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 1.23 2008/05/03 05:28:55 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);
  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, &error);
  99                 hammer_rel_volume(volume, 0);
 100                 if (error)
 101                         break;
 102                 hammer_lock_sh(&node->lock);
 103                 if (node->flags & HAMMER_NODE_DELETED) {
 104                         hammer_unlock(&node->lock);
 105                         hammer_rel_node(node);
 106                         node = NULL;
 107                 }
 108         }
 109
 110         /*
 111          * Step 3 - finish initializing the cursor by acquiring the parent
 112          */
 113         cursor->node = node;
 114         if (error == 0)
 115                 error = hammer_load_cursor_parent(cursor);
 116         KKASSERT(error == 0);
 117         /* if (error) hammer_done_cursor(cursor); */
 118         return(error);
 119 }
 120
 121 #if 0
 122 int
 123 hammer_reinit_cursor(hammer_cursor_t cursor)
 124 {
 125         hammer_transaction_t trans;
 126         hammer_inode_t ip;
 127         struct hammer_node **cache;
 128
 129         trans = cursor->trans;
 130         ip = cursor->ip;
 131         hammer_done_cursor(cursor);
 132         cache = ip ? &ip->cache[0] : NULL;
 133         error = hammer_init_cursor(trans, cursor, cache, ip);
 134         return (error);
 135 }
 136
 137 #endif
 138
 139 /*
 140  * Normalize a cursor.  Sometimes cursors can be left in a state
 141  * where node is NULL.  If the cursor is in this state, cursor up.
 142  */
 143 void
 144 hammer_normalize_cursor(hammer_cursor_t cursor)
 145 {
 146         if (cursor->node == NULL) {
 147                 KKASSERT(cursor->parent != NULL);
 148                 hammer_cursor_up(cursor);
 149         }
 150 }
 151
 152
 153 /*
 154  * We are finished with a cursor.  We NULL out various fields as sanity
 155  * check, in case the structure is inappropriately used afterwords.
 156  */
 157 void
 158 hammer_done_cursor(hammer_cursor_t cursor)
 159 {
 160         hammer_inode_t ip;
 161
 162         if (cursor->parent) {
 163                 hammer_unlock(&cursor->parent->lock);
 164                 hammer_rel_node(cursor->parent);
 165                 cursor->parent = NULL;
 166         }
 167         if (cursor->node) {
 168                 hammer_unlock(&cursor->node->lock);
 169                 hammer_rel_node(cursor->node);
 170                 cursor->node = NULL;
 171         }
 172         if (cursor->data_buffer) {
 173                 hammer_rel_buffer(cursor->data_buffer, 0);
 174                 cursor->data_buffer = NULL;
 175         }
 176         if (cursor->record_buffer) {
 177                 hammer_rel_buffer(cursor->record_buffer, 0);
 178                 cursor->record_buffer = NULL;
 179         }
 180         if ((ip = cursor->ip) != NULL) {
 181                 hammer_mem_done(cursor);
 182                 KKASSERT(ip->cursor_ip_refs > 0);
 183                 --ip->cursor_ip_refs;
 184                 hammer_unlock(&ip->lock);
 185                 cursor->ip = NULL;
 186         }
 187
 188
 189         /*
 190          * If we deadlocked this node will be referenced.  Do a quick
 191          * lock/unlock to wait for the deadlock condition to clear.
 192          */
 193         if (cursor->deadlk_node) {
 194                 hammer_lock_ex(&cursor->deadlk_node->lock);
 195                 hammer_unlock(&cursor->deadlk_node->lock);
 196                 hammer_rel_node(cursor->deadlk_node);
 197                 cursor->deadlk_node = NULL;
 198         }
 199         if (cursor->deadlk_rec) {
 200                 hammer_wait_mem_record(cursor->deadlk_rec);
 201                 hammer_rel_mem_record(cursor->deadlk_rec);
 202                 cursor->deadlk_rec = NULL;
 203         }
 204
 205         cursor->data = NULL;
 206         cursor->record = NULL;
 207         cursor->left_bound = NULL;
 208         cursor->right_bound = NULL;
 209         cursor->trans = NULL;
 210 }
 211
 212 /*
 213  * Upgrade cursor->node and cursor->parent to exclusive locks.  This
 214  * function can return EDEADLK.
