* Time this thread started */
 764
 765         // Register the cleanup function to take care of when we exit.
 766         citthread_cleanup_push(ctdl_internal_thread_cleanup, NULL);
 767         // Get our thread data structure
 768         CtdlThreadAllocTSD();
 769         CT = this_thread;
 770         this_thread->pid = getpid();
 771         memcpy(&this_thread->last_state_change, &this_thread->start_time, sizeof (struct timeval));     /* Changed state so mark it. */
 772         /* Only change to running state if we weren't asked to stop during the create cycle
 773          * Other wise there is a window to allow this threads creation to continue to full grown and
 774          * therby prevent a shutdown of the server.
 775          */
 776         if (!CtdlThreadCheckStop())
 777         {
 778                 citthread_mutex_lock(&this_thread->ThreadMutex);
 779                 this_thread->state = CTDL_THREAD_RUNNING;
 780                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 781         }
 782         end_critical_section(S_THREAD_LIST);
 783
 784         // Register for tracing
 785         #ifdef HAVE_BACKTRACE
 786 ///     eCrash_RegisterThread(this_thread->name, 0);
 787         #endif
 788
 789         // Tell the world we are here
 790 #if defined(HAVE_SYSCALL_H) && defined (SYS_gettid)
 791         this_thread->reltid = syscall(SYS_gettid);
 792 #endif
 793         syslog(LOG_NOTICE, "Created a new thread \"%s\" (0x%08lx).\n",
 794                 this_thread->name, this_thread->tid);
 795
 796         /*
 797          * run the thread to do the work but only if we haven't been asked to stop
 798          */
 799         if (!CtdlThreadCheckStop())
 800                 ret = (this_thread->thread_func)(this_thread->user_args);
 801
 802         /*
 803          * Our thread is exiting either because it wanted to end or because the server is stopping
 804          * We need to clean up
 805          */
 806         citthread_cleanup_pop(1);       // Execute our cleanup routine and remove it
 807
 808         return(ret);
 809 }
 810
 811
 812
 813
 814 /*
 815  * Function to initialise an empty thread structure
 816  */
 817 CtdlThreadNode *ctdl_internal_init_thread_struct(CtdlThreadNode *this_thread, long flags)
 818 {
 819         int ret = 0;
 820
 821         // Ensuring this is zero'd means we make sure the thread doesn't start doing its thing until we are ready.
 822         memset (this_thread, 0, sizeof(CtdlThreadNode));
 823
 824         /* Create the mutex's early so we can use them */
 825         citthread_mutex_init (&(this_thread->ThreadMutex), NULL);
 826         citthread_cond_init (&(this_thread->ThreadCond), NULL);
 827         citthread_mutex_init (&(this_thread->SleepMutex), NULL);
 828         citthread_cond_init (&(this_thread->SleepCond), NULL);
 829
 830         this_thread->state = CTDL_THREAD_CREATE;
 831
 832         if ((ret = citthread_attr_init(&this_thread->attr))) {
 833                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 834                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 835                 citthread_cond_destroy(&(this_thread->ThreadCond));
 836                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 837                 citthread_cond_destroy(&(this_thread->SleepCond));
 838                 syslog(LOG_EMERG, "Thread system, citthread_attr_init: %s\n", strerror(ret));
 839                 free(this_thread);
 840                 return NULL;
 841         }
 842
 843         /* Our per-thread stacks need to be bigger than the default size,
 844          * otherwise the MIME parser crashes on FreeBSD, and the IMAP service
 845          * crashes on 64-bit Linux.
