src/store_dir.cc

   1
   2 /*
   3  * $Id$
   4  *
   5  * DEBUG: section 47    Store Directory Routines
   6  * AUTHOR: Duane Wessels
   7  *
   8  * SQUID Web Proxy Cache          http://www.squid-cache.org/
   9  * ----------------------------------------------------------
  10  *
  11  *  Squid is the result of efforts by numerous individuals from
  12  *  the Internet community; see the CONTRIBUTORS file for full
  13  *  details.   Many organizations have provided support for Squid's
  14  *  development; see the SPONSORS file for full details.  Squid is
  15  *  Copyrighted (C) 2001 by the Regents of the University of
  16  *  California; see the COPYRIGHT file for full details.  Squid
  17  *  incorporates software developed and/or copyrighted by other
  18  *  sources; see the CREDITS file for full details.
  19  *
  20  *  This program is free software; you can redistribute it and/or modify
  21  *  it under the terms of the GNU General Public License as published by
  22  *  the Free Software Foundation; either version 2 of the License, or
  23  *  (at your option) any later version.
  24  *
  25  *  This program is distributed in the hope that it will be useful,
  26  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  27  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  28  *  GNU General Public License for more details.
  29  *
  30  *  You should have received a copy of the GNU General Public License
  31  *  along with this program; if not, write to the Free Software
  32  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
  33  *
  34  */
  35
  36 #include "squid.h"
  37 #include "Store.h"
  38 #include "MemObject.h"
  39 #include "MemStore.h"
  40 #include "mem_node.h"
  41 #include "SquidMath.h"
  42 #include "SquidTime.h"
  43 #include "SwapDir.h"
  44 #include "swap_log_op.h"
  45
  46 #if HAVE_STATVFS
  47 #if HAVE_SYS_STATVFS_H
  48 #include <sys/statvfs.h>
  49 #endif
  50 #endif /* HAVE_STATVFS */
  51 /* statfs() needs <sys/param.h> and <sys/mount.h> on BSD systems */
  52 #if HAVE_SYS_PARAM_H
  53 #include <sys/param.h>
  54 #endif
  55 #if HAVE_SYS_MOUNT_H
  56 #include <sys/mount.h>
  57 #endif
  58 /* Windows and Linux use sys/vfs.h */
  59 #if HAVE_SYS_VFS_H
  60 #include <sys/vfs.h>
  61 #endif
  62
  63 #include "StoreHashIndex.h"
  64
  65 static STDIRSELECT storeDirSelectSwapDirRoundRobin;
  66 static STDIRSELECT storeDirSelectSwapDirLeastLoad;
  67
  68 /*
  69  * store_dirs_rebuilding is initialized to _1_ as a hack so that
  70  * storeDirWriteCleanLogs() doesn't try to do anything unless _all_
  71  * cache_dirs have been read.  For example, without this hack, Squid
  72  * will try to write clean log files if -kparse fails (becasue it
  73  * calls fatal()).
  74  */
  75 int StoreController::store_dirs_rebuilding = 1;
  76
  77 StoreController::StoreController() : swapDir (new StoreHashIndex())
  78     , memStore(NULL)
  79 {}
  80
  81 StoreController::~StoreController()
  82 {
  83     delete memStore;
  84 }
  85
  86 /*
  87  * This function pointer is set according to 'store_dir_select_algorithm'
  88  * in squid.conf.
  89  */
  90 STDIRSELECT *storeDirSelectSwapDir = storeDirSelectSwapDirLeastLoad;
  91
  92 void
  93 StoreController::init()
  94 {
  95     if (UsingSmp() && IamWorkerProcess()) {
  96         memStore = new MemStore;
  97         memStore->init();
  98     }
  99
 100     swapDir->init();
 101
 102     if (0 == strcasecmp(Config.store_dir_select_algorithm, "round-robin")) {
 103         storeDirSelectSwapDir = storeDirSelectSwapDirRoundRobin;
 104         debugs(47, 1, "Using Round Robin store dir selection");
 105     } else {
 106         storeDirSelectSwapDir = storeDirSelectSwapDirLeastLoad;
 107         debugs(47, 1, "Using Least Load store dir selection");
 108     }
 109 }
 110
 111 void
 112 StoreController::createOneStore(Store &aStore)
 113 {
 114     /*
 115      * On Windows, fork() is not available.
 116      * The following is a workaround for create store directories sequentially
 117      * when running on native Windows port.
