sys/vfs/hammer/hammer_object.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_object.c,v 1.28 2008/02/05 07:58:43 dillon Exp $
  35  */
  36
  37 #include "hammer.h"
  38
  39 static int hammer_mem_add(hammer_transaction_t trans,
  40                              hammer_record_t record);
  41 static int hammer_mem_lookup(hammer_cursor_t cursor, hammer_inode_t ip);
  42 static int hammer_mem_first(hammer_cursor_t cursor, hammer_inode_t ip);
  43
  44 /*
  45  * Red-black tree support.
  46  */
  47 static int
  48 hammer_rec_rb_compare(hammer_record_t rec1, hammer_record_t rec2)
  49 {
  50         if (rec1->rec.base.base.rec_type < rec2->rec.base.base.rec_type)
  51                 return(-1);
  52         if (rec1->rec.base.base.rec_type > rec2->rec.base.base.rec_type)
  53                 return(1);
  54
  55         if (rec1->rec.base.base.key < rec2->rec.base.base.key)
  56                 return(-1);
  57         if (rec1->rec.base.base.key > rec2->rec.base.base.key)
  58                 return(1);
  59
  60         if (rec1->rec.base.base.create_tid == 0) {
  61                 if (rec2->rec.base.base.create_tid == 0)
  62                         return(0);
  63                 return(1);
  64         }
  65         if (rec2->rec.base.base.create_tid == 0)
  66                 return(-1);
  67
  68         if (rec1->rec.base.base.create_tid < rec2->rec.base.base.create_tid)
  69                 return(-1);
  70         if (rec1->rec.base.base.create_tid > rec2->rec.base.base.create_tid)
  71                 return(1);
  72         return(0);
  73 }
  74
  75 static int
  76 hammer_rec_compare(hammer_base_elm_t info, hammer_record_t rec)
  77 {
  78         if (info->rec_type < rec->rec.base.base.rec_type)
  79                 return(-3);
  80         if (info->rec_type > rec->rec.base.base.rec_type)
  81                 return(3);
  82
  83         if (info->key < rec->rec.base.base.key)
  84                 return(-2);
  85         if (info->key > rec->rec.base.base.key)
  86                 return(2);
  87
  88         if (info->create_tid == 0) {
  89                 if (rec->rec.base.base.create_tid == 0)
  90                         return(0);
  91                 return(1);
  92         }
  93         if (rec->rec.base.base.create_tid == 0)
  94                 return(-1);
  95         if (info->create_tid < rec->rec.base.base.create_tid)
  96                 return(-1);
  97         if (info->create_tid > rec->rec.base.base.create_tid)
  98                 return(1);
  99         return(0);
 100 }
 101
 102 /*
 103  * RB_SCAN comparison code for hammer_mem_first().  The argument order
 104  * is reversed so the comparison result has to be negated.  key_beg and
 105  * key_end are both range-inclusive.
 106  *
 107  * The creation timestamp can cause hammer_rec_compare() to return -1 or +1.
 108  * These do not stop the scan.
 109  *
 110  * Localized deletions are not cached in-memory.
 111  */
 112 static
 113 int
 114 hammer_rec_scan_cmp(hammer_record_t rec, void *data)
 115 {
 116         hammer_cursor_t cursor = data;
 117         int r;
 118
 119         r = hammer_rec_compare(&cursor->key_beg, rec);
 120         if (r > 1)
 121                 return(-1);
 122         r = hammer_rec_compare(&cursor->key_end, rec);
 123         if (r < -1)
 124                 return(1);
 125         return(0);
 126 }
 127
 128 RB_GENERATE(hammer_rec_rb_tree, hammer_record, rb_node, hammer_rec_rb_compare);
 129 RB_GENERATE_XLOOKUP(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
 130                     hammer_rec_compare, hammer_base_elm_t);
 131
 132 /*
 133  * Allocate a record for the caller to finish filling in.  The record is
 134  * returned referenced.
 135  */
 136 hammer_record_t
 137 hammer_alloc_mem_record(hammer_inode_t ip)
 138 {
 139         hammer_record_t record;
 140
 141         ++hammer_count_records;
 142         record = kmalloc(sizeof(*record), M_HAMMER, M_WAITOK|M_ZERO);
 143         record->ip = ip;
 144         record->rec.base.base.btype = HAMMER_BTREE_TYPE_RECORD;
 145         hammer_ref(&record->lock);
 146         return (record);
 147 }
 148
 149 /*
 150  * Release a memory record.  Records marked for deletion are immediately
 151  * removed from the RB-Tree but otherwise left intact until the last ref
 152  * goes away.
 153  */
 154 void
 155 hammer_rel_mem_record(struct hammer_record *record)
 156 {
 157         hammer_unref(&record->lock);
 158
 159         if (record->flags & HAMMER_RECF_DELETED) {
 160                 if (record->flags & HAMMER_RECF_ONRBTREE) {
 161                         RB_REMOVE(hammer_rec_rb_tree, &record->ip->rec_tree,
 162                                   record);
 163                         record->flags &= ~HAMMER_RECF_ONRBTREE;
 164                 }
 165                 if (record->lock.refs == 0) {
 166                         if (record->flags & HAMMER_RECF_ALLOCDATA) {
 167                                 --hammer_count_record_datas;
 168                                 kfree(record->data, M_HAMMER);
 169                                 record->flags &= ~HAMMER_RECF_ALLOCDATA;
 170                         }
 171                         record->data = NULL;
 172                         --hammer_count_records;
 173                         kfree(record, M_HAMMER);
 174                         return;
 175                 }
 176         }
 177
 178         /*
 179          * If someone wanted the record wake them up.
 180          */
 181         if (record->flags & HAMMER_RECF_WANTED) {
 182                 record->flags &= ~HAMMER_RECF_WANTED;
 183                 wakeup(record);
 184         }
 185 }
 186
 187 /*
 188  * Lookup an in-memory record given the key specified in the cursor.  Works
 189  * just like hammer_btree_lookup() but operates on an inode's in-memory
 190  * record list.
 191  *
 192  * The lookup must fail if the record is marked for deferred deletion.
 193  */
 194 static
 195 int
 196 hammer_mem_lookup(hammer_cursor_t cursor, hammer_inode_t ip)
 197 {
 198         int error;
 199
 200         if (cursor->iprec) {
 201                 hammer_rel_mem_record(cursor->iprec);
 202                 cursor->iprec = NULL;
 203         }
 204         if (cursor->ip) {
 205                 hammer_rec_rb_tree_scan_info_done(&cursor->scan,
 206                                                   &cursor->ip->rec_tree);
 207         }
 208         cursor->ip = ip;
 209         hammer_rec_rb_tree_scan_info_link(&cursor->scan, &ip->rec_tree);
 210         cursor->scan.node = NULL;
 211         cursor->iprec = hammer_rec_rb_tree_RB_LOOKUP_INFO(
 212                                 &ip->rec_tree, &cursor->key_beg);
 213         if (cursor->iprec == NULL) {
 214                 error = ENOENT;
 215         } else {
 216                 hammer_ref(&cursor->iprec->lock);
 217                 error = 0;
 218         }
 219         return(error);
 220 }
 221
 222 /*
 223  * hammer_mem_first() - locate the first in-memory record matching the
 224  * cursor.
