sys/vfs/hammer/hammer_object.c

   1 /*
   2  * Copyright (c) 2007-2008 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 1.95 2008/08/06 15:38:58 dillon Exp $
  35  */
  36
  37 #include "hammer.h"
  38
  39 static int hammer_mem_lookup(hammer_cursor_t cursor);
  40 static int hammer_mem_first(hammer_cursor_t cursor);
  41 static int hammer_frontend_trunc_callback(hammer_record_t record,
  42                                 void *data __unused);
  43 static int hammer_record_needs_overwrite_delete(hammer_record_t record);
  44 static int hammer_delete_general(hammer_cursor_t cursor, hammer_inode_t ip,
  45                       hammer_btree_leaf_elm_t leaf);
  46
  47 struct rec_trunc_info {
  48         u_int16_t       rec_type;
  49         int64_t         trunc_off;
  50 };
  51
  52 /*
  53  * Red-black tree support.  Comparison code for insertion.
  54  */
  55 static int
  56 hammer_rec_rb_compare(hammer_record_t rec1, hammer_record_t rec2)
  57 {
  58         if (rec1->leaf.base.rec_type < rec2->leaf.base.rec_type)
  59                 return(-1);
  60         if (rec1->leaf.base.rec_type > rec2->leaf.base.rec_type)
  61                 return(1);
  62
  63         if (rec1->leaf.base.key < rec2->leaf.base.key)
  64                 return(-1);
  65         if (rec1->leaf.base.key > rec2->leaf.base.key)
  66                 return(1);
  67
  68         /*
  69          * Never match against an item deleted by the front-end.
  70          *
  71          * rec1 is greater then rec2 if rec1 is marked deleted.
  72          * rec1 is less then rec2 if rec2 is marked deleted.
  73          *
  74          * Multiple deleted records may be present, do not return 0
  75          * if both are marked deleted.
  76          */
  77         if (rec1->flags & HAMMER_RECF_DELETED_FE)
  78                 return(1);
  79         if (rec2->flags & HAMMER_RECF_DELETED_FE)
  80                 return(-1);
  81
  82         return(0);
  83 }
  84
  85 /*
  86  * Basic record comparison code similar to hammer_btree_cmp().
  87  */
  88 static int
  89 hammer_rec_cmp(hammer_base_elm_t elm, hammer_record_t rec)
  90 {
  91         if (elm->rec_type < rec->leaf.base.rec_type)
  92                 return(-3);
  93         if (elm->rec_type > rec->leaf.base.rec_type)
  94                 return(3);
  95
  96         if (elm->key < rec->leaf.base.key)
  97                 return(-2);
  98         if (elm->key > rec->leaf.base.key)
  99                 return(2);
 100
 101         /*
 102          * Never match against an item deleted by the front-end.
 103          * elm is less then rec if rec is marked deleted.
 104          */
 105         if (rec->flags & HAMMER_RECF_DELETED_FE)
 106                 return(-1);
 107         return(0);
 108 }
 109
 110 /*
 111  * Special LOOKUP_INFO to locate an overlapping record.  This used by
 112  * the reservation code to implement small-block records (whos keys will
 113  * be different depending on data_len, when representing the same base
 114  * offset).
 115  *
 116  * NOTE: The base file offset of a data record is (key - data_len), not (key).
 117  */
 118 static int
 119 hammer_rec_overlap_compare(hammer_btree_leaf_elm_t leaf, hammer_record_t rec)
 120 {
 121         if (leaf->base.rec_type < rec->leaf.base.rec_type)
 122                 return(-3);
 123         if (leaf->base.rec_type > rec->leaf.base.rec_type)
 124                 return(3);
 125
 126         /*
 127          * Overlap compare
 128          */
 129         if (leaf->base.rec_type == HAMMER_RECTYPE_DATA) {
 130                 /* leaf_end <= rec_beg */
 131                 if (leaf->base.key <= rec->leaf.base.key - rec->leaf.data_len)
 132                         return(-2);
 133                 /* leaf_beg >= rec_end */
 134                 if (leaf->base.key - leaf->data_len >= rec->leaf.base.key)
 135                         return(2);
 136         } else {
 137                 if (leaf->base.key < rec->leaf.base.key)
 138                         return(-2);
 139                 if (leaf->base.key > rec->leaf.base.key)
 140                         return(2);
 141         }
 142
 143         /*
 144          * Never match against an item deleted by the front-end.
 145          * leaf is less then rec if rec is marked deleted.
 146          *
 147          * We must still return the proper code for the scan to continue
 148          * along the correct branches.
 149          */
 150         if (rec->flags & HAMMER_RECF_DELETED_FE) {
 151                 if (leaf->base.key < rec->leaf.base.key)
 152                         return(-2);
 153                 if (leaf->base.key > rec->leaf.base.key)
 154                         return(2);
 155                 return(-1);
 156         }
 157         return(0);
 158 }
 159
 160 /*
 161  * RB_SCAN comparison code for hammer_mem_first().  The argument order
 162  * is reversed so the comparison result has to be negated.  key_beg and
 163  * key_end are both range-inclusive.
 164  *
 165  * Localized deletions are not cached in-memory.
 166  */
 167 static
 168 int
 169 hammer_rec_scan_cmp(hammer_record_t rec, void *data)
 170 {
 171         hammer_cursor_t cursor = data;
 172         int r;
 173
 174         r = hammer_rec_cmp(&cursor->key_beg, rec);
 175         if (r > 1)
 176                 return(-1);
 177         r = hammer_rec_cmp(&cursor->key_end, rec);
 178         if (r < -1)
 179                 return(1);
 180         return(0);
 181 }
 182
 183 /*
 184  * This compare function is used when simply looking up key_beg.
 185  */
 186 static
 187 int
 188 hammer_rec_find_cmp(hammer_record_t rec, void *data)
 189 {
 190         hammer_cursor_t cursor = data;
 191         int r;
 192
 193         r = hammer_rec_cmp(&cursor->key_beg, rec);
 194         if (r > 1)
 195                 return(-1);
 196         if (r < -1)
 197                 return(1);
 198         return(0);
 199 }
 200
 201 /*
 202  * Locate blocks within the truncation range.  Partial blocks do not count.
 203  */
 204 static
 205 int
 206 hammer_rec_trunc_cmp(hammer_record_t rec, void *data)
 207 {
 208         struct rec_trunc_info *info = data;
 209
 210         if (rec->leaf.base.rec_type < info->rec_type)
 211                 return(-1);
 212         if (rec->leaf.base.rec_type > info->rec_type)
 213                 return(1);
 214
 215         switch(rec->leaf.base.rec_type) {
 216         case HAMMER_RECTYPE_DB:
 217                 /*
 218                  * DB record key is not beyond the truncation point, retain.
 219                  */
 220                 if (rec->leaf.base.key < info->trunc_off)
 221                         return(-1);
 222                 break;
 223         case HAMMER_RECTYPE_DATA:
 224                 /*
 225                  * DATA record offset start is not beyond the truncation point,
 226                  * retain.
 227                  */
 228                 if (rec->leaf.base.key - rec->leaf.data_len < info->trunc_off)
 229                         return(-1);
 230                 break;
 231         default:
 232                 panic("hammer_rec_trunc_cmp: unexpected record type");
 233         }
 234
 235         /*
 236          * The record start is >= the truncation point, return match,
 237          * the record should be destroyed.
 238          */
 239         return(0);
 240 }
 241
 242 RB_GENERATE(hammer_rec_rb_tree, hammer_record, rb_node, hammer_rec_rb_compare);
 243 RB_GENERATE_XLOOKUP(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
 244                     hammer_rec_overlap_compare, hammer_btree_leaf_elm_t);
 245
 246 /*
 247  * Allocate a record for the caller to finish filling in.  The record is
 248  * returned referenced.
 249  */
 250 hammer_record_t
 251 hammer_alloc_mem_record(hammer_inode_t ip, int data_len)
 252 {
 253         hammer_record_t record;
 254
 255         ++hammer_count_records;
 256         record = kmalloc(sizeof(*record), M_HAMMER,
 257                          M_WAITOK | M_ZERO | M_USE_RESERVE);
 258         record->flush_state = HAMMER_FST_IDLE;
 259         record->ip = ip;
 260         record->leaf.base.btype = HAMMER_BTREE_TYPE_RECORD;
 261         record->leaf.data_len = data_len;
 262         hammer_ref(&record->lock);
 263
 264         if (data_len) {
 265                 record->data = kmalloc(data_len, M_HAMMER, M_WAITOK | M_ZERO);
 266                 record->flags |= HAMMER_RECF_ALLOCDATA;
 267                 ++hammer_count_record_datas;
 268         }
 269
 270         return (record);
 271 }
 272
 273 void
 274 hammer_wait_mem_record_ident(hammer_record_t record, const char *ident)
 275 {
 276         while (record->flush_state == HAMMER_FST_FLUSH) {
 277                 record->flags |= HAMMER_RECF_WANTED;
 278                 tsleep(record, 0, ident, 0);
 279         }
 280 }
 281
 282 /*
 283  * Called from the backend, hammer_inode.c, after a record has been
 284  * flushed to disk.  The record has been exclusively locked by the
 285  * caller and interlocked with BE.
 286  *
 287  * We clean up the state, unlock, and release the record (the record
 288  * was referenced by the fact that it was in the HAMMER_FST_FLUSH state).
 289  */
 290 void
 291 hammer_flush_record_done(hammer_record_t record, int error)
 292 {
 293         hammer_inode_t target_ip;
 294
 295         KKASSERT(record->flush_state == HAMMER_FST_FLUSH);
 296         KKASSERT(record->flags & HAMMER_RECF_INTERLOCK_BE);
 297
 298         if (error) {
 299                 /*
 300                  * An error occured, the backend was unable to sync the
 301                  * record to its media.  Leave the record intact.
 302                  */
 303                 hammer_critical_error(record->ip->hmp, record->ip, error,
 304                                       "while flushing record");
 305         }
 306
 307         --record->flush_group->refs;
 308         record->flush_group = NULL;
 309
 310         if (record->flags & HAMMER_RECF_DELETED_BE) {
 311                 if ((target_ip = record->target_ip) != NULL) {
 312                         TAILQ_REMOVE(&target_ip->target_list, record,
 313                                      target_entry);
 314                         record->target_ip = NULL;
 315                         hammer_test_inode(target_ip);
 316                 }
 317                 record->flush_state = HAMMER_FST_IDLE;
 318         } else {
 319                 if (record->target_ip) {
 320                         record->flush_state = HAMMER_FST_SETUP;
 321                         hammer_test_inode(record->ip);
 322                         hammer_test_inode(record->target_ip);
 323                 } else {
 324                         record->flush_state = HAMMER_FST_IDLE;
 325                 }
 326         }
 327         record->flags &= ~HAMMER_RECF_INTERLOCK_BE;
 328         if (record->flags & HAMMER_RECF_WANTED) {
 329                 record->flags &= ~HAMMER_RECF_WANTED;
 330                 wakeup(record);
 331         }
 332         hammer_rel_mem_record(record);
 333 }
 334
 335 /*
 336  * Release a memory record.  Records marked for deletion are immediately
 337  * removed from the RB-Tree but otherwise left intact until the last ref
 338  * goes away.