 118      */
 119 #ifndef _SQUID_MSWIN_
 120
 121     if (fork())
 122         return;
 123
 124 #endif
 125
 126     aStore.create();
 127
 128 #ifndef _SQUID_MSWIN_
 129
 130     exit(0);
 131
 132 #endif
 133 }
 134
 135 void
 136 StoreController::create()
 137 {
 138     swapDir->create();
 139
 140 #ifndef _SQUID_MSWIN_
 141
 142     pid_t pid;
 143
 144     do {
 145         int status;
 146 #ifdef _SQUID_NEXT_
 147
 148         pid = wait3(&status, WNOHANG, NULL);
 149 #else
 150
 151         pid = waitpid(-1, &status, 0);
 152 #endif
 153
 154     } while (pid > 0 || (pid < 0 && errno == EINTR));
 155
 156 #endif
 157 }
 158
 159 /**
 160  * Determine whether the given directory can handle this object
 161  * size
 162  *
 163  * Note: if the object size is -1, then the only swapdirs that
 164  * will return true here are ones that have min and max unset,
 165  * ie any-sized-object swapdirs. This is a good thing.
 166  */
 167 bool
 168 SwapDir::objectSizeIsAcceptable(int64_t objsize) const
 169 {
 170     // If the swapdir has no range limits, then it definitely can
 171     if (min_objsize <= 0 && max_objsize == -1)
 172         return true;
 173
 174     /*
 175      * If the object size is -1 and the storedir has limits we
 176      * can't store it there.
 177      */
 178     if (objsize == -1)
 179         return false;
 180
 181     // Else, make sure that the object size will fit.
 182     return min_objsize <= objsize && max_objsize > objsize;
 183 }
 184
 185
 186 /*
 187  * This new selection scheme simply does round-robin on all SwapDirs.
 188  * A SwapDir is skipped if it is over the max_size (100%) limit, or
 189  * overloaded.
 190  */
 191 static int
 192 storeDirSelectSwapDirRoundRobin(const StoreEntry * e)
 193 {
 194     static int dirn = 0;
 195     int i;
 196     int load;
 197     RefCount<SwapDir> sd;
 198
 199     // e->objectLen() is negative at this point when we are still STORE_PENDING
 200     ssize_t objsize = e->mem_obj->expectedReplySize();
 201     if (objsize != -1)
 202         objsize += e->mem_obj->swap_hdr_sz;
 203
 204     for (i = 0; i < Config.cacheSwap.n_configured; i++) {
 205         if (++dirn >= Config.cacheSwap.n_configured)
 206             dirn = 0;
 207
 208         sd = dynamic_cast<SwapDir *>(INDEXSD(dirn));
 209
 210         if (!sd->canStore(*e, objsize, load))
 211             continue;
 212
 213         if (load < 0 || load > 1000) {
 214             continue;
 215         }
 216
 217         return dirn;
 218     }
 219
 220     return -1;
 221 }
 222
 223 /*
 224  * Spread load across all of the store directories
 225  *
 226  * Note: We should modify this later on to prefer sticking objects
 227  * in the *tightest fit* swapdir to conserve space, along with the
 228  * actual swapdir usage. But for now, this hack will do while
 229  * testing, so you should order your swapdirs in the config file
 230  * from smallest maxobjsize to unlimited (-1) maxobjsize.
 231  *
 232  * We also have to choose nleast == nconf since we need to consider
 233  * ALL swapdirs, regardless of state. Again, this is a hack while
 234  * we sort out the real usefulness of this algorithm.
 235  */
 236 static int
 237 storeDirSelectSwapDirLeastLoad(const StoreEntry * e)
 238 {
 239     int64_t most_free = 0;
 240     ssize_t least_objsize = -1;
 241     int least_load = INT_MAX;
 242     int load;
 243     int dirn = -1;
 244     int i;
 245     RefCount<SwapDir> SD;
 246
 247     // e->objectLen() is negative at this point when we are still STORE_PENDING
 248     ssize_t objsize = e->mem_obj->expectedReplySize();
 249
 250     if (objsize != -1)
 251         objsize += e->mem_obj->swap_hdr_sz;
 252
 253     for (i = 0; i < Config.cacheSwap.n_configured; i++) {
 254         SD = dynamic_cast<SwapDir *>(INDEXSD(i));
 255         SD->flags.selected = 0;
 256
 257         if (!SD->canStore(*e, objsize, load))
 258             continue;
 259
 260         if (load < 0 || load > 1000)
 261             continue;
 262
 263         if (load > least_load)
 264             continue;
 265
 266         const int64_t cur_free = SD->maxSize() - SD->currentSize();
 267
 268         /* If the load is equal, then look in more details */
 269         if (load == least_load) {
 270             /* closest max_objsize fit */
 271
 272             if (least_objsize != -1)
 273                 if (SD->max_objsize > least_objsize || SD->max_objsize == -1)
 274                     continue;
 275
 276             /* most free */
 277             if (cur_free < most_free)
 278                 continue;
 279         }
 280
 281         least_load = load;
 282         least_objsize = SD->max_objsize;
 283         most_free = cur_free;
 284         dirn = i;
 285     }
 286
 287     if (dirn >= 0)
 288         dynamic_cast<SwapDir *>(INDEXSD(dirn))->flags.selected = 1;
 289
 290     return dirn;
 291 }
 292
 293 /*
 294  * An entry written to the swap log MUST have the following
 295  * properties.