 225  *
 226  * The RB_SCAN function we use is designed as a callback.  We terminate it
 227  * (return -1) as soon as we get a match.
 228  */
 229 static
 230 int
 231 hammer_rec_scan_callback(hammer_record_t rec, void *data)
 232 {
 233         hammer_cursor_t cursor = data;
 234
 235         /*
 236          * We terminate on success, so this should be NULL on entry.
 237          */
 238         KKASSERT(cursor->iprec == NULL);
 239
 240         /*
 241          * Skip if the record was marked deleted
 242          */
 243         if (rec->flags & HAMMER_RECF_DELETED)
 244                 return(0);
 245
 246         /*
 247          * Skip if not visible due to our as-of TID
 248          */
 249         if (cursor->flags & HAMMER_CURSOR_ASOF) {
 250                 if (cursor->asof < rec->rec.base.base.create_tid)
 251                         return(0);
 252                 if (rec->rec.base.base.delete_tid &&
 253                     cursor->asof >= rec->rec.base.base.delete_tid) {
 254                         return(0);
 255                 }
 256         }
 257
 258         /*
 259          * Block if currently being synchronized to disk, otherwise we
 260          * may get a duplicate.  Wakeup the syncer if it's stuck on
 261          * the record.
 262          */
 263         hammer_ref(&rec->lock);
 264         ++rec->blocked;
 265         while (rec->flags & HAMMER_RECF_SYNCING) {
 266                 rec->flags |= HAMMER_RECF_WANTED;
 267                 tsleep(rec, 0, "hmrrc2", 0);
 268         }
 269         --rec->blocked;
 270
 271         /*
 272          * The record may have been deleted while we were blocked.
 273          */
 274         if (rec->flags & HAMMER_RECF_DELETED) {
 275                 hammer_rel_mem_record(cursor->iprec);
 276                 return(0);
 277         }
 278
 279         /*
 280          * Set the matching record and stop the scan.
 281          */
 282         cursor->iprec = rec;
 283         return(-1);
 284 }
 285
 286 static
 287 int
 288 hammer_mem_first(hammer_cursor_t cursor, hammer_inode_t ip)
 289 {
 290         if (cursor->iprec) {
 291                 hammer_rel_mem_record(cursor->iprec);
 292                 cursor->iprec = NULL;
 293         }
 294         if (cursor->ip) {
 295                 hammer_rec_rb_tree_scan_info_done(&cursor->scan,
 296                                                   &cursor->ip->rec_tree);
 297         }
 298         cursor->ip = ip;
 299         hammer_rec_rb_tree_scan_info_link(&cursor->scan, &ip->rec_tree);
 300
 301         cursor->scan.node = NULL;
 302         hammer_rec_rb_tree_RB_SCAN(&ip->rec_tree, hammer_rec_scan_cmp,
 303                                    hammer_rec_scan_callback, cursor);
 304
 305         /*
 306          * Adjust scan.node and keep it linked into the RB-tree so we can
 307          * hold the cursor through third party modifications of the RB-tree.
 308          */
 309         if (cursor->iprec) {
 310                 cursor->scan.node = hammer_rec_rb_tree_RB_NEXT(cursor->iprec);
 311                 return(0);
 312         }
 313         return(ENOENT);
 314 }
 315
 316 void
 317 hammer_mem_done(hammer_cursor_t cursor)
 318 {
 319         if (cursor->ip) {
 320                 hammer_rec_rb_tree_scan_info_done(&cursor->scan,
 321                                                   &cursor->ip->rec_tree);
 322                 cursor->ip = NULL;
 323         }
 324         if (cursor->iprec) {
 325                 hammer_rel_mem_record(cursor->iprec);
 326                 cursor->iprec = NULL;
 327         }
 328 }
 329
 330 /************************************************************************
 331  *                   HAMMER IN-MEMORY RECORD FUNCTIONS                  *
 332  ************************************************************************
 333  *
 334  * These functions manipulate in-memory records.  Such records typically
 335  * exist prior to being committed to disk or indexed via the on-disk B-Tree.
 336  */
 337
 338 /*
 339  * Add a directory entry (dip,ncp) which references inode (ip).
 340  *
 341  * Note that the low 32 bits of the namekey are set temporarily to create
 342  * a unique in-memory record, and may be modified a second time when the
 343  * record is synchronized to disk.  In particular, the low 32 bits cannot be
 344  * all 0's when synching to disk, which is not handled here.
 345  */
 346 int
 347 hammer_ip_add_directory(struct hammer_transaction *trans,
 348                      struct hammer_inode *dip, struct namecache *ncp,
 349                      struct hammer_inode *ip)
 350 {
 351         hammer_record_t record;
 352         int error;
 353         int bytes;
 354
 355         record = hammer_alloc_mem_record(dip);
 356
 357         bytes = ncp->nc_nlen;   /* NOTE: terminating \0 is NOT included */
 358         if (++trans->hmp->namekey_iterator == 0)
 359                 ++trans->hmp->namekey_iterator;
 360
 361         record->rec.entry.base.base.obj_id = dip->obj_id;
 362         record->rec.entry.base.base.key =
 363                 hammer_directory_namekey(ncp->nc_name, bytes);
 364         record->rec.entry.base.base.key += trans->hmp->namekey_iterator;
 365         record->rec.entry.base.base.create_tid = trans->tid;
 366         record->rec.entry.base.base.rec_type = HAMMER_RECTYPE_DIRENTRY;
 367         record->rec.entry.base.base.obj_type = ip->ino_rec.base.base.obj_type;
 368         record->rec.entry.obj_id = ip->obj_id;
 369         if (bytes <= sizeof(record->rec.entry.den_name)) {
 370                 record->data = (void *)record->rec.entry.den_name;
 371                 record->flags |= HAMMER_RECF_EMBEDDED_DATA;
 372         } else {
 373                 ++hammer_count_record_datas;
 374                 record->data = kmalloc(bytes, M_HAMMER, M_WAITOK);
 375                 record->flags |= HAMMER_RECF_ALLOCDATA;
 376         }
 377         bcopy(ncp->nc_name, record->data, bytes);
 378         record->rec.entry.base.data_len = bytes;
 379         ++ip->ino_rec.ino_nlinks;
 380         hammer_modify_inode(trans, ip, HAMMER_INODE_RDIRTY);
 381         error = hammer_mem_add(trans, record);
 382         return(error);
 383 }
 384
 385 /*
 386  * Delete the directory entry and update the inode link count.  The
 387  * cursor must be seeked to the directory entry record being deleted.
 388  *
 389  * NOTE: HAMMER_CURSOR_DELETE may not have been set.  XXX remove flag.
 390  *
 391  * This function can return EDEADLK requiring the caller to terminate
 392  * the cursor and retry.
 393  */
 394 int
 395 hammer_ip_del_directory(struct hammer_transaction *trans,
 396                      hammer_cursor_t cursor, struct hammer_inode *dip,
 397                      struct hammer_inode *ip)
 398 {
 399         int error;
 400
 401         error = hammer_ip_delete_record(cursor, trans->tid);
 402
 403         /*
 404          * One less link.  The file may still be open in the OS even after
 405          * all links have gone away so we only try to sync if the OS has
 406          * no references and nlinks falls to 0.
 407          *
 408          * We have to terminate the cursor before syncing the inode to
 409          * avoid deadlocking against ourselves.
