sys/vfs/hammer/hammer_ondisk.c

   1 /*
   2  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_ondisk.c,v 1.8 2007/11/27 07:48:52 dillon Exp $
  35  */
  36 /*
  37  * Manage HAMMER's on-disk structures.  These routines are primarily
  38  * responsible for interfacing with the kernel's I/O subsystem and for
  39  * managing in-memory structures.
  40  */
  41
  42 #include "hammer.h"
  43 #include <sys/fcntl.h>
  44 #include <sys/nlookup.h>
  45 #include <sys/buf.h>
  46 #include <sys/buf2.h>
  47
  48 static void hammer_free_volume(hammer_volume_t volume);
  49 static int hammer_load_volume(hammer_volume_t volume);
  50 static int hammer_load_supercl(hammer_supercl_t supercl, int isnew);
  51 static int hammer_load_cluster(hammer_cluster_t cluster, int isnew);
  52 static int hammer_load_buffer(hammer_buffer_t buffer, u_int64_t buf_type);
  53 static void hammer_remove_node_clist(hammer_buffer_t buffer,
  54                         hammer_node_t node);
  55 static void initbuffer(hammer_alist_t live, hammer_fsbuf_head_t head,
  56                         u_int64_t type);
  57 static void alloc_new_buffer(hammer_cluster_t cluster,
  58                         hammer_alist_t live, u_int64_t type, int32_t nelements,
  59                         int32_t start,
  60                         int *errorp, struct hammer_buffer **bufferp);
  61 #if 0
  62 static void readhammerbuf(hammer_volume_t vol, void *data,
  63                         int64_t offset);
  64 static void writehammerbuf(hammer_volume_t vol, const void *data,
  65                         int64_t offset);
  66 #endif
  67 static int64_t calculate_cluster_offset(hammer_volume_t vol, int32_t clu_no);
  68 static int64_t calculate_supercl_offset(hammer_volume_t vol, int32_t scl_no);
  69
  70 struct hammer_alist_config Buf_alist_config;
  71 struct hammer_alist_config Vol_normal_alist_config;
  72 struct hammer_alist_config Vol_super_alist_config;
  73 struct hammer_alist_config Supercl_alist_config;
  74 struct hammer_alist_config Clu_master_alist_config;
  75 struct hammer_alist_config Clu_slave_alist_config;
  76
  77 /*
  78  * Red-Black tree support for various structures
  79  */
  80 static int
  81 hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2)
  82 {
  83         if (ip1->obj_id < ip2->obj_id)
  84                 return(-1);
  85         if (ip1->obj_id > ip2->obj_id)
  86                 return(1);
  87         if (ip1->obj_asof < ip2->obj_asof)
  88                 return(-1);
  89         if (ip1->obj_asof > ip2->obj_asof)
  90                 return(1);
  91         return(0);
  92 }
  93
  94 static int
  95 hammer_inode_info_cmp(hammer_inode_info_t info, hammer_inode_t ip)
  96 {
  97         if (info->obj_id < ip->obj_id)
  98                 return(-1);
  99         if (info->obj_id > ip->obj_id)
 100                 return(1);
 101         if (info->obj_asof < ip->obj_asof)
 102                 return(-1);
 103         if (info->obj_asof > ip->obj_asof)
 104                 return(1);
 105         return(0);
 106 }
 107
 108 static int
 109 hammer_vol_rb_compare(hammer_volume_t vol1, hammer_volume_t vol2)
 110 {
 111         if (vol1->vol_no < vol2->vol_no)
 112                 return(-1);
 113         if (vol1->vol_no > vol2->vol_no)
 114                 return(1);
 115         return(0);
 116 }
 117
 118 static int
 119 hammer_scl_rb_compare(hammer_supercl_t cl1, hammer_supercl_t cl2)
 120 {
 121         if (cl1->scl_no < cl2->scl_no)
 122                 return(-1);
 123         if (cl1->scl_no > cl2->scl_no)
 124                 return(1);
 125         return(0);
 126 }
 127
 128 static int
 129 hammer_clu_rb_compare(hammer_cluster_t cl1, hammer_cluster_t cl2)
 130 {
 131         if (cl1->clu_no < cl2->clu_no)
 132                 return(-1);
 133         if (cl1->clu_no > cl2->clu_no)
 134                 return(1);
 135         return(0);
 136 }
 137
 138 static int
 139 hammer_buf_rb_compare(hammer_buffer_t buf1, hammer_buffer_t buf2)
 140 {
 141         if (buf1->buf_no < buf2->buf_no)
 142                 return(-1);
 143         if (buf1->buf_no > buf2->buf_no)
 144                 return(1);
 145         return(0);
 146 }
 147
 148 static int
 149 hammer_nod_rb_compare(hammer_node_t node1, hammer_node_t node2)
 150 {
 151         if (node1->node_offset < node2->node_offset)
 152                 return(-1);
 153         if (node1->node_offset > node2->node_offset)
 154                 return(1);
 155         return(0);
 156 }
 157
 158 /*
 159  * Note: The lookup function for hammer_ino_rb_tree winds up being named
 160  * hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).  The other lookup
 161  * functions are normal, e.g. hammer_clu_rb_tree_RB_LOOKUP(root, clu_no).
 162  */
 163 RB_GENERATE(hammer_ino_rb_tree, hammer_inode, rb_node, hammer_ino_rb_compare);
 164 RB_GENERATE_XLOOKUP(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
 165                 hammer_inode_info_cmp, hammer_inode_info_t);
 166 RB_GENERATE2(hammer_vol_rb_tree, hammer_volume, rb_node,
 167              hammer_vol_rb_compare, int32_t, vol_no);
 168 RB_GENERATE2(hammer_scl_rb_tree, hammer_supercl, rb_node,
 169              hammer_scl_rb_compare, int32_t, scl_no);
 170 RB_GENERATE2(hammer_clu_rb_tree, hammer_cluster, rb_node,
 171              hammer_clu_rb_compare, int32_t, clu_no);
 172 RB_GENERATE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
 173              hammer_buf_rb_compare, int32_t, buf_no);
 174 RB_GENERATE2(hammer_nod_rb_tree, hammer_node, rb_node,
 175              hammer_nod_rb_compare, int32_t, node_offset);
 176
 177 /************************************************************************
 178  *                              VOLUMES                                 *
 179  ************************************************************************
 180  *
 181  * Load a HAMMER volume by name.  Returns 0 on success or a positive error
 182  * code on failure.  Volumes must be loaded at mount time, get_volume() will
 183  * not load a new volume.
 184  *
 185  * Calls made to hammer_load_volume() or single-threaded
 186  */
 187 int
 188 hammer_install_volume(struct hammer_mount *hmp, const char *volname)
 189 {
 190         struct mount *mp;
 191         hammer_volume_t volume;
 192         struct hammer_volume_ondisk *ondisk;
 193         struct nlookupdata nd;
 194         struct buf *bp = NULL;
 195         int error;
 196         int ronly;
 197
 198         mp = hmp->mp;
 199         ronly = ((mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
 200
 201         /*
 202          * Allocate a volume structure
 203          */
 204         volume = kmalloc(sizeof(*volume), M_HAMMER, M_WAITOK|M_ZERO);
 205         volume->vol_name = kstrdup(volname, M_HAMMER);
 206         volume->hmp = hmp;
 207         volume->io.type = HAMMER_STRUCTURE_VOLUME;
 208         volume->io.offset = 0LL;
 209
 210         /*
 211          * Get the device vnode
 212          */
 213         error = nlookup_init(&nd, volume->vol_name, UIO_SYSSPACE, NLC_FOLLOW);
 214         if (error == 0)
 215                 error = nlookup(&nd);
 216         if (error == 0)
 217                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &volume->devvp);
 218         nlookup_done(&nd);
 219         if (error == 0) {
 220                 vn_isdisk(volume->devvp, &error);
 221         }
 222         if (error == 0) {
 223                 vn_lock(volume->devvp, LK_EXCLUSIVE | LK_RETRY);
 224                 error = VOP_OPEN(volume->devvp, (ronly ? FREAD : FREAD|FWRITE),
 225                                  FSCRED, NULL);
 226                 vn_unlock(volume->devvp);
 227         }
 228         if (error) {
 229                 hammer_free_volume(volume);
 230                 return(error);
 231         }
 232
 233         /*
 234          * Extract the volume number from the volume header and do various
 235          * sanity checks.
 236          */
 237         error = bread(volume->devvp, 0LL, HAMMER_BUFSIZE, &bp);
 238         if (error)
 239                 goto late_failure;
 240         ondisk = (void *)bp->b_data;
 241         if (ondisk->head.buf_type != HAMMER_FSBUF_VOLUME) {
 242                 kprintf("hammer_mount: volume %s has an invalid header\n",
 243                         volume->vol_name);
 244                 error = EFTYPE;
 245                 goto late_failure;
 246         }
 247         volume->vol_no = ondisk->vol_no;
 248         volume->cluster_base = ondisk->vol_clo_beg;
 249         volume->vol_clsize = ondisk->vol_clsize;
 250         volume->vol_flags = ondisk->vol_flags;
 251         RB_INIT(&volume->rb_clus_root);
 252         RB_INIT(&volume->rb_scls_root);
 253
 254         if (RB_EMPTY(&hmp->rb_vols_root)) {
 255                 hmp->fsid = ondisk->vol_fsid;
 256         } else if (bcmp(&hmp->fsid, &ondisk->vol_fsid, sizeof(uuid_t))) {
 257                 kprintf("hammer_mount: volume %s's fsid does not match "
 258                         "other volumes\n", volume->vol_name);
 259                 error = EFTYPE;
 260                 goto late_failure;
 261         }
 262
 263         /*
 264          * Insert the volume structure into the red-black tree.
 265          */
 266         if (RB_INSERT(hammer_vol_rb_tree, &hmp->rb_vols_root, volume)) {
 267                 kprintf("hammer_mount: volume %s has a duplicate vol_no %d\n",
 268                         volume->vol_name, volume->vol_no);
 269                 error = EEXIST;
 270         }
 271
 272         /*
 273          * Set the root volume and load the root cluster.  HAMMER special
 274          * cases rootvol and rootcl and will not deallocate the structures.
 275          * We do not hold a ref because this would prevent related I/O
 276          * from being flushed.