 296  *   1.  It MUST be a public key.  It does no good to log
 297  *       a public ADD, change the key, then log a private
 298  *       DEL.  So we need to log a DEL before we change a
 299  *       key from public to private.
 300  *   2.  It MUST have a valid (> -1) swap_filen.
 301  */
 302 void
 303 storeDirSwapLog(const StoreEntry * e, int op)
 304 {
 305     assert (e);
 306     assert(!EBIT_TEST(e->flags, KEY_PRIVATE));
 307     assert(e->swap_filen >= 0);
 308     /*
 309      * icons and such; don't write them to the swap log
 310      */
 311
 312     if (EBIT_TEST(e->flags, ENTRY_SPECIAL))
 313         return;
 314
 315     assert(op > SWAP_LOG_NOP && op < SWAP_LOG_MAX);
 316
 317     debugs(20, 3, "storeDirSwapLog: " <<
 318            swap_log_op_str[op] << " " <<
 319            e->getMD5Text() << " " <<
 320            e->swap_dirn << " " <<
 321            std::hex << std::uppercase << std::setfill('0') << std::setw(8) << e->swap_filen);
 322
 323     dynamic_cast<SwapDir *>(INDEXSD(e->swap_dirn))->logEntry(*e, op);
 324 }
 325
 326 void
 327 StoreController::stat(StoreEntry &output) const
 328 {
 329     storeAppendPrintf(&output, "Store Directory Statistics:\n");
 330     storeAppendPrintf(&output, "Store Entries          : %lu\n",
 331                       (unsigned long int)StoreEntry::inUseCount());
 332     storeAppendPrintf(&output, "Maximum Swap Size      : %"PRIu64" KB\n",
 333                       maxSize() >> 10);
 334     storeAppendPrintf(&output, "Current Store Swap Size: %.2f KB\n",
 335                       currentSize() / 1024.0);
 336     storeAppendPrintf(&output, "Current Capacity       : %.2f%% used, %.2f%% free\n",
 337                       Math::doublePercent(currentSize(), maxSize()),
 338                       Math::doublePercent((maxSize() - currentSize()), maxSize()));
 339
 340     if (memStore)
 341         memStore->stat(output);
 342
 343     /* now the swapDir */
 344     swapDir->stat(output);
 345 }
 346
 347 /* if needed, this could be taught to cache the result */
 348 uint64_t
 349 StoreController::maxSize() const
 350 {
 351     /* TODO: include memory cache ? */
 352     return swapDir->maxSize();
 353 }
 354
 355 uint64_t
 356 StoreController::minSize() const
 357 {
 358     /* TODO: include memory cache ? */
 359     return swapDir->minSize();
 360 }
 361
 362 uint64_t
 363 StoreController::currentSize() const
 364 {
 365     return swapDir->currentSize();
 366 }
 367
 368 uint64_t
 369 StoreController::currentCount() const
 370 {
 371     return swapDir->currentCount();
 372 }
 373
 374 int64_t
 375 StoreController::maxObjectSize() const
 376 {
 377     return swapDir->maxObjectSize();
 378 }
 379
 380 void
 381 SwapDir::diskFull()
 382 {
 383     if (currentSize() >= maxSize())
 384         return;
 385
 386     max_size = currentSize();
 387
 388     debugs(20, 1, "WARNING: Shrinking cache_dir #" << index << " to " << currentSize() / 1024.0 << " KB");
 389 }
 390
 391 void
 392 storeDirOpenSwapLogs(void)
 393 {
 394     for (int dirn = 0; dirn < Config.cacheSwap.n_configured; ++dirn)
 395         dynamic_cast<SwapDir *>(INDEXSD(dirn))->openLog();
 396 }
 397
 398 void
 399 storeDirCloseSwapLogs(void)
 400 {
 401     for (int dirn = 0; dirn < Config.cacheSwap.n_configured; ++dirn)
 402         dynamic_cast<SwapDir *>(INDEXSD(dirn))->closeLog();
 403 }
 404
 405 /*
 406  *  storeDirWriteCleanLogs
 407  *
 408  *  Writes a "clean" swap log file from in-memory metadata.