 410          */
 411         if (error == 0) {
 412                 --ip->ino_rec.ino_nlinks;
 413                 hammer_modify_inode(trans, ip, HAMMER_INODE_RDIRTY);
 414                 if (ip->ino_rec.ino_nlinks == 0 &&
 415                     (ip->vp == NULL || (ip->vp->v_flag & VINACTIVE))) {
 416                         hammer_done_cursor(cursor);
 417                         hammer_sync_inode(ip, MNT_NOWAIT, 1);
 418                 }
 419
 420         }
 421         return(error);
 422 }
 423
 424 /*
 425  * Add a record to an inode.
 426  *
 427  * The caller must allocate the record with hammer_alloc_mem_record(ip) and
 428  * initialize the following additional fields:
 429  *
 430  * record->rec.entry.base.base.key
 431  * record->rec.entry.base.base.rec_type
 432  * record->rec.entry.base.base.data_len
 433  * record->data         (a copy will be kmalloc'd if not embedded)
 434  */
 435 int
 436 hammer_ip_add_record(struct hammer_transaction *trans, hammer_record_t record)
 437 {
 438         hammer_inode_t ip = record->ip;
 439         int error;
 440         int bytes;
 441         void *data;
 442
 443         record->rec.base.base.obj_id = ip->obj_id;
 444         record->rec.base.base.create_tid = trans->tid;
 445         record->rec.base.base.obj_type = ip->ino_rec.base.base.obj_type;
 446         bytes = record->rec.base.data_len;
 447
 448         if (record->data) {
 449                 if ((char *)record->data < (char *)&record->rec ||
 450                     (char *)record->data >= (char *)(&record->rec + 1)) {
 451                         ++hammer_count_record_datas;
 452                         data = kmalloc(bytes, M_HAMMER, M_WAITOK);
 453                         record->flags |= HAMMER_RECF_ALLOCDATA;
 454                         bcopy(record->data, data, bytes);
 455                         record->data = data;
 456                 } else {
 457                         record->flags |= HAMMER_RECF_EMBEDDED_DATA;
 458                 }
 459         }
 460         hammer_modify_inode(trans, ip, HAMMER_INODE_RDIRTY);
 461         error = hammer_mem_add(trans, record);
 462         return(error);
 463 }
 464
 465 /*
 466  * Sync data from a buffer cache buffer (typically) to the filesystem.  This
 467  * is called via the strategy called from a cached data source.  This code
 468  * is responsible for actually writing a data record out to the disk.
 469  *
 470  * This can only occur non-historically (i.e. 'current' data only).
 471  */
 472 int
 473 hammer_ip_sync_data(hammer_transaction_t trans, hammer_inode_t ip,
 474                        int64_t offset, void *data, int bytes,
 475                        struct hammer_cursor **spike)
 476 {
 477         struct hammer_cursor cursor;
 478         hammer_record_ondisk_t rec;
 479         union hammer_btree_elm elm;
 480         void *bdata;
 481         int error;
 482
 483         KKASSERT((offset & HAMMER_BUFMASK) == 0);
 484         KKASSERT((bytes & HAMMER_BUFMASK) == 0);
 485 retry:
 486         error = hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
 487         if (error)
 488                 return(error);
 489         cursor.key_beg.obj_id = ip->obj_id;
 490         cursor.key_beg.key = offset + bytes;
 491         cursor.key_beg.create_tid = trans->tid;
 492         cursor.key_beg.delete_tid = 0;
 493         cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
 494         cursor.asof = trans->tid;
 495         cursor.flags |= HAMMER_CURSOR_INSERT;
 496
 497         /*
 498          * Issue a lookup to position the cursor and locate the cluster
 499          */
 500         error = hammer_btree_lookup(&cursor);
 501         if (error == 0) {
 502                 kprintf("hammer_ip_sync_data: duplicate data at (%lld,%d)\n",
 503                         offset, bytes);
 504                 hammer_print_btree_elm(&cursor.node->ondisk->elms[cursor.index],
 505                                        HAMMER_BTREE_TYPE_LEAF, cursor.index);
 506                 error = EIO;
 507         }
 508         if (error != ENOENT)
 509                 goto done;
 510
 511         /*
 512          * Allocate record and data space now that we know which cluster
 513          * the B-Tree node ended up in.
 514          */
 515         bdata = hammer_alloc_data(cursor.node->cluster, bytes, &error,
 516                                   &cursor.data_buffer);
 517         if (bdata == NULL)
 518                 goto done;
 519         rec = hammer_alloc_record(cursor.node->cluster, &error,
 520                                   HAMMER_RECTYPE_DATA, &cursor.record_buffer);
 521         if (rec == NULL)
 522                 goto fail1;
 523
 524         /*
 525          * Fill everything in and insert our B-Tree node.
 526          */
 527         hammer_modify_buffer(cursor.record_buffer);
 528         rec->base.base.btype = HAMMER_BTREE_TYPE_RECORD;
 529         rec->base.base.obj_id = ip->obj_id;
 530         rec->base.base.key = offset + bytes;
 531         rec->base.base.create_tid = trans->tid;
 532         rec->base.base.delete_tid = 0;
 533         rec->base.base.rec_type = HAMMER_RECTYPE_DATA;
 534         rec->base.data_crc = crc32(data, bytes);
 535         rec->base.rec_id = hammer_alloc_recid(cursor.node->cluster);
 536         rec->base.data_offset = hammer_bclu_offset(cursor.data_buffer, bdata);
 537         rec->base.data_len = bytes;
 538
 539         hammer_modify_buffer(cursor.data_buffer);
 540         bcopy(data, bdata, bytes);
 541
 542         elm.leaf.base = rec->base.base;
 543         elm.leaf.rec_offset = hammer_bclu_offset(cursor.record_buffer, rec);
 544         elm.leaf.data_offset = rec->base.data_offset;
 545         elm.leaf.data_len = bytes;
 546         elm.leaf.data_crc = rec->base.data_crc;
 547
 548         /*
 549          * Data records can wind up on-disk before the inode itself is
 550          * on-disk.  One must assume data records may be on-disk if either
 551          * HAMMER_INODE_DONDISK or HAMMER_INODE_ONDISK is set
 552          */
 553         ip->flags |= HAMMER_INODE_DONDISK;
 554
 555         error = hammer_btree_insert(&cursor, &elm);
 556         if (error == 0)
 557                 goto done;
 558
 559         hammer_free_record_ptr(cursor.record_buffer, rec, HAMMER_RECTYPE_DATA);
 560 fail1:
 561         hammer_free_data_ptr(cursor.data_buffer, bdata, bytes);
 562 done:
 563         /*
 564          * If ENOSPC in cluster fill in the spike structure and return
 565          * ENOSPC.
 566          */
 567         if (error == ENOSPC)
 568                 hammer_load_spike(&cursor, spike);
 569         hammer_done_cursor(&cursor);
 570         if (error == EDEADLK)
 571                 goto retry;
 572         return(error);
 573 }
 574
 575 /*
 576  * Sync an in-memory record to the disk.  this is typically called via fsync
 577  * from a cached record source.  This code is responsible for actually
 578  * writing a record out to the disk.