 339  */
 340 void
 341 hammer_rel_mem_record(struct hammer_record *record)
 342 {
 343         hammer_mount_t hmp;
 344         hammer_reserve_t resv;
 345         hammer_inode_t ip;
 346         hammer_inode_t target_ip;
 347
 348         hammer_unref(&record->lock);
 349
 350         if (record->lock.refs == 0) {
 351                 /*
 352                  * Upon release of the last reference wakeup any waiters.
 353                  * The record structure may get destroyed so callers will
 354                  * loop up and do a relookup.
 355                  *
 356                  * WARNING!  Record must be removed from RB-TREE before we
 357                  * might possibly block.  hammer_test_inode() can block!
 358                  */
 359                 ip = record->ip;
 360                 hmp = ip->hmp;
 361
 362                 /*
 363                  * Upon release of the last reference a record marked deleted
 364                  * is destroyed.
 365                  */
 366                 if (record->flags & HAMMER_RECF_DELETED_FE) {
 367                         KKASSERT(ip->lock.refs > 0);
 368                         KKASSERT(record->flush_state != HAMMER_FST_FLUSH);
 369
 370                         /*
 371                          * target_ip may have zero refs, we have to ref it
 372                          * to prevent it from being ripped out from under
 373                          * us.
 374                          */
 375                         if ((target_ip = record->target_ip) != NULL) {
 376                                 TAILQ_REMOVE(&target_ip->target_list,
 377                                              record, target_entry);
 378                                 record->target_ip = NULL;
 379                                 hammer_ref(&target_ip->lock);
 380                         }
 381
 382                         if (record->flags & HAMMER_RECF_ONRBTREE) {
 383                                 RB_REMOVE(hammer_rec_rb_tree,
 384                                           &record->ip->rec_tree,
 385                                           record);
 386                                 KKASSERT(ip->rsv_recs > 0);
 387                                 --hmp->rsv_recs;
 388                                 --ip->rsv_recs;
 389                                 hmp->rsv_databytes -= record->leaf.data_len;
 390                                 record->flags &= ~HAMMER_RECF_ONRBTREE;
 391
 392                                 if (RB_EMPTY(&record->ip->rec_tree)) {
 393                                         record->ip->flags &= ~HAMMER_INODE_XDIRTY;
 394                                         record->ip->sync_flags &= ~HAMMER_INODE_XDIRTY;
 395                                         hammer_test_inode(record->ip);
 396                                 }
 397                         }
 398
 399                         /*
 400                          * We must wait for any direct-IO to complete before
 401                          * we can destroy the record because the bio may
 402                          * have a reference to it.
 403                          */
 404                         if (record->flags &
 405                            (HAMMER_RECF_DIRECT_IO | HAMMER_RECF_DIRECT_INVAL)) {
 406                                 hammer_io_direct_wait(record);
 407                         }
 408
 409
 410                         /*
 411                          * Do this test after removing record from the B-Tree.
 412                          */
 413                         if (target_ip) {
 414                                 hammer_test_inode(target_ip);
 415                                 hammer_rel_inode(target_ip, 0);
 416                         }
 417
 418                         if (record->flags & HAMMER_RECF_ALLOCDATA) {
 419                                 --hammer_count_record_datas;
 420                                 kfree(record->data, M_HAMMER);
 421                                 record->flags &= ~HAMMER_RECF_ALLOCDATA;
 422                         }
 423
 424                         /*
 425                          * Release the reservation.  If the record was not
 426                          * committed return the reservation before
 427                          * releasing it.
 428                          */
 429                         if ((resv = record->resv) != NULL) {
 430                                 if ((record->flags & HAMMER_RECF_COMMITTED) == 0) {
 431                                         hammer_blockmap_reserve_undo(
 432                                                 resv,
 433                                                 record->leaf.data_offset,
 434                                                 record->leaf.data_len);
 435                                 }
 436                                 hammer_blockmap_reserve_complete(hmp, resv);
 437                                 record->resv = NULL;
 438                         }
 439                         record->data = NULL;
 440                         --hammer_count_records;
 441                         kfree(record, M_HAMMER);
 442                 }
 443         }
 444 }
 445
 446 /*
 447  * Record visibility depends on whether the record is being accessed by
 448  * the backend or the frontend.
 449  *
 450  * Return non-zero if the record is visible, zero if it isn't or if it is
 451  * deleted.
 452  */
 453 static __inline
 454 int
 455 hammer_ip_iterate_mem_good(hammer_cursor_t cursor, hammer_record_t record)
 456 {
 457         if (cursor->flags & HAMMER_CURSOR_BACKEND) {
 458                 if (record->flags & HAMMER_RECF_DELETED_BE)
 459                         return(0);
 460         } else {
 461                 if (record->flags & HAMMER_RECF_DELETED_FE)
 462                         return(0);
 463         }
 464         return(1);
 465 }
 466
 467 /*
 468  * This callback is used as part of the RB_SCAN function for in-memory
 469  * records.  We terminate it (return -1) as soon as we get a match.
 470  *
 471  * This routine is used by frontend code.
 472  *
 473  * The primary compare code does not account for ASOF lookups.  This
 474  * code handles that case as well as a few others.
 475  */
 476 static
 477 int
 478 hammer_rec_scan_callback(hammer_record_t rec, void *data)
 479 {
 480         hammer_cursor_t cursor = data;
 481
 482         /*
 483          * We terminate on success, so this should be NULL on entry.
 484          */
 485         KKASSERT(cursor->iprec == NULL);
 486
 487         /*
 488          * Skip if the record was marked deleted.
 489          */
 490         if (hammer_ip_iterate_mem_good(cursor, rec) == 0)
 491                 return(0);
 492
 493         /*
 494          * Skip if not visible due to our as-of TID
 495          */
 496         if (cursor->flags & HAMMER_CURSOR_ASOF) {
 497                 if (cursor->asof < rec->leaf.base.create_tid)
 498                         return(0);
 499                 if (rec->leaf.base.delete_tid &&
 500                     cursor->asof >= rec->leaf.base.delete_tid) {
 501                         return(0);
 502                 }
 503         }
 504
 505         /*
 506          * ref the record.  The record is protected from backend B-Tree
 507          * interactions by virtue of the cursor's IP lock.
 508          */
 509         hammer_ref(&rec->lock);
 510
 511         /*
 512          * The record may have been deleted while we were blocked.
 513          */
 514         if (hammer_ip_iterate_mem_good(cursor, rec) == 0) {
 515                 hammer_rel_mem_record(rec);
 516                 return(0);
 517         }
 518
 519         /*
 520          * Set the matching record and stop the scan.
 521          */
 522         cursor->iprec = rec;
 523         return(-1);
 524 }
 525
 526
 527 /*
 528  * Lookup an in-memory record given the key specified in the cursor.  Works
 529  * just like hammer_btree_lookup() but operates on an inode's in-memory
 530  * record list.
 531  *
 532  * The lookup must fail if the record is marked for deferred deletion.
 533  */
 534 static
 535 int
 536 hammer_mem_lookup(hammer_cursor_t cursor)
 537 {
 538         int error;
 539
 540         KKASSERT(cursor->ip);
 541         if (cursor->iprec) {
 542                 hammer_rel_mem_record(cursor->iprec);
 543                 cursor->iprec = NULL;
 544         }
 545         hammer_rec_rb_tree_RB_SCAN(&cursor->ip->rec_tree, hammer_rec_find_cmp,
 546                                    hammer_rec_scan_callback, cursor);
 547
 548         if (cursor->iprec == NULL)
 549                 error = ENOENT;
 550         else
 551                 error = 0;
 552         return(error);
 553 }
 554
 555 /*
 556  * hammer_mem_first() - locate the first in-memory record matching the
 557  * cursor within the bounds of the key range.
 558  */
 559 static
 560 int
 561 hammer_mem_first(hammer_cursor_t cursor)
 562 {
 563         hammer_inode_t ip;
 564
 565         ip = cursor->ip;
 566         KKASSERT(ip != NULL);
 567
 568         if (cursor->iprec) {
 569                 hammer_rel_mem_record(cursor->iprec);
 570                 cursor->iprec = NULL;
 571         }
 572
 573         hammer_rec_rb_tree_RB_SCAN(&ip->rec_tree, hammer_rec_scan_cmp,
 574                                    hammer_rec_scan_callback, cursor);
 575
 576         /*
 577          * Adjust scan.node and keep it linked into the RB-tree so we can
 578          * hold the cursor through third party modifications of the RB-tree.
 579          */
 580         if (cursor->iprec)
 581                 return(0);
 582         return(ENOENT);
 583 }
 584
 585 /************************************************************************
 586  *                   HAMMER IN-MEMORY RECORD FUNCTIONS                  *
 587  ************************************************************************
 588  *
 589  * These functions manipulate in-memory records.  Such records typically
 590  * exist prior to being committed to disk or indexed via the on-disk B-Tree.
 591  */
 592
 593 /*
 594  * Add a directory entry (dip,ncp) which references inode (ip).
 595  *
 596  * Note that the low 32 bits of the namekey are set temporarily to create
 597  * a unique in-memory record, and may be modified a second time when the
 598  * record is synchronized to disk.  In particular, the low 32 bits cannot be
 599  * all 0's when synching to disk, which is not handled here.
 600  *
 601  * NOTE: bytes does not include any terminating \0 on name, and name might
 602  * not be terminated.
