fs/hfs/bnode.c

   1 /*
   2  * linux/fs/hfs/bnode.c
   3  *
   4  * Copyright (C) 1995-1997  Paul H. Hargrove
   5  * This file may be distributed under the terms of the GNU Public License.
   6  *
   7  * This file contains the code to access nodes in the B-tree structure.
   8  *
   9  * "XXX" in a comment is a note to myself to consider changing something.
  10  *
  11  * In function preconditions the term "valid" applied to a pointer to
  12  * a structure means that the pointer is non-NULL and the structure it
  13  * points to has all fields initialized to consistent values.
  14  *
  15  * The code in this file initializes some structures which contain
  16  * pointers by calling memset(&foo, 0, sizeof(foo)).
  17  * This produces the desired behavior only due to the non-ANSI
  18  * assumption that the machine representation of NULL is all zeros.
  19  */
  20
  21 #include "hfs_btree.h"
  22
  23 /*================ File-local variables ================*/
  24
  25 /* debugging statistics */
  26 #if defined(DEBUG_BNODES) || defined(DEBUG_ALL)
  27 int bnode_count = 0;
  28 #endif
  29
  30 /*================ Global functions ================*/
  31
  32 /*
  33  * hfs_bnode_delete()
  34  *
  35  * Description:
  36  *   This function is called to remove a bnode from the cache and
  37  *   release its resources.
  38  * Input Variable(s):
  39  *   struct hfs_bnode *bn: Pointer to the (struct hfs_bnode) to be
  40  *   removed from the cache.
  41  * Output Variable(s):
  42  *   NONE
  43  * Returns:
  44  *   void
  45  * Preconditions:
  46  *   'bn' points to a "valid" (struct hfs_bnode).
  47  * Postconditions:
  48  *   The node 'bn' is removed from the cache, its memory freed and its
  49  *   buffer (if any) released.
  50  */
  51 void hfs_bnode_delete(struct hfs_bnode *bn)
  52 {
  53 #if defined(DEBUG_BNODES) || defined(DEBUG_ALL)
  54         --bnode_count;
  55 #endif
  56         /* join neighbors */
  57         if (bn->next) {
  58                 bn->next->prev = bn->prev;
  59         }
  60         if (bn->prev) {
  61                 bn->prev->next = bn->next;
  62         }
  63         /* fix cache slot if necessary */
  64         if (bhash(bn->tree, bn->node) == bn) {
  65                 bhash(bn->tree, bn->node) = bn->next;
  66         }
  67         /* release resources */
  68         hfs_buffer_put(bn->buf); /* safe: checks for NULL argument */
  69         HFS_DELETE(bn);
  70 }
  71
  72
  73 /*
  74  * hfs_bnode_read()
  75  *
  76  * Description:
  77  *   This function creates a (struct hfs_bnode) and, if appropriate,
  78  *   inserts it in the cache.
  79  * Input Variable(s):
  80  *   struct hfs_bnode *bnode: pointer to the new bnode.
  81  *   struct hfs_btree *tree: pointer to the (struct hfs_btree)
  82  *    containing the desired node
  83  *   hfs_u32 node: the number of the desired node.
  84  *   int sticky: the value to assign to the 'sticky' field.
  85  * Output Variable(s):
  86  *   NONE
  87  * Returns:
  88  *   (struct hfs_bnode *) pointing to the newly created bnode or NULL.
  89  * Preconditions:
  90  *   'bnode' points to a "valid" (struct hfs_bnode).
  91  *   'tree' points to a "valid" (struct hfs_btree).
  92  *   'node' is an existing node number in the B-tree.
  93  * Postconditions:
  94  *   The following are true of 'bnode' upon return:
  95  *    The 'magic' field is set to indicate a valid (struct hfs_bnode).
  96  *    The 'sticky', 'tree' and 'node' fields are initialized to the
  97  *    values of the of the corresponding arguments.
  98  *    If the 'sticky' argument is zero then the fields 'prev' and
  99  *    'next' are initialized by inserting the (struct hfs_bnode) in the
 100  *    linked list of the appropriate cache slot; otherwise they are
 101  *    initialized to NULL.
 102  *    The data is read from disk (or buffer cache) and the 'buf' field
 103  *    points to the buffer for that data.
 104  *    If no other processes tried to access this node while this
 105  *    process was waiting on disk I/O (if necessary) then the
 106  *    remaining fields are zero ('count', 'resrv', 'lock') or NULL
 107  *    ('wqueue', 'rqueue') corresponding to no accesses.