 579  */
 580 int
 581 hammer_ip_sync_record(hammer_record_t record, struct hammer_cursor **spike)
 582 {
 583         struct hammer_cursor cursor;
 584         hammer_record_ondisk_t rec;
 585         hammer_mount_t hmp;
 586         union hammer_btree_elm elm;
 587         void *bdata;
 588         int error;
 589
 590 retry:
 591         /*
 592          * If the record has been deleted or is being synchronized, stop.
 593          * Interlock with the syncing flag.
 594          */
 595         if (record->flags & (HAMMER_RECF_DELETED | HAMMER_RECF_SYNCING))
 596                 return(0);
 597         record->flags |= HAMMER_RECF_SYNCING;
 598
 599         /*
 600          * If someone other then us is referencing the record and not
 601          * blocking waiting for us, we have to wait until they finish.
 602          *
 603          * It is possible the record got destroyed while we were blocked.
 604          */
 605         if (record->lock.refs > record->blocked + 1) {
 606                 while (record->lock.refs > record->blocked + 1) {
 607                         record->flags |= HAMMER_RECF_WANTED;
 608                         tsleep(record, 0, "hmrrc1", 0);
 609                 }
 610                 if (record->flags & HAMMER_RECF_DELETED)
 611                         return(0);
 612         }
 613
 614         /*
 615          * Get a cursor
 616          */
 617         error = hammer_init_cursor_hmp(&cursor, &record->ip->cache[0],
 618                                        record->ip->hmp);
 619         if (error)
 620                 return(error);
 621         cursor.key_beg = record->rec.base.base;
 622         cursor.flags |= HAMMER_CURSOR_INSERT;
 623
 624         /*
 625          * Issue a lookup to position the cursor and locate the cluster.  The
 626          * target key should not exist.  If we are creating a directory entry
 627          * we may have to iterate the low 32 bits of the key to find an unused
 628          * key.
 629          */
 630         for (;;) {
 631                 error = hammer_btree_lookup(&cursor);
 632                 if (error)
 633                         break;
 634                 if (record->rec.base.base.rec_type != HAMMER_RECTYPE_DIRENTRY) {
 635                         kprintf("hammer_ip_sync_record: duplicate rec "
 636                                 "at (%016llx)\n", record->rec.base.base.key);
 637                         Debugger("duplicate record1");
 638                         error = EIO;
 639                         break;
 640                 }
 641                 hmp = cursor.node->cluster->volume->hmp;
 642                 if (++hmp->namekey_iterator == 0)
 643                         ++hmp->namekey_iterator;
 644                 record->rec.base.base.key &= ~(0xFFFFFFFFLL);
 645                 record->rec.base.base.key |= hmp->namekey_iterator;
 646                 cursor.key_beg.key = record->rec.base.base.key;
 647         }
 648         if (error != ENOENT)
 649                 goto done;
 650
 651         /*
 652          * Mark the record as undergoing synchronization.  Our cursor is
 653          * holding a locked B-Tree node for the insertion which interlocks
 654          * anyone trying to access this record.
 655          *
 656          * XXX There is still a race present related to iterations.  An
 657          * iteration may process the record, a sync may occur, and then
 658          * later process the B-Tree element for the same record.
 659          *
 660          * We do not try to synchronize a deleted record.
 661          */
 662         if (record->flags & HAMMER_RECF_DELETED) {
 663                 error = 0;
 664                 goto done;
 665         }
 666
 667         /*
 668          * Allocate record and data space now that we know which cluster
 669          * the B-Tree node ended up in.
 670          */
 671         if (record->data == NULL ||
 672             (record->flags & HAMMER_RECF_EMBEDDED_DATA)) {
 673                 bdata = record->data;
 674         } else {
 675                 bdata = hammer_alloc_data(cursor.node->cluster,
 676                                           record->rec.base.data_len, &error,
 677                                           &cursor.data_buffer);
 678                 if (bdata == NULL)
 679                         goto done;
 680         }
 681         rec = hammer_alloc_record(cursor.node->cluster, &error,
 682                                   record->rec.base.base.rec_type,
 683                                   &cursor.record_buffer);
 684         if (rec == NULL)
 685                 goto fail1;
 686
 687         /*
 688          * Fill everything in and insert our B-Tree node.
 689          */
 690         hammer_modify_buffer(cursor.record_buffer);
 691         *rec = record->rec;
 692         if (bdata) {
 693                 rec->base.data_crc = crc32(record->data,
 694                                            record->rec.base.data_len);
 695                 if (record->flags & HAMMER_RECF_EMBEDDED_DATA) {
 696                         /*
 697                          * Data embedded in record
 698                          */
 699                         rec->base.data_offset = ((char *)bdata -
 700                                                  (char *)&record->rec);
 701                         KKASSERT(rec->base.data_offset >= 0 &&
 702                                  rec->base.data_offset + rec->base.data_len <=
 703                                   sizeof(*rec));
 704                         rec->base.data_offset += hammer_bclu_offset(cursor.record_buffer, rec);
 705                 } else {
 706                         /*
 707                          * Data separate from record
 708                          */
 709                         rec->base.data_offset = hammer_bclu_offset(cursor.data_buffer,bdata);
 710                         hammer_modify_buffer(cursor.data_buffer);
 711                         bcopy(record->data, bdata, rec->base.data_len);
 712                 }
 713         }
 714         rec->base.rec_id = hammer_alloc_recid(cursor.node->cluster);
 715
 716         elm.leaf.base = record->rec.base.base;
 717         elm.leaf.rec_offset = hammer_bclu_offset(cursor.record_buffer, rec);
 718         elm.leaf.data_offset = rec->base.data_offset;
 719         elm.leaf.data_len = rec->base.data_len;
 720         elm.leaf.data_crc = rec->base.data_crc;
 721
 722         error = hammer_btree_insert(&cursor, &elm);
 723
 724         /*
 725          * Clean up on success, or fall through on error.
 726          */
 727         if (error == 0) {
 728                 record->flags |= HAMMER_RECF_DELETED;
 729                 goto done;
 730         }
 731
 732         hammer_free_record_ptr(cursor.record_buffer, rec,
 733                                record->rec.base.base.rec_type);
 734 fail1:
 735         if (record->data && (record->flags & HAMMER_RECF_EMBEDDED_DATA) == 0) {
 736                 hammer_free_data_ptr(cursor.data_buffer, bdata,
 737                                      record->rec.base.data_len);
 738         }
 739 done:
 740         record->flags &= ~HAMMER_RECF_SYNCING;
 741         /*
 742          * If ENOSPC in cluster fill in the spike structure and return
 743          * ENOSPC.
 744          */
 745         if (error == ENOSPC)
 746                 hammer_load_spike(&cursor, spike);
 747         hammer_done_cursor(&cursor);
 748         if (error == EDEADLK)
 749                 goto retry;
 750         return(error);
 751 }
 752
 753 /*
 754  * Write out a record using the specified cursor.  The caller does not have
 755  * to seek the cursor.  The flags are used to determine whether the data
 756  * (if any) is embedded in the record or not.
 757  *
 758  * The cursor must be initialized, including its key_beg element.  The B-Tree
 759  * key may be different from the base element in the record (e.g. for spikes).
 760  *
 761  * NOTE: This can return EDEADLK, requiring the caller to release its cursor
 762  * and retry the operation.