 409  *  This is a rewrite of the original function to troll each
 410  *  StoreDir and write the logs, and flush at the end of
 411  *  the run. Thanks goes to Eric Stern, since this solution
 412  *  came out of his COSS code.
 413  */
 414 int
 415 storeDirWriteCleanLogs(int reopen)
 416 {
 417     const StoreEntry *e = NULL;
 418     int n = 0;
 419
 420     struct timeval start;
 421     double dt;
 422     RefCount<SwapDir> sd;
 423     int dirn;
 424     int notdone = 1;
 425
 426     if (StoreController::store_dirs_rebuilding) {
 427         debugs(20, 1, "Not currently OK to rewrite swap log.");
 428         debugs(20, 1, "storeDirWriteCleanLogs: Operation aborted.");
 429         return 0;
 430     }
 431
 432     debugs(20, 1, "storeDirWriteCleanLogs: Starting...");
 433     getCurrentTime();
 434     start = current_time;
 435
 436     for (dirn = 0; dirn < Config.cacheSwap.n_configured; dirn++) {
 437         sd = dynamic_cast<SwapDir *>(INDEXSD(dirn));
 438
 439         if (sd->writeCleanStart() < 0) {
 440             debugs(20, 1, "log.clean.start() failed for dir #" << sd->index);
 441             continue;
 442         }
 443     }
 444
 445     /*
 446      * This may look inefficient as CPU wise it is more efficient to do this
 447      * sequentially, but I/O wise the parallellism helps as it allows more
 448      * hdd spindles to be active.
 449      */
 450     while (notdone) {
 451         notdone = 0;
 452
 453         for (dirn = 0; dirn < Config.cacheSwap.n_configured; dirn++) {
 454             sd = dynamic_cast<SwapDir *>(INDEXSD(dirn));
 455
 456             if (NULL == sd->cleanLog)
 457                 continue;
 458
 459             e = sd->cleanLog->nextEntry();
 460
 461             if (!e)
 462                 continue;
 463
 464             notdone = 1;
 465
 466             if (!sd->canLog(*e))
 467                 continue;
 468
 469             sd->cleanLog->write(*e);
 470
 471             if ((++n & 0xFFFF) == 0) {
 472                 getCurrentTime();
 473                 debugs(20, 1, "  " << std::setw(7) << n  <<
 474                        " entries written so far.");
 475             }
 476         }
 477     }
 478
 479     /* Flush */
 480     for (dirn = 0; dirn < Config.cacheSwap.n_configured; dirn++)
 481         dynamic_cast<SwapDir *>(INDEXSD(dirn))->writeCleanDone();
 482
 483     if (reopen)
 484         storeDirOpenSwapLogs();
 485
 486     getCurrentTime();
 487
 488     dt = tvSubDsec(start, current_time);
 489
 490     debugs(20, 1, "  Finished.  Wrote " << n << " entries.");
 491     debugs(20, 1, "  Took "<< std::setw(3)<< std::setprecision(2) << dt <<
 492            " seconds ("<< std::setw(6) << ((double) n / (dt > 0.0 ? dt : 1.0)) << " entries/sec).");
 493
 494
 495     return n;
 496 }
 497
 498 StoreSearch *
 499 StoreController::search(String const url, HttpRequest *request)
 500 {
 501     /* cheat, for now you can't search the memory hot cache */
 502     return swapDir->search(url, request);
 503 }
 504
 505 StorePointer
 506 StoreHashIndex::store(int const x) const
 507 {
 508     return INDEXSD(x);
 509 }
 510
 511 SwapDir &
 512 StoreHashIndex::dir(const int i) const
 513 {
 514     SwapDir *sd = dynamic_cast<SwapDir*>(INDEXSD(i));
 515     assert(sd);
 516     return *sd;
 517 }
 518
 519 void
 520 StoreController::sync(void)
 521 {
 522     if (memStore)
 523         memStore->sync();
 524     swapDir->sync();
 525 }
 526
 527 /*
 528  * handle callbacks all avaliable fs'es
 529  */
 530 int
 531 StoreController::callback()
 532 {
 533     /* This will likely double count. Thats ok. */
 534     PROF_start(storeDirCallback);
 535
 536     /* mem cache callbacks ? */
 537     int result = swapDir->callback();
 538
 539     PROF_stop(storeDirCallback);
 540
 541     return result;
 542 }
 543
 544 int
 545 storeDirGetBlkSize(const char *path, int *blksize)
 546 {
 547 #if HAVE_STATVFS
 548
 549     struct statvfs sfs;
 550
 551     if (statvfs(path, &sfs)) {
 552         debugs(50, 1, "" << path << ": " << xstrerror());
 553         *blksize = 2048;
 554         return 1;
 555     }
 556
 557     *blksize = (int) sfs.f_frsize;
 558 #else
 559
 560     struct statfs sfs;
 561
 562     if (statfs(path, &sfs)) {
 563         debugs(50, 1, "" << path << ": " << xstrerror());
 564         *blksize = 2048;
 565         return 1;
 566     }
 567
 568     *blksize = (int) sfs.f_bsize;
 569 #endif
 570     /*
 571      * Sanity check; make sure we have a meaningful value.
