src/comm/ModPoll.cc

   1 /*
   2  * $Id$
   3  *
   4  * DEBUG: section 05    Socket Functions
   5  *
   6  * SQUID Web Proxy Cache          http://www.squid-cache.org/
   7  * ----------------------------------------------------------
   8  *
   9  *  Squid is the result of efforts by numerous individuals from
  10  *  the Internet community; see the CONTRIBUTORS file for full
  11  *  details.   Many organizations have provided support for Squid's
  12  *  development; see the SPONSORS file for full details.  Squid is
  13  *  Copyrighted (C) 2001 by the Regents of the University of
  14  *  California; see the COPYRIGHT file for full details.  Squid
  15  *  incorporates software developed and/or copyrighted by other
  16  *  sources; see the CREDITS file for full details.
  17  *
  18  *  This program is free software; you can redistribute it and/or modify
  19  *  it under the terms of the GNU General Public License as published by
  20  *  the Free Software Foundation; either version 2 of the License, or
  21  *  (at your option) any later version.
  22  *
  23  *  This program is distributed in the hope that it will be useful,
  24  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  25  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  26  *  GNU General Public License for more details.
  27  *
  28  *  You should have received a copy of the GNU General Public License
  29  *  along with this program; if not, write to the Free Software
  30  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
  31  *
  32  */
  33 #include "config.h"
  34
  35 #if USE_POLL
  36
  37 #include "squid.h"
  38 #include "comm/Connection.h"
  39 #include "comm/Loops.h"
  40 #include "fde.h"
  41 #include "ICP.h"
  42 #include "mgr/Registration.h"
  43 #include "SquidTime.h"
  44 #include "Store.h"
  45
  46 #if HAVE_POLL_H
  47 #include <poll.h>
  48 #endif
  49
  50 /* Needed for poll() on Linux at least */
  51 #if USE_POLL
  52 #ifndef POLLRDNORM
  53 #define POLLRDNORM POLLIN
  54 #endif
  55 #ifndef POLLWRNORM
  56 #define POLLWRNORM POLLOUT
  57 #endif
  58 #endif
  59
  60 static int MAX_POLL_TIME = 1000;        /* see also Comm::QuickPollRequired() */
  61
  62 #ifndef        howmany
  63 #define howmany(x, y)   (((x)+((y)-1))/(y))
  64 #endif
  65 #ifndef        NBBY
  66 #define        NBBY    8
  67 #endif
  68 #define FD_MASK_BYTES sizeof(fd_mask)
  69 #define FD_MASK_BITS (FD_MASK_BYTES*NBBY)
  70
  71 /* STATIC */
  72 static int fdIsHttp(int fd);
  73 static int fdIsIcp(int fd);
  74 static int fdIsDns(int fd);
  75 static OBJH commIncomingStats;
  76 static int comm_check_incoming_poll_handlers(int nfds, int *fds);
  77 static void comm_poll_dns_incoming(void);
  78
  79 /*
  80  * Automatic tuning for incoming requests:
  81  *
  82  * INCOMING sockets are the ICP and HTTP ports.  We need to check these
  83  * fairly regularly, but how often?  When the load increases, we
  84  * want to check the incoming sockets more often.  If we have a lot
  85  * of incoming ICP, then we need to check these sockets more than
  86  * if we just have HTTP.
  87  *
  88  * The variables 'incoming_icp_interval' and 'incoming_http_interval'
  89  * determine how many normal I/O events to process before checking
  90  * incoming sockets again.  Note we store the incoming_interval
  91  * multipled by a factor of (2^INCOMING_FACTOR) to have some
  92  * pseudo-floating point precision.
  93  *
  94  * The variable 'icp_io_events' and 'http_io_events' counts how many normal
  95  * I/O events have been processed since the last check on the incoming
  96  * sockets.  When io_events > incoming_interval, its time to check incoming
  97  * sockets.
  98  *
  99  * Every time we check incoming sockets, we count how many new messages
 100  * or connections were processed.  This is used to adjust the
 101  * incoming_interval for the next iteration.  The new incoming_interval
 102  * is calculated as the current incoming_interval plus what we would
 103  * like to see as an average number of events minus the number of
 104  * events just processed.