 215  *
 216  * The lock must already be either held shared or already held exclusively
 217  * by us.
 218  *
 219  * If we fail to upgrade the lock and cursor->deadlk_node is NULL,
 220  * we add another reference to the node that failed and set
 221  * cursor->deadlk_node so hammer_done_cursor() can block on it.
 222  */
 223 int
 224 hammer_cursor_upgrade(hammer_cursor_t cursor)
 225 {
 226         int error;
 227
 228         error = hammer_lock_upgrade(&cursor->node->lock);
 229         if (error && cursor->deadlk_node == NULL) {
 230                 cursor->deadlk_node = cursor->node;
 231                 hammer_ref_node(cursor->deadlk_node);
 232         } else if (error == 0 && cursor->parent) {
 233                 error = hammer_lock_upgrade(&cursor->parent->lock);
 234                 if (error && cursor->deadlk_node == NULL) {
 235                         cursor->deadlk_node = cursor->parent;
 236                         hammer_ref_node(cursor->deadlk_node);
 237                 }
 238         }
 239         return(error);
 240 }
 241
 242 /*
 243  * Downgrade cursor->node and cursor->parent to shared locks.  This
 244  * function can return EDEADLK.
 245  */
 246 void
 247 hammer_cursor_downgrade(hammer_cursor_t cursor)
 248 {
 249         if (hammer_lock_excl_owned(&cursor->node->lock, curthread))
 250                 hammer_lock_downgrade(&cursor->node->lock);
 251         if (cursor->parent &&
 252             hammer_lock_excl_owned(&cursor->parent->lock, curthread)) {
 253                 hammer_lock_downgrade(&cursor->parent->lock);
 254         }
 255 }
 256
 257 /*
 258  * Seek the cursor to the specified node and index.
 259  */
 260 int
 261 hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, int index)
 262 {
 263         int error;
 264
 265         hammer_cursor_downgrade(cursor);
 266         error = 0;
 267
 268         if (cursor->node != node) {
 269                 hammer_unlock(&cursor->node->lock);
 270                 hammer_rel_node(cursor->node);
 271                 cursor->node = node;
 272                 hammer_ref_node(node);
 273                 hammer_lock_sh(&node->lock);
 274
 275                 if (cursor->parent) {
 276                         hammer_unlock(&cursor->parent->lock);
 277                         hammer_rel_node(cursor->parent);
 278                         cursor->parent = NULL;
 279                         cursor->parent_index = 0;
 280                 }
 281                 error = hammer_load_cursor_parent(cursor);
 282         }
 283         cursor->index = index;
 284         return (error);
 285 }
 286
 287 /*
 288  * Load the parent of cursor->node into cursor->parent.
 289  */
 290 static
 291 int
 292 hammer_load_cursor_parent(hammer_cursor_t cursor)
 293 {
 294         hammer_mount_t hmp;
 295         hammer_node_t parent;
 296         hammer_node_t node;
 297         hammer_btree_elm_t elm;
 298         int error;
 299         int i;
 300
 301         hmp = cursor->trans->hmp;
 302
 303         if (cursor->node->ondisk->parent) {
 304                 node = cursor->node;
 305                 parent = hammer_get_node(hmp, node->ondisk->parent, &error);
 306                 if (error)
 307                         return(error);
 308                 hammer_lock_sh(&parent->lock);
 309                 elm = NULL;
 310                 for (i = 0; i < parent->ondisk->count; ++i) {
 311                         elm = &parent->ondisk->elms[i];
 312                         if (parent->ondisk->elms[i].internal.subtree_offset ==
 313                             node->node_offset) {
 314                                 break;
 315                         }
 316                 }
 317                 if (i == parent->ondisk->count) {
 318                         hammer_unlock(&parent->lock);
 319                         panic("Bad B-Tree link: parent %p node %p\n", parent, node);
 320                 }
 321                 KKASSERT(i != parent->ondisk->count);
 322                 cursor->parent = parent;
 323                 cursor->parent_index = i;
 324                 cursor->left_bound = &elm[0].internal.base;
 325                 cursor->right_bound = &elm[1].internal.base;
 326                 return(error);
 327         } else {
 328                 cursor->parent = NULL;
 329                 cursor->parent_index = 0;
 330                 cursor->left_bound = &hmp->root_btree_beg;
 331                 cursor->right_bound = &hmp->root_btree_end;
 332                 error = 0;
 333         }
 334         return(error);
 335 }
 336
 337 /*
 338  * Cursor up to our parent node.  Return ENOENT if we are at the root of
 339  * the filesystem.