 846          */
 847         if (flags & CTDLTHREAD_BIGSTACK)
 848         {
 849 #ifdef WITH_THREADLOG
 850                 syslog(LOG_INFO, "Thread system. Creating BIG STACK thread.\n");
 851 #endif
 852                 if ((ret = citthread_attr_setstacksize(&this_thread->attr, THREADSTACKSIZE))) {
 853                         citthread_mutex_unlock(&this_thread->ThreadMutex);
 854                         citthread_mutex_destroy(&(this_thread->ThreadMutex));
 855                         citthread_cond_destroy(&(this_thread->ThreadCond));
 856                         citthread_mutex_destroy(&(this_thread->SleepMutex));
 857                         citthread_cond_destroy(&(this_thread->SleepCond));
 858                         citthread_attr_destroy(&this_thread->attr);
 859                         syslog(LOG_EMERG, "Thread system, citthread_attr_setstacksize: %s\n",
 860                                 strerror(ret));
 861                         free(this_thread);
 862                         return NULL;
 863                 }
 864         }
 865
 866         /* Set this new thread with an avg_blocked of 2. We do this so that its creation affects the
 867          * load average for the system. If we don't do this then we create a mass of threads at the same time
 868          * because the creation didn't affect the load average.
 869          */
 870         this_thread->avg_blocked = 2;
 871
 872         return (this_thread);
 873 }
 874
 875
 876
 877
 878 /*
 879  * Internal function to create a thread.
 880  */
 881 CtdlThreadNode *ctdl_internal_create_thread(char *name, long flags, void *(*thread_func) (void *arg), void *args)
 882 {
 883         int ret = 0;
 884         CtdlThreadNode *this_thread;
 885
 886         if (num_threads >= 32767)
 887         {
 888                 syslog(LOG_EMERG, "Thread system. Thread list full.\n");
 889                 return NULL;
 890         }
 891
 892         this_thread = malloc(sizeof(CtdlThreadNode));
 893         if (this_thread == NULL) {
 894                 syslog(LOG_EMERG, "Thread system, can't allocate CtdlThreadNode, exiting\n");
 895                 return NULL;
 896         }
 897
 898         /* Initialise the thread structure */
 899         if (ctdl_internal_init_thread_struct(this_thread, flags) == NULL)
 900         {
 901                 free(this_thread);
 902                 syslog(LOG_EMERG, "Thread system, can't initialise CtdlThreadNode, exiting\n");
 903                 return NULL;
 904         }
 905         /*
 906          * If we got here we are going to create the thread so we must initilise the structure
 907          * first because most implimentations of threading can't create it in a stopped state
 908          * and it might want to do things with its structure that aren't initialised otherwise.
 909          */
 910         if(name)
 911         {
 912                 this_thread->name = name;
 913         }
 914         else
 915         {
 916                 this_thread->name = "Un-named Thread";
 917         }
 918
 919         this_thread->flags = flags;
 920         this_thread->thread_func = thread_func;
 921         this_thread->user_args = args;
 922
 923         begin_critical_section(S_THREAD_LIST);
 924         /*
 925          * We pass this_thread into the thread as its args so that it can find out information
 926          * about itself and it has a bit of storage space for itself, not to mention that the REAL
 927          * thread function needs to finish off the setup of the structure
 928          */
 929         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
 930         {
 931                 end_critical_section(S_THREAD_LIST);
 932                 syslog(LOG_ALERT, "Thread system, Can't create thread: %s\n",
 933                         strerror(ret));
 934                 citthread_mutex_unlock(&this_thread->ThreadMutex);
 935                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 936                 citthread_cond_destroy(&(this_thread->ThreadCond));
 937                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 938                 citthread_cond_destroy(&(this_thread->SleepCond));
 939                 citthread_attr_destroy(&this_thread->attr);
 940                 free(this_thread);
 941                 return NULL;
 942         }
 943         num_threads++;  // Increase the count of threads in the system.
 944         if(this_thread->flags & CTDLTHREAD_WORKER)
 945                 num_workers++;
 946
 947         this_thread->next = CtdlThreadList;
 948         CtdlThreadList = this_thread;
 949         if (this_thread->next)
 950                 this_thread->next->prev = this_thread;
 951
 952         end_critical_section(S_THREAD_LIST);
 953
 954         return this_thread;
 955 }
 956
 957 /*
 958  * Wrapper function to create a thread
 959  * ensures the critical section and other protections are in place.