 108  *    If there were access attempts during I/O then they were blocked
 109  *    until the I/O was complete, and the fields 'count', 'resrv',
 110  *    'lock', 'wqueue' and 'rqueue' reflect the results of unblocking
 111  *    those processes when the I/O was completed.
 112  */
 113 void hfs_bnode_read(struct hfs_bnode *bnode, struct hfs_btree *tree,
 114                     hfs_u32 node, int sticky)
 115 {
 116         struct NodeDescriptor *nd;
 117         int block, lcv;
 118         hfs_u16 curr, prev, limit;
 119
 120         /* Initialize the structure */
 121         memset(bnode, 0, sizeof(*bnode));
 122         bnode->magic = HFS_BNODE_MAGIC;
 123         bnode->tree = tree;
 124         bnode->node = node;
 125         bnode->sticky = sticky;
 126         hfs_init_waitqueue(&bnode->rqueue);
 127         hfs_init_waitqueue(&bnode->wqueue);
 128
 129         if (sticky == HFS_NOT_STICKY) {
 130                 /* Insert it in the cache if appropriate */
 131                 if ((bnode->next = bhash(tree, node))) {
 132                         bnode->next->prev = bnode;
 133                 }
 134                 bhash(tree, node) = bnode;
 135         }
 136
 137         /* Make the bnode look like it is being
 138            modified so other processes will wait for
 139            the I/O to complete */
 140         bnode->count = bnode->resrv = bnode->lock = 1;
 141
 142         /* Read in the node, possibly causing a schedule()
 143            call.  If the I/O fails then emit a warning.  Each
 144            process that was waiting on the bnode (including
 145            the current one) will notice the failure and
 146            hfs_bnode_relse() the node.  The last hfs_bnode_relse()
 147            will call hfs_bnode_delete() and discard the bnode.  */
 148
 149         block = hfs_extent_map(&tree->entry.u.file.data_fork, node, 0);
 150         if (!block) {
 151                 hfs_warn("hfs_bnode_read: bad node number 0x%08x\n", node);
 152         } else if (hfs_buffer_ok(bnode->buf =
 153                                  hfs_buffer_get(tree->sys_mdb, block, 1))) {
 154                 /* read in the NodeDescriptor */
 155                 nd = (struct NodeDescriptor *)hfs_buffer_data(bnode->buf);
 156                 bnode->ndFLink    = hfs_get_hl(nd->ndFLink);
 157                 bnode->ndBLink    = hfs_get_hl(nd->ndBLink);
 158                 bnode->ndType     = nd->ndType;
 159                 bnode->ndNHeight  = nd->ndNHeight;
 160                 bnode->ndNRecs    = hfs_get_hs(nd->ndNRecs);
 161
 162                 /* verify the integrity of the node */
 163                 prev = sizeof(struct NodeDescriptor);
 164                 limit = HFS_SECTOR_SIZE - sizeof(hfs_u16)*(bnode->ndNRecs + 1);
 165                 for (lcv=1; lcv <= (bnode->ndNRecs + 1); ++lcv) {
 166                         curr = hfs_get_hs(RECTBL(bnode, lcv));
 167                         if ((curr < prev) || (curr > limit)) {
 168                                 hfs_warn("hfs_bnode_read: corrupt node "
 169                                          "number 0x%08x\n", node);
 170                                 hfs_buffer_put(bnode->buf);
 171                                 bnode->buf = NULL;
 172                                 break;
 173                         }
 174                         prev = curr;
 175                 }
 176         }
 177
 178         /* Undo our fakery with the lock state and
 179            hfs_wake_up() anyone who we managed to trick */
 180         --bnode->count;
 181         bnode->resrv = bnode->lock = 0;
 182         hfs_wake_up(&bnode->rqueue);
 183 }
 184
 185 /*
 186  * hfs_bnode_lock()
 187  *
 188  * Description:
 189  *   This function does the locking of a bnode.
 190  * Input Variable(s):
 191  *   struct hfs_bnode *bn: pointer to the (struct hfs_bnode) to lock
 192  *   int lock_type: the type of lock desired
 193  * Output Variable(s):
 194  *   NONE
 195  * Returns:
 196  *   void
 197  * Preconditions:
 198  *   'bn' points to a "valid" (struct hfs_bnode).
 199  *   'lock_type' is a valid hfs_lock_t
 200  * Postconditions:
 201  *   The 'count' field of 'bn' is incremented by one.  If 'lock_type'
 202  *   is HFS_LOCK_RESRV the 'resrv' field is also incremented.