 603  */
 604 int
 605 hammer_ip_add_directory(struct hammer_transaction *trans,
 606                      struct hammer_inode *dip, const char *name, int bytes,
 607                      struct hammer_inode *ip)
 608 {
 609         struct hammer_cursor cursor;
 610         hammer_record_t record;
 611         int error;
 612         int count;
 613         u_int32_t iterator;
 614
 615         record = hammer_alloc_mem_record(dip, HAMMER_ENTRY_SIZE(bytes));
 616         if (++trans->hmp->namekey_iterator == 0)
 617                 ++trans->hmp->namekey_iterator;
 618
 619         record->type = HAMMER_MEM_RECORD_ADD;
 620         record->leaf.base.localization = dip->obj_localization +
 621                                          HAMMER_LOCALIZE_MISC;
 622         record->leaf.base.obj_id = dip->obj_id;
 623         record->leaf.base.key = hammer_directory_namekey(name, bytes);
 624         record->leaf.base.key += trans->hmp->namekey_iterator;
 625         record->leaf.base.rec_type = HAMMER_RECTYPE_DIRENTRY;
 626         record->leaf.base.obj_type = ip->ino_leaf.base.obj_type;
 627         record->data->entry.obj_id = ip->obj_id;
 628         record->data->entry.localization = ip->obj_localization;
 629         bcopy(name, record->data->entry.name, bytes);
 630
 631         ++ip->ino_data.nlinks;
 632         hammer_modify_inode(ip, HAMMER_INODE_DDIRTY);
 633
 634         /*
 635          * Find an unused namekey.  Both the in-memory record tree and
 636          * the B-Tree are checked.  Exact matches also match create_tid
 637          * so use an ASOF search to (mostly) ignore it.
 638          *
 639          * delete-visibility is set so pending deletions do not give us
 640          * a false-negative on our ability to use an iterator.
 641          */
 642         hammer_init_cursor(trans, &cursor, &dip->cache[1], dip);
 643         cursor.key_beg = record->leaf.base;
 644         cursor.flags |= HAMMER_CURSOR_ASOF;
 645         cursor.flags |= HAMMER_CURSOR_DELETE_VISIBILITY;
 646         cursor.asof = ip->obj_asof;
 647
 648         count = 0;
 649         while (hammer_ip_lookup(&cursor) == 0) {
 650                 iterator = (u_int32_t)record->leaf.base.key + 1;
 651                 if (iterator == 0)
 652                         iterator = 1;
 653                 record->leaf.base.key &= ~0xFFFFFFFFLL;
 654                 record->leaf.base.key |= iterator;
 655                 cursor.key_beg.key = record->leaf.base.key;
 656                 if (++count == 1000000000) {
 657                         hammer_rel_mem_record(record);
 658                         error = ENOSPC;
 659                         goto failed;
 660                 }
 661         }
 662
 663         /*
 664          * The target inode and the directory entry are bound together.
 665          */
 666         record->target_ip = ip;
 667         record->flush_state = HAMMER_FST_SETUP;
 668         TAILQ_INSERT_TAIL(&ip->target_list, record, target_entry);
 669
 670         /*
 671          * The inode now has a dependancy and must be taken out of the idle
 672          * state.  An inode not in an idle state is given an extra reference.
 673          *
 674          * When transitioning to a SETUP state flag for an automatic reflush
 675          * when the dependancies are disposed of if someone is waiting on
 676          * the inode.
 677          */
 678         if (ip->flush_state == HAMMER_FST_IDLE) {
 679                 hammer_ref(&ip->lock);
 680                 ip->flush_state = HAMMER_FST_SETUP;
 681                 if (ip->flags & HAMMER_INODE_FLUSHW)
 682                         ip->flags |= HAMMER_INODE_REFLUSH;
 683         }
 684         error = hammer_mem_add(record);
 685         if (error == 0) {
 686                 dip->ino_data.mtime = trans->time;
 687                 hammer_modify_inode(dip, HAMMER_INODE_MTIME);
 688         }
 689 failed:
 690         hammer_done_cursor(&cursor);
 691         return(error);
 692 }
 693
 694 /*
 695  * Delete the directory entry and update the inode link count.  The
 696  * cursor must be seeked to the directory entry record being deleted.
 697  *
 698  * The related inode should be share-locked by the caller.  The caller is
 699  * on the frontend.
 700  *
 701  * This function can return EDEADLK requiring the caller to terminate
 702  * the cursor, any locks, wait on the returned record, and retry.
 703  */
 704 int
 705 hammer_ip_del_directory(struct hammer_transaction *trans,
 706                      hammer_cursor_t cursor, struct hammer_inode *dip,
 707                      struct hammer_inode *ip)
 708 {
 709         hammer_record_t record;
 710         int error;
 711
 712         if (hammer_cursor_inmem(cursor)) {
 713                 /*
 714                  * In-memory (unsynchronized) records can simply be freed.
 715                  * Even though the HAMMER_RECF_DELETED_FE flag is ignored
 716                  * by the backend, we must still avoid races against the
 717                  * backend potentially syncing the record to the media.
 718                  *
 719                  * We cannot call hammer_ip_delete_record(), that routine may
 720                  * only be called from the backend.
 721                  */
 722                 record = cursor->iprec;
 723                 if (record->flags & HAMMER_RECF_INTERLOCK_BE) {
 724                         KKASSERT(cursor->deadlk_rec == NULL);
 725                         hammer_ref(&record->lock);
 726                         cursor->deadlk_rec = record;
 727                         error = EDEADLK;
 728                 } else {
 729                         KKASSERT(record->type == HAMMER_MEM_RECORD_ADD);
 730                         record->flags |= HAMMER_RECF_DELETED_FE;
 731                         error = 0;
 732                 }
 733         } else {
 734                 /*
 735                  * If the record is on-disk we have to queue the deletion by
 736                  * the record's key.  This also causes lookups to skip the
 737                  * record.
 738                  */
 739                 KKASSERT(dip->flags &
 740                          (HAMMER_INODE_ONDISK | HAMMER_INODE_DONDISK));
 741                 record = hammer_alloc_mem_record(dip, 0);
 742                 record->type = HAMMER_MEM_RECORD_DEL;
 743                 record->leaf.base = cursor->leaf->base;
 744
 745                 record->target_ip = ip;
 746                 record->flush_state = HAMMER_FST_SETUP;
 747                 TAILQ_INSERT_TAIL(&ip->target_list, record, target_entry);
 748
 749                 /*
 750                  * The inode now has a dependancy and must be taken out of
 751                  * the idle state.  An inode not in an idle state is given
 752                  * an extra reference.
 753                  *
 754                  * When transitioning to a SETUP state flag for an automatic
 755                  * reflush when the dependancies are disposed of if someone
 756                  * is waiting on the inode.
 757                  */
 758                 if (ip->flush_state == HAMMER_FST_IDLE) {
 759                         hammer_ref(&ip->lock);
 760                         ip->flush_state = HAMMER_FST_SETUP;
 761                         if (ip->flags & HAMMER_INODE_FLUSHW)
 762                                 ip->flags |= HAMMER_INODE_REFLUSH;
 763                 }
 764
 765                 error = hammer_mem_add(record);
 766         }
 767
 768         /*
 769          * One less link.  The file may still be open in the OS even after
 770          * all links have gone away.
 771          *
 772          * We have to terminate the cursor before syncing the inode to
 773          * avoid deadlocking against ourselves.  XXX this may no longer
 774          * be true.
 775          *
 776          * If nlinks drops to zero and the vnode is inactive (or there is
 777          * no vnode), call hammer_inode_unloadable_check() to zonk the
 778          * inode.  If we don't do this here the inode will not be destroyed
 779          * on-media until we unmount.
 780          */
 781         if (error == 0) {
 782                 --ip->ino_data.nlinks;
 783                 hammer_modify_inode(ip, HAMMER_INODE_DDIRTY);
 784                 if (ip->ino_data.nlinks == 0 &&
 785                     (ip->vp == NULL || (ip->vp->v_flag & VINACTIVE))) {
 786                         hammer_done_cursor(cursor);
 787                         hammer_inode_unloadable_check(ip, 1);
 788                         hammer_flush_inode(ip, 0);
 789                 }
 790                 dip->ino_data.mtime = trans->time;
 791                 hammer_modify_inode(dip, HAMMER_INODE_MTIME);
 792
 793         }
 794         return(error);
 795 }
 796
 797 /*
 798  * Add a record to an inode.
 799  *
 800  * The caller must allocate the record with hammer_alloc_mem_record(ip) and
 801  * initialize the following additional fields:
 802  *
 803  * The related inode should be share-locked by the caller.  The caller is
 804  * on the frontend.
 805  *
 806  * record->rec.entry.base.base.key
 807  * record->rec.entry.base.base.rec_type
 808  * record->rec.entry.base.base.data_len
 809  * record->data         (a copy will be kmalloc'd if it cannot be embedded)
 810  */
 811 int
 812 hammer_ip_add_record(struct hammer_transaction *trans, hammer_record_t record)
 813 {
 814         hammer_inode_t ip = record->ip;
 815         int error;
 816
 817         KKASSERT(record->leaf.base.localization != 0);
 818         record->leaf.base.obj_id = ip->obj_id;
 819         record->leaf.base.obj_type = ip->ino_leaf.base.obj_type;
 820         error = hammer_mem_add(record);
 821         return(error);
 822 }
 823
 824 /*
 825  * Locate a bulk record in-memory.  Bulk records allow disk space to be
 826  * reserved so the front-end can flush large data writes without having
 827  * to queue the BIO to the flusher.  Only the related record gets queued
 828  * to the flusher.
 829  */
 830 static hammer_record_t
 831 hammer_ip_get_bulk(hammer_inode_t ip, off_t file_offset, int bytes)
 832 {
 833         hammer_record_t record;
 834         struct hammer_btree_leaf_elm leaf;
 835
 836         bzero(&leaf, sizeof(leaf));
 837         leaf.base.obj_id = ip->obj_id;
 838         leaf.base.key = file_offset + bytes;
 839         leaf.base.create_tid = 0;
 840         leaf.base.delete_tid = 0;
 841         leaf.base.rec_type = HAMMER_RECTYPE_DATA;
 842         leaf.base.obj_type = 0;                 /* unused */
 843         leaf.base.btype = HAMMER_BTREE_TYPE_RECORD;     /* unused */
 844         leaf.base.localization = ip->obj_localization + HAMMER_LOCALIZE_MISC;
 845         leaf.data_len = bytes;
 846
 847         record = hammer_rec_rb_tree_RB_LOOKUP_INFO(&ip->rec_tree, &leaf);
 848         if (record)
 849                 hammer_ref(&record->lock);
 850         return(record);
 851 }
 852
 853 /*
 854  * Reserve blockmap space placemarked with an in-memory record.
 855  *
 856  * This routine is called by the frontend in order to be able to directly
 857  * flush a buffer cache buffer.  The frontend has locked the related buffer
 858  * cache buffers and we should be able to manipulate any overlapping
 859  * in-memory records.
 860  *
 861  * The caller is responsible for adding the returned record.
 862  */
 863 hammer_record_t
 864 hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset, void *data, int bytes,
 865                    int *errorp)
 866 {
 867         hammer_record_t record;
 868         hammer_record_t conflict;
 869         int zone;
 870
 871         /*
 872          * Deal with conflicting in-memory records.  We cannot have multiple
 873          * in-memory records for the same offset without seriously confusing
 874          * the backend, including but not limited to the backend issuing
 875          * delete-create-delete sequences and asserting on the delete_tid
 876          * being the same as the create_tid.