 277          */
 278         if (error == 0 && ondisk->vol_rootvol == ondisk->vol_no) {
 279                 hmp->rootvol = volume;
 280                 hmp->rootcl = hammer_get_cluster(volume,
 281                                                  ondisk->vol0_root_clu_no,
 282                                                  &error, 0);
 283                 hammer_rel_cluster(hmp->rootcl, 0);
 284                 hmp->fsid_udev = dev2udev(vn_todev(volume->devvp));
 285         }
 286 late_failure:
 287         if (bp)
 288                 brelse(bp);
 289         if (error) {
 290                 /*vinvalbuf(volume->devvp, V_SAVE, 0, 0);*/
 291                 VOP_CLOSE(volume->devvp, ronly ? FREAD : FREAD|FWRITE);
 292                 hammer_free_volume(volume);
 293         }
 294         return (error);
 295 }
 296
 297 /*
 298  * Unload and free a HAMMER volume.  Must return >= 0 to continue scan
 299  * so returns -1 on failure.
 300  */
 301 int
 302 hammer_unload_volume(hammer_volume_t volume, void *data __unused)
 303 {
 304         struct hammer_mount *hmp = volume->hmp;
 305         hammer_cluster_t rootcl;
 306         int ronly = ((hmp->mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
 307
 308         /*
 309          * Sync clusters, sync volume
 310          */
 311
 312
 313         /*
 314          * Clean up the root cluster, which is held unlocked in the root
 315          * volume.
 316          */
 317         if (hmp->rootvol == volume) {
 318                 if ((rootcl = hmp->rootcl) != NULL)
 319                         hmp->rootcl = NULL;
 320                 hmp->rootvol = NULL;
 321         }
 322
 323         /*
 324          * Unload clusters and super-clusters.  Unloading a super-cluster
 325          * also unloads related clusters, but the filesystem may not be
 326          * using super-clusters so unload clusters anyway.
 327          */
 328         RB_SCAN(hammer_clu_rb_tree, &volume->rb_clus_root, NULL,
 329                         hammer_unload_cluster, NULL);
 330         RB_SCAN(hammer_scl_rb_tree, &volume->rb_scls_root, NULL,
 331                         hammer_unload_supercl, NULL);
 332
 333         /*
 334          * Release our buffer and flush anything left in the buffer cache.
 335          */
 336         hammer_io_release(&volume->io, 1);
 337
 338         /*
 339          * There should be no references on the volume.
 340          */
 341         KKASSERT(volume->io.lock.refs == 0);
 342
 343         volume->ondisk = NULL;
 344         if (volume->devvp) {
 345                 if (ronly) {
 346                         vinvalbuf(volume->devvp, 0, 0, 0);
 347                         VOP_CLOSE(volume->devvp, FREAD);
 348                 } else {
 349                         vinvalbuf(volume->devvp, V_SAVE, 0, 0);
 350                         VOP_CLOSE(volume->devvp, FREAD|FWRITE);
 351                 }
 352         }
 353
 354         /*
 355          * Destroy the structure
 356          */
 357         RB_REMOVE(hammer_vol_rb_tree, &hmp->rb_vols_root, volume);
 358         hammer_free_volume(volume);
 359         return(0);
 360 }
 361
 362 static
 363 void
 364 hammer_free_volume(hammer_volume_t volume)
 365 {
 366         if (volume->vol_name) {
 367                 kfree(volume->vol_name, M_HAMMER);
 368                 volume->vol_name = NULL;
 369         }
 370         if (volume->devvp) {
 371                 vrele(volume->devvp);
 372                 volume->devvp = NULL;
 373         }
 374         kfree(volume, M_HAMMER);
 375 }
 376
 377 /*
 378  * Get a HAMMER volume.  The volume must already exist.
 379  */
 380 hammer_volume_t
 381 hammer_get_volume(struct hammer_mount *hmp, int32_t vol_no, int *errorp)
 382 {
 383         struct hammer_volume *volume;
 384
 385         /*
 386          * Locate the volume structure
 387          */
 388         volume = RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, vol_no);
 389         if (volume == NULL) {
 390                 *errorp = ENOENT;
 391                 return(NULL);
 392         }
 393         hammer_ref(&volume->io.lock);
 394
 395         /*
 396          * Deal with on-disk info
 397          */
 398         if (volume->ondisk == NULL) {
 399                 *errorp = hammer_load_volume(volume);
 400                 if (*errorp) {
 401                         hammer_rel_volume(volume, 1);
 402                         volume = NULL;
 403                 }
 404         } else {
 405                 *errorp = 0;
 406         }
 407         return(volume);
 408 }
 409
 410 hammer_volume_t
 411 hammer_get_root_volume(struct hammer_mount *hmp, int *errorp)
 412 {
 413         hammer_volume_t volume;
 414
 415         volume = hmp->rootvol;
 416         KKASSERT(volume != NULL);
 417         hammer_ref(&volume->io.lock);
 418
 419         /*
 420          * Deal with on-disk info
 421          */
 422         if (volume->ondisk == NULL) {
 423                 *errorp = hammer_load_volume(volume);
 424                 if (*errorp) {
 425                         hammer_rel_volume(volume, 1);
 426                         volume = NULL;
 427                 }
 428         } else {
 429                 *errorp = 0;
 430         }
 431         return (volume);
 432 }
 433
 434 /*
 435  * Load a volume's on-disk information.  The volume must be referenced and
 436  * not locked.  We temporarily acquire an exclusive lock to interlock
 437  * against releases or multiple get's.
 438  */
 439 static int
 440 hammer_load_volume(hammer_volume_t volume)
 441 {
 442         struct hammer_volume_ondisk *ondisk;
 443         int error;
 444
 445         hammer_lock_ex(&volume->io.lock);
 446         if (volume->ondisk == NULL) {
 447                 error = hammer_io_read(volume->devvp, &volume->io);
 448                 if (error) {
 449                         hammer_unlock(&volume->io.lock);
 450                         return (error);
 451                 }
 452                 volume->ondisk = ondisk = (void *)volume->io.bp->b_data;
 453
 454                 /*
 455                  * Configure the volume's A-lists.  These are used to
 456                  * allocate clusters.
 457                  */
 458                 if (volume->vol_flags & HAMMER_VOLF_USINGSUPERCL) {
 459                         volume->alist.config = &Vol_super_alist_config;
 460                         volume->alist.meta = ondisk->vol_almeta.super;
 461                         volume->alist.info = volume;
 462                 } else {
 463                         volume->alist.config = &Vol_normal_alist_config;
 464                         volume->alist.meta = ondisk->vol_almeta.normal;
 465                         volume->alist.info = NULL;
 466                 }
 467                 hammer_alist_init(&volume->alist);
 468         } else {
 469                 error = 0;
 470         }
 471         hammer_unlock(&volume->io.lock);
 472         return(0);
 473 }
 474
 475 /*
 476  * Release a volume.  Call hammer_io_release on the last reference.  We have
 477  * to acquire an exclusive lock to interlock against volume->ondisk tests
 478  * in hammer_load_volume().
 479  */
 480 void
 481 hammer_rel_volume(hammer_volume_t volume, int flush)
 482 {
 483         if (hammer_islastref(&volume->io.lock)) {
 484                 hammer_lock_ex(&volume->io.lock);
 485                 if (hammer_islastref(&volume->io.lock)) {
 486                         volume->ondisk = NULL;
 487                         hammer_io_release(&volume->io, flush);
 488                 }
 489                 hammer_unlock(&volume->io.lock);
 490         }
 491         hammer_unref(&volume->io.lock);
 492 }
 493
 494 /************************************************************************
 495  *                              SUPER-CLUSTERS                          *
 496  ************************************************************************
 497  *
 498  * Manage super-clusters.  Note that a supercl holds a reference to its
 499  * associated volume.
 500  */
 501 hammer_supercl_t
 502 hammer_get_supercl(hammer_volume_t volume, int32_t scl_no,
 503                    int *errorp, int isnew)
 504 {
 505         hammer_supercl_t supercl;
 506
 507         /*
 508          * Locate and lock the super-cluster structure, creating one
 509          * if necessary.
 510          */
 511 again:
 512         supercl = RB_LOOKUP(hammer_scl_rb_tree, &volume->rb_scls_root, scl_no);
 513         if (supercl == NULL) {
 514                 supercl = kmalloc(sizeof(*supercl), M_HAMMER, M_WAITOK|M_ZERO);
 515                 supercl->scl_no = scl_no;
 516                 supercl->volume = volume;
 517                 supercl->io.offset = calculate_supercl_offset(volume, scl_no);
 518                 supercl->io.type = HAMMER_STRUCTURE_SUPERCL;
 519                 hammer_ref(&supercl->io.lock);
 520
 521                 /*
 522                  * Insert the cluster into the RB tree and handle late
 523                  * collisions.
 524                  */
 525                 if (RB_INSERT(hammer_scl_rb_tree, &volume->rb_scls_root, supercl)) {
 526                         hammer_unref(&supercl->io.lock);
 527                         kfree(supercl, M_HAMMER);
 528                         goto again;
 529                 }
 530                 hammer_ref(&volume->io.lock);
 531         } else {
 532                 hammer_ref(&supercl->io.lock);
 533         }
 534
 535         /*
 536          * Deal with on-disk info
 537          */
 538         if (supercl->ondisk == NULL || isnew) {
 539                 *errorp = hammer_load_supercl(supercl, isnew);
 540                 if (*errorp) {
 541                         hammer_rel_supercl(supercl, 1);
 542                         supercl = NULL;
 543                 }
 544         } else {
 545                 *errorp = 0;
 546         }
 547         return(supercl);
 548 }
 549
 550 static int
 551 hammer_load_supercl(hammer_supercl_t supercl, int isnew)
 552 {
 553         struct hammer_supercl_ondisk *ondisk;
 554         hammer_volume_t volume = supercl->volume;
 555         int error;
 556
 557         hammer_lock_ex(&supercl->io.lock);
 558         if (supercl->ondisk == NULL) {
 559                 if (isnew)
 560                         error = hammer_io_new(volume->devvp, &supercl->io);
 561                 else
 562                         error = hammer_io_read(volume->devvp, &supercl->io);
 563                 if (error) {
 564                         hammer_unlock(&supercl->io.lock);
 565                         return (error);
 566                 }
 567                 supercl->ondisk = ondisk = (void *)supercl->io.bp->b_data;
 568
 569                 supercl->alist.config = &Supercl_alist_config;
 570                 supercl->alist.meta = ondisk->scl_meta;
 571                 supercl->alist.info = NULL;
 572         } else if (isnew) {
 573                 error = hammer_io_new(volume->devvp, &supercl->io);
 574         } else {
 575                 error = 0;
 576         }
 577         if (error == 0 && isnew) {
 578                 /*
 579                  * If this is a new super-cluster we have to initialize
 580                  * various ondisk structural elements.  The caller is
 581                  * responsible for the remainder.