 105  *
 106  *  incoming_interval = incoming_interval + target_average - number_of_events_processed
 107  *
 108  * There are separate incoming_interval counters for both HTTP and ICP events
 109  *
 110  * You can see the current values of the incoming_interval's, as well as
 111  * a histogram of 'incoming_events' by asking the cache manager
 112  * for 'comm_incoming', e.g.:
 113  *
 114  *      % ./client mgr:comm_poll_incoming
 115  *
 116  * Caveats:
 117  *
 118  *      - We have MAX_INCOMING_INTEGER as a magic upper limit on
 119  *        incoming_interval for both types of sockets.  At the
 120  *        largest value the cache will effectively be idling.
 121  *
 122  *      - The higher the INCOMING_FACTOR, the slower the algorithm will
 123  *        respond to load spikes/increases/decreases in demand. A value
 124  *        between 3 and 8 is recommended.
 125  */
 126
 127 #define MAX_INCOMING_INTEGER 256
 128 #define INCOMING_FACTOR 5
 129 #define MAX_INCOMING_INTERVAL (MAX_INCOMING_INTEGER << INCOMING_FACTOR)
 130 static int icp_io_events = 0;
 131 static int dns_io_events = 0;
 132 static int http_io_events = 0;
 133 static int incoming_icp_interval = 16 << INCOMING_FACTOR;
 134 static int incoming_dns_interval = 16 << INCOMING_FACTOR;
 135 static int incoming_http_interval = 16 << INCOMING_FACTOR;
 136 #define commCheckICPIncoming (++icp_io_events > (incoming_icp_interval>> INCOMING_FACTOR))
 137 #define commCheckDNSIncoming (++dns_io_events > (incoming_dns_interval>> INCOMING_FACTOR))
 138 #define commCheckHTTPIncoming (++http_io_events > (incoming_http_interval>> INCOMING_FACTOR))
 139
 140
 141 void
 142 Comm::SetSelect(int fd, unsigned int type, PF * handler, void *client_data, time_t timeout)
 143 {
 144     fde *F = &fd_table[fd];
 145     assert(fd >= 0);
 146     assert(F->flags.open);
 147     debugs(5, 5, "commSetSelect: FD " << fd << " type " << type);
 148
 149     if (type & COMM_SELECT_READ) {
 150         F->read_handler = handler;
 151         F->read_data = client_data;
 152     }
 153
 154     if (type & COMM_SELECT_WRITE) {
 155         F->write_handler = handler;
 156         F->write_data = client_data;
 157     }
 158
 159     if (timeout)
 160         F->timeout = squid_curtime + timeout;
 161 }
 162
 163 void
 164 Comm::ResetSelect(int fd)
 165 {
 166 }
 167
 168 static int
 169 fdIsIcp(int fd)
 170 {
 171     if (icpIncomingConn != NULL && icpIncomingConn->fd == fd)
 172         return 1;
 173
 174     if (icpOutgoingConn != NULL && icpOutgoingConn->fd == fd)
 175         return 1;
 176
 177     return 0;
 178 }
 179
 180 static int
 181 fdIsDns(int fd)
 182 {
 183     if (fd == DnsSocketA)
 184         return 1;
 185
 186     if (fd == DnsSocketB)
 187         return 1;
 188
 189     return 0;
 190 }
 191
 192 static int
 193 fdIsHttp(int fd)
 194 {
 195     int j;
 196
 197     for (j = 0; j < NHttpSockets; j++) {
 198         if (fd == HttpSockets[j])
 199             return 1;
 200     }
 201
 202     return 0;
 203 }
 204
 205 static int
 206 comm_check_incoming_poll_handlers(int nfds, int *fds)
 207 {
 208     int i;
 209     int fd;
 210     PF *hdl = NULL;