 203  */
 204 void hfs_bnode_lock(struct hfs_bnode_ref *bnr, int lock_type)
 205 {
 206         struct hfs_bnode *bn = bnr->bn;
 207
 208         if ((lock_type == bnr->lock_type) || !bn) {
 209                 return;
 210         }
 211
 212         if (bnr->lock_type == HFS_LOCK_WRITE) {
 213                 hfs_bnode_commit(bnr->bn);
 214         }
 215
 216         switch (lock_type) {
 217         default:
 218                 goto bail;
 219                 break;
 220
 221         case HFS_LOCK_READ:
 222                 /* We may not obtain read access if any process is
 223                    currently modifying or waiting to modify this node.
 224                    If we can't obtain access we wait on the rqueue
 225                    wait queue to be woken up by the modifying process
 226                    when it relinquishes its lock. */
 227                 switch (bnr->lock_type) {
 228                 default:
 229                         goto bail;
 230                         break;
 231
 232                 case HFS_LOCK_NONE:
 233                         while (bn->lock || waitqueue_active(&bn->wqueue)) {
 234                                 hfs_sleep_on(&bn->rqueue);
 235                         }
 236                         ++bn->count;
 237                         break;
 238                 }
 239                 break;
 240
 241         case HFS_LOCK_RESRV:
 242                 /* We may not obtain a reservation (read access with
 243                    an option to write later), if any process currently
 244                    holds a reservation on this node.  That includes
 245                    any process which is currently modifying this node.
 246                    If we can't obtain access, then we wait on the
 247                    rqueue wait queue to e woken up by the
 248                    reservation-holder when it calls hfs_bnode_relse. */
 249                 switch (bnr->lock_type) {
 250                 default:
 251                         goto bail;
 252                         break;
 253
 254                 case HFS_LOCK_NONE:
 255                         while (bn->resrv) {
 256                                 hfs_sleep_on(&bn->rqueue);
 257                         }
 258                         bn->resrv = 1;
 259                         ++bn->count;
 260                         break;
 261
 262                 case HFS_LOCK_WRITE:
 263                         bn->lock = 0;
 264                         hfs_wake_up(&bn->rqueue);
 265                         break;
 266                 }
 267                 break;
 268
 269         case HFS_LOCK_WRITE:
 270                 switch (bnr->lock_type) {
 271                 default:
 272                         goto bail;
 273                         break;
 274
 275                 case HFS_LOCK_NONE:
 276                         while (bn->resrv) {
 277                                 hfs_sleep_on(&bn->rqueue);
 278                         }
 279                         bn->resrv = 1;
 280                         ++bn->count;
 281                 case HFS_LOCK_RESRV:
 282                         while (bn->count > 1) {
 283                                 hfs_sleep_on(&bn->wqueue);
 284                         }
 285                         bn->lock = 1;
 286                         break;
 287                 }
 288                 break;
 289
 290         case HFS_LOCK_NONE:
 291                 switch (bnr->lock_type) {
 292                 default:
 293                         goto bail;
 294                         break;
 295
 296                 case HFS_LOCK_READ:
 297                         /* This process was reading this node.  If
 298                            there is now exactly one other process using
 299                            the node then hfs_wake_up() a (potentially
 300                            nonexistent) waiting process.  Note that I
 301                            refer to "a" process since the reservation
 302                            system ensures that only one process can
 303                            get itself on the wait queue.  */
 304                         if (bn->count == 2) {
 305                                 hfs_wake_up(&bn->wqueue);
 306                         }
 307                         break;
 308
 309                 case HFS_LOCK_WRITE:
 310                         /* This process was modifying this node.
 311                            Unlock the node and fall-through to the
 312                            HFS_LOCK_RESRV case, since a 'reservation'
 313                            is a prerequisite for HFS_LOCK_WRITE.  */
 314                         bn->lock = 0;
 315                 case HFS_LOCK_RESRV:
 316                         /* This process had placed a 'reservation' on
 317                            this node, indicating an intention to
 318                            possibly modify the node.  We can get to
 319                            this spot directly (if the 'reservation'
 320                            not converted to a HFS_LOCK_WRITE), or by
 321                            falling through from the above case if the
 322                            reservation was converted.
