sys/vfs/hammer/hammer_cursor.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_cursor.c,v 1.30 2008/06/17 04:02:38 dillon Exp $
  35  */
  36
  37 /*
  38  * HAMMER B-Tree index - cursor support routines
  39  */
  40 #include "hammer.h"
  41
  42 static int hammer_load_cursor_parent(hammer_cursor_t cursor, int try_exclusive);
  43
  44 /*
  45  * Initialize a fresh cursor using the B-Tree node cache.  If the cache
  46  * is not available initialize a fresh cursor at the root of the filesystem.
  47  */
  48 int
  49 hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
  50                    struct hammer_node **cache, hammer_inode_t ip)
  51 {
  52         hammer_volume_t volume;
  53         hammer_node_t node;
  54         int error;
  55
  56         bzero(cursor, sizeof(*cursor));
  57
  58         cursor->trans = trans;
  59
  60         /*
  61          * If the cursor operation is on behalf of an inode, lock
  62          * the inode.
  63          */
  64         if ((cursor->ip = ip) != NULL) {
  65                 ++ip->cursor_ip_refs;
  66                 if (trans->type == HAMMER_TRANS_FLS)
  67                         hammer_lock_ex(&ip->lock);
  68                 else
  69                         hammer_lock_sh(&ip->lock);
  70         }
  71
  72         /*
  73          * Step 1 - acquire a locked node from the cache if possible
  74          */
  75         if (cache && *cache) {
  76                 node = hammer_ref_node_safe(trans->hmp, cache, &error);
  77                 if (error == 0) {
  78                         hammer_lock_sh(&node->lock);
  79                         if (node->flags & HAMMER_NODE_DELETED) {
  80                                 hammer_unlock(&node->lock);
  81                                 hammer_rel_node(node);
  82                                 node = NULL;
  83                         }
  84                 }
  85         } else {
  86                 node = NULL;
  87         }
  88
  89         /*
  90          * Step 2 - If we couldn't get a node from the cache, get
  91          * the one from the root of the filesystem.
  92          */
  93         while (node == NULL) {
  94                 volume = hammer_get_root_volume(trans->hmp, &error);
  95                 if (error)
  96                         break;
  97                 node = hammer_get_node(trans->hmp,
  98                                        volume->ondisk->vol0_btree_root,
  99                                        0, &error);
 100                 hammer_rel_volume(volume, 0);
 101                 if (error)
 102                         break;
 103                 hammer_lock_sh(&node->lock);
 104
 105                 /*
 106                  * If someone got in before we could lock the node, retry.
 107                  */
 108                 if (node->flags & HAMMER_NODE_DELETED) {
 109                         hammer_unlock(&node->lock);
 110                         hammer_rel_node(node);
 111                         node = NULL;
 112                         continue;
 113                 }
 114                 if (volume->ondisk->vol0_btree_root != node->node_offset) {
 115                         hammer_unlock(&node->lock);
 116                         hammer_rel_node(node);
 117                         node = NULL;
 118                         continue;
 119                 }
 120         }
 121
 122         /*
 123          * Step 3 - finish initializing the cursor by acquiring the parent
 124          */
 125         cursor->node = node;
 126         if (error == 0)
 127                 error = hammer_load_cursor_parent(cursor, 0);
 128         KKASSERT(error == 0);
 129         /* if (error) hammer_done_cursor(cursor); */
 130         return(error);
 131 }
 132
 133 #if 0
 134 int
 135 hammer_reinit_cursor(hammer_cursor_t cursor)
 136 {
 137         hammer_transaction_t trans;
 138         hammer_inode_t ip;
 139         struct hammer_node **cache;
 140
 141         trans = cursor->trans;
 142         ip = cursor->ip;
 143         hammer_done_cursor(cursor);
 144         cache = ip ? &ip->cache[0] : NULL;
 145         error = hammer_init_cursor(trans, cursor, cache, ip);
 146         return (error);
 147 }
 148
 149 #endif
 150
 151 /*
 152  * Normalize a cursor.  Sometimes cursors can be left in a state
 153  * where node is NULL.  If the cursor is in this state, cursor up.
 154  */
 155 void
 156 hammer_normalize_cursor(hammer_cursor_t cursor)
 157 {
 158         if (cursor->node == NULL) {
 159                 KKASSERT(cursor->parent != NULL);
 160                 hammer_cursor_up(cursor);
 161         }
 162 }
 163
 164
 165 /*
 166  * We are finished with a cursor.  We NULL out various fields as sanity
 167  * check, in case the structure is inappropriately used afterwords.