 582                  */
 583                 struct hammer_alist_live dummy;
 584
 585                 ondisk = supercl->ondisk;
 586                 dummy.config = &Buf_alist_config;
 587                 dummy.meta = ondisk->head.buf_almeta;
 588                 dummy.info = NULL;
 589                 initbuffer(&dummy, &ondisk->head, HAMMER_FSBUF_SUPERCL);
 590                 hammer_alist_init(&supercl->alist);
 591         }
 592         hammer_unlock(&supercl->io.lock);
 593         return (error);
 594 }
 595
 596 /*
 597  * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
 598  */
 599 int
 600 hammer_unload_supercl(hammer_supercl_t supercl, void *data __unused)
 601 {
 602         KKASSERT(supercl->io.lock.refs == 0);
 603         hammer_ref(&supercl->io.lock);
 604         hammer_io_release(&supercl->io, 1);
 605         hammer_rel_supercl(supercl, 1);
 606         return(0);
 607 }
 608
 609 /*
 610  * Release a super-cluster.  We have to deal with several places where
 611  * another thread can ref the super-cluster.
 612  *
 613  * Only destroy the structure itself if the related buffer cache buffer
 614  * was disassociated from it.  This ties the management of the structure
 615  * to the buffer cache subsystem.
 616  */
 617 void
 618 hammer_rel_supercl(hammer_supercl_t supercl, int flush)
 619 {
 620         hammer_volume_t volume;
 621
 622         if (hammer_islastref(&supercl->io.lock)) {
 623                 hammer_lock_ex(&supercl->io.lock);
 624                 if (hammer_islastref(&supercl->io.lock)) {
 625                         hammer_io_release(&supercl->io, flush);
 626                         if (supercl->io.bp == NULL &&
 627                             hammer_islastref(&supercl->io.lock)) {
 628                                 volume = supercl->volume;
 629                                 RB_REMOVE(hammer_scl_rb_tree,
 630                                           &volume->rb_scls_root, supercl);
 631                                 supercl->volume = NULL; /* sanity */
 632                                 kfree(supercl, M_HAMMER);
 633                                 hammer_rel_volume(volume, 0);
 634                                 return;
 635                         }
 636                 }
 637                 hammer_unlock(&supercl->io.lock);
 638         }
 639         hammer_unref(&supercl->io.lock);
 640 }
 641
 642 /************************************************************************
 643  *                              CLUSTERS                                *
 644  ************************************************************************
 645  *
 646  * Manage clusters.  Note that a cluster holds a reference to its
 647  * associated volume.
 648  */
 649 hammer_cluster_t
 650 hammer_get_cluster(hammer_volume_t volume, int32_t clu_no,
 651                    int *errorp, int isnew)
 652 {
 653         hammer_cluster_t cluster;
 654
 655 again:
 656         cluster = RB_LOOKUP(hammer_clu_rb_tree, &volume->rb_clus_root, clu_no);
 657         if (cluster == NULL) {
 658                 cluster = kmalloc(sizeof(*cluster), M_HAMMER, M_WAITOK|M_ZERO);
 659                 cluster->clu_no = clu_no;
 660                 cluster->volume = volume;
 661                 cluster->io.offset = calculate_cluster_offset(volume, clu_no);
 662                 RB_INIT(&cluster->rb_bufs_root);
 663                 RB_INIT(&cluster->rb_nods_root);
 664                 cluster->io.type = HAMMER_STRUCTURE_CLUSTER;
 665                 hammer_ref(&cluster->io.lock);
 666
 667                 /*
 668                  * Insert the cluster into the RB tree and handle late
 669                  * collisions.
 670                  */
 671                 if (RB_INSERT(hammer_clu_rb_tree, &volume->rb_clus_root, cluster)) {
 672                         hammer_unref(&cluster->io.lock);
 673                         kfree(cluster, M_HAMMER);
 674                         goto again;
 675                 }
 676                 hammer_ref(&volume->io.lock);
 677         } else {
 678                 hammer_ref(&cluster->io.lock);
 679         }
 680
 681         /*
 682          * Deal with on-disk info
 683          */
 684         if (cluster->ondisk == NULL || isnew) {
 685                 *errorp = hammer_load_cluster(cluster, isnew);
 686                 if (*errorp) {
 687                         hammer_rel_cluster(cluster, 1);
 688                         cluster = NULL;
 689                 }
 690         } else {
 691                 *errorp = 0;
 692         }
 693         return (cluster);
 694 }
 695
 696 hammer_cluster_t
 697 hammer_get_root_cluster(struct hammer_mount *hmp, int *errorp)
 698 {
 699         hammer_cluster_t cluster;
 700
 701         cluster = hmp->rootcl;
 702         KKASSERT(cluster != NULL);
 703         hammer_ref(&cluster->io.lock);
 704
 705         /*
 706          * Deal with on-disk info
 707          */
 708         if (cluster->ondisk == NULL) {
 709                 *errorp = hammer_load_cluster(cluster, 0);
 710                 if (*errorp) {
 711                         hammer_rel_cluster(cluster, 1);
 712                         cluster = NULL;
 713                 }
 714         } else {
 715                 *errorp = 0;
 716         }
 717         return (cluster);
 718 }
 719
 720 static
 721 int
 722 hammer_load_cluster(hammer_cluster_t cluster, int isnew)
 723 {
 724         hammer_volume_t volume = cluster->volume;
 725         struct hammer_cluster_ondisk *ondisk;
 726         int error;
 727
 728         /*
 729          * Load the cluster's on-disk info
 730          */
 731         hammer_lock_ex(&cluster->io.lock);
 732         if (cluster->ondisk == NULL) {
 733                 if (isnew)
 734                         error = hammer_io_new(volume->devvp, &cluster->io);
 735                 else
 736                         error = hammer_io_read(volume->devvp, &cluster->io);
 737                 if (error) {
 738                         hammer_unlock(&cluster->io.lock);
 739                         return (error);
 740                 }
 741                 cluster->ondisk = ondisk = (void *)cluster->io.bp->b_data;
 742
 743                 cluster->alist_master.config = &Clu_master_alist_config;
 744                 cluster->alist_master.meta = ondisk->clu_master_meta;
 745                 cluster->alist_btree.config = &Clu_slave_alist_config;
 746                 cluster->alist_btree.meta = ondisk->clu_btree_meta;
 747                 cluster->alist_btree.info = cluster;
 748                 cluster->alist_record.config = &Clu_slave_alist_config;
 749                 cluster->alist_record.meta = ondisk->clu_record_meta;
 750                 cluster->alist_record.info = cluster;
 751                 cluster->alist_mdata.config = &Clu_slave_alist_config;
 752                 cluster->alist_mdata.meta = ondisk->clu_mdata_meta;
 753                 cluster->alist_mdata.info = cluster;
 754
 755                 cluster->clu_btree_beg = ondisk->clu_btree_beg;
 756                 cluster->clu_btree_end = ondisk->clu_btree_end;
 757         } else if (isnew) {
 758                 error = hammer_io_new(volume->devvp, &cluster->io);
 759         } else {
 760                 error = 0;
 761         }
 762         if (error == 0 && isnew) {
 763                 /*
 764                  * If this is a new cluster we have to initialize
 765                  * various ondisk structural elements.  The caller is
 766                  * responsible for the remainder.
 767                  */
 768                 struct hammer_alist_live dummy;
 769
 770                 ondisk = cluster->ondisk;
 771
 772                 dummy.config = &Buf_alist_config;
 773                 dummy.meta = ondisk->head.buf_almeta;
 774                 dummy.info = NULL;
 775                 initbuffer(&dummy, &ondisk->head, HAMMER_FSBUF_CLUSTER);
 776
 777                 hammer_alist_init(&cluster->alist_master);
 778                 hammer_alist_init(&cluster->alist_btree);
 779                 hammer_alist_init(&cluster->alist_record);
 780                 hammer_alist_init(&cluster->alist_mdata);
 781         }
 782         hammer_unlock(&cluster->io.lock);
 783         return (error);
 784 }
 785
 786 /*
 787  * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
 788  */
 789 int
 790 hammer_unload_cluster(hammer_cluster_t cluster, void *data __unused)
 791 {
 792         hammer_ref(&cluster->io.lock);
 793         RB_SCAN(hammer_buf_rb_tree, &cluster->rb_bufs_root, NULL,
 794                         hammer_unload_buffer, NULL);
 795         KKASSERT(cluster->io.lock.refs == 1);
 796         hammer_io_release(&cluster->io, 1);
 797         hammer_rel_cluster(cluster, 1);
 798         return(0);
 799 }
 800
 801 /*
 802  * Reference a cluster that is either already referenced or via a specially
 803  * handled pointer (aka rootcl).
 804  */
 805 int
 806 hammer_ref_cluster(hammer_cluster_t cluster)
 807 {
 808         int error;
 809
 810         KKASSERT(cluster != NULL);
 811         hammer_ref(&cluster->io.lock);
 812
 813         /*
 814          * Deal with on-disk info
 815          */
 816         if (cluster->ondisk == NULL) {
 817                 error = hammer_load_cluster(cluster, 0);
 818                 if (error)
 819                         hammer_rel_cluster(cluster, 1);
 820         } else {
 821                 error = 0;
 822         }
 823         return(error);
 824 }
 825
 826 /*
 827  * Release a cluster.  We have to deal with several places where
 828  * another thread can ref the cluster.
 829  *
 830  * Only destroy the structure itself if the related buffer cache buffer
 831  * was disassociated from it.  This ties the management of the structure
 832  * to the buffer cache subsystem.
 833  */
 834 void
 835 hammer_rel_cluster(hammer_cluster_t cluster, int flush)
 836 {
 837         hammer_node_t node;
 838         hammer_volume_t volume;
 839
 840         if (hammer_islastref(&cluster->io.lock)) {
 841                 hammer_lock_ex(&cluster->io.lock);
 842                 if (hammer_islastref(&cluster->io.lock)) {
 843                         hammer_io_release(&cluster->io, flush);
 844
 845                         /*
 846                          * Clean out the B-Tree node cache, if any, then
 847                          * clean up the volume ref and free the cluster.
 848                          *
 849                          * If the cluster acquires a new reference while we
 850                          * are trying to clean it out, abort the cleaning.
 851                          *
 852                          * There really shouldn't be any nodes at this point
 853                          * but we allow a node with no buffer association
 854                          * so handle the case.