 323                            Since HFS_LOCK_RESRV and HFS_LOCK_WRITE
 324                            both block processes that want access
 325                            (HFS_LOCK_RESRV blocks other processes that
 326                            want reservations but allow HFS_LOCK_READ
 327                            accesses, while HFS_LOCK_WRITE must have
 328                            exclusive access and thus blocks both
 329                            types) we hfs_wake_up() any processes that
 330                            might be waiting for access.  If multiple
 331                            processes are waiting for a reservation
 332                            then the magic of process scheduling will
 333                            settle the dispute. */
 334                         bn->resrv = 0;
 335                         hfs_wake_up(&bn->rqueue);
 336                         break;
 337                 }
 338                 --bn->count;
 339                 break;
 340         }
 341         bnr->lock_type = lock_type;
 342         return;
 343
 344 bail:
 345         hfs_warn("hfs_bnode_lock: invalid lock change: %d->%d.\n",
 346                 bnr->lock_type, lock_type);
 347         return;
 348 }
 349
 350 /*
 351  * hfs_bnode_relse()
 352  *
 353  * Description:
 354  *   This function is called when a process is done using a bnode.  If
 355  *   the proper conditions are met then we call hfs_bnode_delete() to remove
 356  *   it from the cache.  If it is not deleted then we update its state
 357  *   to reflect one less process using it.
 358  * Input Variable(s):
 359  *   struct hfs_bnode *bn: pointer to the (struct hfs_bnode) to release.
 360  *   int lock_type: The type of lock held by the process releasing this node.
 361  * Output Variable(s):
 362  *   NONE
 363  * Returns:
 364  *   void
 365  * Preconditions:
 366  *   'bn' is NULL or points to a "valid" (struct hfs_bnode).
 367  * Postconditions:
 368  *   If 'bn' meets the appropriate conditions (see below) then it is
 369  *   kept in the cache and all fields are set to consistent values
 370  *   which reflect one less process using the node than upon entry.
 371  *   If 'bn' does not meet the conditions then it is deleted (see
 372  *   hfs_bnode_delete() for postconditions).
 373  *   In either case, if 'lock_type' is HFS_LOCK_WRITE
 374  *   then the corresponding buffer is dirtied.
 375  */
 376 void hfs_bnode_relse(struct hfs_bnode_ref *bnr)
 377 {
 378         struct hfs_bnode *bn;
 379
 380         if (!bnr || !(bn = bnr->bn)) {
 381                 return;
 382         }
 383
 384         /* We update the lock state of the node if it is still in use
 385            or if it is "sticky" (such as the B-tree head and root).
 386            Otherwise we just delete it.  */
 387         if ((bn->count > 1) || (waitqueue_active(&bn->rqueue)) || (bn->sticky != HFS_NOT_STICKY)) {
 388                 hfs_bnode_lock(bnr, HFS_LOCK_NONE);
 389         } else {
 390                 /* dirty buffer if we (might) have modified it */
 391                 if (bnr->lock_type == HFS_LOCK_WRITE) {
 392                         hfs_bnode_commit(bn);
 393                 }
 394                 hfs_bnode_delete(bn);
 395                 bnr->lock_type = HFS_LOCK_NONE;
 396         }
 397         bnr->bn = NULL;
 398 }
 399
 400 /*
 401  * hfs_bnode_find()
 402  *
 403  * Description:
 404  *   This function is called to obtain a bnode.  The cache is
 405  *   searched for the node.  If it not found there it is added to
 406  *   the cache by hfs_bnode_read().  There are two special cases node=0
 407  *   (the header node) and node='tree'->bthRoot (the root node), in
 408  *   which the nodes are obtained from fields of 'tree' without
 409  *   consulting or modifying the cache.
 410  * Input Variable(s):
 411  *   struct hfs_tree *tree: pointer to the (struct hfs_btree) from
 412  *    which to get a node.
 413  *   int node: the node number to get from 'tree'.
 414  *   int lock_type: The kind of access (HFS_LOCK_READ, or
 415  *    HFS_LOCK_RESRV) to obtain to the node
 416  * Output Variable(s):
 417  *   NONE
 418  * Returns:
 419  *   (struct hfs_bnode_ref) Reference to the requested node.
 420  * Preconditions:
 421  *   'tree' points to a "valid" (struct hfs_btree).
 422  * Postconditions:
 423  *   If 'node' refers to a valid node in 'tree' and 'lock_type' has
 424  *   one of the values listed above and no I/O errors occur then the
 425  *   value returned refers to a valid (struct hfs_bnode) corresponding
 426  *   to the requested node with the requested access type.  The node
 427  *   is also added to the cache if not previously present and not the
 428  *   root or header.
 429  *   If the conditions given above are not met, the bnode in the
 430  *   returned reference is NULL.