 960  * char *name = name to give to thread, if NULL, use generic name
 961  * int flags = flags to determine type of thread and standard facilities
 962  */
 963 CtdlThreadNode *CtdlThreadCreate(char *name, long flags, void *(*thread_func) (void *arg), void *args)
 964 {
 965         CtdlThreadNode *ret = NULL;
 966
 967         ret = ctdl_internal_create_thread(name, flags, thread_func, args);
 968         return ret;
 969 }
 970
 971
 972
 973 CtdlThreadNode *ctdl_thread_internal_start_scheduled (CtdlThreadNode *this_thread)
 974 {
 975         int ret = 0;
 976
 977         begin_critical_section(S_THREAD_LIST);
 978         /*
 979          * We pass this_thread into the thread as its args so that it can find out information
 980          * about itself and it has a bit of storage space for itself, not to mention that the REAL
 981          * thread function needs to finish off the setup of the structure
 982          */
 983         if ((ret = citthread_create(&this_thread->tid, &this_thread->attr, ctdl_internal_thread_func, this_thread) != 0))
 984         {
 985                 end_critical_section(S_THREAD_LIST);
 986                 syslog(LOG_DEBUG, "Failed to start scheduled thread \"%s\": %s\n", this_thread->name, strerror(ret));
 987                 citthread_mutex_destroy(&(this_thread->ThreadMutex));
 988                 citthread_cond_destroy(&(this_thread->ThreadCond));
 989                 citthread_mutex_destroy(&(this_thread->SleepMutex));
 990                 citthread_cond_destroy(&(this_thread->SleepCond));
 991                 citthread_attr_destroy(&this_thread->attr);
 992                 free(this_thread);
 993                 return NULL;
 994         }
 995
 996
 997         num_threads++;  // Increase the count of threads in the system.
 998         if(this_thread->flags & CTDLTHREAD_WORKER)
 999                 num_workers++;
1000
1001         this_thread->next = CtdlThreadList;
1002         CtdlThreadList = this_thread;
1003         if (this_thread->next)
1004                 this_thread->next->prev = this_thread;
1005
1006         end_critical_section(S_THREAD_LIST);
1007
1008
1009         return this_thread;
1010 }
1011
1012
1013
1014 void ctdl_thread_internal_check_scheduled(void)
1015 {
1016         CtdlThreadNode *this_thread, *that_thread;
1017         time_t now;
1018
1019         /* Don't start scheduled threads if the system wants single user mode */
1020         if (CtdlWantSingleUser())
1021                 return;
1022
1023         if (try_critical_section(S_SCHEDULE_LIST))
1024                 return; /* If this list is locked we wait till the next chance */
1025
1026         now = time(NULL);
1027
1028 #ifdef WITH_THREADLOG
1029         syslog(LOG_DEBUG, "Checking for scheduled threads to start.\n");
1030 #endif
1031
1032         this_thread = CtdlThreadSchedList;
1033         while(this_thread)
1034         {
1035                 that_thread = this_thread;
1036                 this_thread = this_thread->next;
1037
1038                 if (now > that_thread->when)
1039                 {
1040                         /* Unlink from schedule list */
1041                         if (that_thread->prev)
1042                                 that_thread->prev->next = that_thread->next;
1043                         else
1044                                 CtdlThreadSchedList = that_thread->next;
1045                         if (that_thread->next)
1046                                 that_thread->next->prev = that_thread->prev;
1047
1048                         that_thread->next = that_thread->prev = NULL;
1049 #ifdef WITH_THREADLOG
1050                         syslog(LOG_DEBUG, "About to start scheduled thread \"%s\".\n", that_thread->name);
1051 #endif
1052                         if (CT->state > CTDL_THREAD_STOP_REQ)
1053                         {       /* Only start it if the system is not stopping */
1054                                 if (ctdl_thread_internal_start_scheduled (that_thread))
1055                                 {
1056 #ifdef WITH_THREADLOG