 877          *
 878          * If we encounter a record with the backend interlock set we cannot
 879          * immediately delete it without confusing the backend.
 880          */
 881         while ((conflict = hammer_ip_get_bulk(ip, file_offset, bytes)) !=NULL) {
 882                 if (conflict->flags & HAMMER_RECF_INTERLOCK_BE) {
 883                         conflict->flags |= HAMMER_RECF_WANTED;
 884                         tsleep(conflict, 0, "hmrrc3", 0);
 885                 } else {
 886                         conflict->flags |= HAMMER_RECF_DELETED_FE;
 887                 }
 888                 hammer_rel_mem_record(conflict);
 889         }
 890
 891         /*
 892          * Create a record to cover the direct write.  This is called with
 893          * the related BIO locked so there should be no possible conflict.
 894          *
 895          * The backend is responsible for finalizing the space reserved in
 896          * this record.
 897          *
 898          * XXX bytes not aligned, depend on the reservation code to
 899          * align the reservation.
 900          */
 901         record = hammer_alloc_mem_record(ip, 0);
 902         zone = (bytes >= HAMMER_BUFSIZE) ? HAMMER_ZONE_LARGE_DATA_INDEX :
 903                                            HAMMER_ZONE_SMALL_DATA_INDEX;
 904         record->resv = hammer_blockmap_reserve(ip->hmp, zone, bytes,
 905                                                &record->leaf.data_offset,
 906                                                errorp);
 907         if (record->resv == NULL) {
 908                 kprintf("hammer_ip_add_bulk: reservation failed\n");
 909                 hammer_rel_mem_record(record);
 910                 return(NULL);
 911         }
 912         record->type = HAMMER_MEM_RECORD_DATA;
 913         record->leaf.base.rec_type = HAMMER_RECTYPE_DATA;
 914         record->leaf.base.obj_type = ip->ino_leaf.base.obj_type;
 915         record->leaf.base.obj_id = ip->obj_id;
 916         record->leaf.base.key = file_offset + bytes;
 917         record->leaf.base.localization = ip->obj_localization +
 918                                          HAMMER_LOCALIZE_MISC;
 919         record->leaf.data_len = bytes;
 920         hammer_crc_set_leaf(data, &record->leaf);
 921         KKASSERT(*errorp == 0);
 922         return(record);
 923 }
 924
 925 /*
 926  * Frontend truncation code.  Scan in-memory records only.  On-disk records
 927  * and records in a flushing state are handled by the backend.  The vnops
 928  * setattr code will handle the block containing the truncation point.
 929  *
 930  * Partial blocks are not deleted.
 931  */
 932 int
 933 hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size)
 934 {
 935         struct rec_trunc_info info;
 936
 937         switch(ip->ino_data.obj_type) {
 938         case HAMMER_OBJTYPE_REGFILE:
 939                 info.rec_type = HAMMER_RECTYPE_DATA;
 940                 break;
 941         case HAMMER_OBJTYPE_DBFILE:
 942                 info.rec_type = HAMMER_RECTYPE_DB;
 943                 break;
 944         default:
 945                 return(EINVAL);
 946         }
 947         info.trunc_off = file_size;
 948         hammer_rec_rb_tree_RB_SCAN(&ip->rec_tree, hammer_rec_trunc_cmp,
 949                                    hammer_frontend_trunc_callback, &info);
 950         return(0);
 951 }
 952
 953 static int
 954 hammer_frontend_trunc_callback(hammer_record_t record, void *data __unused)
 955 {
 956         if (record->flags & HAMMER_RECF_DELETED_FE)
 957                 return(0);
 958         if (record->flush_state == HAMMER_FST_FLUSH)
 959                 return(0);
 960         KKASSERT((record->flags & HAMMER_RECF_INTERLOCK_BE) == 0);
 961         hammer_ref(&record->lock);
 962         record->flags |= HAMMER_RECF_DELETED_FE;
 963         hammer_rel_mem_record(record);
 964         return(0);
 965 }
 966
 967 /*
 968  * Return 1 if the caller must check for and delete existing records
 969  * before writing out a new data record.
 970  *
 971  * Return 0 if the caller can just insert the record into the B-Tree without
 972  * checking.
 973  */
 974 static int
 975 hammer_record_needs_overwrite_delete(hammer_record_t record)
 976 {
 977         hammer_inode_t ip = record->ip;
 978         int64_t file_offset;
 979         int r;
 980
 981         if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE)
 982                 file_offset = record->leaf.base.key;
 983         else
 984                 file_offset = record->leaf.base.key - record->leaf.data_len;
 985         r = (file_offset < ip->save_trunc_off);
 986         if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
 987                 if (ip->save_trunc_off <= record->leaf.base.key)
 988                         ip->save_trunc_off = record->leaf.base.key + 1;
 989         } else {
 990                 if (ip->save_trunc_off < record->leaf.base.key)
 991                         ip->save_trunc_off = record->leaf.base.key;
 992         }
 993         return(r);
 994 }
 995
 996 /*
 997  * Backend code.  Sync a record to the media.
 998  */
 999 int
1000 hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t record)
1001 {
1002         hammer_transaction_t trans = cursor->trans;
1003         int64_t file_offset;
1004         int bytes;
1005         void *bdata;
1006         int error;
1007         int doprop;
1008
1009         KKASSERT(record->flush_state == HAMMER_FST_FLUSH);
1010         KKASSERT(record->flags & HAMMER_RECF_INTERLOCK_BE);
1011         KKASSERT(record->leaf.base.localization != 0);
1012
1013         /*
1014          * Any direct-write related to the record must complete before we
1015          * can sync the record to the on-disk media.
1016          */
1017         if (record->flags & (HAMMER_RECF_DIRECT_IO | HAMMER_RECF_DIRECT_INVAL))
1018                 hammer_io_direct_wait(record);
1019
1020         /*
1021          * If this is a bulk-data record placemarker there may be an existing
1022          * record on-disk, indicating a data overwrite.  If there is the
1023          * on-disk record must be deleted before we can insert our new record.
1024          *
1025          * We've synthesized this record and do not know what the create_tid
1026          * on-disk is, nor how much data it represents.
1027          *
1028          * Keep in mind that (key) for data records is (base_offset + len),
1029          * not (base_offset).  Also, we only want to get rid of on-disk
1030          * records since we are trying to sync our in-memory record, call
1031          * hammer_ip_delete_range() with truncating set to 1 to make sure
1032          * it skips in-memory records.
1033          *
1034          * It is ok for the lookup to return ENOENT.
1035          *
1036          * NOTE OPTIMIZATION: sync_trunc_off is used to determine if we have
1037          * to call hammer_ip_delete_range() or not.  This also means we must
1038          * update sync_trunc_off() as we write.
1039          */
1040         if (record->type == HAMMER_MEM_RECORD_DATA &&
1041             hammer_record_needs_overwrite_delete(record)) {
1042                 file_offset = record->leaf.base.key - record->leaf.data_len;
1043                 bytes = (record->leaf.data_len + HAMMER_BUFMASK) &
1044                         ~HAMMER_BUFMASK;
1045                 KKASSERT((file_offset & HAMMER_BUFMASK) == 0);
1046                 error = hammer_ip_delete_range(
1047                                 cursor, record->ip,
1048                                 file_offset, file_offset + bytes - 1,
1049                                 1);
1050                 if (error && error != ENOENT)
1051                         goto done;
1052         }
1053
1054         /*
1055          * If this is a general record there may be an on-disk version
1056          * that must be deleted before we can insert the new record.
1057          */
1058         if (record->type == HAMMER_MEM_RECORD_GENERAL) {
1059                 error = hammer_delete_general(cursor, record->ip,
1060                                               &record->leaf);
1061                 if (error && error != ENOENT)
1062                         goto done;
1063         }
1064
1065         /*
1066          * Setup the cursor.
1067          */
1068         hammer_normalize_cursor(cursor);
1069         cursor->key_beg = record->leaf.base;
1070         cursor->flags &= ~HAMMER_CURSOR_INITMASK;
1071         cursor->flags |= HAMMER_CURSOR_BACKEND;
1072         cursor->flags &= ~HAMMER_CURSOR_INSERT;
1073
1074         /*
1075          * Records can wind up on-media before the inode itself is on-media.
1076          * Flag the case.
1077          */
1078         record->ip->flags |= HAMMER_INODE_DONDISK;
1079
1080         /*
1081          * If we are deleting a directory entry an exact match must be
1082          * found on-disk.
1083          */
1084         if (record->type == HAMMER_MEM_RECORD_DEL) {
1085                 error = hammer_btree_lookup(cursor);
1086                 if (error == 0) {
1087                         KKASSERT(cursor->iprec == NULL);
1088                         error = hammer_ip_delete_record(cursor, record->ip,
1089                                                         trans->tid);
1090                         if (error == 0) {
1091                                 record->flags |= HAMMER_RECF_DELETED_FE;
1092                                 record->flags |= HAMMER_RECF_DELETED_BE;
1093                                 record->flags |= HAMMER_RECF_COMMITTED;
1094                         }
1095                 }
1096                 goto done;
1097         }
1098
1099         /*
1100          * We are inserting.
1101          *
1102          * Issue a lookup to position the cursor and locate the cluster.  The
1103          * target key should not exist.  If we are creating a directory entry
1104          * we may have to iterate the low 32 bits of the key to find an unused
1105          * key.
1106          */
1107         hammer_sync_lock_sh(trans);
1108         cursor->flags |= HAMMER_CURSOR_INSERT;
1109         error = hammer_btree_lookup(cursor);
1110         if (hammer_debug_inode)
1111                 kprintf("DOINSERT LOOKUP %d\n", error);
1112         if (error == 0) {
1113                 kprintf("hammer_ip_sync_record: duplicate rec "
1114                         "at (%016llx)\n", record->leaf.base.key);
1115                 Debugger("duplicate record1");
1116                 error = EIO;
1117         }
1118 #if 0
1119         if (record->type == HAMMER_MEM_RECORD_DATA)
1120                 kprintf("sync_record  %016llx ---------------- %016llx %d\n",
1121                         record->leaf.base.key - record->leaf.data_len,
1122                         record->leaf.data_offset, error);
1123 #endif
1124
1125         if (error != ENOENT)
1126                 goto done_unlock;
1127
1128         /*
1129          * Allocate the record and data.  The result buffers will be
1130          * marked as being modified and further calls to
1131          * hammer_modify_buffer() will result in unneeded UNDO records.