 431  */
 432 struct hfs_bnode_ref hfs_bnode_find(struct hfs_btree *tree,
 433                                     hfs_u32 node, int lock_type)
 434 {
 435         struct hfs_bnode *bn;
 436         struct hfs_bnode *empty = NULL;
 437         struct hfs_bnode_ref bnr;
 438
 439         bnr.lock_type = HFS_LOCK_NONE;
 440         bnr.bn = NULL;
 441
 442 #if defined(DEBUG_BNODES) || defined(DEBUG_ALL)
 443         hfs_warn("hfs_bnode_find: %c %d:%d\n",
 444                  lock_type==HFS_LOCK_READ?'R':
 445                         (lock_type==HFS_LOCK_RESRV?'V':'W'),
 446                  (int)ntohl(tree->entry.cnid), node);
 447 #endif
 448
 449         /* check special cases */
 450         if (!node) {
 451                 bn = &tree->head;
 452                 goto return_it;
 453         } else if (node == tree->bthRoot) {
 454                 bn = tree->root;
 455                 goto return_it;
 456         }
 457
 458 restart:
 459         /* look for the node in the cache. */
 460         bn = bhash(tree, node);
 461         while (bn && (bn->magic == HFS_BNODE_MAGIC)) {
 462                 if (bn->node == node) {
 463                         goto found_it;
 464                 }
 465                 bn = bn->next;
 466         }
 467
 468         if (!empty) {
 469 #if defined(DEBUG_BNODES) || defined(DEBUG_ALL)
 470                 ++bnode_count;
 471 #endif
 472                 if (HFS_NEW(empty)) {
 473                         goto restart;
 474                 }
 475                 return bnr;
 476         }
 477         bn = empty;
 478         hfs_bnode_read(bn, tree, node, HFS_NOT_STICKY);
 479         goto return_it;
 480
 481 found_it:
 482         /* check validity */
 483         if (bn->magic != HFS_BNODE_MAGIC) {
 484                 /* If we find a corrupt bnode then we return
 485                    NULL.  However, we don't try to remove it
 486                    from the cache or release its resources
 487                    since we have no idea what kind of trouble
 488                    we could get into that way. */
 489                 hfs_warn("hfs_bnode_find: bnode cache is corrupt.\n");
 490                 return bnr;
 491         }
 492         if (empty) {
 493 #if defined(DEBUG_BNODES) || defined(DEBUG_ALL)
 494                 --bnode_count;
 495 #endif
 496                 HFS_DELETE(empty);
 497         }
 498
 499 return_it:
 500         /* Wait our turn */
 501         bnr.bn = bn;
 502         hfs_bnode_lock(&bnr, lock_type);
 503
 504         /* Check for failure to read the node from disk */
 505         if (!hfs_buffer_ok(bn->buf)) {
 506                 hfs_bnode_relse(&bnr);
 507         }
 508
 509 #if defined(DEBUG_BNODES) || defined(DEBUG_ALL)
 510         if (!bnr.bn) {
 511                 hfs_warn("hfs_bnode_find: failed\n");
 512         } else {
 513                 hfs_warn("hfs_bnode_find: use %d(%d) lvl %d [%d]\n", bn->count,
 514                          bn->buf->b_count, bn->ndNHeight, bnode_count);
 515                 hfs_warn("hfs_bnode_find: blnk %u flnk %u recs %u\n",
 516                          bn->ndBLink, bn->ndFLink, bn->ndNRecs);
 517         }
 518 #endif
 519
 520         return bnr;
 521 }
 522
 523 /*
 524  * hfs_bnode_commit()
 525  *
 526  * Called to write a possibly dirty bnode back to disk.
 527  */
 528 void hfs_bnode_commit(struct hfs_bnode *bn)
 529 {
 530         if (hfs_buffer_ok(bn->buf)) {
 531                 struct NodeDescriptor *nd;
 532                 nd = (struct NodeDescriptor *)hfs_buffer_data(bn->buf);
 533
 534                 hfs_put_hl(bn->ndFLink, nd->ndFLink);
 535                 hfs_put_hl(bn->ndBLink, nd->ndBLink);
 536                 nd->ndType    = bn->ndType;
 537                 nd->ndNHeight = bn->ndNHeight;
 538                 hfs_put_hs(bn->ndNRecs, nd->ndNRecs);
 539                 hfs_buffer_dirty(bn->buf);
 540
 541                 /* increment write count */
 542                 hfs_mdb_dirty(bn->tree->sys_mdb);
 543         }
 544 }