 763  */
 764 int
 765 hammer_write_record(hammer_cursor_t cursor, hammer_record_ondisk_t orec,
 766                     void *data, int cursor_flags)
 767 {
 768         union hammer_btree_elm elm;
 769         hammer_record_ondisk_t nrec;
 770         hammer_cluster_t ncluster;
 771         hammer_volume_t nvolume;
 772         int32_t nrec_offset;
 773         void *bdata;
 774         int error;
 775
 776         KKASSERT(cursor->flags & HAMMER_CURSOR_INSERT);
 777
 778         /*
 779          * Issue a lookup to position the cursor and locate the cluster.  The
 780          * target key should not exist.
 781          *
 782          * If we run out of space trying to adjust the B-Tree for the
 783          * insert, re-lookup without the insert flag so the cursor
 784          * is properly positioned for the spike.
 785          */
 786         error = hammer_btree_lookup(cursor);
 787         if (error == 0) {
 788                 kprintf("hammer_ip_sync_record: duplicate rec at (%016llx)\n",
 789                         orec->base.base.key);
 790                 Debugger("duplicate record2");
 791                 error = EIO;
 792         }
 793         if (error != ENOENT)
 794                 goto done;
 795
 796         /*
 797          * Allocate record and data space now that we know which cluster
 798          * the B-Tree node ended up in.
 799          */
 800         if (data == NULL ||
 801             (cursor_flags & HAMMER_RECF_EMBEDDED_DATA)) {
 802                 bdata = data;
 803         } else {
 804                 bdata = hammer_alloc_data(cursor->node->cluster,
 805                                           orec->base.data_len, &error,
 806                                           &cursor->data_buffer);
 807                 if (bdata == NULL)
 808                         goto done;
 809         }
 810         nrec = hammer_alloc_record(cursor->node->cluster, &error,
 811                                   orec->base.base.rec_type,
 812                                   &cursor->record_buffer);
 813         if (nrec == NULL)
 814                 goto fail1;
 815
 816         /*
 817          * Fill everything in and insert our B-Tree node.
 818          */
 819         hammer_modify_buffer(cursor->record_buffer);
 820         *nrec = *orec;
 821         nrec->base.data_offset = 0;
 822         if (bdata) {
 823                 nrec->base.data_crc = crc32(bdata, nrec->base.data_len);
 824                 if (cursor_flags & HAMMER_RECF_EMBEDDED_DATA) {
 825                         /*
 826                          * Data embedded in record
 827                          */
 828                         nrec->base.data_offset = ((char *)bdata - (char *)orec);
 829                         KKASSERT(nrec->base.data_offset >= 0 &&
 830                                  nrec->base.data_offset + nrec->base.data_len <
 831                                   sizeof(*nrec));
 832                         nrec->base.data_offset += hammer_bclu_offset(cursor->record_buffer, nrec);
 833                 } else {
 834                         /*
 835                          * Data separate from record
 836                          */
 837                         nrec->base.data_offset = hammer_bclu_offset(cursor->data_buffer, bdata);
 838                         hammer_modify_buffer(cursor->data_buffer);
 839                         bcopy(data, bdata, nrec->base.data_len);
 840                 }
 841         }
 842         nrec->base.rec_id = hammer_alloc_recid(cursor->node->cluster);
 843         nrec_offset = hammer_bclu_offset(cursor->record_buffer, nrec);
 844
 845         switch(cursor->key_beg.btype) {
 846         case HAMMER_BTREE_TYPE_RECORD:
 847                 elm.leaf.base = cursor->key_beg;
 848                 elm.leaf.rec_offset = nrec_offset;
 849                 elm.leaf.data_offset = nrec->base.data_offset;
 850                 elm.leaf.data_len = nrec->base.data_len;
 851                 elm.leaf.data_crc = nrec->base.data_crc;
 852                 error = hammer_btree_insert(cursor, &elm);
 853                 break;
 854         case HAMMER_BTREE_TYPE_SPIKE_BEG:
 855 #if 0
 856                 Debugger("SpikeSpike");
 857 #endif
 858                 nvolume = hammer_get_volume(cursor->node->cluster->volume->hmp,
 859                                             nrec->spike.vol_no, &error);
 860                 if (error)
 861                         break;
 862                 ncluster = hammer_get_cluster(nvolume, nrec->spike.clu_no,
 863                                               &error, GET_CLUSTER_NORECOVER);
 864                 hammer_rel_volume(nvolume, 0);
 865                 if (error)
 866                         break;
 867                 hammer_modify_cluster(ncluster);
 868                 ncluster->ondisk->clu_btree_parent_offset =
 869                         cursor->node->node_offset;
 870                 hammer_rel_cluster(ncluster, 0);
 871                 error = hammer_btree_insert_cluster(cursor, ncluster,
 872                                                     nrec_offset);
 873                 break;
 874         case HAMMER_BTREE_TYPE_SPIKE_END:
 875                 panic("hammer_write_record: cannot write spike-end elms");
 876                 break;
 877         default:
 878                 panic("hammer_write_record: unknown btype %02x",
 879                       elm.leaf.base.btype);
 880                 break;
 881         }
 882         if (error == 0)
 883                 goto done;
 884
 885         hammer_free_record_ptr(cursor->record_buffer, nrec,
 886                                orec->base.base.rec_type);
 887 fail1:
 888         if (data && (cursor_flags & HAMMER_RECF_EMBEDDED_DATA) == 0) {
 889                 hammer_free_data_ptr(cursor->data_buffer, bdata,
 890                                      orec->base.data_len);
 891         }
 892 done:
 893         /* leave cursor intact */
 894         return(error);
 895 }
 896
 897 /*
 898  * Add the record to the inode's rec_tree.  The low 32 bits of a directory
 899  * entry's key is used to deal with hash collisions in the upper 32 bits.
 900  * A unique 64 bit key is generated in-memory and may be regenerated a
 901  * second time when the directory record is flushed to the on-disk B-Tree.
 902  *
 903  * A referenced record is passed to this function.  This function
 904  * eats the reference.  If an error occurs the record will be deleted.
 905  */
 906 static
 907 int
 908 hammer_mem_add(struct hammer_transaction *trans, hammer_record_t record)
 909 {
 910         while (RB_INSERT(hammer_rec_rb_tree, &record->ip->rec_tree, record)) {
 911                 if (record->rec.base.base.rec_type != HAMMER_RECTYPE_DIRENTRY){
 912                         record->flags |= HAMMER_RECF_DELETED;
 913                         hammer_rel_mem_record(record);
 914                         return (EEXIST);
 915                 }
 916                 if (++trans->hmp->namekey_iterator == 0)
 917                         ++trans->hmp->namekey_iterator;
 918                 record->rec.base.base.key &= ~(0xFFFFFFFFLL);
 919                 record->rec.base.base.key |= trans->hmp->namekey_iterator;
 920         }
 921         record->flags |= HAMMER_RECF_ONRBTREE;
 922         hammer_modify_inode(trans, record->ip, HAMMER_INODE_XDIRTY);
 923         hammer_rel_mem_record(record);
 924         return(0);
 925 }
 926
 927 /************************************************************************
 928  *                   HAMMER INODE MERGED-RECORD FUNCTIONS               *
 929  ************************************************************************
 930  *
 931  * These functions augment the B-Tree scanning functions in hammer_btree.c
 932  * by merging in-memory records with on-disk records.