1132          *
1133          * Support zero-fill records (data == NULL and data_len != 0)
1134          */
1135         if (record->type == HAMMER_MEM_RECORD_DATA) {
1136                 /*
1137                  * The data portion of a bulk-data record has already been
1138                  * committed to disk, we need only adjust the layer2
1139                  * statistics in the same transaction as our B-Tree insert.
1140                  */
1141                 KKASSERT(record->leaf.data_offset != 0);
1142                 error = hammer_blockmap_finalize(trans,
1143                                                  record->leaf.data_offset,
1144                                                  record->leaf.data_len);
1145         } else if (record->data && record->leaf.data_len) {
1146                 /*
1147                  * Wholely cached record, with data.  Allocate the data.
1148                  */
1149                 bdata = hammer_alloc_data(trans, record->leaf.data_len,
1150                                           record->leaf.base.rec_type,
1151                                           &record->leaf.data_offset,
1152                                           &cursor->data_buffer, &error);
1153                 if (bdata == NULL)
1154                         goto done_unlock;
1155                 hammer_crc_set_leaf(record->data, &record->leaf);
1156                 hammer_modify_buffer(trans, cursor->data_buffer, NULL, 0);
1157                 bcopy(record->data, bdata, record->leaf.data_len);
1158                 hammer_modify_buffer_done(cursor->data_buffer);
1159         } else {
1160                 /*
1161                  * Wholely cached record, without data.
1162                  */
1163                 record->leaf.data_offset = 0;
1164                 record->leaf.data_crc = 0;
1165         }
1166
1167         error = hammer_btree_insert(cursor, &record->leaf, &doprop);
1168         if (hammer_debug_inode && error)
1169                 kprintf("BTREE INSERT error %d @ %016llx:%d key %016llx\n", error, cursor->node->node_offset, cursor->index, record->leaf.base.key);
1170
1171         /*
1172          * Our record is on-disk, normally mark the in-memory version as
1173          * deleted.  If the record represented a directory deletion but
1174          * we had to sync a valid directory entry to disk we must convert
1175          * the record to a covering delete so the frontend does not have
1176          * visibility on the synced entry.
1177          */
1178         if (error == 0) {
1179                 if (doprop) {
1180                         hammer_btree_do_propagation(cursor,
1181                                                     record->ip->pfsm,
1182                                                     &record->leaf);
1183                 }
1184                 if (record->flags & HAMMER_RECF_CONVERT_DELETE) {
1185                         KKASSERT(record->type == HAMMER_MEM_RECORD_ADD);
1186                         record->flags &= ~HAMMER_RECF_DELETED_FE;
1187                         record->type = HAMMER_MEM_RECORD_DEL;
1188                         KKASSERT(record->flush_state == HAMMER_FST_FLUSH);
1189                         record->flags &= ~HAMMER_RECF_CONVERT_DELETE;
1190                         /* hammer_flush_record_done takes care of the rest */
1191                 } else {
1192                         record->flags |= HAMMER_RECF_DELETED_FE;
1193                         record->flags |= HAMMER_RECF_DELETED_BE;
1194                 }
1195                 record->flags |= HAMMER_RECF_COMMITTED;
1196         } else {
1197                 if (record->leaf.data_offset) {
1198                         hammer_blockmap_free(trans, record->leaf.data_offset,
1199                                              record->leaf.data_len);
1200                 }
1201         }
1202 done_unlock:
1203         hammer_sync_unlock(trans);
1204 done:
1205         return(error);
1206 }
1207
1208 /*
1209  * Add the record to the inode's rec_tree.  The low 32 bits of a directory
1210  * entry's key is used to deal with hash collisions in the upper 32 bits.
1211  * A unique 64 bit key is generated in-memory and may be regenerated a
1212  * second time when the directory record is flushed to the on-disk B-Tree.
1213  *
1214  * A referenced record is passed to this function.  This function
1215  * eats the reference.  If an error occurs the record will be deleted.
1216  *
1217  * A copy of the temporary record->data pointer provided by the caller
1218  * will be made.
1219  */
1220 int
1221 hammer_mem_add(hammer_record_t record)
1222 {
1223         hammer_mount_t hmp = record->ip->hmp;
1224
1225         /*
1226          * Make a private copy of record->data
1227          */
1228         if (record->data)
1229                 KKASSERT(record->flags & HAMMER_RECF_ALLOCDATA);
1230
1231         /*
1232          * Insert into the RB tree.  A unique key should have already
1233          * been selected if this is a directory entry.
1234          */
1235         if (RB_INSERT(hammer_rec_rb_tree, &record->ip->rec_tree, record)) {
1236                 record->flags |= HAMMER_RECF_DELETED_FE;
1237                 hammer_rel_mem_record(record);
1238                 return (EEXIST);
1239         }
1240         ++hmp->count_newrecords;
1241         ++hmp->rsv_recs;
1242         ++record->ip->rsv_recs;
1243         record->ip->hmp->rsv_databytes += record->leaf.data_len;
1244         record->flags |= HAMMER_RECF_ONRBTREE;
1245         hammer_modify_inode(record->ip, HAMMER_INODE_XDIRTY);
1246         hammer_rel_mem_record(record);
1247         return(0);
1248 }
1249
1250 /************************************************************************
1251  *                   HAMMER INODE MERGED-RECORD FUNCTIONS               *
1252  ************************************************************************
1253  *
1254  * These functions augment the B-Tree scanning functions in hammer_btree.c
1255  * by merging in-memory records with on-disk records.
1256  */
1257
1258 /*
1259  * Locate a particular record either in-memory or on-disk.
1260  *
1261  * NOTE: This is basically a standalone routine, hammer_ip_next() may
1262  * NOT be called to iterate results.
1263  */
1264 int
1265 hammer_ip_lookup(hammer_cursor_t cursor)
1266 {
1267         int error;
1268
1269         /*
1270          * If the element is in-memory return it without searching the
1271          * on-disk B-Tree
1272          */
1273         KKASSERT(cursor->ip);
1274         error = hammer_mem_lookup(cursor);
1275         if (error == 0) {
1276                 cursor->leaf = &cursor->iprec->leaf;
1277                 return(error);
1278         }
1279         if (error != ENOENT)
1280                 return(error);
1281
1282         /*
1283          * If the inode has on-disk components search the on-disk B-Tree.
1284          */
1285         if ((cursor->ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DONDISK)) == 0)
1286                 return(error);
1287         error = hammer_btree_lookup(cursor);
1288         if (error == 0)
1289                 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_LEAF);
1290         return(error);
1291 }
1292
1293 /*
1294  * Locate the first record within the cursor's key_beg/key_end range,
1295  * restricted to a particular inode.  0 is returned on success, ENOENT
1296  * if no records matched the requested range, or some other error.
1297  *
1298  * When 0 is returned hammer_ip_next() may be used to iterate additional
1299  * records within the requested range.
1300  *
1301  * This function can return EDEADLK, requiring the caller to terminate
1302  * the cursor and try again.
1303  */
1304 int
1305 hammer_ip_first(hammer_cursor_t cursor)
1306 {
1307         hammer_inode_t ip = cursor->ip;
1308         int error;
1309
1310         KKASSERT(ip != NULL);
1311
1312         /*
1313          * Clean up fields and setup for merged scan
1314          */
1315         cursor->flags &= ~HAMMER_CURSOR_RETEST;
1316         cursor->flags |= HAMMER_CURSOR_ATEDISK | HAMMER_CURSOR_ATEMEM;
1317         cursor->flags |= HAMMER_CURSOR_DISKEOF | HAMMER_CURSOR_MEMEOF;
1318         if (cursor->iprec) {
1319                 hammer_rel_mem_record(cursor->iprec);
1320                 cursor->iprec = NULL;
1321         }
1322
1323         /*
1324          * Search the on-disk B-Tree.  hammer_btree_lookup() only does an
1325          * exact lookup so if we get ENOENT we have to call the iterate
1326          * function to validate the first record after the begin key.
1327          *
1328          * The ATEDISK flag is used by hammer_btree_iterate to determine
1329          * whether it must index forwards or not.  It is also used here
1330          * to select the next record from in-memory or on-disk.
1331          *
1332          * EDEADLK can only occur if the lookup hit an empty internal
1333          * element and couldn't delete it.  Since this could only occur
1334          * in-range, we can just iterate from the failure point.
1335          */
1336         if (ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DONDISK)) {
1337                 error = hammer_btree_lookup(cursor);
1338                 if (error == ENOENT || error == EDEADLK) {
1339                         cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1340                         if (hammer_debug_general & 0x2000)
1341                                 kprintf("error %d node %p %016llx index %d\n", error, cursor->node, cursor->node->node_offset, cursor->index);
1342                         error = hammer_btree_iterate(cursor);
1343                 }
1344                 if (error && error != ENOENT)
1345                         return(error);
1346                 if (error == 0) {
1347                         cursor->flags &= ~HAMMER_CURSOR_DISKEOF;
1348                         cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1349                 } else {
1350                         cursor->flags |= HAMMER_CURSOR_ATEDISK;
1351                 }
1352         }
1353
1354         /*
1355          * Search the in-memory record list (Red-Black tree).  Unlike the
1356          * B-Tree search, mem_first checks for records in the range.
1357          */
1358         error = hammer_mem_first(cursor);
1359         if (error && error != ENOENT)
1360                 return(error);
1361         if (error == 0) {
1362                 cursor->flags &= ~HAMMER_CURSOR_MEMEOF;
1363                 cursor->flags &= ~HAMMER_CURSOR_ATEMEM;
1364                 if (hammer_ip_iterate_mem_good(cursor, cursor->iprec) == 0)
1365                         cursor->flags |= HAMMER_CURSOR_ATEMEM;
1366         }
1367
1368         /*
1369          * This will return the first matching record.
1370          */
1371         return(hammer_ip_next(cursor));
1372 }
1373
1374 /*
1375  * Retrieve the next record in a merged iteration within the bounds of the
1376  * cursor.  This call may be made multiple times after the cursor has been
1377  * initially searched with hammer_ip_first().
1378  *
1379  * 0 is returned on success, ENOENT if no further records match the
1380  * requested range, or some other error code is returned.
1381  */
1382 int
1383 hammer_ip_next(hammer_cursor_t cursor)
1384 {
1385         hammer_btree_elm_t elm;
1386         hammer_record_t rec, save;
1387         int error;
1388         int r;
1389
1390 next_btree:
1391         /*
1392          * Load the current on-disk and in-memory record.  If we ate any
1393          * records we have to get the next one.
1394          *
1395          * If we deleted the last on-disk record we had scanned ATEDISK will
1396          * be clear and RETEST will be set, forcing a call to iterate.  The
1397          * fact that ATEDISK is clear causes iterate to re-test the 'current'
1398          * element.  If ATEDISK is set, iterate will skip the 'current'
1399          * element.