 855                          */
 856                         while (cluster->io.bp == NULL &&
 857                                hammer_islastref(&cluster->io.lock) &&
 858                                (node = RB_ROOT(&cluster->rb_nods_root)) != NULL
 859                         ) {
 860                                 KKASSERT(node->lock.refs == 0);
 861                                 hammer_flush_node(node);
 862                         }
 863                         if (cluster->io.bp == NULL &&
 864                             hammer_islastref(&cluster->io.lock)) {
 865                                 volume = cluster->volume;
 866                                 RB_REMOVE(hammer_clu_rb_tree,
 867                                           &volume->rb_clus_root, cluster);
 868                                 cluster->volume = NULL; /* sanity */
 869                                 kfree(cluster, M_HAMMER);
 870                                 hammer_rel_volume(volume, 0);
 871                                 return;
 872                         }
 873                 }
 874                 hammer_unlock(&cluster->io.lock);
 875         }
 876         hammer_unref(&cluster->io.lock);
 877 }
 878
 879 /************************************************************************
 880  *                              BUFFERS                                 *
 881  ************************************************************************
 882  *
 883  * Manage buffers.  Note that a buffer holds a reference to its associated
 884  * cluster, and its cluster will hold a reference to the cluster's volume.
 885  *
 886  * A non-zero buf_type indicates that a new buffer should be created and
 887  * zero'd.
 888  */
 889 hammer_buffer_t
 890 hammer_get_buffer(hammer_cluster_t cluster, int32_t buf_no,
 891                   u_int64_t buf_type, int *errorp)
 892 {
 893         hammer_buffer_t buffer;
 894
 895         /*
 896          * Find the buffer.  Note that buffer 0 corresponds to the cluster
 897          * header and should never be requested.
 898          */
 899         KKASSERT(buf_no >= cluster->ondisk->clu_start / HAMMER_BUFSIZE &&
 900                  buf_no < cluster->ondisk->clu_limit / HAMMER_BUFSIZE);
 901
 902         /*
 903          * Locate and lock the buffer structure, creating one if necessary.
 904          */
 905 again:
 906         buffer = RB_LOOKUP(hammer_buf_rb_tree, &cluster->rb_bufs_root, buf_no);
 907         if (buffer == NULL) {
 908                 buffer = kmalloc(sizeof(*cluster), M_HAMMER, M_WAITOK|M_ZERO);
 909                 buffer->buf_no = buf_no;
 910                 buffer->cluster = cluster;
 911                 buffer->volume = cluster->volume;
 912                 buffer->io.offset = cluster->io.offset +
 913                                     (buf_no * HAMMER_BUFSIZE);
 914                 buffer->io.type = HAMMER_STRUCTURE_BUFFER;
 915                 TAILQ_INIT(&buffer->clist);
 916                 hammer_ref(&buffer->io.lock);
 917
 918                 /*
 919                  * Insert the cluster into the RB tree and handle late
 920                  * collisions.
 921                  */
 922                 if (RB_INSERT(hammer_buf_rb_tree, &cluster->rb_bufs_root, buffer)) {
 923                         hammer_unref(&buffer->io.lock);
 924                         kfree(buffer, M_HAMMER);
 925                         goto again;
 926                 }
 927                 hammer_ref(&cluster->io.lock);
 928         } else {
 929                 hammer_ref(&buffer->io.lock);
 930         }
 931
 932         /*
 933          * Deal with on-disk info
 934          */
 935         if (buffer->ondisk == NULL || buf_type) {
 936                 *errorp = hammer_load_buffer(buffer, buf_type);
 937                 if (*errorp) {
 938                         hammer_rel_buffer(buffer, 1);
 939                         buffer = NULL;
 940                 }
 941         } else {
 942                 *errorp = 0;
 943         }
 944         return(buffer);
 945 }
 946
 947 static int
 948 hammer_load_buffer(hammer_buffer_t buffer, u_int64_t buf_type)
 949 {
 950         hammer_volume_t volume;
 951         hammer_fsbuf_ondisk_t ondisk;
 952         int error;
 953
 954         /*
 955          * Load the buffer's on-disk info
 956          */
 957         volume = buffer->volume;
 958         hammer_lock_ex(&buffer->io.lock);
 959         if (buffer->ondisk == NULL) {
 960                 if (buf_type) {
 961                         error = hammer_io_new(volume->devvp, &buffer->io);
 962                 } else {
 963                         error = hammer_io_read(volume->devvp, &buffer->io);
 964                 }
 965                 if (error) {
 966                         hammer_unlock(&buffer->io.lock);
 967                         return (error);
 968                 }
 969                 buffer->ondisk = ondisk = (void *)buffer->io.bp->b_data;
 970                 buffer->alist.config = &Buf_alist_config;
 971                 buffer->alist.meta = ondisk->head.buf_almeta;
 972                 buffer->buf_type = ondisk->head.buf_type;
 973         } else if (buf_type) {
 974                 error = hammer_io_new(volume->devvp, &buffer->io);
 975         } else {
 976                 error = 0;
 977         }
 978         if (error == 0 && buf_type) {
 979                 ondisk = buffer->ondisk;
 980                 initbuffer(&buffer->alist, &ondisk->head, buf_type);
 981                 buffer->buf_type = ondisk->head.buf_type;
 982         }
 983         hammer_unlock(&buffer->io.lock);
 984         return (error);
 985 }
 986
 987 /*
 988  * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
 989  */
 990 int
 991 hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused)
 992 {
 993         hammer_ref(&buffer->io.lock);
 994         hammer_flush_buffer_nodes(buffer);
 995         hammer_io_release(&buffer->io, 1);
 996         KKASSERT(buffer->io.lock.refs == 1);
 997         hammer_rel_buffer(buffer, 1);
 998         return(0);
 999 }
1000
1001 /*
1002  * Reference a buffer that is either already referenced or via a specially
1003  * handled pointer (aka cursor->buffer).
1004  */
1005 int
1006 hammer_ref_buffer(hammer_buffer_t buffer)
1007 {
1008         int error;
1009
1010         hammer_ref(&buffer->io.lock);
1011         if (buffer->ondisk == NULL) {
1012                 error = hammer_load_buffer(buffer, 0);
1013                 if (error) {
1014                         hammer_rel_buffer(buffer, 1);
1015                         /*
1016                          * NOTE: buffer pointer can become stale after
1017                          * the above release.
1018                          */
1019                 } else {
1020                         KKASSERT(buffer->buf_type ==
1021                                  buffer->ondisk->head.buf_type);
1022                 }
1023         } else {
1024                 error = 0;
1025         }
1026         return(error);
1027 }
1028
1029 /*
1030  * Release a buffer.  We have to deal with several places where
1031  * another thread can ref the buffer.
1032  *
1033  * Only destroy the structure itself if the related buffer cache buffer
1034  * was disassociated from it.  This ties the management of the structure
1035  * to the buffer cache subsystem.  buffer->ondisk determines whether the
1036  * embedded io is referenced or not.
1037  */
1038 void
1039 hammer_rel_buffer(hammer_buffer_t buffer, int flush)
1040 {
1041         hammer_cluster_t cluster;
1042         hammer_node_t node;
1043
1044         if (hammer_islastref(&buffer->io.lock)) {
1045                 hammer_lock_ex(&buffer->io.lock);
1046                 if (hammer_islastref(&buffer->io.lock)) {
1047                         hammer_io_release(&buffer->io, flush);
1048
1049                         /*
1050                          * Clean out the B-Tree node cache, if any, then
1051                          * clean up the cluster ref and free the buffer.
1052                          *
1053                          * If the buffer acquires a new reference while we
1054                          * are trying to clean it out, abort the cleaning.
1055                          */
1056                         while (buffer->io.bp == NULL &&
1057                                hammer_islastref(&buffer->io.lock) &&
1058                                (node = TAILQ_FIRST(&buffer->clist)) != NULL
1059                         ) {
1060                                 KKASSERT(node->lock.refs == 0);
1061                                 hammer_flush_node(node);
1062                         }
1063                         if (buffer->io.bp == NULL &&
1064                             hammer_islastref(&buffer->io.lock)) {
1065                                 cluster = buffer->cluster;
1066                                 RB_REMOVE(hammer_buf_rb_tree,
1067                                           &cluster->rb_bufs_root, buffer);
1068                                 buffer->cluster = NULL; /* sanity */
1069                                 kfree(buffer, M_HAMMER);
1070                                 hammer_rel_cluster(cluster, 0);
1071                                 return;
1072                         }
1073                 }
1074                 hammer_unlock(&buffer->io.lock);
1075         }
1076         hammer_unref(&buffer->io.lock);
1077 }
1078
1079 /*
1080  * Flush passively cached B-Tree nodes associated with this buffer.
1081  *
1082  * NOTE: The buffer is referenced and locked.
1083  */
1084 void
1085 hammer_flush_buffer_nodes(hammer_buffer_t buffer)
1086 {
1087         hammer_node_t node;
1088
1089         node = TAILQ_FIRST(&buffer->clist);
1090         while (node) {
1091                 buffer->save_scan = TAILQ_NEXT(node, entry);
1092                 if (node->lock.refs == 0)
1093                         hammer_flush_node(node);
1094                 node = buffer->save_scan;
1095         }
1096 }
1097
1098 /************************************************************************
1099  *                              NODES                                   *
1100  ************************************************************************
1101  *
1102  * Manage B-Tree nodes.  B-Tree nodes represent the primary indexing
1103  * method used by the HAMMER filesystem.
1104  *
1105  * Unlike other HAMMER structures, a hammer_node can be PASSIVELY
1106  * associated with its buffer.  It can have an active buffer reference
1107  * even when the node itself has no references.  The node also passively
1108  * associates itself with its cluster without holding any cluster refs.
1109  * The cluster ref is indirectly maintained by the active buffer ref when
1110  * a node is acquired.
1111  *
1112  * A hammer_node can also be passively associated with other HAMMER
1113  * structures, such as inodes, while retaining 0 references.  These
1114  * associations can be cleared backwards using a pointer-to-pointer in
1115  * the hammer_node.
1116  *
1117  * This allows the HAMMER implementation to cache hammer_node's long-term
1118  * and short-cut a great deal of the infrastructure's complexity.  In
1119  * most cases a cached node can be reacquired without having to dip into
1120  * either the buffer or cluster management code.
1121  *
1122  * The caller must pass a referenced cluster on call and will retain
1123  * ownership of the reference on return.  The node will acquire its own
1124  * additional references, if necessary.
1125  */
1126 hammer_node_t
1127 hammer_get_node(hammer_cluster_t cluster, int32_t node_offset, int *errorp)
1128 {
1129         hammer_node_t node;
1130
1131         /*
1132          * Locate the structure, allocating one if necessary.