 168  */
 169 void
 170 hammer_done_cursor(hammer_cursor_t cursor)
 171 {
 172         hammer_inode_t ip;
 173
 174         if (cursor->parent) {
 175                 hammer_unlock(&cursor->parent->lock);
 176                 hammer_rel_node(cursor->parent);
 177                 cursor->parent = NULL;
 178         }
 179         if (cursor->node) {
 180                 hammer_unlock(&cursor->node->lock);
 181                 hammer_rel_node(cursor->node);
 182                 cursor->node = NULL;
 183         }
 184         if (cursor->data_buffer) {
 185                 hammer_rel_buffer(cursor->data_buffer, 0);
 186                 cursor->data_buffer = NULL;
 187         }
 188         if (cursor->record_buffer) {
 189                 hammer_rel_buffer(cursor->record_buffer, 0);
 190                 cursor->record_buffer = NULL;
 191         }
 192         if ((ip = cursor->ip) != NULL) {
 193                 hammer_mem_done(cursor);
 194                 KKASSERT(ip->cursor_ip_refs > 0);
 195                 --ip->cursor_ip_refs;
 196                 hammer_unlock(&ip->lock);
 197                 cursor->ip = NULL;
 198         }
 199
 200
 201         /*
 202          * If we deadlocked this node will be referenced.  Do a quick
 203          * lock/unlock to wait for the deadlock condition to clear.
 204          */
 205         if (cursor->deadlk_node) {
 206                 hammer_lock_ex_ident(&cursor->deadlk_node->lock, "hmrdlk");
 207                 hammer_unlock(&cursor->deadlk_node->lock);
 208                 hammer_rel_node(cursor->deadlk_node);
 209                 cursor->deadlk_node = NULL;
 210         }
 211         if (cursor->deadlk_rec) {
 212                 hammer_wait_mem_record_ident(cursor->deadlk_rec, "hmmdlr");
 213                 hammer_rel_mem_record(cursor->deadlk_rec);
 214                 cursor->deadlk_rec = NULL;
 215         }
 216
 217         cursor->data = NULL;
 218         cursor->leaf = NULL;
 219         cursor->left_bound = NULL;
 220         cursor->right_bound = NULL;
 221         cursor->trans = NULL;
 222 }
 223
 224 /*
 225  * Upgrade cursor->node and cursor->parent to exclusive locks.  This
 226  * function can return EDEADLK.
 227  *
 228  * The lock must already be either held shared or already held exclusively
 229  * by us.
 230  *
 231  * If we fail to upgrade the lock and cursor->deadlk_node is NULL,
 232  * we add another reference to the node that failed and set
 233  * cursor->deadlk_node so hammer_done_cursor() can block on it.
 234  */
 235 int
 236 hammer_cursor_upgrade(hammer_cursor_t cursor)
 237 {
 238         int error;
 239
 240         error = hammer_lock_upgrade(&cursor->node->lock);
 241         if (error && cursor->deadlk_node == NULL) {
 242                 cursor->deadlk_node = cursor->node;
 243                 hammer_ref_node(cursor->deadlk_node);
 244         } else if (error == 0 && cursor->parent) {
 245                 error = hammer_lock_upgrade(&cursor->parent->lock);
 246                 if (error && cursor->deadlk_node == NULL) {
 247                         cursor->deadlk_node = cursor->parent;
 248                         hammer_ref_node(cursor->deadlk_node);
 249                 }
 250         }
 251         return(error);
 252 }
 253
 254 int
 255 hammer_cursor_upgrade_node(hammer_cursor_t cursor)
 256 {
 257         int error;
 258
 259         error = hammer_lock_upgrade(&cursor->node->lock);
 260         if (error && cursor->deadlk_node == NULL) {
 261                 cursor->deadlk_node = cursor->node;
 262                 hammer_ref_node(cursor->deadlk_node);
 263         }
 264         return(error);
 265 }
 266
 267 /*
 268  * Downgrade cursor->node and cursor->parent to shared locks.  This
 269  * function can return EDEADLK.
 270  */
 271 void
 272 hammer_cursor_downgrade(hammer_cursor_t cursor)
 273 {
 274         if (hammer_lock_excl_owned(&cursor->node->lock, curthread))
 275                 hammer_lock_downgrade(&cursor->node->lock);
 276         if (cursor->parent &&
 277             hammer_lock_excl_owned(&cursor->parent->lock, curthread)) {
 278                 hammer_lock_downgrade(&cursor->parent->lock);
 279         }
 280 }
 281
 282 /*
 283  * Seek the cursor to the specified node and index.