1400          *
1401          * Get the next on-disk record
1402          */
1403         if (cursor->flags & (HAMMER_CURSOR_ATEDISK|HAMMER_CURSOR_RETEST)) {
1404                 if ((cursor->flags & HAMMER_CURSOR_DISKEOF) == 0) {
1405                         error = hammer_btree_iterate(cursor);
1406                         cursor->flags &= ~HAMMER_CURSOR_RETEST;
1407                         if (error == 0) {
1408                                 cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
1409                                 hammer_cache_node(&cursor->ip->cache[1],
1410                                                   cursor->node);
1411                         } else {
1412                                 cursor->flags |= HAMMER_CURSOR_DISKEOF |
1413                                                  HAMMER_CURSOR_ATEDISK;
1414                         }
1415                 }
1416         }
1417
1418 next_memory:
1419         /*
1420          * Get the next in-memory record.
1421          *
1422          * hammer_rec_scan_cmp:  Is the record still in our general range,
1423          *                       (non-inclusive of snapshot exclusions)?
1424          * hammer_rec_scan_callback: Is the record in our snapshot?
1425          */
1426         if (cursor->flags & HAMMER_CURSOR_ATEMEM) {
1427                 if ((cursor->flags & HAMMER_CURSOR_MEMEOF) == 0) {
1428                         save = cursor->iprec;
1429                         cursor->iprec = NULL;
1430                         rec = save ? hammer_rec_rb_tree_RB_NEXT(save) : NULL;
1431                         while (rec) {
1432                                 if (hammer_rec_scan_cmp(rec, cursor) != 0)
1433                                         break;
1434                                 if (hammer_rec_scan_callback(rec, cursor) != 0)
1435                                         break;
1436                                 rec = hammer_rec_rb_tree_RB_NEXT(rec);
1437                         }
1438                         if (save)
1439                                 hammer_rel_mem_record(save);
1440                         if (cursor->iprec) {
1441                                 KKASSERT(cursor->iprec == rec);
1442                                 cursor->flags &= ~HAMMER_CURSOR_ATEMEM;
1443                         } else {
1444                                 cursor->flags |= HAMMER_CURSOR_MEMEOF;
1445                         }
1446                 }
1447         }
1448
1449         /*
1450          * The memory record may have become stale while being held in
1451          * cursor->iprec.  We are interlocked against the backend on
1452          * with regards to B-Tree entries.
1453          */
1454         if ((cursor->flags & HAMMER_CURSOR_ATEMEM) == 0) {
1455                 if (hammer_ip_iterate_mem_good(cursor, cursor->iprec) == 0) {
1456                         cursor->flags |= HAMMER_CURSOR_ATEMEM;
1457                         goto next_memory;
1458                 }
1459         }
1460
1461         /*
1462          * Extract either the disk or memory record depending on their
1463          * relative position.
1464          */
1465         error = 0;
1466         switch(cursor->flags & (HAMMER_CURSOR_ATEDISK | HAMMER_CURSOR_ATEMEM)) {
1467         case 0:
1468                 /*
1469                  * Both entries valid.   Compare the entries and nominally
1470                  * return the first one in the sort order.  Numerous cases
1471                  * require special attention, however.
1472                  */
1473                 elm = &cursor->node->ondisk->elms[cursor->index];
1474                 r = hammer_btree_cmp(&elm->base, &cursor->iprec->leaf.base);
1475
1476                 /*
1477                  * If the two entries differ only by their key (-2/2) or
1478                  * create_tid (-1/1), and are DATA records, we may have a
1479                  * nominal match.  We have to calculate the base file
1480                  * offset of the data.
1481                  */
1482                 if (r <= 2 && r >= -2 && r != 0 &&
1483                     cursor->ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE &&
1484                     cursor->iprec->type == HAMMER_MEM_RECORD_DATA) {
1485                         int64_t base1 = elm->leaf.base.key - elm->leaf.data_len;
1486                         int64_t base2 = cursor->iprec->leaf.base.key -
1487                                         cursor->iprec->leaf.data_len;
1488                         if (base1 == base2)
1489                                 r = 0;
1490                 }
1491
1492                 if (r < 0) {
1493                         error = hammer_btree_extract(cursor,
1494                                                      HAMMER_CURSOR_GET_LEAF);
1495                         cursor->flags |= HAMMER_CURSOR_ATEDISK;
1496                         break;
1497                 }
1498
1499                 /*
1500                  * If the entries match exactly the memory entry is either
1501                  * an on-disk directory entry deletion or a bulk data
1502                  * overwrite.  If it is a directory entry deletion we eat
1503                  * both entries.
1504                  *
1505                  * For the bulk-data overwrite case it is possible to have
1506                  * visibility into both, which simply means the syncer
1507                  * hasn't gotten around to doing the delete+insert sequence
1508                  * on the B-Tree.  Use the memory entry and throw away the
1509                  * on-disk entry.
1510                  *
1511                  * If the in-memory record is not either of these we
1512                  * probably caught the syncer while it was syncing it to
1513                  * the media.  Since we hold a shared lock on the cursor,
1514                  * the in-memory record had better be marked deleted at
1515                  * this point.
1516                  */
1517                 if (r == 0) {
1518                         if (cursor->iprec->type == HAMMER_MEM_RECORD_DEL) {
1519                                 if ((cursor->flags & HAMMER_CURSOR_DELETE_VISIBILITY) == 0) {
1520                                         cursor->flags |= HAMMER_CURSOR_ATEDISK;
1521                                         cursor->flags |= HAMMER_CURSOR_ATEMEM;
1522                                         goto next_btree;
1523                                 }
1524                         } else if (cursor->iprec->type == HAMMER_MEM_RECORD_DATA) {
1525                                 if ((cursor->flags & HAMMER_CURSOR_DELETE_VISIBILITY) == 0) {
1526                                         cursor->flags |= HAMMER_CURSOR_ATEDISK;
1527                                 }
1528                                 /* fall through to memory entry */
1529                         } else {
1530                                 panic("hammer_ip_next: duplicate mem/b-tree entry %p %d %08x", cursor->iprec, cursor->iprec->type, cursor->iprec->flags);
1531                                 cursor->flags |= HAMMER_CURSOR_ATEMEM;
1532                                 goto next_memory;
1533                         }
1534                 }
1535                 /* fall through to the memory entry */
1536         case HAMMER_CURSOR_ATEDISK:
1537                 /*
1538                  * Only the memory entry is valid.
1539                  */
1540                 cursor->leaf = &cursor->iprec->leaf;
1541                 cursor->flags |= HAMMER_CURSOR_ATEMEM;
1542
1543                 /*
1544                  * If the memory entry is an on-disk deletion we should have
1545                  * also had found a B-Tree record.  If the backend beat us
1546                  * to it it would have interlocked the cursor and we should
1547                  * have seen the in-memory record marked DELETED_FE.
1548                  */
1549                 if (cursor->iprec->type == HAMMER_MEM_RECORD_DEL &&
1550                     (cursor->flags & HAMMER_CURSOR_DELETE_VISIBILITY) == 0) {
1551                         panic("hammer_ip_next: del-on-disk with no b-tree entry iprec %p flags %08x", cursor->iprec, cursor->iprec->flags);
1552                 }
1553                 break;
1554         case HAMMER_CURSOR_ATEMEM:
1555                 /*
1556                  * Only the disk entry is valid
1557                  */
1558                 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_LEAF);
1559                 cursor->flags |= HAMMER_CURSOR_ATEDISK;
1560                 break;
1561         default:
1562                 /*
1563                  * Neither entry is valid
1564                  *
1565                  * XXX error not set properly
1566                  */
1567                 cursor->leaf = NULL;
1568                 error = ENOENT;
1569                 break;
1570         }
1571         return(error);
1572 }
1573
1574 /*
1575  * Resolve the cursor->data pointer for the current cursor position in
1576  * a merged iteration.
1577  */
1578 int
1579 hammer_ip_resolve_data(hammer_cursor_t cursor)
1580 {
1581         hammer_record_t record;
1582         int error;
1583
1584         if (hammer_cursor_inmem(cursor)) {
1585                 /*
1586                  * The data associated with an in-memory record is usually
1587                  * kmalloced, but reserve-ahead data records will have an
1588                  * on-disk reference.
1589                  *
1590                  * NOTE: Reserve-ahead data records must be handled in the
1591                  * context of the related high level buffer cache buffer
1592                  * to interlock against async writes.
1593                  */
1594                 record = cursor->iprec;
1595                 cursor->data = record->data;
1596                 error = 0;
1597                 if (cursor->data == NULL) {
1598                         KKASSERT(record->leaf.base.rec_type ==
1599                                  HAMMER_RECTYPE_DATA);
1600                         cursor->data = hammer_bread_ext(cursor->trans->hmp,
1601                                                     record->leaf.data_offset,
1602                                                     record->leaf.data_len,
1603                                                     &error,
1604                                                     &cursor->data_buffer);
1605                 }
1606         } else {
1607                 cursor->leaf = &cursor->node->ondisk->elms[cursor->index].leaf;
1608                 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA);
1609         }
1610         return(error);
1611 }
1612
1613 /*
1614  * Backend truncation / record replacement - delete records in range.
1615  *
1616  * Delete all records within the specified range for inode ip.  In-memory
1617  * records still associated with the frontend are ignored.
1618  *
1619  * If truncating is non-zero in-memory records associated with the back-end
1620  * are ignored.  If truncating is > 1 we can return EWOULDBLOCK.
1621  *
1622  * NOTES:
1623  *
1624  *      * An unaligned range will cause new records to be added to cover
1625  *        the edge cases. (XXX not implemented yet).
1626  *
1627  *      * Replacement via reservations (see hammer_ip_sync_record_cursor())
1628  *        also do not deal with unaligned ranges.
1629  *
1630  *      * ran_end is inclusive (e.g. 0,1023 instead of 0,1024).
1631  *
1632  *      * Record keys for regular file data have to be special-cased since
1633  *        they indicate the end of the range (key = base + bytes).
1634  *
1635  *      * This function may be asked to delete ridiculously huge ranges, for
1636  *        example if someone truncates or removes a 1TB regular file.  We
1637  *        must be very careful on restarts and we may have to stop w/
1638  *        EWOULDBLOCK to avoid blowing out the buffer cache.