 572      */
 573
 574     if (*blksize < 512)
 575         *blksize = 2048;
 576
 577     return 0;
 578 }
 579
 580 #define fsbtoblk(num, fsbs, bs) \
 581     (((fsbs) != 0 && (fsbs) < (bs)) ? \
 582             (num) / ((bs) / (fsbs)) : (num) * ((fsbs) / (bs)))
 583 int
 584 storeDirGetUFSStats(const char *path, int *totl_kb, int *free_kb, int *totl_in, int *free_in)
 585 {
 586 #if HAVE_STATVFS
 587
 588     struct statvfs sfs;
 589
 590     if (statvfs(path, &sfs)) {
 591         debugs(50, 1, "" << path << ": " << xstrerror());
 592         return 1;
 593     }
 594
 595     *totl_kb = (int) fsbtoblk(sfs.f_blocks, sfs.f_frsize, 1024);
 596     *free_kb = (int) fsbtoblk(sfs.f_bfree, sfs.f_frsize, 1024);
 597     *totl_in = (int) sfs.f_files;
 598     *free_in = (int) sfs.f_ffree;
 599 #else
 600
 601     struct statfs sfs;
 602
 603     if (statfs(path, &sfs)) {
 604         debugs(50, 1, "" << path << ": " << xstrerror());
 605         return 1;
 606     }
 607
 608     *totl_kb = (int) fsbtoblk(sfs.f_blocks, sfs.f_bsize, 1024);
 609     *free_kb = (int) fsbtoblk(sfs.f_bfree, sfs.f_bsize, 1024);
 610     *totl_in = (int) sfs.f_files;
 611     *free_in = (int) sfs.f_ffree;
 612 #endif
 613
 614     return 0;
 615 }
 616
 617 void
 618 allocate_new_swapdir(SquidConfig::_cacheSwap * swap)
 619 {
 620     if (swap->swapDirs == NULL) {
 621         swap->n_allocated = 4;
 622         swap->swapDirs = static_cast<SwapDir::Pointer *>(xcalloc(swap->n_allocated, sizeof(SwapDir::Pointer)));
 623     }
 624
 625     if (swap->n_allocated == swap->n_configured) {
 626         swap->n_allocated <<= 1;
 627         SwapDir::Pointer *const tmp = static_cast<SwapDir::Pointer *>(xcalloc(swap->n_allocated, sizeof(SwapDir::Pointer)));
 628         memcpy(tmp, swap->swapDirs, swap->n_configured * sizeof(SwapDir *));
 629         xfree(swap->swapDirs);
 630         swap->swapDirs = tmp;
 631     }
 632 }
 633
 634 void
 635 free_cachedir(SquidConfig::_cacheSwap * swap)
 636 {
 637     int i;
 638     /* DON'T FREE THESE FOR RECONFIGURE */
 639
 640     if (reconfiguring)
 641         return;
 642
 643     for (i = 0; i < swap->n_configured; i++) {
 644         /* TODO XXX this lets the swapdir free resources asynchronously
 645         * swap->swapDirs[i]->deactivate();
 646         * but there may be such a means already.