 211     int npfds;
 212
 213     struct pollfd pfds[3 + MAXHTTPPORTS];
 214     PROF_start(comm_check_incoming);
 215     incoming_sockets_accepted = 0;
 216
 217     for (i = npfds = 0; i < nfds; i++) {
 218         int events;
 219         fd = fds[i];
 220         events = 0;
 221
 222         if (fd_table[fd].read_handler)
 223             events |= POLLRDNORM;
 224
 225         if (fd_table[fd].write_handler)
 226             events |= POLLWRNORM;
 227
 228         if (events) {
 229             pfds[npfds].fd = fd;
 230             pfds[npfds].events = events;
 231             pfds[npfds].revents = 0;
 232             npfds++;
 233         }
 234     }
 235
 236     if (!nfds) {
 237         PROF_stop(comm_check_incoming);
 238         return -1;
 239     }
 240
 241     getCurrentTime();
 242     statCounter.syscalls.selects++;
 243
 244     if (poll(pfds, npfds, 0) < 1) {
 245         PROF_stop(comm_check_incoming);
 246         return incoming_sockets_accepted;
 247     }
 248
 249     for (i = 0; i < npfds; i++) {
 250         int revents;
 251
 252         if (((revents = pfds[i].revents) == 0) || ((fd = pfds[i].fd) == -1))
 253             continue;
 254
 255         if (revents & (POLLRDNORM | POLLIN | POLLHUP | POLLERR)) {
 256             if ((hdl = fd_table[fd].read_handler)) {
 257                 fd_table[fd].read_handler = NULL;
 258                 hdl(fd, fd_table[fd].read_data);
 259             } else if (pfds[i].events & POLLRDNORM)
 260                 debugs(5, 1, "comm_poll_incoming: FD " << fd << " NULL read handler");
 261         }
 262
 263         if (revents & (POLLWRNORM | POLLOUT | POLLHUP | POLLERR)) {
 264             if ((hdl = fd_table[fd].write_handler)) {
 265                 fd_table[fd].write_handler = NULL;
 266                 hdl(fd, fd_table[fd].write_data);
 267             } else if (pfds[i].events & POLLWRNORM)
 268                 debugs(5, 1, "comm_poll_incoming: FD " << fd << " NULL write_handler");
 269         }
 270     }
 271
 272     PROF_stop(comm_check_incoming);
 273     return incoming_sockets_accepted;
 274 }
 275
 276 static void
 277 comm_poll_icp_incoming(void)
 278 {
 279     int nfds = 0;
 280     int fds[2];
 281     int nevents;
 282     icp_io_events = 0;
 283
 284     if (Comm::IsConnOpen(icpIncomingConn))
 285         fds[nfds++] = icpIncomingConn->fd;
 286
 287     if (icpIncomingConn != icpOutgoingConn && Comm::IsConnOpen(icpOutgoingConn))
 288             fds[nfds++] = icpOutgoingConn->fd;
 289
 290     if (nfds == 0)
 291         return;
 292
 293     nevents = comm_check_incoming_poll_handlers(nfds, fds);
 294
 295     incoming_icp_interval += Config.comm_incoming.icp_average - nevents;
 296
 297     if (incoming_icp_interval < Config.comm_incoming.icp_min_poll)
 298         incoming_icp_interval = Config.comm_incoming.icp_min_poll;
 299
 300     if (incoming_icp_interval > MAX_INCOMING_INTERVAL)
 301         incoming_icp_interval = MAX_INCOMING_INTERVAL;
 302
 303     if (nevents > INCOMING_ICP_MAX)
 304         nevents = INCOMING_ICP_MAX;
 305
 306     statHistCount(&statCounter.comm_icp_incoming, nevents);
 307 }
 308
 309 static void