1639  */
1640 int
1641 hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1642                        int64_t ran_beg, int64_t ran_end, int truncating)
1643 {
1644         hammer_transaction_t trans = cursor->trans;
1645         hammer_btree_leaf_elm_t leaf;
1646         int error;
1647         int64_t off;
1648         int64_t tmp64;
1649
1650 #if 0
1651         kprintf("delete_range %p %016llx-%016llx\n", ip, ran_beg, ran_end);
1652 #endif
1653
1654         KKASSERT(trans->type == HAMMER_TRANS_FLS);
1655 retry:
1656         hammer_normalize_cursor(cursor);
1657         cursor->key_beg.localization = ip->obj_localization +
1658                                        HAMMER_LOCALIZE_MISC;
1659         cursor->key_beg.obj_id = ip->obj_id;
1660         cursor->key_beg.create_tid = 0;
1661         cursor->key_beg.delete_tid = 0;
1662         cursor->key_beg.obj_type = 0;
1663
1664         if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1665                 cursor->key_beg.key = ran_beg;
1666                 cursor->key_beg.rec_type = HAMMER_RECTYPE_DB;
1667         } else {
1668                 /*
1669                  * The key in the B-Tree is (base+bytes), so the first possible
1670                  * matching key is ran_beg + 1.
1671                  */
1672                 cursor->key_beg.key = ran_beg + 1;
1673                 cursor->key_beg.rec_type = HAMMER_RECTYPE_DATA;
1674         }
1675
1676         cursor->key_end = cursor->key_beg;
1677         if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1678                 cursor->key_end.key = ran_end;
1679         } else {
1680                 tmp64 = ran_end + MAXPHYS + 1;  /* work around GCC-4 bug */
1681                 if (tmp64 < ran_end)
1682                         cursor->key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1683                 else
1684                         cursor->key_end.key = ran_end + MAXPHYS + 1;
1685         }
1686
1687         cursor->asof = ip->obj_asof;
1688         cursor->flags &= ~HAMMER_CURSOR_INITMASK;
1689         cursor->flags |= HAMMER_CURSOR_ASOF;
1690         cursor->flags |= HAMMER_CURSOR_DELETE_VISIBILITY;
1691         cursor->flags |= HAMMER_CURSOR_BACKEND;
1692         cursor->flags |= HAMMER_CURSOR_END_INCLUSIVE;
1693
1694         error = hammer_ip_first(cursor);
1695
1696         /*
1697          * Iterate through matching records and mark them as deleted.
1698          */
1699         while (error == 0) {
1700                 leaf = cursor->leaf;
1701
1702                 KKASSERT(leaf->base.delete_tid == 0);
1703                 KKASSERT(leaf->base.obj_id == ip->obj_id);
1704
1705                 /*
1706                  * There may be overlap cases for regular file data.  Also
1707                  * remember the key for a regular file record is (base + len),
1708                  * NOT (base).
1709                  *
1710                  * Note that do to duplicates (mem & media) allowed by
1711                  * DELETE_VISIBILITY, off can wind up less then ran_beg.
1712                  */
1713                 if (leaf->base.rec_type == HAMMER_RECTYPE_DATA) {
1714                         off = leaf->base.key - leaf->data_len;
1715                         /*
1716                          * Check the left edge case.  We currently do not
1717                          * split existing records.
1718                          */
1719                         if (off < ran_beg && leaf->base.key > ran_beg) {
1720                                 panic("hammer left edge case %016llx %d\n",
1721                                         leaf->base.key, leaf->data_len);
1722                         }
1723
1724                         /*
1725                          * Check the right edge case.  Note that the
1726                          * record can be completely out of bounds, which
1727                          * terminates the search.
1728                          *
1729                          * base->key is exclusive of the right edge while
1730                          * ran_end is inclusive of the right edge.  The
1731                          * (key - data_len) left boundary is inclusive.
1732                          *
1733                          * XXX theory-check this test at some point, are
1734                          * we missing a + 1 somewhere?  Note that ran_end
1735                          * could overflow.
1736                          */
1737                         if (leaf->base.key - 1 > ran_end) {
1738                                 if (leaf->base.key - leaf->data_len > ran_end)
1739                                         break;
1740                                 panic("hammer right edge case\n");
1741                         }
1742                 } else {
1743                         off = leaf->base.key;
1744                 }
1745
1746                 /*
1747                  * Delete the record.  When truncating we do not delete
1748                  * in-memory (data) records because they represent data
1749                  * written after the truncation.
1750                  *
1751                  * This will also physically destroy the B-Tree entry and
1752                  * data if the retention policy dictates.  The function
1753                  * will set HAMMER_CURSOR_RETEST to cause hammer_ip_next()
1754                  * to retest the new 'current' element.
1755                  */
1756                 if (truncating == 0 || hammer_cursor_ondisk(cursor)) {
1757                         error = hammer_ip_delete_record(cursor, ip, trans->tid);
1758                         /*
1759                          * If we have built up too many meta-buffers we risk
1760                          * deadlocking the kernel and must stop.  This can
1761                          * occur when deleting ridiculously huge files.
1762                          * sync_trunc_off is updated so the next cycle does
1763                          * not re-iterate records we have already deleted.
1764                          *
1765                          * This is only done with formal truncations.
1766                          */
1767                         if (truncating > 1 && error == 0 &&
1768                             hammer_flusher_meta_limit(ip->hmp)) {
1769                                 ip->sync_trunc_off = off;
1770                                 error = EWOULDBLOCK;
1771                         }
1772                 }
1773                 if (error)
1774                         break;
1775                 ran_beg = off;  /* for restart */
1776                 error = hammer_ip_next(cursor);
1777         }
1778         if (cursor->node)
1779                 hammer_cache_node(&ip->cache[1], cursor->node);
1780
1781         if (error == EDEADLK) {
1782                 hammer_done_cursor(cursor);
1783                 error = hammer_init_cursor(trans, cursor, &ip->cache[1], ip);
1784                 if (error == 0)
1785                         goto retry;
1786         }
1787         if (error == ENOENT)
1788                 error = 0;
1789         return(error);
1790 }
1791
1792 /*
1793  * This backend function deletes the specified record on-disk, similar to
1794  * delete_range but for a specific record.  Unlike the exact deletions
1795  * used when deleting a directory entry this function uses an ASOF search
1796  * like delete_range.
1797  *
1798  * This function may be called with ip->obj_asof set for a slave snapshot,
1799  * so don't use it.  We always delete non-historical records only.
1800  */
1801 static int
1802 hammer_delete_general(hammer_cursor_t cursor, hammer_inode_t ip,
1803                       hammer_btree_leaf_elm_t leaf)
1804 {
1805         hammer_transaction_t trans = cursor->trans;
1806         int error;
1807
1808         KKASSERT(trans->type == HAMMER_TRANS_FLS);
1809 retry:
1810         hammer_normalize_cursor(cursor);
1811         cursor->key_beg = leaf->base;
1812         cursor->asof = HAMMER_MAX_TID;
1813         cursor->flags &= ~HAMMER_CURSOR_INITMASK;
1814         cursor->flags |= HAMMER_CURSOR_ASOF;
1815         cursor->flags |= HAMMER_CURSOR_BACKEND;
1816         cursor->flags &= ~HAMMER_CURSOR_INSERT;
1817
1818         error = hammer_btree_lookup(cursor);
1819         if (error == 0) {
1820                 error = hammer_ip_delete_record(cursor, ip, trans->tid);
1821         }
1822         if (error == EDEADLK) {
1823                 hammer_done_cursor(cursor);
1824                 error = hammer_init_cursor(trans, cursor, &ip->cache[1], ip);
1825                 if (error == 0)
1826                         goto retry;
1827         }
1828         return(error);
1829 }
1830
1831 /*
1832  * This function deletes remaining auxillary records when an inode is
1833  * being deleted.  This function explicitly does not delete the
1834  * inode record, directory entry, data, or db records.  Those must be
1835  * properly disposed of prior to this call.
1836  */
1837 int
1838 hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip, int *countp)
1839 {
1840         hammer_transaction_t trans = cursor->trans;
1841         hammer_btree_leaf_elm_t leaf;
1842         int error;
1843
1844         KKASSERT(trans->type == HAMMER_TRANS_FLS);
1845 retry:
1846         hammer_normalize_cursor(cursor);
1847         cursor->key_beg.localization = ip->obj_localization +
1848                                        HAMMER_LOCALIZE_MISC;
1849         cursor->key_beg.obj_id = ip->obj_id;
1850         cursor->key_beg.create_tid = 0;
1851         cursor->key_beg.delete_tid = 0;
1852         cursor->key_beg.obj_type = 0;
1853         cursor->key_beg.rec_type = HAMMER_RECTYPE_CLEAN_START;
1854         cursor->key_beg.key = HAMMER_MIN_KEY;
1855
1856         cursor->key_end = cursor->key_beg;
1857         cursor->key_end.rec_type = HAMMER_RECTYPE_MAX;
1858         cursor->key_end.key = HAMMER_MAX_KEY;
1859
1860         cursor->asof = ip->obj_asof;
1861         cursor->flags &= ~HAMMER_CURSOR_INITMASK;
1862         cursor->flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1863         cursor->flags |= HAMMER_CURSOR_DELETE_VISIBILITY;
1864         cursor->flags |= HAMMER_CURSOR_BACKEND;
1865
1866         error = hammer_ip_first(cursor);
1867
1868         /*
1869          * Iterate through matching records and mark them as deleted.
1870          */
1871         while (error == 0) {
1872                 leaf = cursor->leaf;
1873
1874                 KKASSERT(leaf->base.delete_tid == 0);
1875
1876                 /*
1877                  * Mark the record and B-Tree entry as deleted.  This will
1878                  * also physically delete the B-Tree entry, record, and
1879                  * data if the retention policy dictates.  The function
1880                  * will set HAMMER_CURSOR_RETEST to cause hammer_ip_next()
1881                  * to retest the new 'current' element.
1882                  *
1883                  * Directory entries (and delete-on-disk directory entries)
1884                  * must be synced and cannot be deleted.
1885                  */
1886                 error = hammer_ip_delete_record(cursor, ip, trans->tid);
1887                 ++*countp;
1888                 if (error)
1889                         break;
1890                 error = hammer_ip_next(cursor);
1891         }
1892         if (cursor->node)
1893                 hammer_cache_node(&ip->cache[1], cursor->node);
1894         if (error == EDEADLK) {
1895                 hammer_done_cursor(cursor);
1896                 error = hammer_init_cursor(trans, cursor, &ip->cache[1], ip);
1897                 if (error == 0)
1898                         goto retry;
1899         }
1900         if (error == ENOENT)
1901                 error = 0;
1902         return(error);
1903 }
1904
1905 /*
1906  * Delete the record at the current cursor.  On success the cursor will
1907  * be positioned appropriately for an iteration but may no longer be at
1908  * a leaf node.
1909  *
1910  * This routine is only called from the backend.
1911  *
1912  * NOTE: This can return EDEADLK, requiring the caller to terminate the
1913  * cursor and retry.