 647         * RBC 20041225
 648         */
 649         swap->swapDirs[i] = NULL;
 650     }
 651
 652     safe_free(swap->swapDirs);
 653     swap->swapDirs = NULL;
 654     swap->n_allocated = 0;
 655     swap->n_configured = 0;
 656 }
 657
 658 /* this should be a virtual method on StoreEntry,
 659  * i.e. e->referenced()
 660  * so that the entry can notify the creating Store
 661  */
 662 void
 663 StoreController::reference(StoreEntry &e)
 664 {
 665     /* Notify the fs that we're referencing this object again */
 666
 667     if (e.swap_dirn > -1)
 668         e.store()->reference(e);
 669
 670     // Notify the memory cache that we're referencing this object again
 671     if (memStore && e.mem_status == IN_MEMORY)
 672         memStore->reference(e);
 673
 674     // TODO: move this code to a non-shared memory cache class when we have it
 675     if (e.mem_obj) {
 676         if (mem_policy->Referenced)
 677             mem_policy->Referenced(mem_policy, &e, &e.mem_obj->repl);
 678     }
 679 }
 680
 681 void
 682 StoreController::dereference(StoreEntry & e)
 683 {
 684     /* Notify the fs that we're not referencing this object any more */
 685
 686     if (e.swap_filen > -1)
 687         e.store()->dereference(e);
 688
 689     // Notify the memory cache that we're not referencing this object any more
 690     if (memStore && e.mem_status == IN_MEMORY)
 691         memStore->dereference(e);
 692
 693     // TODO: move this code to a non-shared memory cache class when we have it
 694     if (e.mem_obj) {
 695         if (mem_policy->Dereferenced)
 696             mem_policy->Dereferenced(mem_policy, &e, &e.mem_obj->repl);
 697     }
 698 }
 699
 700 StoreEntry *
 701 StoreController::get(const cache_key *key)
 702 {
 703     if (StoreEntry *e = swapDir->get(key)) {
 704         // TODO: ignore and maybe handleIdleEntry() unlocked intransit entries
 705         // because their backing store slot may be gone already.
 706         debugs(20, 3, HERE << "got in-transit entry: " << *e);
 707         return e;
 708     }
 709
 710     if (memStore) {
 711         if (StoreEntry *e = memStore->get(key)) {
 712             debugs(20, 3, HERE << "got mem-cached entry: " << *e);
 713             return e;
 714         }
 715     }
 716
 717     // TODO: this disk iteration is misplaced; move to StoreHashIndex
 718     if (const int cacheDirs = Config.cacheSwap.n_configured) {
 719         // ask each cache_dir until the entry is found; use static starting
 720         // point to avoid asking the same subset of disks more often
 721         // TODO: coordinate with put() to be able to guess the right disk often
 722         static int idx = 0;
 723         for (int n = 0; n < cacheDirs; ++n) {
 724             idx = (idx + 1) % cacheDirs;
 725             SwapDir *sd = dynamic_cast<SwapDir*>(INDEXSD(idx));
 726             if (!sd->active())
 727                 continue;
 728
 729             if (StoreEntry *e = sd->get(key)) {
 730                 debugs(20, 3, HERE << "cache_dir " << idx <<
 731                     " got cached entry: " << *e);
 732                 return e;
 733             }
 734         }
 735     }
 736
 737     debugs(20, 4, HERE << "none of " << Config.cacheSwap.n_configured <<
 738         " cache_dirs have " << storeKeyText(key));
 739     return NULL;
 740 }
 741
 742 void
 743 StoreController::get(String const key, STOREGETCLIENT aCallback, void *aCallbackData)
 744 {
 745     fatal("not implemented");
 746 }
 747
 748 void
 749 StoreController::handleIdleEntry(StoreEntry &e)
 750 {
 751     bool keepInLocalMemory = false;
 752     if (memStore) {
 753         memStore->considerKeeping(e);
 754         // leave keepInLocalMemory false; memStore maintains its own cache
 755     } else {
 756         keepInLocalMemory = e.memoryCachable() && // entry is in good shape and
 757             // the local memory cache is not overflowing
 758             (mem_node::InUseCount() <= store_pages_max);
 759     }
 760
 761     dereference(e);
 762
 763     // XXX: Rock store specific: Since each SwapDir controls its index,
 764     // unlocked entries should not stay in the global store_table.
