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 1.33 2008/06/26 04:06:22 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                    hammer_node_cache_t 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->node) {
  76                 node = hammer_ref_node_safe(trans->hmp, cache, &error);
  77                 if (error == 0) {
  78                         hammer_lock_sh_lowpri(&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_lowpri(&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 /*
 134  * Normalize a cursor.  Sometimes cursors can be left in a state
 135  * where node is NULL.  If the cursor is in this state, cursor up.
 136  */
 137 void
 138 hammer_normalize_cursor(hammer_cursor_t cursor)
 139 {
 140         if (cursor->node == NULL) {
 141                 KKASSERT(cursor->parent != NULL);
 142                 hammer_cursor_up(cursor);
 143         }
 144 }
 145
 146
 147 /*
 148  * We are finished with a cursor.  We NULL out various fields as sanity
 149  * check, in case the structure is inappropriately used afterwords.
 150  */
 151 void
 152 hammer_done_cursor(hammer_cursor_t cursor)
 153 {
 154         hammer_inode_t ip;
 155
 156         if (cursor->parent) {
 157                 hammer_unlock(&cursor->parent->lock);
 158                 hammer_rel_node(cursor->parent);
 159                 cursor->parent = NULL;
 160         }
 161         if (cursor->node) {
 162                 hammer_unlock(&cursor->node->lock);
 163                 hammer_rel_node(cursor->node);
 164                 cursor->node = NULL;
 165         }
 166         if (cursor->data_buffer) {
 167                 hammer_rel_buffer(cursor->data_buffer, 0);
 168                 cursor->data_buffer = NULL;
 169         }
 170         if (cursor->record_buffer) {
 171                 hammer_rel_buffer(cursor->record_buffer, 0);
 172                 cursor->record_buffer = NULL;
 173         }
 174         if ((ip = cursor->ip) != NULL) {
 175                 hammer_mem_done(cursor);
 176                 KKASSERT(ip->cursor_ip_refs > 0);
 177                 --ip->cursor_ip_refs;
 178                 hammer_unlock(&ip->lock);
 179                 cursor->ip = NULL;
 180         }
 181
 182
 183         /*
 184          * If we deadlocked this node will be referenced.  Do a quick
 185          * lock/unlock to wait for the deadlock condition to clear.
 186          */
 187         if (cursor->deadlk_node) {
 188                 hammer_lock_ex_ident(&cursor->deadlk_node->lock, "hmrdlk");
 189                 hammer_unlock(&cursor->deadlk_node->lock);
 190                 hammer_rel_node(cursor->deadlk_node);
 191                 cursor->deadlk_node = NULL;
 192         }
 193         if (cursor->deadlk_rec) {
 194                 hammer_wait_mem_record_ident(cursor->deadlk_rec, "hmmdlr");
 195                 hammer_rel_mem_record(cursor->deadlk_rec);
 196                 cursor->deadlk_rec = NULL;
 197         }
 198
 199         cursor->data = NULL;
 200         cursor->leaf = NULL;
 201         cursor->left_bound = NULL;
 202         cursor->right_bound = NULL;
 203         cursor->trans = NULL;
 204 }
 205
 206 /*
 207  * Upgrade cursor->node and cursor->parent to exclusive locks.  This
 208  * function can return EDEADLK.
 209  *
 210  * The lock must already be either held shared or already held exclusively
 211  * by us.
 212  *
 213  * If we fail to upgrade the lock and cursor->deadlk_node is NULL,
 214  * we add another reference to the node that failed and set
 215  * cursor->deadlk_node so hammer_done_cursor() can block on it.
 216  */
 217 int
 218 hammer_cursor_upgrade(hammer_cursor_t cursor)
 219 {
 220         int error;
 221
 222         error = hammer_lock_upgrade(&cursor->node->lock);
 223         if (error && cursor->deadlk_node == NULL) {
 224                 cursor->deadlk_node = cursor->node;
 225                 hammer_ref_node(cursor->deadlk_node);
 226         } else if (error == 0 && cursor->parent) {
 227                 error = hammer_lock_upgrade(&cursor->parent->lock);
 228                 if (error && cursor->deadlk_node == NULL) {
 229                         cursor->deadlk_node = cursor->parent;
 230                         hammer_ref_node(cursor->deadlk_node);
 231                 }
 232         }
 233         return(error);
 234 }
 235
 236 int
 237 hammer_cursor_upgrade_node(hammer_cursor_t cursor)
 238 {
 239         int error;
 240
 241         error = hammer_lock_upgrade(&cursor->node->lock);
 242         if (error && cursor->deadlk_node == NULL) {
 243                 cursor->deadlk_node = cursor->node;
 244                 hammer_ref_node(cursor->deadlk_node);
 245         }
 246         return(error);
 247 }
 248
 249 /*
 250  * Downgrade cursor->node and cursor->parent to shared locks.  This
 251  * function can return EDEADLK.