1133          */
1134 again:
1135         node = RB_LOOKUP(hammer_nod_rb_tree, &cluster->rb_nods_root,
1136                          node_offset);
1137         if (node == NULL) {
1138                 node = kmalloc(sizeof(*node), M_HAMMER, M_WAITOK|M_ZERO);
1139                 node->node_offset = node_offset;
1140                 node->cluster = cluster;
1141                 if (RB_INSERT(hammer_nod_rb_tree, &cluster->rb_nods_root,
1142                               node)) {
1143                         kfree(node, M_HAMMER);
1144                         goto again;
1145                 }
1146         }
1147         *errorp = hammer_ref_node(node);
1148         if (*errorp) {
1149                 /*
1150                  * NOTE: The node pointer may be stale on error return.
1151                  * In fact, its probably been destroyed.
1152                  */
1153                 node = NULL;
1154         }
1155         return(node);
1156 }
1157
1158 /*
1159  * Reference the node to prevent disassociations, then associate and
1160  * load the related buffer.  This routine can also be called to reference
1161  * a node from a cache pointer.
1162  *
1163  * NOTE: Because the caller does not have a ref on the node, the caller's
1164  * node pointer will be stale if an error is returned.  We may also wind
1165  * up clearing the related cache pointers.
1166  *
1167  * NOTE: The cluster is indirectly referenced by our buffer ref.
1168  */
1169 int
1170 hammer_ref_node(hammer_node_t node)
1171 {
1172         hammer_buffer_t buffer;
1173         int32_t buf_no;
1174         int error;
1175
1176         hammer_ref(&node->lock);
1177         error = 0;
1178         if (node->ondisk == NULL) {
1179                 hammer_lock_ex(&node->lock);
1180                 if (node->ondisk == NULL) {
1181                         /*
1182                          * This is a little confusing but the jist is that
1183                          * node->buffer determines whether the node is on
1184                          * the buffer's clist and node->ondisk determines
1185                          * whether the buffer is referenced.
1186                          */
1187                         if ((buffer = node->buffer) != NULL) {
1188                                 error = hammer_ref_buffer(buffer);
1189                         } else {
1190                                 buf_no = node->node_offset / HAMMER_BUFSIZE;
1191                                 buffer = hammer_get_buffer(node->cluster,
1192                                                            buf_no, 0, &error);
1193                                 if (buffer) {
1194                                         KKASSERT(error == 0);
1195                                         TAILQ_INSERT_TAIL(&buffer->clist,
1196                                                           node, entry);
1197                                         node->buffer = buffer;
1198                                 }
1199                         }
1200                         if (error == 0) {
1201                                 node->ondisk = (void *)((char *)buffer->ondisk +
1202                                        (node->node_offset & HAMMER_BUFMASK));
1203                         }
1204                 }
1205                 hammer_unlock(&node->lock);
1206         }
1207         if (error)
1208                 hammer_rel_node(node);
1209         return (error);
1210 }
1211
1212 /*
1213  * Release a hammer_node.  The node retains a passive association with
1214  * its cluster, buffer and caches.
1215  *
1216  * However, to avoid cluttering up kernel memory with tons of B-Tree
1217  * node cache structures we destroy the node if no passive cache or
1218  * (instantiated) buffer references exist.
1219  */
1220 void
1221 hammer_rel_node(hammer_node_t node)
1222 {
1223         hammer_cluster_t cluster;
1224         hammer_buffer_t buffer;
1225
1226         if (hammer_islastref(&node->lock)) {
1227                 cluster = node->cluster;
1228                 /*
1229                  * Clutter control, this case only occurs after a failed
1230                  * load since otherwise ondisk will be non-NULL.
1231                  */
1232                 if (node->cache1 == NULL && node->cache2 == NULL &&
1233                     node->ondisk == NULL) {
1234                         RB_REMOVE(hammer_nod_rb_tree, &cluster->rb_nods_root,
1235                                   node);
1236                         if ((buffer = node->buffer) != NULL) {
1237                                 node->buffer = NULL;
1238                                 hammer_remove_node_clist(buffer, node);
1239                         }
1240                         kfree(node, M_HAMMER);
1241                         return;
1242                 }
1243
1244                 /*
1245                  * node->ondisk determines whether we have a buffer reference
1246                  * to get rid of or not.  Only get rid of the reference if
1247                  * the kernel tried to flush the buffer.
1248                  *
1249                  * NOTE: Once unref'd the node can be physically destroyed,
1250                  * so our node is stale afterwords.
1251                  *
1252                  * This case occurs if the node still has cache references.
1253                  * We could remove the references and free the structure
1254                  * but for now we allow them (and the node structure) to
1255                  * remain intact.
1256                  */
1257                 if (node->ondisk && hammer_io_checkflush(&node->buffer->io)) {
1258                         buffer = node->buffer;
1259                         node->buffer = NULL;
1260                         node->ondisk = NULL;
1261                         hammer_remove_node_clist(buffer, node);
1262                         hammer_unref(&node->lock);
1263                         hammer_rel_buffer(buffer, 0);
1264                 } else {
1265                         hammer_unref(&node->lock);
1266                 }
1267         } else {
1268                 hammer_unref(&node->lock);
1269         }
1270 }
1271
1272 /*
1273  * Cache-and-release a hammer_node.  Kinda like catching and releasing a
1274  * fish, but keeping an eye on him.  The node is passively cached in *cache.
1275  *
1276  * NOTE!  HAMMER may NULL *cache at any time, even after you have
1277  * referenced the node!
1278  */
1279 void
1280 hammer_cache_node(hammer_node_t node, struct hammer_node **cache)
1281 {
1282         if (node->cache1 != cache) {
1283                 if (node->cache2 == cache) {
1284                         struct hammer_node **tmp;
1285                         tmp = node->cache1;
1286                         node->cache1 = node->cache2;
1287                         node->cache2 = tmp;
1288                 } else {
1289                         if (node->cache2)
1290                                 *node->cache2 = NULL;
1291                         node->cache2 = node->cache1;
1292                         node->cache1 = cache;
1293                         *cache = node;
1294                 }
1295         }
1296 }
1297
1298 void
1299 hammer_uncache_node(struct hammer_node **cache)
1300 {
1301         hammer_node_t node;
1302
1303         if ((node = *cache) != NULL) {
1304                 *cache = NULL;
1305                 if (node->cache1 == cache) {
1306                         node->cache1 = node->cache2;
1307                         node->cache2 = NULL;
1308                 } else if (node->cache2 == cache) {
1309                         node->cache2 = NULL;
1310                 } else {
1311                         panic("hammer_uncache_node: missing cache linkage");
1312                 }
1313                 if (node->cache1 == NULL && node->cache2 == NULL &&
1314                     node->lock.refs == 0) {
1315                         hammer_flush_node(node);
1316                 }
1317         }
1318 }
1319
1320 /*
1321  * Remove a node's cache references and destroy the node if it has no
1322  * references.  This is typically called from the buffer handling code.
1323  *
1324  * The node may have an active buffer reference (ondisk != NULL) even
1325  * if the node itself has no references.
1326  *
1327  * Note that a caller iterating through nodes via a buffer must have its
1328  * own reference on the buffer or our hammer_rel_buffer() call below may
1329  * rip it out from under the caller.
1330  */
1331 void
1332 hammer_flush_node(hammer_node_t node)
1333 {
1334         hammer_buffer_t buffer;
1335
1336         if (node->cache1)
1337                 *node->cache1 = NULL;
1338         if (node->cache2)
1339                 *node->cache2 = NULL;
1340         if (node->lock.refs == 0) {
1341                 RB_REMOVE(hammer_nod_rb_tree, &node->cluster->rb_nods_root,
1342                           node);
1343                 if ((buffer = node->buffer) != NULL) {
1344                         node->buffer = NULL;
1345                         hammer_remove_node_clist(buffer, node);
1346                         if (node->ondisk) {
1347                                 node->ondisk = NULL;
1348                                 hammer_rel_buffer(buffer, 0);
1349                         }
1350                 }
1351                 kfree(node, M_HAMMER);
1352         }
1353 }
1354
1355 /*
1356  * Remove a node from the buffer's clist.  Adjust save_scan as appropriate.
1357  * This is in its own little routine to properly handle interactions with
1358  * save_scan, so it is possible to block while scanning a buffer's node list.
1359  */
1360 static
1361 void
1362 hammer_remove_node_clist(hammer_buffer_t buffer, hammer_node_t node)
1363 {
1364         if (buffer->save_scan == node)
1365                 buffer->save_scan = TAILQ_NEXT(node, entry);
1366         TAILQ_REMOVE(&buffer->clist, node, entry);
1367 }
1368
1369 /************************************************************************
1370  *                              A-LIST ALLOCATORS                       *
1371  ************************************************************************/
1372
1373 /*
1374  * Allocate HAMMER elements - btree nodes, data storage, and record elements
1375  *
1376  * The passed *bufferp should be initialized to NULL.  On successive calls
1377  * *bufferp caches the most recent buffer used until put away by the caller.
1378  * Note that previously returned pointers using the cached buffer become
1379  * invalid on successive calls which reuse *bufferp.
1380  *
1381  * All allocations first attempt to use the block found at the specified
1382  * iterator.  If that fails the first available block is used.  If that
1383  * fails a new buffer is allocated and associated with the buffer type
1384  * A-list and the element is allocated out of the new buffer.
1385  */
1386
1387 hammer_node_t
1388 hammer_alloc_btree(hammer_cluster_t cluster, int *errorp)
1389 {
1390         hammer_buffer_t buffer;
1391         hammer_alist_t live;
1392         hammer_node_t node;
1393         int32_t elm_no;
1394         int32_t buf_no;
1395         int32_t node_offset;
1396
1397         /*
1398          * Allocate a B-Tree element
1399          */
1400         buffer = NULL;
1401         live = &cluster->alist_btree;
1402         elm_no = hammer_alist_alloc_fwd(live, 1, cluster->ondisk->idx_index);
1403         if (elm_no == HAMMER_ALIST_BLOCK_NONE)
1404                 elm_no = hammer_alist_alloc_fwd(live, 1, 0);
1405         if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
1406                 alloc_new_buffer(cluster, live,
1407                                  HAMMER_FSBUF_BTREE, HAMMER_BTREE_NODES,
1408                                  cluster->ondisk->idx_index, errorp, &buffer);
1409                 elm_no = hammer_alist_alloc(live, 1);
1410                 if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
1411                         *errorp = ENOSPC;
1412                         if (buffer)
1413                                 hammer_rel_buffer(buffer, 0);
1414                         hammer_modify_cluster(cluster);
1415                         return(NULL);
1416                 }
1417         }
1418         cluster->ondisk->idx_index = elm_no;
1419         KKASSERT((elm_no & HAMMER_FSBUF_BLKMASK) < HAMMER_BTREE_NODES);
1420         hammer_modify_cluster(cluster);
1421
1422         /*
1423          * Load and return the B-Tree element
1424          */
1425         buf_no = elm_no / HAMMER_FSBUF_MAXBLKS;
1426         node_offset = buf_no * HAMMER_BUFSIZE +
1427                       offsetof(union hammer_fsbuf_ondisk,
1428                                btree.nodes[elm_no & HAMMER_FSBUF_BLKMASK]);
1429         node = hammer_get_node(cluster, node_offset, errorp);
1430         if (node) {
1431                 bzero(node->ondisk, sizeof(*node->ondisk));
1432         } else {
1433                 hammer_alist_free(live, elm_no, 1);
1434                 hammer_rel_node(node);
1435                 node = NULL;
1436         }
1437         if (buffer)
1438                 hammer_rel_buffer(buffer, 0);
1439         return(node);
1440 }
1441
1442 void *
1443 hammer_alloc_data(hammer_cluster_t cluster, int32_t bytes,
1444                   int *errorp, struct hammer_buffer **bufferp)
1445 {
1446         hammer_buffer_t buffer;
1447         hammer_alist_t live;
1448         int32_t elm_no;
1449         int32_t buf_no;
1450         int32_t nblks;
1451         void *item;
1452
1453         /*
1454          * Deal with large data blocks.  The blocksize is HAMMER_BUFSIZE
1455          * for these allocations.