 284  *
 285  * The caller must ref the node prior to calling this routine and release
 286  * it after it returns.  If the seek succeeds the cursor will gain its own
 287  * ref on the node.
 288  */
 289 int
 290 hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, int index)
 291 {
 292         int error;
 293
 294         hammer_cursor_downgrade(cursor);
 295         error = 0;
 296
 297         if (cursor->node != node) {
 298                 hammer_unlock(&cursor->node->lock);
 299                 hammer_rel_node(cursor->node);
 300                 cursor->node = node;
 301                 hammer_ref_node(node);
 302                 hammer_lock_sh(&node->lock);
 303                 KKASSERT ((node->flags & HAMMER_NODE_DELETED) == 0);
 304
 305                 if (cursor->parent) {
 306                         hammer_unlock(&cursor->parent->lock);
 307                         hammer_rel_node(cursor->parent);
 308                         cursor->parent = NULL;
 309                         cursor->parent_index = 0;
 310                 }
 311                 error = hammer_load_cursor_parent(cursor, 0);
 312         }
 313         cursor->index = index;
 314         return (error);
 315 }
 316
 317 /*
 318  * Load the parent of cursor->node into cursor->parent.
 319  */
 320 static
 321 int
 322 hammer_load_cursor_parent(hammer_cursor_t cursor, int try_exclusive)
 323 {
 324         hammer_mount_t hmp;
 325         hammer_node_t parent;
 326         hammer_node_t node;
 327         hammer_btree_elm_t elm;
 328         int error;
 329         int i;
 330
 331         hmp = cursor->trans->hmp;
 332
 333         if (cursor->node->ondisk->parent) {
 334                 node = cursor->node;
 335                 parent = hammer_get_node(hmp, node->ondisk->parent, 0, &error);
 336                 if (error)
 337                         return(error);
 338                 if (try_exclusive) {
 339                         if (hammer_lock_ex_try(&parent->lock)) {
 340                                 hammer_rel_node(parent);
 341                                 return(EDEADLK);
 342                         }
 343                 } else {
 344                         hammer_lock_sh(&parent->lock);
 345                 }
 346                 KKASSERT ((parent->flags & HAMMER_NODE_DELETED) == 0);
 347                 elm = NULL;
 348                 for (i = 0; i < parent->ondisk->count; ++i) {
 349                         elm = &parent->ondisk->elms[i];
 350                         if (parent->ondisk->elms[i].internal.subtree_offset ==
 351                             node->node_offset) {
 352                                 break;
 353                         }
 354                 }
 355                 if (i == parent->ondisk->count) {
 356                         hammer_unlock(&parent->lock);
 357                         panic("Bad B-Tree link: parent %p node %p\n", parent, node);
 358                 }
 359                 KKASSERT(i != parent->ondisk->count);
 360                 cursor->parent = parent;
 361                 cursor->parent_index = i;
 362                 cursor->left_bound = &elm[0].internal.base;
 363                 cursor->right_bound = &elm[1].internal.base;
 364                 return(error);
 365         } else {
 366                 cursor->parent = NULL;
 367                 cursor->parent_index = 0;
 368                 cursor->left_bound = &hmp->root_btree_beg;
 369                 cursor->right_bound = &hmp->root_btree_end;
 370                 error = 0;
 371         }
 372         return(error);
 373 }
 374
 375 /*
 376  * Cursor up to our parent node.  Return ENOENT if we are at the root of
 377  * the filesystem.
 378  */
 379 int
 380 hammer_cursor_up(hammer_cursor_t cursor)
 381 {
 382         int error;
 383
 384         hammer_cursor_downgrade(cursor);
 385
 386         /*
 387          * If the parent is NULL we are at the root of the B-Tree and
 388          * return ENOENT.
 389          */
 390         if (cursor->parent == NULL)
 391                 return (ENOENT);
 392
 393         /*
 394          * Set the node to its parent.