 252  */
 253 void
 254 hammer_cursor_downgrade(hammer_cursor_t cursor)
 255 {
 256         if (hammer_lock_excl_owned(&cursor->node->lock, curthread))
 257                 hammer_lock_downgrade(&cursor->node->lock);
 258         if (cursor->parent &&
 259             hammer_lock_excl_owned(&cursor->parent->lock, curthread)) {
 260                 hammer_lock_downgrade(&cursor->parent->lock);
 261         }
 262 }
 263
 264 /*
 265  * Seek the cursor to the specified node and index.
 266  *
 267  * The caller must ref the node prior to calling this routine and release
 268  * it after it returns.  If the seek succeeds the cursor will gain its own
 269  * ref on the node.
 270  */
 271 int
 272 hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, int index)
 273 {
 274         int error;
 275
 276         hammer_cursor_downgrade(cursor);
 277         error = 0;
 278
 279         if (cursor->node != node) {
 280                 hammer_unlock(&cursor->node->lock);
 281                 hammer_rel_node(cursor->node);
 282                 cursor->node = node;
 283                 hammer_ref_node(node);
 284                 hammer_lock_sh(&node->lock);
 285                 KKASSERT ((node->flags & HAMMER_NODE_DELETED) == 0);
 286
 287                 if (cursor->parent) {
 288                         hammer_unlock(&cursor->parent->lock);
 289                         hammer_rel_node(cursor->parent);
 290                         cursor->parent = NULL;
 291                         cursor->parent_index = 0;
 292                 }
 293                 error = hammer_load_cursor_parent(cursor, 0);
 294         }
 295         cursor->index = index;
 296         return (error);
 297 }
 298
 299 /*
 300  * Load the parent of cursor->node into cursor->parent.
 301  */
 302 static
 303 int
 304 hammer_load_cursor_parent(hammer_cursor_t cursor, int try_exclusive)
 305 {
 306         hammer_mount_t hmp;
 307         hammer_node_t parent;
 308         hammer_node_t node;
 309         hammer_btree_elm_t elm;
 310         int error;
 311         int parent_index;
 312
 313         hmp = cursor->trans->hmp;
 314
 315         if (cursor->node->ondisk->parent) {
 316                 node = cursor->node;
 317                 parent = hammer_btree_get_parent(node, &parent_index,
 318                                                  &error, try_exclusive);
 319                 if (error == 0) {
 320                         elm = &parent->ondisk->elms[parent_index];
 321                         cursor->parent = parent;
 322                         cursor->parent_index = parent_index;
 323                         cursor->left_bound = &elm[0].internal.base;
 324                         cursor->right_bound = &elm[1].internal.base;
 325                 }
 326         } else {
 327                 cursor->parent = NULL;
 328                 cursor->parent_index = 0;
 329                 cursor->left_bound = &hmp->root_btree_beg;
 330                 cursor->right_bound = &hmp->root_btree_end;
 331                 error = 0;
 332         }
 333         return(error);
 334 }
 335
 336 /*
 337  * Cursor up to our parent node.  Return ENOENT if we are at the root of
 338  * the filesystem.
 339  */
 340 int
 341 hammer_cursor_up(hammer_cursor_t cursor)
 342 {
 343         int error;
 344
 345         hammer_cursor_downgrade(cursor);
 346
 347         /*
 348          * If the parent is NULL we are at the root of the B-Tree and
 349          * return ENOENT.
 350          */
 351         if (cursor->parent == NULL)
 352                 return (ENOENT);
 353
 354         /*
 355          * Set the node to its parent.
 356          */
 357         hammer_unlock(&cursor->node->lock);
 358         hammer_rel_node(cursor->node);
 359         cursor->node = cursor->parent;
 360         cursor->index = cursor->parent_index;
 361         cursor->parent = NULL;
 362         cursor->parent_index = 0;
 363
 364         error = hammer_load_cursor_parent(cursor, 0);
 365         return(error);
 366 }
 367
 368 /*
 369  * Special cursor up given a locked cursor.  The orignal node is not
 370  * unlocked and released and the cursor is not downgraded.  If we are
 371  * unable to acquire and lock the parent, EDEADLK is returned.