 310 comm_poll_http_incoming(void)
 311 {
 312     int nfds = 0;
 313     int fds[MAXHTTPPORTS];
 314     int j;
 315     int nevents;
 316     http_io_events = 0;
 317
 318     /* only poll sockets that won't be deferred */
 319
 320     for (j = 0; j < NHttpSockets; j++) {
 321         if (HttpSockets[j] < 0)
 322             continue;
 323
 324         fds[nfds++] = HttpSockets[j];
 325     }
 326
 327     nevents = comm_check_incoming_poll_handlers(nfds, fds);
 328     incoming_http_interval = incoming_http_interval
 329                              + Config.comm_incoming.http_average - nevents;
 330
 331     if (incoming_http_interval < Config.comm_incoming.http_min_poll)
 332         incoming_http_interval = Config.comm_incoming.http_min_poll;
 333
 334     if (incoming_http_interval > MAX_INCOMING_INTERVAL)
 335         incoming_http_interval = MAX_INCOMING_INTERVAL;
 336
 337     if (nevents > INCOMING_HTTP_MAX)
 338         nevents = INCOMING_HTTP_MAX;
 339
 340     statHistCount(&statCounter.comm_http_incoming, nevents);
 341 }
 342
 343 /* poll all sockets; call handlers for those that are ready. */
 344 comm_err_t
 345 Comm::DoSelect(int msec)
 346 {
 347     struct pollfd pfds[SQUID_MAXFD];
 348
 349     PF *hdl = NULL;
 350     int fd;
 351     int maxfd;
 352     unsigned long nfds;
 353     unsigned long npending;
 354     int num;
 355     int callicp = 0, callhttp = 0;
 356     int calldns = 0;
 357     double timeout = current_dtime + (msec / 1000.0);
 358
 359     do {
 360         double start;
 361         getCurrentTime();
 362         start = current_dtime;
 363
 364         if (commCheckICPIncoming)
 365             comm_poll_icp_incoming();
 366
 367         if (commCheckDNSIncoming)
 368             comm_poll_dns_incoming();
 369
 370         if (commCheckHTTPIncoming)
 371             comm_poll_http_incoming();
 372
 373         PROF_start(comm_poll_prep_pfds);
 374
 375         callicp = calldns = callhttp = 0;
 376
 377         nfds = 0;
 378
 379         npending = 0;
 380
 381         maxfd = Biggest_FD + 1;
 382
 383         for (int i = 0; i < maxfd; i++) {
 384             int events;
 385             events = 0;
 386             /* Check each open socket for a handler. */
 387
 388             if (fd_table[i].read_handler)
 389                 events |= POLLRDNORM;
 390
 391             if (fd_table[i].write_handler)
 392                 events |= POLLWRNORM;
 393
 394             if (events) {
 395                 pfds[nfds].fd = i;
 396                 pfds[nfds].events = events;
 397                 pfds[nfds].revents = 0;
 398                 nfds++;
 399
 400                 if ((events & POLLRDNORM) && fd_table[i].flags.read_pending)
 401                     npending++;
 402             }
 403         }
 404
 405         PROF_stop(comm_poll_prep_pfds);
 406
 407         if (npending)
 408             msec = 0;
 409
 410         if (msec > MAX_POLL_TIME)
 411             msec = MAX_POLL_TIME;
 412
 413         /* nothing to do
 414          *
 415          * Note that this will only ever trigger when there are no log files
 416          * and stdout/err/in are all closed too.