 395          */
 396         hammer_unlock(&cursor->node->lock);
 397         hammer_rel_node(cursor->node);
 398         cursor->node = cursor->parent;
 399         cursor->index = cursor->parent_index;
 400         cursor->parent = NULL;
 401         cursor->parent_index = 0;
 402
 403         error = hammer_load_cursor_parent(cursor, 0);
 404         return(error);
 405 }
 406
 407 /*
 408  * Special cursor up given a locked cursor.  The orignal node is not
 409  * unlocked and released and the cursor is not downgraded.  If we are
 410  * unable to acquire and lock the parent, EDEADLK is returned.
 411  */
 412 int
 413 hammer_cursor_up_locked(hammer_cursor_t cursor)
 414 {
 415         hammer_node_t save;
 416         int error;
 417
 418         /*
 419          * If the parent is NULL we are at the root of the B-Tree and
 420          * return ENOENT.
 421          */
 422         if (cursor->parent == NULL)
 423                 return (ENOENT);
 424
 425         save = cursor->node;
 426
 427         /*
 428          * Set the node to its parent.
 429          */
 430         cursor->node = cursor->parent;
 431         cursor->index = cursor->parent_index;
 432         cursor->parent = NULL;
 433         cursor->parent_index = 0;
 434
 435         /*
 436          * load the new parent, attempt to exclusively lock it.  Note that
 437          * we are still holding the old parent (now cursor->node) exclusively
 438          * locked.  This can return EDEADLK.
 439          */
 440         error = hammer_load_cursor_parent(cursor, 1);
 441         if (error) {
 442                 cursor->parent = cursor->node;
 443                 cursor->parent_index = cursor->index;
 444                 cursor->node = save;
 445                 cursor->index = 0;
 446         }
 447         return(error);
 448 }
 449
 450
 451 /*
 452  * Cursor down through the current node, which must be an internal node.
 453  *
 454  * This routine adjusts the cursor and sets index to 0.
 455  */
 456 int
 457 hammer_cursor_down(hammer_cursor_t cursor)
 458 {
 459         hammer_node_t node;
 460         hammer_btree_elm_t elm;
 461         int error;
 462
 463         /*
 464          * The current node becomes the current parent
 465          */
 466         hammer_cursor_downgrade(cursor);
 467         node = cursor->node;
 468         KKASSERT(cursor->index >= 0 && cursor->index < node->ondisk->count);
 469         if (cursor->parent) {
 470                 hammer_unlock(&cursor->parent->lock);
 471                 hammer_rel_node(cursor->parent);
 472         }
 473         cursor->parent = node;
 474         cursor->parent_index = cursor->index;
 475         cursor->node = NULL;
 476         cursor->index = 0;
 477
 478         /*
 479          * Extract element to push into at (node,index), set bounds.
 480          */
 481         elm = &node->ondisk->elms[cursor->parent_index];
 482
 483         /*
 484          * Ok, push down into elm.  If elm specifies an internal or leaf
 485          * node the current node must be an internal node.  If elm specifies
 486          * a spike then the current node must be a leaf node.
 487          */
 488         switch(elm->base.btype) {
 489         case HAMMER_BTREE_TYPE_INTERNAL:
 490         case HAMMER_BTREE_TYPE_LEAF:
 491                 KKASSERT(node->ondisk->type == HAMMER_BTREE_TYPE_INTERNAL);
 492                 KKASSERT(elm->internal.subtree_offset != 0);
 493                 cursor->left_bound = &elm[0].internal.base;
 494                 cursor->right_bound = &elm[1].internal.base;
 495                 node = hammer_get_node(cursor->trans->hmp,
 496                                        elm->internal.subtree_offset, 0, &error);
 497                 if (error == 0) {
 498                         KASSERT(elm->base.btype == node->ondisk->type, ("BTYPE MISMATCH %c %c NODE %p\n", elm->base.btype, node->ondisk->type, node));
 499                         if (node->ondisk->parent != cursor->parent->node_offset)
 500                                 panic("node %p %016llx vs %016llx\n", node, node->ondisk->parent, cursor->parent->node_offset);
 501                         KKASSERT(node->ondisk->parent == cursor->parent->node_offset);
 502                 }
 503                 break;
 504         default:
 505                 panic("hammer_cursor_down: illegal btype %02x (%c)\n",
 506                       elm->base.btype,
 507                       (elm->base.btype ? elm->base.btype : '?'));
 508                 break;
 509         }
 510         if (error == 0) {
 511                 hammer_lock_sh(&node->lock);
 512                 KKASSERT ((node->flags & HAMMER_NODE_DELETED) == 0);
 513                 cursor->node = node;
 514                 cursor->index = 0;
 515         }
 516         return(error);
 517 }
 518