 933  */
 934
 935 /*
 936  * Locate a particular record either in-memory or on-disk.
 937  *
 938  * NOTE: This is basically a standalone routine, hammer_ip_next() may
 939  * NOT be called to iterate results.
 940  */
 941 int
 942 hammer_ip_lookup(hammer_cursor_t cursor, struct hammer_inode *ip)
 943 {
 944         int error;
 945
 946         /*
 947          * If the element is in-memory return it without searching the
 948          * on-disk B-Tree
 949          */
 950         error = hammer_mem_lookup(cursor, ip);
 951         if (error == 0) {
 952                 cursor->record = &cursor->iprec->rec;
 953                 return(error);
 954         }
 955         if (error != ENOENT)
 956                 return(error);
 957
 958         /*
 959          * If the inode has on-disk components search the on-disk B-Tree.
 960          */
 961         if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DONDISK)) == 0)
 962                 return(error);
 963         error = hammer_btree_lookup(cursor);
 964         if (error == 0)
 965                 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_RECORD);
 966         return(error);
 967 }
 968
 969 /*
 970  * Locate the first record within the cursor's key_beg/key_end range,
 971  * restricted to a particular inode.  0 is returned on success, ENOENT
 972  * if no records matched the requested range, or some other error.
 973  *
 974  * When 0 is returned hammer_ip_next() may be used to iterate additional
 975  * records within the requested range.
 976  *
 977  * This function can return EDEADLK, requiring the caller to terminate
 978  * the cursor and try again.
 979  */
 980 int
 981 hammer_ip_first(hammer_cursor_t cursor, struct hammer_inode *ip)
 982 {
 983         int error;
 984
 985         /*
 986          * Clean up fields and setup for merged scan
 987          */
 988         cursor->flags &= ~HAMMER_CURSOR_DELBTREE;
 989         cursor->flags |= HAMMER_CURSOR_ATEDISK | HAMMER_CURSOR_ATEMEM;
 990         cursor->flags |= HAMMER_CURSOR_DISKEOF | HAMMER_CURSOR_MEMEOF;
 991         if (cursor->iprec) {
 992                 hammer_rel_mem_record(cursor->iprec);
 993                 cursor->iprec = NULL;
 994         }
 995
 996         /*
 997          * Search the on-disk B-Tree.  hammer_btree_lookup() only does an
 998          * exact lookup so if we get ENOENT we have to call the iterate
 999          * function to validate the first record after the begin key.
1000          *
1001          * The ATEDISK flag is used by hammer_btree_iterate to determine
1002          * whether it must index forwards or not.  It is also used here
1003          * to select the next record from in-memory or on-disk.
1004          *
1005          * EDEADLK can only occur if the lookup hit an empty internal
1006          * element and couldn't delete it.  Since this could only occur
1007          * in-range, we can just iterate from the failure point.
1008          */
1009         if (ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DONDISK)) {
1010                 error = hammer_btree_lookup(cursor);
1011                 if (error == ENOENT || error == EDEADLK) {
1012                         cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1013                         error = hammer_btree_iterate(cursor);
1014                 }
1015                 if (error && error != ENOENT)
1016                         return(error);
1017                 if (error == 0) {
1018                         cursor->flags &= ~HAMMER_CURSOR_DISKEOF;
1019                         cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1020                 } else {
1021                         cursor->flags |= HAMMER_CURSOR_ATEDISK;
1022                 }
1023         }
1024
1025         /*
1026          * Search the in-memory record list (Red-Black tree).  Unlike the
1027          * B-Tree search, mem_first checks for records in the range.
1028          */
1029         error = hammer_mem_first(cursor, ip);
1030         if (error && error != ENOENT)
1031                 return(error);
1032         if (error == 0) {
1033                 cursor->flags &= ~HAMMER_CURSOR_MEMEOF;
1034                 cursor->flags &= ~HAMMER_CURSOR_ATEMEM;
1035         }
1036
1037         /*
1038          * This will return the first matching record.
1039          */
1040         return(hammer_ip_next(cursor));
1041 }
1042
1043 /*
1044  * Retrieve the next record in a merged iteration within the bounds of the
1045  * cursor.  This call may be made multiple times after the cursor has been
1046  * initially searched with hammer_ip_first().
1047  *
1048  * 0 is returned on success, ENOENT if no further records match the
1049  * requested range, or some other error code is returned.
1050  */
1051 int
1052 hammer_ip_next(hammer_cursor_t cursor)
1053 {
1054         hammer_btree_elm_t elm;
1055         hammer_record_t rec;
1056         int error;
1057         int r;
1058
1059         /*
1060          * Load the current on-disk and in-memory record.  If we ate any
1061          * records we have to get the next one.
1062          *
1063          * If we deleted the last on-disk record we had scanned ATEDISK will
1064          * be clear and DELBTREE will be set, forcing a call to iterate. The
1065          * fact that ATEDISK is clear causes iterate to re-test the 'current'
1066          * element.  If ATEDISK is set, iterate will skip the 'current'
1067          * element.
1068          *
1069          * Get the next on-disk record
1070          */
1071         if (cursor->flags & (HAMMER_CURSOR_ATEDISK|HAMMER_CURSOR_DELBTREE)) {
1072                 if ((cursor->flags & HAMMER_CURSOR_DISKEOF) == 0) {
1073                         error = hammer_btree_iterate(cursor);
1074                         cursor->flags &= ~HAMMER_CURSOR_DELBTREE;
1075                         if (error == 0)
1076                                 cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1077                         else
1078                                 cursor->flags |= HAMMER_CURSOR_DISKEOF |
1079                                                  HAMMER_CURSOR_ATEDISK;
1080                 }
1081         }
1082
1083         /*
1084          * Get the next in-memory record.  The record can be ripped out
1085          * of the RB tree so we maintain a scan_info structure to track
1086          * the next node.
1087          *
1088          * hammer_rec_scan_cmp:  Is the record still in our general range,
1089          *                       (non-inclusive of snapshot exclusions)?
1090          * hammer_rec_scan_callback: Is the record in our snapshot?
1091          */
1092         if (cursor->flags & HAMMER_CURSOR_ATEMEM) {
1093                 if ((cursor->flags & HAMMER_CURSOR_MEMEOF) == 0) {
1094                         if (cursor->iprec) {
1095                                 hammer_rel_mem_record(cursor->iprec);
1096                                 cursor->iprec = NULL;
1097                         }
1098                         rec = cursor->scan.node;        /* next node */
1099                         while (rec) {
1100                                 if (hammer_rec_scan_cmp(rec, cursor) != 0)
1101                                         break;
1102                                 if (hammer_rec_scan_callback(rec, cursor) != 0)
1103                                         break;
1104                                 rec = hammer_rec_rb_tree_RB_NEXT(rec);
1105                         }
1106                         if (cursor->iprec) {
1107                                 KKASSERT(cursor->iprec == rec);
1108                                 cursor->flags &= ~HAMMER_CURSOR_ATEMEM;
1109                                 cursor->scan.node =
1110                                         hammer_rec_rb_tree_RB_NEXT(rec);
1111                         } else {
1112                                 cursor->flags |= HAMMER_CURSOR_MEMEOF;
1113                         }
1114                 }
1115         }
1116
1117         /*
1118          * Extract either the disk or memory record depending on their
1119          * relative position.