 417          */
 418         if (nfds == 0 && npending == 0) {
 419             if (shutting_down)
 420                 return COMM_SHUTDOWN;
 421             else
 422                 return COMM_IDLE;
 423         }
 424
 425         for (;;) {
 426             PROF_start(comm_poll_normal);
 427             ++statCounter.syscalls.selects;
 428             num = poll(pfds, nfds, msec);
 429             ++statCounter.select_loops;
 430             PROF_stop(comm_poll_normal);
 431
 432             if (num >= 0 || npending > 0)
 433                 break;
 434
 435             if (ignoreErrno(errno))
 436                 continue;
 437
 438             debugs(5, 0, "comm_poll: poll failure: " << xstrerror());
 439
 440             assert(errno != EINVAL);
 441
 442             return COMM_ERROR;
 443
 444             /* NOTREACHED */
 445         }
 446
 447         getCurrentTime();
 448
 449         debugs(5, num ? 5 : 8, "comm_poll: " << num << "+" << npending << " FDs ready");
 450         statHistCount(&statCounter.select_fds_hist, num);
 451
 452         if (num == 0 && npending == 0)
 453             continue;
 454
 455         /* scan each socket but the accept socket. Poll this
 456          * more frequently to minimize losses due to the 5 connect
 457          * limit in SunOS */
 458         PROF_start(comm_handle_ready_fd);
 459
 460         for (size_t loopIndex = 0; loopIndex < nfds; loopIndex++) {
 461             fde *F;
 462             int revents = pfds[loopIndex].revents;
 463             fd = pfds[loopIndex].fd;
 464
 465             if (fd == -1)
 466                 continue;
 467
 468             if (fd_table[fd].flags.read_pending)
 469                 revents |= POLLIN;
 470
 471             if (revents == 0)
 472                 continue;
 473
 474             if (fdIsIcp(fd)) {
 475                 callicp = 1;
 476                 continue;
 477             }
 478
 479             if (fdIsDns(fd)) {
 480                 calldns = 1;
 481                 continue;
 482             }
 483
 484             if (fdIsHttp(fd)) {
 485                 callhttp = 1;
 486                 continue;
 487             }
 488
 489             F = &fd_table[fd];
 490
 491             if (revents & (POLLRDNORM | POLLIN | POLLHUP | POLLERR)) {
 492                 debugs(5, 6, "comm_poll: FD " << fd << " ready for reading");
 493
 494                 if (NULL == (hdl = F->read_handler))
 495                     (void) 0;
 496                 else {
 497                     PROF_start(comm_read_handler);
 498                     F->read_handler = NULL;
 499                     F->flags.read_pending = 0;
 500                     hdl(fd, F->read_data);
 501                     PROF_stop(comm_read_handler);
 502                     statCounter.select_fds++;
 503
 504                     if (commCheckICPIncoming)
 505                         comm_poll_icp_incoming();
 506
 507                     if (commCheckDNSIncoming)
 508                         comm_poll_dns_incoming();
 509
 510                     if (commCheckHTTPIncoming)
 511                         comm_poll_http_incoming();
 512                 }
 513             }
 514
 515             if (revents & (POLLWRNORM | POLLOUT | POLLHUP | POLLERR)) {
 516                 debugs(5, 5, "comm_poll: FD " << fd << " ready for writing");
 517
 518                 if ((hdl = F->write_handler)) {
 519                     PROF_start(comm_write_handler);
 520                     F->write_handler = NULL;
 521                     hdl(fd, F->write_data);
 522                     PROF_stop(comm_write_handler);
 523                     statCounter.select_fds++;
 524
 525                     if (commCheckICPIncoming)
 526                         comm_poll_icp_incoming();
 527
 528                     if (commCheckDNSIncoming)
 529                         comm_poll_dns_incoming();
 530
 531                     if (commCheckHTTPIncoming)
 532                         comm_poll_http_incoming();
 533                 }
 534             }
 535
 536             if (revents & POLLNVAL) {
 537                 AsyncCall::Pointer ch;
 538                 debugs(5, 0, "WARNING: FD " << fd << " has handlers, but it's invalid.");
 539                 debugs(5, 0, "FD " << fd << " is a " << fdTypeStr[F->type]);