1914  */
1915 int
1916 hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1917                         hammer_tid_t tid)
1918 {
1919         hammer_record_t iprec;
1920         hammer_mount_t hmp;
1921         int error;
1922
1923         KKASSERT(cursor->flags & HAMMER_CURSOR_BACKEND);
1924         KKASSERT(tid != 0);
1925         hmp = cursor->node->hmp;
1926
1927         /*
1928          * In-memory (unsynchronized) records can simply be freed.  This
1929          * only occurs in range iterations since all other records are
1930          * individually synchronized.  Thus there should be no confusion with
1931          * the interlock.
1932          *
1933          * An in-memory record may be deleted before being committed to disk,
1934          * but could have been accessed in the mean time.  The reservation
1935          * code will deal with the case.
1936          */
1937         if (hammer_cursor_inmem(cursor)) {
1938                 iprec = cursor->iprec;
1939                 KKASSERT((iprec->flags & HAMMER_RECF_INTERLOCK_BE) ==0);
1940                 iprec->flags |= HAMMER_RECF_DELETED_FE;
1941                 iprec->flags |= HAMMER_RECF_DELETED_BE;
1942                 return(0);
1943         }
1944
1945         /*
1946          * On-disk records are marked as deleted by updating their delete_tid.
1947          * This does not effect their position in the B-Tree (which is based
1948          * on their create_tid).
1949          *
1950          * Frontend B-Tree operations track inodes so we tell
1951          * hammer_delete_at_cursor() not to.
1952          */
1953         error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_LEAF);
1954
1955         if (error == 0) {
1956                 error = hammer_delete_at_cursor(
1957                                 cursor,
1958                                 HAMMER_DELETE_ADJUST | hammer_nohistory(ip),
1959                                 cursor->trans->tid,
1960                                 cursor->trans->time32,
1961                                 0, NULL);
1962         }
1963         return(error);
1964 }
1965
1966 /*
1967  * Delete the B-Tree element at the current cursor and do any necessary
1968  * mirror propagation.
1969  *
1970  * The cursor must be properly positioned for an iteration on return but
1971  * may be pointing at an internal element.
1972  *
1973  * An element can be un-deleted by passing a delete_tid of 0 with
1974  * HAMMER_DELETE_ADJUST.
1975  */
1976 int
1977 hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1978                         hammer_tid_t delete_tid, u_int32_t delete_ts,
1979                         int track, int64_t *stat_bytes)
1980 {
1981         struct hammer_btree_leaf_elm save_leaf;
1982         hammer_transaction_t trans;
1983         hammer_btree_leaf_elm_t leaf;
1984         hammer_node_t node;
1985         hammer_btree_elm_t elm;
1986         hammer_off_t data_offset;
1987         int32_t data_len;
1988         u_int16_t rec_type;
1989         int error;
1990         int icount;
1991         int doprop;
1992
1993         error = hammer_cursor_upgrade(cursor);
1994         if (error)
1995                 return(error);
1996
1997         trans = cursor->trans;
1998         node = cursor->node;
1999         elm = &node->ondisk->elms[cursor->index];
2000         leaf = &elm->leaf;
2001         KKASSERT(elm->base.btype == HAMMER_BTREE_TYPE_RECORD);
2002
2003         hammer_sync_lock_sh(trans);
2004         doprop = 0;
2005         icount = 0;
2006
2007         /*
2008          * Adjust the delete_tid.  Update the mirror_tid propagation field
2009          * as well.  delete_tid can be 0 (undelete -- used by mirroring).
2010          */
2011         if (delete_flags & HAMMER_DELETE_ADJUST) {
2012                 if (elm->base.rec_type == HAMMER_RECTYPE_INODE) {
2013                         if (elm->leaf.base.delete_tid == 0 && delete_tid)
2014                                 icount = -1;
2015                         if (elm->leaf.base.delete_tid && delete_tid == 0)
2016                                 icount = 1;
2017                 }
2018
2019                 hammer_modify_node(trans, node, elm, sizeof(*elm));
2020                 elm->leaf.base.delete_tid = delete_tid;
2021                 elm->leaf.delete_ts = delete_ts;
2022                 hammer_modify_node_done(node);
2023
2024                 if (elm->leaf.base.delete_tid > node->ondisk->mirror_tid) {
2025                         hammer_modify_node_field(trans, node, mirror_tid);
2026                         node->ondisk->mirror_tid = elm->leaf.base.delete_tid;
2027                         hammer_modify_node_done(node);
2028                         doprop = 1;
2029                         if (hammer_debug_general & 0x0002) {
2030                                 kprintf("delete_at_cursor: propagate %016llx"
2031                                         " @%016llx\n",
2032                                         elm->leaf.base.delete_tid,
2033                                         node->node_offset);
2034                         }
2035                 }
2036
2037                 /*
2038                  * Adjust for the iteration.  We have deleted the current
2039                  * element and want to clear ATEDISK so the iteration does
2040                  * not skip the element after, which now becomes the current
2041                  * element.  This element must be re-tested if doing an
2042                  * iteration, which is handled by the RETEST flag.
2043                  */
2044                 if ((cursor->flags & HAMMER_CURSOR_DISKEOF) == 0) {
2045                         cursor->flags |= HAMMER_CURSOR_RETEST;
2046                         cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
2047                 }
2048
2049                 /*
2050                  * An on-disk record cannot have the same delete_tid
2051                  * as its create_tid.  In a chain of record updates
2052                  * this could result in a duplicate record.
2053                  */
2054                 KKASSERT(elm->leaf.base.delete_tid !=
2055                          elm->leaf.base.create_tid);
2056         }
2057
2058         /*
2059          * Destroy the B-Tree element if asked (typically if a nohistory
2060          * file or mount, or when called by the pruning code).
2061          *
2062          * Adjust the ATEDISK flag to properly support iterations.
2063          */
2064         if (delete_flags & HAMMER_DELETE_DESTROY) {
2065                 data_offset = elm->leaf.data_offset;
2066                 data_len = elm->leaf.data_len;
2067                 rec_type = elm->leaf.base.rec_type;
2068                 if (doprop) {
2069                         save_leaf = elm->leaf;
2070                         leaf = &save_leaf;
2071                 }
2072                 if (elm->base.rec_type == HAMMER_RECTYPE_INODE &&
2073                     elm->leaf.base.delete_tid == 0) {
2074                         icount = -1;
2075                 }
2076
2077                 error = hammer_btree_delete(cursor);
2078                 if (error == 0) {
2079                         /*
2080                          * The deletion moves the next element (if any) to
2081                          * the current element position.  We must clear
2082                          * ATEDISK so this element is not skipped and we
2083                          * must set RETEST to force any iteration to re-test
2084                          * the element.
2085                          */
2086                         if ((cursor->flags & HAMMER_CURSOR_DISKEOF) == 0) {
2087                                 cursor->flags |= HAMMER_CURSOR_RETEST;
2088                                 cursor->flags &= ~HAMMER_CURSOR_ATEDISK;
2089                         }
2090                 }
2091                 if (error == 0) {
2092                         switch(data_offset & HAMMER_OFF_ZONE_MASK) {
2093                         case HAMMER_ZONE_LARGE_DATA:
2094                         case HAMMER_ZONE_SMALL_DATA:
2095                         case HAMMER_ZONE_META:
2096                                 hammer_blockmap_free(trans,
2097                                                      data_offset, data_len);
2098                                 break;
2099                         default:
2100                                 break;
2101                         }
2102                 }
2103         }
2104
2105         /*
2106          * Track inode count and next_tid.  This is used by the mirroring
2107          * and PFS code.  icount can be negative, zero, or positive.
2108          */
2109         if (error == 0 && track) {
2110                 if (icount) {
2111                         hammer_modify_volume_field(trans, trans->rootvol,
2112                                                    vol0_stat_inodes);
2113                         trans->rootvol->ondisk->vol0_stat_inodes += icount;
2114                         hammer_modify_volume_done(trans->rootvol);
2115                 }
2116                 if (trans->rootvol->ondisk->vol0_next_tid < delete_tid) {
2117                         hammer_modify_volume(trans, trans->rootvol, NULL, 0);
2118                         trans->rootvol->ondisk->vol0_next_tid = delete_tid;
2119                         hammer_modify_volume_done(trans->rootvol);
2120                 }
2121         }
2122
2123         /*
2124          * mirror_tid propagation occurs if the node's mirror_tid had to be
2125          * updated while adjusting the delete_tid.
2126          *
2127          * This occurs when deleting even in nohistory mode, but does not
2128          * occur when pruning an already-deleted node.
2129          *
2130          * cursor->ip is NULL when called from the pruning, mirroring,
2131          * and pfs code.  If non-NULL propagation will be conditionalized
2132          * on whether the PFS is in no-history mode or not.
2133          */
2134         if (doprop) {
2135                 if (cursor->ip)
2136                         hammer_btree_do_propagation(cursor, cursor->ip->pfsm, leaf);
2137                 else
2138                         hammer_btree_do_propagation(cursor, NULL, leaf);
2139         }
2140         hammer_sync_unlock(trans);
2141         return (error);
2142 }
2143
2144 /*
2145  * Determine whether we can remove a directory.  This routine checks whether
2146  * a directory is empty or not and enforces flush connectivity.
2147  *
2148  * Flush connectivity requires that we block if the target directory is
2149  * currently flushing, otherwise it may not end up in the same flush group.
2150  *
2151  * Returns 0 on success, ENOTEMPTY or EDEADLK (or other errors) on failure.
2152  */
2153 int
2154 hammer_ip_check_directory_empty(hammer_transaction_t trans, hammer_inode_t ip)
2155 {
2156         struct hammer_cursor cursor;
2157         int error;
2158
2159         /*
2160          * Check directory empty
2161          */
2162         hammer_init_cursor(trans, &cursor, &ip->cache[1], ip);
2163
2164         cursor.key_beg.localization = ip->obj_localization +
2165                                       HAMMER_LOCALIZE_MISC;
2166         cursor.key_beg.obj_id = ip->obj_id;
2167         cursor.key_beg.create_tid = 0;
2168         cursor.key_beg.delete_tid = 0;
2169         cursor.key_beg.obj_type = 0;
2170         cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE + 1;
2171         cursor.key_beg.key = HAMMER_MIN_KEY;
2172
2173         cursor.key_end = cursor.key_beg;
2174         cursor.key_end.rec_type = 0xFFFF;
2175         cursor.key_end.key = HAMMER_MAX_KEY;
2176
2177         cursor.asof = ip->obj_asof;
2178         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2179
2180         error = hammer_ip_first(&cursor);
2181         if (error == ENOENT)
2182                 error = 0;
2183         else if (error == 0)
2184                 error = ENOTEMPTY;
2185         hammer_done_cursor(&cursor);
2186         return(error);
2187 }
2188