1120          */
1121         error = 0;
1122         switch(cursor->flags & (HAMMER_CURSOR_ATEDISK | HAMMER_CURSOR_ATEMEM)) {
1123         case 0:
1124                 /*
1125                  * Both entries valid
1126                  */
1127                 elm = &cursor->node->ondisk->elms[cursor->index];
1128                 r = hammer_btree_cmp(&elm->base, &cursor->iprec->rec.base.base);
1129                 if (r < 0) {
1130                         error = hammer_btree_extract(cursor,
1131                                                      HAMMER_CURSOR_GET_RECORD);
1132                         cursor->flags |= HAMMER_CURSOR_ATEDISK;
1133                         break;
1134                 }
1135                 /* fall through to the memory entry */
1136         case HAMMER_CURSOR_ATEDISK:
1137                 /*
1138                  * Only the memory entry is valid
1139                  */
1140                 cursor->record = &cursor->iprec->rec;
1141                 cursor->flags |= HAMMER_CURSOR_ATEMEM;
1142                 break;
1143         case HAMMER_CURSOR_ATEMEM:
1144                 /*
1145                  * Only the disk entry is valid
1146                  */
1147                 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_RECORD);
1148                 cursor->flags |= HAMMER_CURSOR_ATEDISK;
1149                 break;
1150         default:
1151                 /*
1152                  * Neither entry is valid
1153                  *
1154                  * XXX error not set properly
1155                  */
1156                 cursor->record = NULL;
1157                 error = ENOENT;
1158                 break;
1159         }
1160         return(error);
1161 }
1162
1163 /*
1164  * Resolve the cursor->data pointer for the current cursor position in
1165  * a merged iteration.
1166  */
1167 int
1168 hammer_ip_resolve_data(hammer_cursor_t cursor)
1169 {
1170         int error;
1171
1172         if (cursor->iprec && cursor->record == &cursor->iprec->rec) {
1173                 cursor->data = cursor->iprec->data;
1174                 error = 0;
1175         } else {
1176                 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA);
1177         }
1178         return(error);
1179 }
1180
1181 /*
1182  * Delete all records within the specified range for inode ip.
1183  *
1184  * NOTE: An unaligned range will cause new records to be added to cover
1185  * the edge cases. (XXX not implemented yet).
1186  *
1187  * NOTE: ran_end is inclusive (e.g. 0,1023 instead of 0,1024).
1188  *
1189  * NOTE: Record keys for regular file data have to be special-cased since
1190  * they indicate the end of the range (key = base + bytes).
1191  *
1192  * NOTE: The spike structure must be filled in if we return ENOSPC.
1193  */
1194 int
1195 hammer_ip_delete_range(hammer_transaction_t trans, hammer_inode_t ip,
1196                        int64_t ran_beg, int64_t ran_end,
1197                        struct hammer_cursor **spike)
1198 {
1199         struct hammer_cursor cursor;
1200         hammer_record_ondisk_t rec;
1201         hammer_base_elm_t base;
1202         int error;
1203         int64_t off;
1204
1205 retry:
1206         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1207
1208         cursor.key_beg.obj_id = ip->obj_id;
1209         cursor.key_beg.create_tid = 0;
1210         cursor.key_beg.delete_tid = 0;
1211         cursor.key_beg.obj_type = 0;
1212         cursor.asof = ip->obj_asof;
1213         cursor.flags |= HAMMER_CURSOR_ASOF;
1214
1215         cursor.key_end = cursor.key_beg;
1216         if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1217                 cursor.key_beg.key = ran_beg;
1218                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1219                 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1220                 cursor.key_end.key = ran_end;
1221         } else {
1222                 /*
1223                  * The key in the B-Tree is (base+bytes), so the first possible
1224                  * matching key is ran_beg + 1.
1225                  */
1226                 int64_t tmp64;
1227
1228                 cursor.key_beg.key = ran_beg + 1;
1229                 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1230                 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1231
1232                 tmp64 = ran_end + MAXPHYS + 1;  /* work around GCC-4 bug */
1233                 if (tmp64 < ran_end)
1234                         cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1235                 else
1236                         cursor.key_end.key = ran_end + MAXPHYS + 1;
1237         }
1238         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1239
1240         error = hammer_ip_first(&cursor, ip);
1241
1242         /*
1243          * Iterate through matching records and mark them as deleted.
1244          */
1245         while (error == 0) {
1246                 rec = cursor.record;
1247                 base = &rec->base.base;
1248
1249                 KKASSERT(base->delete_tid == 0);
1250
1251                 /*
1252                  * There may be overlap cases for regular file data.  Also
1253                  * remember the key for a regular file record is the offset
1254                  * of the last byte of the record (base + len - 1), NOT the
1255                  * base offset.
1256                  */
1257 #if 0
1258                 kprintf("delete_range rec_type %02x\n", base->rec_type);
1259 #endif
1260                 if (base->rec_type == HAMMER_RECTYPE_DATA) {
1261 #if 0
1262                         kprintf("delete_range loop key %016llx\n",
1263                                 base->key - rec->base.data_len);
1264 #endif
1265                         off = base->key - rec->base.data_len;
1266                         /*
1267                          * Check the left edge case.  We currently do not
1268                          * split existing records.
1269                          */
1270                         if (off < ran_beg) {
1271                                 panic("hammer left edge case %016llx %d\n",
1272                                         base->key, rec->base.data_len);
1273                         }
1274
1275                         /*
1276                          * Check the right edge case.  Note that the
1277                          * record can be completely out of bounds, which
1278                          * terminates the search.
1279                          *
1280                          * base->key is exclusive of the right edge while
1281                          * ran_end is inclusive of the right edge.  The
1282                          * (key - data_len) left boundary is inclusive.
1283                          *
1284                          * XXX theory-check this test at some point, are
1285                          * we missing a + 1 somewhere?  Note that ran_end
1286                          * could overflow.
1287                          */
1288                         if (base->key - 1 > ran_end) {
1289                                 if (base->key - rec->base.data_len > ran_end)
1290                                         break;
1291                                 panic("hammer right edge case\n");
1292                         }
1293                 }
1294
1295                 /*
1296                  * Mark the record and B-Tree entry as deleted.  This will
1297                  * also physically delete the B-Tree entry, record, and
1298                  * data if the retention policy dictates.  The function
1299                  * will set HAMMER_CURSOR_DELBTREE which hammer_ip_next()
1300                  * uses to perform a fixup.
1301                  */
1302                 error = hammer_ip_delete_record(&cursor, trans->tid);
1303                 if (error)
1304                         break;
1305                 error = hammer_ip_next(&cursor);
1306         }
1307         hammer_done_cursor(&cursor);
1308         if (error == EDEADLK)
1309                 goto retry;
1310         if (error == ENOENT)
1311                 error = 0;
1312         return(error);
1313 }
1314
1315 /*
1316  * Delete all records associated with an inode except the inode record
1317  * itself.