 765     if (fileno >= 0) {
 766         debugs(20, 5, HERE << "destroying unlocked entry: " << &e << ' ' << e);
 767         destroyStoreEntry(static_cast<hash_link*>(&e));
 768         return;
 769     }
 770
 771     // TODO: move this into [non-shared] memory cache class when we have one
 772     if (keepInLocalMemory) {
 773         e.setMemStatus(IN_MEMORY);
 774         e.mem_obj->unlinkRequest();
 775     } else {
 776         e.purgeMem(); // may free e
 777     }
 778 }
 779
 780 StoreHashIndex::StoreHashIndex()
 781 {
 782     if (store_table)
 783         abort();
 784     assert (store_table == NULL);
 785 }
 786
 787 StoreHashIndex::~StoreHashIndex()
 788 {
 789     if (store_table) {
 790         hashFreeItems(store_table, destroyStoreEntry);
 791         hashFreeMemory(store_table);
 792         store_table = NULL;
 793     }
 794 }
 795
 796 int
 797 StoreHashIndex::callback()
 798 {
 799     int result = 0;
 800     int j;
 801     static int ndir = 0;
 802
 803     do {
 804         j = 0;
 805
 806         for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 807             if (ndir >= Config.cacheSwap.n_configured)
 808                 ndir = ndir % Config.cacheSwap.n_configured;
 809
 810             int temp_result = store(ndir)->callback();
 811
 812             ++ndir;
 813
 814             j += temp_result;
 815
 816             result += temp_result;
 817
 818             if (j > 100)
 819                 fatal ("too much io\n");
 820         }
 821     } while (j > 0);
 822
 823     ndir++;
 824
 825     return result;
 826 }
 827
 828 void
 829 StoreHashIndex::create()
 830 {
 831     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 832         if (dir(i).active())
 833             store(i)->create();
 834     }
 835 }
 836
 837 /* Lookup an object in the cache.
 838  * return just a reference to object, don't start swapping in yet. */
 839 StoreEntry *
 840 StoreHashIndex::get(const cache_key *key)
 841 {
 842     PROF_start(storeGet);
 843     debugs(20, 3, "storeGet: looking up " << storeKeyText(key));
 844     StoreEntry *p = static_cast<StoreEntry *>(hash_lookup(store_table, key));
 845     PROF_stop(storeGet);
 846     return p;
 847 }
 848
 849 void
 850 StoreHashIndex::get(String const key, STOREGETCLIENT aCallback, void *aCallbackData)
 851 {
 852     fatal("not implemented");
 853 }
 854
 855 void
 856 StoreHashIndex::init()
 857 {
 858     /* Calculate size of hash table (maximum currently 64k buckets).  */
 859     /* this is very bogus, its specific to the any Store maintaining an
 860      * in-core index, not global */
 861     size_t buckets = (Store::Root().maxSize() + Config.memMaxSize) / Config.Store.avgObjectSize;
 862     debugs(20, 1, "Swap maxSize " << (Store::Root().maxSize() >> 10) <<
 863            " + " << ( Config.memMaxSize >> 10) << " KB, estimated " << buckets << " objects");
 864     buckets /= Config.Store.objectsPerBucket;
 865     debugs(20, 1, "Target number of buckets: " << buckets);
 866     /* ideally the full scan period should be configurable, for the
 867      * moment it remains at approximately 24 hours.  */
 868     store_hash_buckets = storeKeyHashBuckets(buckets);
 869     debugs(20, 1, "Using " << store_hash_buckets << " Store buckets");
 870     debugs(20, 1, "Max Mem  size: " << ( Config.memMaxSize >> 10) << " KB");
 871     debugs(20, 1, "Max Swap size: " << (Store::Root().maxSize() >> 10) << " KB");
 872
 873     store_table = hash_create(storeKeyHashCmp,
 874                               store_hash_buckets, storeKeyHashHash);
 875
 876     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 877         /* this starts a search of the store dirs, loading their
 878          * index. under the new Store api this should be
 879          * driven by the StoreHashIndex, not by each store.
 880         *
 881         * That is, the HashIndex should perform a search of each dir it is
 882         * indexing to do the hash insertions. The search is then able to
 883         * decide 'from-memory', or 'from-clean-log' or 'from-dirty-log' or
 884         * 'from-no-log'.
 885         *
 886          * Step 1: make the store rebuilds use a search internally
 887         * Step 2: change the search logic to use the four modes described
 888         *         above
 889         * Step 3: have the hash index walk the searches itself.