 540                 debugs(5, 0, "--> " << F->desc);
 541                 debugs(5, 0, "tmout:" << F->timeoutHandler << "read:" <<
 542                        F->read_handler << " write:" << F->write_handler);
 543
 544                 for (ch = F->closeHandler; ch != NULL; ch = ch->Next())
 545                     debugs(5, 0, " close handler: " << ch);
 546
 547                 if (F->closeHandler != NULL) {
 548                     commCallCloseHandlers(fd);
 549                 } else if (F->timeoutHandler != NULL) {
 550                     debugs(5, 0, "comm_poll: Calling Timeout Handler");
 551                     ScheduleCallHere(F->timeoutHandler);
 552                 }
 553
 554                 F->closeHandler = NULL;
 555                 F->timeoutHandler = NULL;
 556                 F->read_handler = NULL;
 557                 F->write_handler = NULL;
 558
 559                 if (F->flags.open)
 560                     fd_close(fd);
 561             }
 562         }
 563
 564         PROF_stop(comm_handle_ready_fd);
 565
 566         if (callicp)
 567             comm_poll_icp_incoming();
 568
 569         if (calldns)
 570             comm_poll_dns_incoming();
 571
 572         if (callhttp)
 573             comm_poll_http_incoming();
 574
 575         getCurrentTime();
 576
 577         statCounter.select_time += (current_dtime - start);
 578
 579         return COMM_OK;
 580     } while (timeout > current_dtime);
 581
 582     debugs(5, 8, "comm_poll: time out: " << squid_curtime << ".");
 583
 584     return COMM_TIMEOUT;
 585 }
 586
 587
 588 static void
 589 comm_poll_dns_incoming(void)
 590 {
 591     int nfds = 0;
 592     int fds[2];
 593     int nevents;
 594     dns_io_events = 0;
 595
 596     if (DnsSocketA < 0 && DnsSocketB < 0)
 597         return;
 598
 599     if (DnsSocketA >= 0)
 600         fds[nfds++] = DnsSocketA;
 601
 602     if (DnsSocketB >= 0)
 603         fds[nfds++] = DnsSocketB;
 604
 605     nevents = comm_check_incoming_poll_handlers(nfds, fds);
 606
 607     if (nevents < 0)
 608         return;
 609
 610     incoming_dns_interval += Config.comm_incoming.dns_average - nevents;
 611
 612     if (incoming_dns_interval < Config.comm_incoming.dns_min_poll)
 613         incoming_dns_interval = Config.comm_incoming.dns_min_poll;
 614
 615     if (incoming_dns_interval > MAX_INCOMING_INTERVAL)
 616         incoming_dns_interval = MAX_INCOMING_INTERVAL;
 617
 618     if (nevents > INCOMING_DNS_MAX)
 619         nevents = INCOMING_DNS_MAX;
 620
 621     statHistCount(&statCounter.comm_dns_incoming, nevents);
 622 }
 623
 624
 625 static void
 626 commPollRegisterWithCacheManager(void)
 627 {
 628     Mgr::RegisterAction("comm_poll_incoming",
 629                         "comm_incoming() stats",
 630                         commIncomingStats, 0, 1);
 631 }
 632
 633 void
 634 Comm::SelectLoopInit(void)
 635 {
 636     commPollRegisterWithCacheManager();
 637 }
 638
 639 static void
 640 commIncomingStats(StoreEntry * sentry)
 641 {
 642     StatCounters *f = &statCounter;
 643     storeAppendPrintf(sentry, "Current incoming_icp_interval: %d\n",
 644                       incoming_icp_interval >> INCOMING_FACTOR);
 645     storeAppendPrintf(sentry, "Current incoming_dns_interval: %d\n",
 646                       incoming_dns_interval >> INCOMING_FACTOR);
 647     storeAppendPrintf(sentry, "Current incoming_http_interval: %d\n",
 648                       incoming_http_interval >> INCOMING_FACTOR);
 649     storeAppendPrintf(sentry, "\n");
 650     storeAppendPrintf(sentry, "Histogram of events per incoming socket type\n");
 651     storeAppendPrintf(sentry, "ICP Messages handled per comm_poll_icp_incoming() call:\n");
 652     statHistDump(&f->comm_icp_incoming, sentry, statHistIntDumper);
 653     storeAppendPrintf(sentry, "DNS Messages handled per comm_poll_dns_incoming() call:\n");
 654     statHistDump(&f->comm_dns_incoming, sentry, statHistIntDumper);
 655     storeAppendPrintf(sentry, "HTTP Messages handled per comm_poll_http_incoming() call:\n");
 656     statHistDump(&f->comm_http_incoming, sentry, statHistIntDumper);
 657 }
 658
 659 /* Called by async-io or diskd to speed up the polling */
 660 void
 661 Comm::QuickPollRequired(void)
 662 {
 663     MAX_POLL_TIME = 10;
 664 }
 665
 666 #endif /* USE_POLL */