1456          */
1457         if ((bytes & HAMMER_BUFMASK) == 0) {
1458                 nblks = bytes / HAMMER_BUFSIZE;
1459                 /* only one block allowed for now (so buffer can hold it) */
1460                 KKASSERT(nblks == 1);
1461
1462                 buf_no = hammer_alist_alloc_fwd(&cluster->alist_master,
1463                                                 nblks,
1464                                                 cluster->ondisk->idx_ldata);
1465                 if (buf_no == HAMMER_ALIST_BLOCK_NONE) {
1466                         buf_no = hammer_alist_alloc_fwd(&cluster->alist_master,
1467                                                 nblks,
1468                                                 0);
1469                 }
1470                 hammer_modify_cluster(cluster);
1471                 if (buf_no == HAMMER_ALIST_BLOCK_NONE) {
1472                         *errorp = ENOSPC;
1473                         return(NULL);
1474                 }
1475                 cluster->ondisk->idx_ldata = buf_no;
1476                 buffer = *bufferp;
1477                 *bufferp = hammer_get_buffer(cluster, buf_no, -1, errorp);
1478                 if (buffer)
1479                         hammer_rel_buffer(buffer, 0);
1480                 buffer = *bufferp;
1481                 return(buffer->ondisk);
1482         }
1483
1484         /*
1485          * Allocate a data element.  The block size is HAMMER_DATA_BLKSIZE
1486          * (64 bytes) for these allocations.
1487          */
1488         nblks = (bytes + HAMMER_DATA_BLKMASK) & ~HAMMER_DATA_BLKMASK;
1489         nblks /= HAMMER_DATA_BLKSIZE;
1490         live = &cluster->alist_mdata;
1491         elm_no = hammer_alist_alloc_fwd(live, nblks, cluster->ondisk->idx_data);
1492         if (elm_no == HAMMER_ALIST_BLOCK_NONE)
1493                 elm_no = hammer_alist_alloc_fwd(live, 1, 0);
1494         if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
1495                 alloc_new_buffer(cluster, live,
1496                                  HAMMER_FSBUF_DATA, HAMMER_DATA_NODES,
1497                                  cluster->ondisk->idx_data, errorp, bufferp);
1498                 elm_no = hammer_alist_alloc(live, nblks);
1499                 if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
1500                         *errorp = ENOSPC;
1501                         hammer_modify_cluster(cluster);
1502                         return(NULL);
1503                 }
1504         }
1505         cluster->ondisk->idx_index = elm_no;
1506         hammer_modify_cluster(cluster);
1507
1508         /*
1509          * Load and return the B-Tree element
1510          */
1511         buf_no = elm_no / HAMMER_FSBUF_MAXBLKS;
1512         buffer = *bufferp;
1513         if (buffer == NULL || buffer->cluster != cluster ||
1514             buffer->buf_no != buf_no) {
1515                 if (buffer)
1516                         hammer_rel_buffer(buffer, 0);
1517                 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
1518                 *bufferp = buffer;
1519         }
1520         KKASSERT(buffer->ondisk->head.buf_type == HAMMER_FSBUF_DATA);
1521         KKASSERT((elm_no & HAMMER_FSBUF_BLKMASK) < HAMMER_DATA_NODES);
1522         item = &buffer->ondisk->data.data[elm_no & HAMMER_FSBUF_BLKMASK];
1523         bzero(item, nblks * HAMMER_DATA_BLKSIZE);
1524         *errorp = 0;
1525         return(item);
1526 }
1527
1528 void *
1529 hammer_alloc_record(hammer_cluster_t cluster,
1530                     int *errorp, struct hammer_buffer **bufferp)
1531 {
1532         hammer_buffer_t buffer;
1533         hammer_alist_t live;
1534         int32_t elm_no;
1535         int32_t buf_no;
1536         void *item;
1537
1538         /*
1539          * Allocate a record element
1540          */
1541         live = &cluster->alist_record;
1542         kprintf("IDX_RECORD %d\n", cluster->ondisk->idx_record);
1543         elm_no = hammer_alist_alloc_rev(live, 1, cluster->ondisk->idx_record);
1544         if (elm_no == HAMMER_ALIST_BLOCK_NONE)
1545                 elm_no = hammer_alist_alloc_rev(live, 1,HAMMER_ALIST_BLOCK_MAX);
1546         kprintf("hammer_alloc_record elm %08x\n", elm_no);
1547         if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
1548                 alloc_new_buffer(cluster, live,
1549                                  HAMMER_FSBUF_RECORDS, HAMMER_RECORD_NODES,
1550                                  cluster->ondisk->idx_record, errorp, bufferp);
1551                 elm_no = hammer_alist_alloc_rev(live, 1,HAMMER_ALIST_BLOCK_MAX);
1552                 kprintf("hammer_alloc_record elm again %08x\n", elm_no);
1553                 if (elm_no == HAMMER_ALIST_BLOCK_NONE) {
1554                         *errorp = ENOSPC;
1555                         hammer_modify_cluster(cluster);
1556                         return(NULL);
1557                 }
1558         }
1559         cluster->ondisk->idx_record = elm_no;
1560         hammer_modify_cluster(cluster);
1561
1562         /*
1563          * Load and return the B-Tree element
1564          */
1565         buf_no = elm_no / HAMMER_FSBUF_MAXBLKS;
1566         buffer = *bufferp;
1567         if (buffer == NULL || buffer->cluster != cluster ||
1568             buffer->buf_no != buf_no) {
1569                 if (buffer)
1570                         hammer_rel_buffer(buffer, 0);
1571                 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
1572                 *bufferp = buffer;
1573         }
1574         KKASSERT(buffer->ondisk->head.buf_type == HAMMER_FSBUF_RECORDS);
1575         KKASSERT((elm_no & HAMMER_FSBUF_BLKMASK) < HAMMER_RECORD_NODES);
1576         item = &buffer->ondisk->record.recs[elm_no & HAMMER_FSBUF_BLKMASK];
1577         bzero(item, sizeof(union hammer_record_ondisk));
1578         *errorp = 0;
1579         return(item);
1580 }
1581
1582 /*
1583  * Free HAMMER elements based on either a hammer_buffer and element pointer
1584  * or a cluster-relative byte offset.
1585  */
1586 void
1587 hammer_free_btree_ptr(hammer_buffer_t buffer, hammer_node_ondisk_t node)
1588 {
1589         int32_t elm_no;
1590         hammer_alist_t live;
1591
1592         elm_no = node - &buffer->ondisk->btree.nodes[0];
1593         KKASSERT(elm_no >= 0 && elm_no < HAMMER_BTREE_NODES);
1594         elm_no += buffer->buf_no * HAMMER_FSBUF_MAXBLKS;
1595         live = &buffer->cluster->alist_btree;
1596         hammer_alist_free(live, elm_no, 1);
1597         hammer_modify_cluster(buffer->cluster);
1598 }
1599
1600 void
1601 hammer_free_data_ptr(hammer_buffer_t buffer, void *data, int bytes)
1602 {
1603         int32_t elm_no;
1604         int32_t nblks;
1605         hammer_alist_t live;
1606
1607         if ((bytes & HAMMER_BUFMASK) == 0) {
1608                 nblks = bytes / HAMMER_BUFSIZE;
1609                 KKASSERT(nblks == 1 && data == (void *)buffer->ondisk);
1610                 hammer_alist_free(&buffer->cluster->alist_master,
1611                                   buffer->buf_no, nblks);
1612                 hammer_modify_cluster(buffer->cluster);
1613                 return;
1614         }
1615
1616         elm_no = ((char *)data - (char *)buffer->ondisk->data.data) /
1617                  HAMMER_DATA_BLKSIZE;
1618         KKASSERT(elm_no >= 0 && elm_no < HAMMER_DATA_NODES);
1619         elm_no += buffer->buf_no * HAMMER_FSBUF_MAXBLKS;
1620         nblks = (bytes + HAMMER_DATA_BLKMASK) & ~HAMMER_DATA_BLKMASK;
1621         nblks /= HAMMER_DATA_BLKSIZE;
1622         live = &buffer->cluster->alist_mdata;
1623         hammer_alist_free(live, elm_no, nblks);
1624         hammer_modify_cluster(buffer->cluster);
1625 }
1626
1627 void
1628 hammer_free_record_ptr(hammer_buffer_t buffer, union hammer_record_ondisk *rec)
1629 {
1630         int32_t elm_no;
1631         hammer_alist_t live;
1632
1633         elm_no = rec - &buffer->ondisk->record.recs[0];
1634         KKASSERT(elm_no >= 0 && elm_no < HAMMER_BTREE_NODES);
1635         elm_no += buffer->buf_no * HAMMER_FSBUF_MAXBLKS;
1636         live = &buffer->cluster->alist_record;
1637         hammer_alist_free(live, elm_no, 1);
1638         hammer_modify_cluster(buffer->cluster);
1639 }
1640
1641 void
1642 hammer_free_btree(hammer_cluster_t cluster, int32_t bclu_offset)
1643 {
1644         const int32_t blksize = sizeof(struct hammer_node_ondisk);
1645         int32_t fsbuf_offset = bclu_offset & HAMMER_BUFMASK;
1646         hammer_alist_t live;
1647         int32_t elm_no;
1648
1649         elm_no = bclu_offset / HAMMER_BUFSIZE * HAMMER_FSBUF_MAXBLKS;
1650         fsbuf_offset -= offsetof(union hammer_fsbuf_ondisk, btree.nodes[0]);
1651         live = &cluster->alist_btree;
1652         KKASSERT(fsbuf_offset >= 0 && fsbuf_offset % blksize == 0);
1653         elm_no += fsbuf_offset / blksize;
1654         hammer_alist_free(live, elm_no, 1);
1655         hammer_modify_cluster(cluster);
1656 }
1657
1658 void
1659 hammer_free_data(hammer_cluster_t cluster, int32_t bclu_offset, int32_t bytes)
1660 {
1661         const int32_t blksize = HAMMER_DATA_BLKSIZE;
1662         int32_t fsbuf_offset = bclu_offset & HAMMER_BUFMASK;
1663         hammer_alist_t live;
1664         int32_t elm_no;
1665         int32_t buf_no;
1666         int32_t nblks;
1667
1668         if ((bytes & HAMMER_BUFMASK) == 0) {
1669                 nblks = bytes / HAMMER_BUFSIZE;
1670                 KKASSERT(nblks == 1 && (bclu_offset & HAMMER_BUFMASK) == 0);
1671                 buf_no = bclu_offset / HAMMER_BUFSIZE;
1672                 hammer_alist_free(&cluster->alist_master, buf_no, nblks);
1673                 hammer_modify_cluster(cluster);
1674                 return;
1675         }
1676
1677         elm_no = bclu_offset / HAMMER_BUFSIZE * HAMMER_FSBUF_MAXBLKS;
1678         fsbuf_offset -= offsetof(union hammer_fsbuf_ondisk, data.data[0][0]);
1679         live = &cluster->alist_mdata;
1680         nblks = (bytes + HAMMER_DATA_BLKMASK) & ~HAMMER_DATA_BLKMASK;
1681         nblks /= HAMMER_DATA_BLKSIZE;
1682         KKASSERT(fsbuf_offset >= 0 && fsbuf_offset % blksize == 0);
1683         elm_no += fsbuf_offset / blksize;
1684         hammer_alist_free(live, elm_no, nblks);
1685         hammer_modify_cluster(cluster);
1686 }
1687
1688 void
1689 hammer_free_record(hammer_cluster_t cluster, int32_t bclu_offset)
1690 {
1691         const int32_t blksize = sizeof(union hammer_record_ondisk);
1692         int32_t fsbuf_offset = bclu_offset & HAMMER_BUFMASK;
1693         hammer_alist_t live;
1694         int32_t elm_no;
1695
1696         elm_no = bclu_offset / HAMMER_BUFSIZE * HAMMER_FSBUF_MAXBLKS;
1697         fsbuf_offset -= offsetof(union hammer_fsbuf_ondisk, record.recs[0]);
1698         live = &cluster->alist_record;
1699         KKASSERT(fsbuf_offset >= 0 && fsbuf_offset % blksize == 0);
1700         elm_no += fsbuf_offset / blksize;
1701         hammer_alist_free(live, elm_no, 1);
1702         hammer_modify_cluster(cluster);
1703 }
1704
1705
1706 /*
1707  * Allocate a new filesystem buffer and assign it to the specified
1708  * filesystem buffer type.  The new buffer will be added to the
1709  * type-specific A-list and initialized.