1057                                         syslog(LOG_INFO, "Thread system, Started a scheduled thread \"%s\" (0x%08lx).\n",
1058                                                 that_thread->name, that_thread->tid);
1059 #endif
1060                                 }
1061                         }
1062                 }
1063 #ifdef WITH_THREADLOG
1064                 else
1065                 {
1066                         syslog(LOG_DEBUG, "Thread \"%s\" will start in %ld seconds.\n",
1067                                 that_thread->name, that_thread->when - time(NULL));
1068                 }
1069 #endif
1070         }
1071         end_critical_section(S_SCHEDULE_LIST);
1072 }
1073
1074
1075 /*
1076  * A warapper function for select so we can show a thread as blocked
1077  */
1078 int CtdlThreadSelect(int n, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, struct timeval *timeout)
1079 {
1080         int ret = 0;
1081
1082         ctdl_thread_internal_change_state(CT, CTDL_THREAD_BLOCKED);
1083         if (!CtdlThreadCheckStop())
1084                 ret = select(n, readfds, writefds, exceptfds, timeout);
1085         /**
1086          * If the select returned <= 0 then it failed due to an error
1087          * or timeout so this thread could stop if asked to do so.
1088          * Anything else means it needs to continue unless the system is shutting down
1089          */
1090         if (ret > 0)
1091         {
1092                 /**
1093                  * The select says this thread needs to do something useful.
1094                  * This thread was in an idle state so it may have been asked to stop
1095                  * but if the system isn't shutting down this thread is no longer
1096                  * idle and select has given it a task to do so it must not stop
1097                  * In this condition we need to force it into the running state.
1098                  * CtdlThreadGC will clear its ticker for us.
1099                  *
1100                  * FIXME: there is still a small hole here. It is possible for the sequence of locking
1101                  * to allow the state to get changed to STOP_REQ just after this code if the other thread
1102                  * has decided to change the state before this lock, it there fore has to wait till the lock
1103                  * completes but it will continue to change the state. We need something a bit better here.
1104                  */
1105                 citthread_mutex_lock(&CT->ThreadMutex); /* To prevent race condition of a sleeping thread */
1106                 if (GC_thread->state > CTDL_THREAD_STOP_REQ && CT->state <= CTDL_THREAD_STOP_REQ)
1107                 {
1108                         syslog(LOG_DEBUG, "Thread %s (0x%08lx) refused stop request.\n", CT->name, CT->tid);
1109                         CT->state = CTDL_THREAD_RUNNING;
1110                 }
1111                 citthread_mutex_unlock(&CT->ThreadMutex);
1112         }
1113
1114         ctdl_thread_internal_change_state(CT, CTDL_THREAD_RUNNING);
1115
1116         return ret;
1117 }
1118
1119
1120
1121 void *new_worker_thread(void *arg);
1122 extern void close_masters (void);
1123
1124
1125
1126 void go_threading(void)
1127 {
1128         int i;
1129         CtdlThreadNode *last_worker;
1130
1131         /*
1132          * Initialise the thread system
1133          */
1134         ctdl_thread_internal_init();
1135
1136         /* Second call to module init functions now that threading is up */
1137         initialise_modules(1);
1138         CtdlThreadCreate("select_on_master", CTDLTHREAD_BIGSTACK, select_on_master, NULL);
1139
1140         /*
1141          * This thread is now used for garbage collection of other threads in the thread list
1142          */
1143         syslog(LOG_INFO, "Startup thread %ld becoming garbage collector,\n", (long) citthread_self());
1144
1145         /*
1146          * We do a lot of locking and unlocking of the thread list in here.
1147          * We do this so that we can repeatedly release time for other threads
1148          * that may be waiting on the thread list.