 890          */
 891         if (dir(i).active())
 892             store(i)->init();
 893     }
 894 }
 895
 896 uint64_t
 897 StoreHashIndex::maxSize() const
 898 {
 899     uint64_t result = 0;
 900
 901     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 902         if (dir(i).doReportStat())
 903             result += store(i)->maxSize();
 904     }
 905
 906     return result;
 907 }
 908
 909 uint64_t
 910 StoreHashIndex::minSize() const
 911 {
 912     uint64_t result = 0;
 913
 914     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 915         if (dir(i).doReportStat())
 916             result += store(i)->minSize();
 917     }
 918
 919     return result;
 920 }
 921
 922 uint64_t
 923 StoreHashIndex::currentSize() const
 924 {
 925     uint64_t result = 0;
 926
 927     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 928         if (dir(i).doReportStat())
 929             result += store(i)->currentSize();
 930     }
 931
 932     return result;
 933 }
 934
 935 uint64_t
 936 StoreHashIndex::currentCount() const
 937 {
 938     uint64_t result = 0;
 939
 940     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 941         if (dir(i).doReportStat())
 942             result += store(i)->currentCount();
 943     }
 944
 945     return result;
 946 }
 947
 948 int64_t
 949 StoreHashIndex::maxObjectSize() const
 950 {
 951     int64_t result = -1;
 952
 953     for (int i = 0; i < Config.cacheSwap.n_configured; i++) {
 954         if (dir(i).active() && store(i)->maxObjectSize() > result)
 955             result = store(i)->maxObjectSize();
 956     }
 957
 958     return result;
 959 }
 960
 961 void
 962 StoreHashIndex::stat(StoreEntry & output) const
 963 {
 964     int i;
 965
 966     /* Now go through each store, calling its stat routine */
 967
 968     for (i = 0; i < Config.cacheSwap.n_configured; i++) {
 969         storeAppendPrintf(&output, "\n");
 970         store(i)->stat(output);
 971     }
 972 }
 973
 974 void
 975 StoreHashIndex::reference(StoreEntry&)
 976 {}
 977
 978 void
 979 StoreHashIndex::dereference(StoreEntry&)
 980 {}
 981
 982 void
 983 StoreHashIndex::maintain()
 984 {
 985     int i;
 986     /* walk each fs */
 987
 988     for (i = 0; i < Config.cacheSwap.n_configured; i++) {
 989         /* XXX FixMe: This should be done "in parallell" on the different
 990          * cache_dirs, not one at a time.
 991          */
 992         /* call the maintain function .. */
 993         store(i)->maintain();
 994     }
 995 }
 996
 997 void
 998 StoreHashIndex::sync()
 999 {
1000     for (int i = 0; i < Config.cacheSwap.n_configured; ++i)
1001         store(i)->sync();
1002 }
1003
1004 StoreSearch *
1005 StoreHashIndex::search(String const url, HttpRequest *)
1006 {
1007     if (url.size())
1008         fatal ("Cannot search by url yet\n");
1009
1010     return new StoreSearchHashIndex (this);
1011 }
1012
1013 CBDATA_CLASS_INIT(StoreSearchHashIndex);
1014
1015 StoreSearchHashIndex::StoreSearchHashIndex(RefCount<StoreHashIndex> aSwapDir) : sd(aSwapDir), _done (false), bucket (0)
1016 {}
1017
1018 /* do not link
1019 StoreSearchHashIndex::StoreSearchHashIndex(StoreSearchHashIndex const &);
1020 */
1021
1022 StoreSearchHashIndex::~StoreSearchHashIndex()
1023 {}
1024
1025 void
1026 StoreSearchHashIndex::next(void (aCallback)(void *), void *aCallbackData)
1027 {
1028     next();
1029     aCallback (aCallbackData);
1030 }
1031
1032 bool
1033 StoreSearchHashIndex::next()
1034 {
1035     if (entries.size())
1036         entries.pop_back();
1037
1038     while (!isDone() && !entries.size())
1039         copyBucket();
1040
1041     return currentItem() != NULL;
1042 }
1043
1044 bool
1045 StoreSearchHashIndex::error() const
1046 {
1047     return false;
1048 }
1049
1050 bool
1051 StoreSearchHashIndex::isDone() const
1052 {
1053     return bucket >= store_hash_buckets || _done;
1054 }
1055
1056 StoreEntry *
1057 StoreSearchHashIndex::currentItem()
1058 {
1059     if (!entries.size())
1060         return NULL;
1061
1062     return entries.back();
1063 }
1064
1065 void
1066 StoreSearchHashIndex::copyBucket()
1067 {
1068     /* probably need to lock the store entries...
1069      * we copy them all to prevent races on the links. */
1070     debugs(47, 3, "StoreSearchHashIndex::copyBucket #" << bucket);
1071     assert (!entries.size());
1072     hash_link *link_ptr = NULL;
1073     hash_link *link_next = NULL;
1074     link_next = hash_get_bucket(store_table, bucket);
1075
1076     while (NULL != (link_ptr = link_next)) {
1077         link_next = link_ptr->next;
1078         StoreEntry *e = (StoreEntry *) link_ptr;
1079
1080         entries.push_back(e);
1081     }
1082
1083     bucket++;
1084     debugs(47,3, "got entries: " << entries.size());
1085 }