1710  */
1711 static void
1712 alloc_new_buffer(hammer_cluster_t cluster, hammer_alist_t live,
1713                  u_int64_t type, int32_t nelements,
1714                  int start, int *errorp, struct hammer_buffer **bufferp)
1715 {
1716         hammer_buffer_t buffer;
1717         int32_t buf_no;
1718
1719         start = start / HAMMER_FSBUF_MAXBLKS;   /* convert to buf_no */
1720
1721         if (type == HAMMER_FSBUF_RECORDS) {
1722                 buf_no = hammer_alist_alloc_rev(&cluster->alist_master,
1723                                                 1, start);
1724                 if (buf_no == HAMMER_ALIST_BLOCK_NONE) {
1725                         buf_no = hammer_alist_alloc_rev(&cluster->alist_master,
1726                                                 1, HAMMER_ALIST_BLOCK_MAX);
1727                 }
1728         } else {
1729                 buf_no = hammer_alist_alloc_fwd(&cluster->alist_master,
1730                                                 1, start);
1731                 if (buf_no == HAMMER_ALIST_BLOCK_NONE) {
1732                         buf_no = hammer_alist_alloc_fwd(&cluster->alist_master,
1733                                                 1, 0);
1734                 }
1735         }
1736         KKASSERT(buf_no != HAMMER_ALIST_BLOCK_NONE); /* XXX */
1737         hammer_modify_cluster(cluster);
1738
1739         /*
1740          * The new buffer must be initialized (type != 0) regardless of
1741          * whether we already have it cached or not, so don't try to
1742          * optimize the cached buffer check.  Just call hammer_get_buffer().
1743          */
1744         buffer = hammer_get_buffer(cluster, buf_no, type, errorp);
1745         if (*bufferp)
1746                 hammer_rel_buffer(*bufferp, 0);
1747         *bufferp = buffer;
1748
1749         /*
1750          * Finally, do a meta-free of the buffer's elements.
1751          */
1752         if (buffer) {
1753                 kprintf("alloc_new_buffer buf_no %d type %016llx nelms %d\n",
1754                         buf_no, type, nelements);
1755                 hammer_alist_free(live, buf_no * HAMMER_FSBUF_MAXBLKS,
1756                                   nelements);
1757         }
1758 }
1759
1760 #if 0
1761
1762 /*
1763  * Flush various tracking structures to disk
1764  */
1765
1766 /*
1767  * Flush various tracking structures to disk
1768  */
1769 void
1770 flush_all_volumes(void)
1771 {
1772         hammer_volume_t vol;
1773
1774         for (vol = VolBase; vol; vol = vol->next)
1775                 flush_volume(vol);
1776 }
1777
1778 void
1779 flush_volume(hammer_volume_t vol)
1780 {
1781         hammer_supercl_t supercl;
1782         hammer_cluster_t cl;
1783
1784         for (supercl = vol->supercl_base; supercl; supercl = supercl->next)
1785                 flush_supercl(supercl);
1786         for (cl = vol->cluster_base; cl; cl = cl->next)
1787                 flush_cluster(cl);
1788         writehammerbuf(vol, vol->ondisk, 0);
1789 }
1790
1791 void
1792 flush_supercl(hammer_supercl_t supercl)
1793 {
1794         int64_t supercl_offset;
1795
1796         supercl_offset = supercl->scl_offset;
1797         writehammerbuf(supercl->volume, supercl->ondisk, supercl_offset);
1798 }
1799
1800 void
1801 flush_cluster(hammer_cluster_t cl)
1802 {
1803         hammer_buffer_t buf;
1804         int64_t cluster_offset;
1805
1806         for (buf = cl->buffer_base; buf; buf = buf->next)
1807                 flush_buffer(buf);
1808         cluster_offset = cl->clu_offset;
1809         writehammerbuf(cl->volume, cl->ondisk, cluster_offset);
1810 }
1811
1812 void
1813 flush_buffer(hammer_buffer_t buf)
1814 {
1815         int64_t buffer_offset;
1816
1817         buffer_offset = buf->buf_offset + buf->cluster->clu_offset;
1818         writehammerbuf(buf->volume, buf->ondisk, buffer_offset);
1819 }
1820
1821 #endif
1822
1823 /*
1824  * Generic buffer initialization
1825  */
1826 static void
1827 initbuffer(hammer_alist_t live, hammer_fsbuf_head_t head, u_int64_t type)
1828 {
1829         head->buf_type = type;
1830         hammer_alist_init(live);
1831 }
1832
1833 /*
1834  * Calculate the cluster's offset in the volume.  This calculation is
1835  * slightly more complex when using superclusters because superclusters
1836  * are grouped in blocks of 16, followed by 16 x N clusters where N
1837  * is the number of clusters a supercluster can manage.
1838  */
1839 static int64_t
1840 calculate_cluster_offset(hammer_volume_t volume, int32_t clu_no)
1841 {
1842         int32_t scl_group;
1843         int64_t scl_group_size;
1844         int64_t off;
1845
1846         if (volume->vol_flags & HAMMER_VOLF_USINGSUPERCL) {
1847                 scl_group = clu_no / HAMMER_VOL_SUPERCLUSTER_GROUP /
1848                             HAMMER_SCL_MAXCLUSTERS;
1849                 scl_group_size =
1850                             ((int64_t)HAMMER_BUFSIZE *
1851                              HAMMER_VOL_SUPERCLUSTER_GROUP) +
1852                             ((int64_t)HAMMER_VOL_SUPERCLUSTER_GROUP *
1853                              volume->vol_clsize * HAMMER_SCL_MAXCLUSTERS);
1854                 scl_group_size +=
1855                             HAMMER_VOL_SUPERCLUSTER_GROUP * HAMMER_BUFSIZE;
1856
1857                 off = volume->cluster_base +
1858                       scl_group * scl_group_size +
1859                       (HAMMER_BUFSIZE * HAMMER_VOL_SUPERCLUSTER_GROUP) +
1860                       ((int64_t)clu_no % ((int64_t)HAMMER_SCL_MAXCLUSTERS *
1861                        HAMMER_VOL_SUPERCLUSTER_GROUP))
1862                       * HAMMER_BUFSIZE;
1863         } else {
1864                 off = volume->cluster_base +
1865                       (int64_t)clu_no * volume->vol_clsize;
1866         }
1867         return(off);
1868 }
1869
1870 /*
1871  * Calculate a super-cluster's offset in the volume.
1872  */
1873 static int64_t
1874 calculate_supercl_offset(hammer_volume_t volume, int32_t scl_no)
1875 {
1876         int64_t off;
1877         int32_t scl_group;
1878         int64_t scl_group_size;
1879
1880         KKASSERT (volume->vol_flags & HAMMER_VOLF_USINGSUPERCL);
1881         scl_group = scl_no / HAMMER_VOL_SUPERCLUSTER_GROUP;
1882         if (scl_group) {
1883                 scl_group_size =
1884                             ((int64_t)HAMMER_BUFSIZE *
1885                              HAMMER_VOL_SUPERCLUSTER_GROUP) +
1886                             ((int64_t)HAMMER_VOL_SUPERCLUSTER_GROUP *
1887                              volume->vol_clsize * HAMMER_SCL_MAXCLUSTERS);
1888                 scl_group_size +=
1889                             HAMMER_VOL_SUPERCLUSTER_GROUP * HAMMER_BUFSIZE;
1890                 off = volume->cluster_base + (scl_group * scl_group_size) +
1891                       (scl_no % HAMMER_VOL_SUPERCLUSTER_GROUP) * HAMMER_BUFSIZE;
1892         } else {
1893                 off = volume->cluster_base + (scl_no * HAMMER_BUFSIZE);
1894         }
1895         return(off);
1896 }
1897
1898 /*
1899  * A-LIST SUPPORT
1900  *
1901  * Setup the parameters for the various A-lists we use in hammer.  The
1902  * supercluster A-list must be chained to the cluster A-list and cluster
1903  * slave A-lists are chained to buffer A-lists.