 372  */
 373 int
 374 hammer_cursor_up_locked(hammer_cursor_t cursor)
 375 {
 376         hammer_node_t save;
 377         int error;
 378
 379         /*
 380          * If the parent is NULL we are at the root of the B-Tree and
 381          * return ENOENT.
 382          */
 383         if (cursor->parent == NULL)
 384                 return (ENOENT);
 385
 386         save = cursor->node;
 387
 388         /*
 389          * Set the node to its parent.
 390          */
 391         cursor->node = cursor->parent;
 392         cursor->index = cursor->parent_index;
 393         cursor->parent = NULL;
 394         cursor->parent_index = 0;
 395
 396         /*
 397          * load the new parent, attempt to exclusively lock it.  Note that
 398          * we are still holding the old parent (now cursor->node) exclusively
 399          * locked.  This can return EDEADLK.
 400          */
 401         error = hammer_load_cursor_parent(cursor, 1);
 402         if (error) {
 403                 cursor->parent = cursor->node;
 404                 cursor->parent_index = cursor->index;
 405                 cursor->node = save;
 406                 cursor->index = 0;
 407         }
 408         return(error);
 409 }
 410
 411
 412 /*
 413  * Cursor down through the current node, which must be an internal node.
 414  *
 415  * This routine adjusts the cursor and sets index to 0.
 416  */
 417 int
 418 hammer_cursor_down(hammer_cursor_t cursor)
 419 {
 420         hammer_node_t node;
 421         hammer_btree_elm_t elm;
 422         int error;
 423
 424         /*
 425          * The current node becomes the current parent
 426          */
 427         hammer_cursor_downgrade(cursor);
 428         node = cursor->node;
 429         KKASSERT(cursor->index >= 0 && cursor->index < node->ondisk->count);
 430         if (cursor->parent) {
 431                 hammer_unlock(&cursor->parent->lock);
 432                 hammer_rel_node(cursor->parent);
 433         }
 434         cursor->parent = node;
 435         cursor->parent_index = cursor->index;
 436         cursor->node = NULL;
 437         cursor->index = 0;
 438
 439         /*
 440          * Extract element to push into at (node,index), set bounds.
 441          */
 442         elm = &node->ondisk->elms[cursor->parent_index];
 443
 444         /*
 445          * Ok, push down into elm.  If elm specifies an internal or leaf
 446          * node the current node must be an internal node.  If elm specifies
 447          * a spike then the current node must be a leaf node.
 448          */
 449         switch(elm->base.btype) {
 450         case HAMMER_BTREE_TYPE_INTERNAL:
 451         case HAMMER_BTREE_TYPE_LEAF:
 452                 KKASSERT(node->ondisk->type == HAMMER_BTREE_TYPE_INTERNAL);
 453                 KKASSERT(elm->internal.subtree_offset != 0);
 454                 cursor->left_bound = &elm[0].internal.base;
 455                 cursor->right_bound = &elm[1].internal.base;
 456                 node = hammer_get_node(cursor->trans->hmp,
 457                                        elm->internal.subtree_offset, 0, &error);
 458                 if (error == 0) {
 459                         KASSERT(elm->base.btype == node->ondisk->type, ("BTYPE MISMATCH %c %c NODE %p\n", elm->base.btype, node->ondisk->type, node));
 460                         if (node->ondisk->parent != cursor->parent->node_offset)
 461                                 panic("node %p %016llx vs %016llx\n", node, node->ondisk->parent, cursor->parent->node_offset);
 462                         KKASSERT(node->ondisk->parent == cursor->parent->node_offset);
 463                 }
 464                 break;
 465         default:
 466                 panic("hammer_cursor_down: illegal btype %02x (%c)\n",
 467                       elm->base.btype,
 468                       (elm->base.btype ? elm->base.btype : '?'));
 469                 break;
 470         }
 471         if (error == 0) {
 472                 hammer_lock_sh(&node->lock);
 473                 KKASSERT ((node->flags & HAMMER_NODE_DELETED) == 0);
 474                 cursor->node = node;
 475                 cursor->index = 0;
 476         }
 477         return(error);
 478 }
 479