 340  *
 341  * If doing a nonblocking cursor-up and we are unable to acquire the lock,
 342  * the cursor remains unchanged.
 343  */
 344 int
 345 hammer_cursor_up(hammer_cursor_t cursor)
 346 {
 347         int error;
 348
 349         hammer_cursor_downgrade(cursor);
 350
 351         /*
 352          * If the parent is NULL we are at the root of the B-Tree and
 353          * return ENOENT.
 354          */
 355         if (cursor->parent == NULL)
 356                 return (ENOENT);
 357
 358         /*
 359          * Set the node to its parent.
 360          */
 361         hammer_unlock(&cursor->node->lock);
 362         hammer_rel_node(cursor->node);
 363         cursor->node = cursor->parent;
 364         cursor->index = cursor->parent_index;
 365         cursor->parent = NULL;
 366         cursor->parent_index = 0;
 367
 368         error = hammer_load_cursor_parent(cursor);
 369         return(error);
 370 }
 371
 372 /*
 373  * Cursor down through the current node, which must be an internal node.
 374  *
 375  * This routine adjusts the cursor and sets index to 0.
 376  */
 377 int
 378 hammer_cursor_down(hammer_cursor_t cursor)
 379 {
 380         hammer_node_t node;
 381         hammer_btree_elm_t elm;
 382         int error;
 383
 384         /*
 385          * The current node becomes the current parent
 386          */
 387         hammer_cursor_downgrade(cursor);
 388         node = cursor->node;
 389         KKASSERT(cursor->index >= 0 && cursor->index < node->ondisk->count);
 390         if (cursor->parent) {
 391                 hammer_unlock(&cursor->parent->lock);
 392                 hammer_rel_node(cursor->parent);
 393         }
 394         cursor->parent = node;
 395         cursor->parent_index = cursor->index;
 396         cursor->node = NULL;
 397         cursor->index = 0;
 398
 399         /*
 400          * Extract element to push into at (node,index), set bounds.
 401          */
 402         elm = &node->ondisk->elms[cursor->parent_index];
 403
 404         /*
 405          * Ok, push down into elm.  If elm specifies an internal or leaf
 406          * node the current node must be an internal node.  If elm specifies
 407          * a spike then the current node must be a leaf node.
 408          */
 409         switch(elm->base.btype) {
 410         case HAMMER_BTREE_TYPE_INTERNAL:
 411         case HAMMER_BTREE_TYPE_LEAF:
 412                 KKASSERT(node->ondisk->type == HAMMER_BTREE_TYPE_INTERNAL);
 413                 KKASSERT(elm->internal.subtree_offset != 0);
 414                 cursor->left_bound = &elm[0].internal.base;
 415                 cursor->right_bound = &elm[1].internal.base;
 416                 node = hammer_get_node(cursor->trans->hmp,
 417                                        elm->internal.subtree_offset, &error);
 418                 if (error == 0) {
 419                         KASSERT(elm->base.btype == node->ondisk->type, ("BTYPE MISMATCH %c %c NODE %p\n", elm->base.btype, node->ondisk->type, node));
 420                         if (node->ondisk->parent != cursor->parent->node_offset)
 421                                 panic("node %p %016llx vs %016llx\n", node, node->ondisk->parent, cursor->parent->node_offset);
 422                         KKASSERT(node->ondisk->parent == cursor->parent->node_offset);
 423                 }
 424                 break;
 425         default:
 426                 panic("hammer_cursor_down: illegal btype %02x (%c)\n",
 427                       elm->base.btype,
 428                       (elm->base.btype ? elm->base.btype : '?'));
 429                 break;
 430         }
 431         if (error == 0) {
 432                 hammer_lock_sh(&node->lock);
 433                 cursor->node = node;
 434                 cursor->index = 0;
 435         }
 436         return(error);
 437 }
 438