1904  *
1905  * See hammer_init_alist_config() below.
1906  */
1907
1908 /*
1909  * A-LIST - cluster recursion into a filesystem buffer
1910  */
1911 static int
1912 buffer_alist_init(void *info, int32_t blk, int32_t radix)
1913 {
1914         hammer_cluster_t cluster = info;
1915         hammer_buffer_t buffer;
1916         int32_t buf_no;
1917         int error = 0;
1918
1919         /*
1920          * Calculate the buffer number, initialize based on the buffer type.
1921          * The buffer has already been allocated so assert that it has been
1922          * initialized.
1923          */
1924         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1925         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1926         if (buffer)
1927                 hammer_rel_buffer(buffer, 0);
1928         return (error);
1929 }
1930
1931 static int
1932 buffer_alist_destroy(void *info, int32_t blk, int32_t radix)
1933 {
1934         return (0);
1935 }
1936
1937 /*
1938  * Note: atblk can be negative and atblk - blk can go negative.
1939  */
1940 static int
1941 buffer_alist_alloc_fwd(void *info, int32_t blk, int32_t radix,
1942                       int32_t count, int32_t atblk, int32_t *fullp)
1943 {
1944         hammer_cluster_t cluster = info;
1945         hammer_buffer_t buffer;
1946         int32_t buf_no;
1947         int32_t r;
1948         int error = 0;
1949
1950         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1951         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1952         if (buffer) {
1953                 KKASSERT(buffer->ondisk->head.buf_type != 0);
1954
1955                 r = hammer_alist_alloc_fwd(&buffer->alist, count, atblk - blk);
1956                 if (r != HAMMER_ALIST_BLOCK_NONE)
1957                         r += blk;
1958                 *fullp = hammer_alist_isfull(&buffer->alist);
1959                 hammer_modify_buffer(buffer);
1960                 hammer_rel_buffer(buffer, 0);
1961         } else {
1962                 r = HAMMER_ALIST_BLOCK_NONE;
1963         }
1964         return(r);
1965 }
1966
1967 static int
1968 buffer_alist_alloc_rev(void *info, int32_t blk, int32_t radix,
1969                       int32_t count, int32_t atblk, int32_t *fullp)
1970 {
1971         hammer_cluster_t cluster = info;
1972         hammer_buffer_t buffer;
1973         int32_t buf_no;
1974         int32_t r;
1975         int error = 0;
1976
1977         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
1978         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
1979         if (buffer) {
1980                 KKASSERT(buffer->ondisk->head.buf_type != 0);
1981
1982                 r = hammer_alist_alloc_rev(&buffer->alist, count, atblk - blk);
1983                 if (r != HAMMER_ALIST_BLOCK_NONE)
1984                         r += blk;
1985                 *fullp = hammer_alist_isfull(&buffer->alist);
1986                 hammer_modify_buffer(buffer);
1987                 hammer_rel_buffer(buffer, 0);
1988         } else {
1989                 r = HAMMER_ALIST_BLOCK_NONE;
1990                 *fullp = 0;
1991         }
1992         return(r);
1993 }
1994
1995 static void
1996 buffer_alist_free(void *info, int32_t blk, int32_t radix,
1997                  int32_t base_blk, int32_t count, int32_t *emptyp)
1998 {
1999         hammer_cluster_t cluster = info;
2000         hammer_buffer_t buffer;
2001         int32_t buf_no;
2002         int error = 0;
2003
2004         buf_no = blk / HAMMER_FSBUF_MAXBLKS;
2005         buffer = hammer_get_buffer(cluster, buf_no, 0, &error);
2006         if (buffer) {
2007                 KKASSERT(buffer->ondisk->head.buf_type != 0);
2008                 hammer_alist_free(&buffer->alist, base_blk, count);
2009                 *emptyp = hammer_alist_isempty(&buffer->alist);
2010                 hammer_modify_buffer(buffer);
2011                 hammer_rel_buffer(buffer, 0);
2012         } else {
2013                 *emptyp = 0;
2014         }
2015 }
2016
2017 static void
2018 buffer_alist_print(void *info, int32_t blk, int32_t radix, int tab)
2019 {
2020 }
2021
2022 /*
2023  * A-LIST - super-cluster recursion into a cluster and cluster recursion
2024  * into a filesystem buffer.  A-List's are mostly self-contained entities,
2025  * but callbacks must be installed to recurse from one A-List to another.
2026  *
2027  * Implementing these callbacks allows us to operate a multi-layered A-List
2028  * as a single entity.
2029  */
2030 static int
2031 super_alist_init(void *info, int32_t blk, int32_t radix)
2032 {
2033         hammer_volume_t volume = info;
2034         hammer_supercl_t supercl;
2035         int32_t scl_no;
2036         int error = 0;
2037
2038         /*
2039          * Calculate the super-cluster number containing the cluster (blk)
2040          * and obtain the super-cluster buffer.
2041          */
2042         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
2043         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
2044         if (supercl)
2045                 hammer_rel_supercl(supercl, 0);
2046         return (error);
2047 }
2048
2049 static int
2050 super_alist_destroy(void *info, int32_t blk, int32_t radix)
2051 {
2052         return(0);
2053 }
2054
2055 static int
2056 super_alist_alloc_fwd(void *info, int32_t blk, int32_t radix,
2057                       int32_t count, int32_t atblk, int32_t *fullp)
2058 {
2059         hammer_volume_t volume = info;
2060         hammer_supercl_t supercl;
2061         int32_t scl_no;
2062         int32_t r;
2063         int error = 0;
2064
2065         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
2066         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
2067         if (supercl) {
2068                 r = hammer_alist_alloc_fwd(&supercl->alist, count, atblk - blk);
2069                 if (r != HAMMER_ALIST_BLOCK_NONE)
2070                         r += blk;
2071                 *fullp = hammer_alist_isfull(&supercl->alist);
2072                 hammer_modify_supercl(supercl);
2073                 hammer_rel_supercl(supercl, 0);
2074         } else {
2075                 r = HAMMER_ALIST_BLOCK_NONE;
2076                 *fullp = 0;
2077         }
2078         return(r);
2079 }
2080
2081 static int
2082 super_alist_alloc_rev(void *info, int32_t blk, int32_t radix,
2083                       int32_t count, int32_t atblk, int32_t *fullp)
2084 {
2085         hammer_volume_t volume = info;
2086         hammer_supercl_t supercl;
2087         int32_t scl_no;
2088         int32_t r;
2089         int error = 0;
2090
2091         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
2092         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
2093         if (supercl) {
2094                 r = hammer_alist_alloc_rev(&supercl->alist, count, atblk - blk);
2095                 if (r != HAMMER_ALIST_BLOCK_NONE)
2096                         r += blk;
2097                 *fullp = hammer_alist_isfull(&supercl->alist);
2098                 hammer_modify_supercl(supercl);
2099                 hammer_rel_supercl(supercl, 0);
2100         } else {
2101                 r = HAMMER_ALIST_BLOCK_NONE;
2102                 *fullp = 0;
2103         }
2104         return(r);
2105 }
2106
2107 static void
2108 super_alist_free(void *info, int32_t blk, int32_t radix,
2109                  int32_t base_blk, int32_t count, int32_t *emptyp)
2110 {
2111         hammer_volume_t volume = info;
2112         hammer_supercl_t supercl;
2113         int32_t scl_no;
2114         int error = 0;
2115
2116         scl_no = blk / HAMMER_SCL_MAXCLUSTERS;
2117         supercl = hammer_get_supercl(volume, scl_no, &error, 1);
2118         if (supercl) {
2119                 hammer_alist_free(&supercl->alist, base_blk, count);
2120                 *emptyp = hammer_alist_isempty(&supercl->alist);
2121                 hammer_modify_supercl(supercl);
2122                 hammer_rel_supercl(supercl, 0);
2123         } else {
2124                 *emptyp = 0;
2125         }
2126 }
2127
2128 static void
2129 super_alist_print(void *info, int32_t blk, int32_t radix, int tab)
2130 {
2131 }
2132
2133 void
2134 hammer_init_alist_config(void)
2135 {
2136         hammer_alist_config_t config;
2137
2138         hammer_alist_template(&Buf_alist_config, HAMMER_FSBUF_MAXBLKS,
2139                               1, HAMMER_FSBUF_METAELMS);
2140         hammer_alist_template(&Vol_normal_alist_config, HAMMER_VOL_MAXCLUSTERS,
2141                               1, HAMMER_VOL_METAELMS_1LYR);
2142         hammer_alist_template(&Vol_super_alist_config,
2143                           HAMMER_VOL_MAXSUPERCLUSTERS * HAMMER_SCL_MAXCLUSTERS,
2144                               HAMMER_SCL_MAXCLUSTERS, HAMMER_VOL_METAELMS_2LYR);
2145         hammer_alist_template(&Supercl_alist_config, HAMMER_VOL_MAXCLUSTERS,
2146                               1, HAMMER_SUPERCL_METAELMS);
2147         hammer_alist_template(&Clu_master_alist_config, HAMMER_CLU_MAXBUFFERS,
2148                               1, HAMMER_CLU_MASTER_METAELMS);
2149         hammer_alist_template(&Clu_slave_alist_config,
2150                               HAMMER_CLU_MAXBUFFERS * HAMMER_FSBUF_MAXBLKS,
2151                               HAMMER_FSBUF_MAXBLKS, HAMMER_CLU_SLAVE_METAELMS);
2152
2153         config = &Vol_super_alist_config;
2154         config->bl_radix_init = super_alist_init;
2155         config->bl_radix_destroy = super_alist_destroy;
2156         config->bl_radix_alloc_fwd = super_alist_alloc_fwd;
2157         config->bl_radix_alloc_rev = super_alist_alloc_rev;
2158         config->bl_radix_free = super_alist_free;
2159         config->bl_radix_print = super_alist_print;
2160
2161         config = &Clu_slave_alist_config;
2162         config->bl_radix_init = buffer_alist_init;
2163         config->bl_radix_destroy = buffer_alist_destroy;
2164         config->bl_radix_alloc_fwd = buffer_alist_alloc_fwd;
2165         config->bl_radix_alloc_rev = buffer_alist_alloc_rev;
2166         config->bl_radix_free = buffer_alist_free;
2167         config->bl_radix_print = buffer_alist_print;
2168 }
2169