1149          * We are a low priority thread so we can afford to do this
1150          */
1151
1152         while (CtdlThreadGetCount())
1153         {
1154                 if (CT->signal)
1155                         exit_signal = CT->signal;
1156                 if (exit_signal)
1157                 {
1158                         CtdlThreadStopAll();
1159                 }
1160                 check_sched_shutdown();
1161                 if (CT->state > CTDL_THREAD_STOP_REQ)
1162                 {
1163                         begin_critical_section(S_THREAD_LIST);
1164                         end_critical_section(S_THREAD_LIST);
1165
1166                         ctdl_thread_internal_check_scheduled(); /* start scheduled threads */
1167                 }
1168
1169                 /* Reduce the size of the worker thread pool if necessary. */
1170                 if ((CtdlThreadGetWorkers() > config.c_min_workers + 1) && (CtdlThreadWorkerAvg < 20) && (CT->state > CTDL_THREAD_STOP_REQ))
1171                 {
1172                         /* Ask a worker thread to stop as we no longer need it */
1173                         begin_critical_section(S_THREAD_LIST);
1174                         last_worker = CtdlThreadList;
1175                         while (last_worker)
1176                         {
1177                                 citthread_mutex_lock(&last_worker->ThreadMutex);
1178                                 if (last_worker->flags & CTDLTHREAD_WORKER && (last_worker->state > CTDL_THREAD_STOPPING) && (last_worker->Context == NULL))
1179                                 {
1180                                         citthread_mutex_unlock(&last_worker->ThreadMutex);
1181                                         break;
1182                                 }
1183                                 citthread_mutex_unlock(&last_worker->ThreadMutex);
1184                                 last_worker = last_worker->next;
1185                         }
1186                         end_critical_section(S_THREAD_LIST);
1187                         if (last_worker)
1188                         {
1189 #ifdef WITH_THREADLOG
1190                                 syslog(LOG_DEBUG, "Thread system, stopping excess worker thread \"%s\" (0x%08lx).\n",
1191                                         last_worker->name,
1192                                         last_worker->tid
1193                                         );
1194 #endif
1195                                 CtdlThreadStop(last_worker);
1196                         }
1197                 }
1198
1199                 /*
1200                  * If all our workers are working hard, start some more to help out
1201                  * with things
1202                  */
1203                 /* FIXME: come up with a better way to dynamically alter the number of threads
1204                  * based on the system load
1205                  */
1206                 if (    (((CtdlThreadGetWorkers() < config.c_max_workers)
1207                         && (CtdlThreadGetWorkerAvg() > 60))
1208                         || CtdlThreadGetWorkers() < config.c_min_workers)
1209                         && (CT->state > CTDL_THREAD_STOP_REQ)
1210                 )
1211                 {
1212                         /* Only start new threads if we are not going to overload the machine */
1213                         /* Temporarily set to 10 should be enough to make sure we don't stranglew the server
1214                          * at least until we make this a config option */
1215                         if (CtdlThreadGetLoadAvg() < ((double)10.00)) {
1216                                 for (i=0; i<5 ; i++) {
1217                                         CtdlThreadCreate("Worker Thread",
1218                                                 CTDLTHREAD_BIGSTACK + CTDLTHREAD_WORKER,
1219                                                 worker_thread,
1220                                                 NULL
1221                                                 );
1222                                 }
1223                         }
1224                         else
1225                                 syslog(LOG_WARNING, "Server strangled due to machine load average too high.\n");
1226                 }
1227
1228                 CtdlThreadGC();
1229
1230                 if (CtdlThreadGetCount() <= 1) // Shutting down clean up the garbage collector
1231                 {
1232                         CtdlThreadGC();
1233                 }
1234
1235 #ifdef THREADS_USESIGNALS
1236                 if (CtdlThreadGetCount() && CT->state > CTDL_THREAD_STOP_REQ)
1237 #else
1238                 if (CtdlThreadGetCount())
1239 #endif
1240                         CtdlThreadSleep(1);
1241         }
1242         /*
1243          * If the above loop exits we must be shutting down since we obviously have no threads
1244          */
1245         ctdl_thread_internal_cleanup();
1246 }
1247
1248
1249
1250