1318  */
1319 int
1320 hammer_ip_delete_range_all(hammer_transaction_t trans, hammer_inode_t ip)
1321 {
1322         struct hammer_cursor cursor;
1323         hammer_record_ondisk_t rec;
1324         hammer_base_elm_t base;
1325         int error;
1326
1327 retry:
1328         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1329
1330         cursor.key_beg.obj_id = ip->obj_id;
1331         cursor.key_beg.create_tid = 0;
1332         cursor.key_beg.delete_tid = 0;
1333         cursor.key_beg.obj_type = 0;
1334         cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE + 1;
1335         cursor.key_beg.key = HAMMER_MIN_KEY;
1336
1337         cursor.key_end = cursor.key_beg;
1338         cursor.key_end.rec_type = 0xFFFF;
1339         cursor.key_end.key = HAMMER_MAX_KEY;
1340
1341         cursor.asof = ip->obj_asof;
1342         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1343
1344         error = hammer_ip_first(&cursor, ip);
1345
1346         /*
1347          * Iterate through matching records and mark them as deleted.
1348          */
1349         while (error == 0) {
1350                 rec = cursor.record;
1351                 base = &rec->base.base;
1352
1353                 KKASSERT(base->delete_tid == 0);
1354
1355                 /*
1356                  * Mark the record and B-Tree entry as deleted.  This will
1357                  * also physically delete the B-Tree entry, record, and
1358                  * data if the retention policy dictates.  The function
1359                  * will set HAMMER_CURSOR_DELBTREE which hammer_ip_next()
1360                  * uses to perform a fixup.
1361                  */
1362                 error = hammer_ip_delete_record(&cursor, trans->tid);
1363                 if (error)
1364                         break;
1365                 error = hammer_ip_next(&cursor);
1366         }
1367         hammer_done_cursor(&cursor);
1368         if (error == EDEADLK)
1369                 goto retry;
1370         if (error == ENOENT)
1371                 error = 0;
1372         return(error);
1373 }
1374
1375 /*
1376  * Delete the record at the current cursor.  On success the cursor will
1377  * be positioned appropriately for an iteration but may no longer be at
1378  * a leaf node.
1379  *
1380  * NOTE: This can return EDEADLK, requiring the caller to terminate the
1381  * cursor and retry.
1382  */
1383 int
1384 hammer_ip_delete_record(hammer_cursor_t cursor, hammer_tid_t tid)
1385 {
1386         hammer_btree_elm_t elm;
1387         hammer_mount_t hmp;
1388         int error;
1389         int dodelete;
1390
1391         /*
1392          * In-memory (unsynchronized) records can simply be freed.
1393          */
1394         if (cursor->record == &cursor->iprec->rec) {
1395                 cursor->iprec->flags |= HAMMER_RECF_DELETED;
1396                 return(0);
1397         }
1398
1399         /*
1400          * On-disk records are marked as deleted by updating their delete_tid.
1401          * This does not effect their position in the B-Tree (which is based
1402          * on their create_tid).
1403          */
1404         error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_RECORD);
1405         elm = NULL;
1406         hmp = cursor->node->cluster->volume->hmp;
1407
1408         dodelete = 0;
1409         if (error == 0) {
1410                 error = hammer_cursor_upgrade(cursor);
1411                 if (error == 0) {
1412                         hammer_modify_node(cursor->node);
1413                         elm = &cursor->node->ondisk->elms[cursor->index];
1414                         elm->leaf.base.delete_tid = tid;
1415                         hammer_modify_buffer(cursor->record_buffer);
1416                         cursor->record->base.base.delete_tid = tid;
1417                 }
1418         }
1419
1420         /*
1421          * If we were mounted with the nohistory option, we physically
1422          * delete the record.
1423          */
1424         if (hmp->hflags & HMNT_NOHISTORY)
1425                 dodelete = 1;
1426
1427         if (error == 0 && dodelete) {
1428                 error = hammer_delete_at_cursor(cursor, NULL);
1429                 if (error) {
1430                         panic("hammer_ip_delete_record: unable to physically delete the record!\n");
1431                         error = 0;
1432                 }
1433         }
1434         return(error);
1435 }
1436
1437 int
1438 hammer_delete_at_cursor(hammer_cursor_t cursor, int64_t *stat_bytes)
1439 {
1440         hammer_btree_elm_t elm;
1441         int32_t rec_offset;
1442         int32_t data_offset;
1443         int32_t data_len;
1444         u_int8_t rec_type;
1445         hammer_cluster_t cluster;
1446         int error;
1447
1448         elm = &cursor->node->ondisk->elms[cursor->index];
1449         KKASSERT(elm->base.btype == HAMMER_BTREE_TYPE_RECORD);
1450
1451         rec_offset = elm->leaf.rec_offset;
1452         data_offset = elm->leaf.data_offset;
1453         data_len = elm->leaf.data_len;
1454         rec_type = elm->leaf.base.rec_type;
1455
1456         /*
1457          * We must ref the cluster to prevent it from being
1458          * freed prior to our freeing the last record.
1459          */
1460         cluster = cursor->node->cluster;
1461         hammer_ref_cluster(cluster);
1462
1463         error = hammer_btree_delete(cursor);
1464         if (error == 0) {
1465                 /*
1466                  * This forces a fixup for the iteration because
1467                  * the cursor is now either sitting at the 'next'
1468                  * element or sitting at the end of a leaf.
1469                  */
1470                 if ((cursor->flags & HAMMER_CURSOR_DISKEOF) == 0) {
1471                         cursor->flags |= HAMMER_CURSOR_DELBTREE;
1472                         cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1473                 }
1474                 hammer_free_record(cluster, rec_offset, rec_type);
1475                 if (data_offset && (data_offset - rec_offset < 0 ||
1476                     data_offset - rec_offset >= HAMMER_RECORD_SIZE)) {
1477                         hammer_free_data(cluster, data_offset,data_len);
1478                         if (stat_bytes)
1479                                 *stat_bytes += data_len;
1480                 }
1481         }
1482 #if 0
1483         kprintf("hammer_delete_at_cursor: %d:%d:%08x %08x/%d "
1484                 "(%d remain in cluster)\n",
1485                 cluster->volume->vol_no, cluster->clu_no,
1486                 rec_offset, data_offset, data_len,
1487                 cluster->ondisk->stat_records);
1488 #endif
1489         hammer_rel_cluster(cluster, 0);
1490         return (error);
1491 }
1492
1493 /*
1494  * Determine whether a directory is empty or not.  Returns 0 if the directory
1495  * is empty, ENOTEMPTY if it isn't, plus other possible errors.
1496  */
1497 int
1498 hammer_ip_check_directory_empty(hammer_transaction_t trans, hammer_inode_t ip)
1499 {
1500         struct hammer_cursor cursor;
1501         int error;
1502
1503         hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp);
1504
1505         cursor.key_beg.obj_id = ip->obj_id;
1506         cursor.key_beg.create_tid = 0;
1507         cursor.key_beg.delete_tid = 0;
1508         cursor.key_beg.obj_type = 0;
1509         cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE + 1;
1510         cursor.key_beg.key = HAMMER_MIN_KEY;
1511
1512         cursor.key_end = cursor.key_beg;
1513         cursor.key_end.rec_type = 0xFFFF;
1514         cursor.key_end.key = HAMMER_MAX_KEY;
1515
1516         cursor.asof = ip->obj_asof;
1517         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1518
1519         error = hammer_ip_first(&cursor, ip);
1520         if (error == ENOENT)
1521                 error = 0;
1522         else if (error == 0)
1523                 error = ENOTEMPTY;
1524         hammer_done_cursor(&cursor);
1525         return(error);
1526 }
1527