git.oblomov.eu Git - linux-2.6/blob - fs/xfs/xfs_da_btree.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_btree.h"
  41 #include "xfs_bmap.h"
  42 #include "xfs_attr.h"
  43 #include "xfs_attr_leaf.h"
  44 #include "xfs_dir2_data.h"
  45 #include "xfs_dir2_leaf.h"
  46 #include "xfs_dir2_block.h"
  47 #include "xfs_dir2_node.h"
  48 #include "xfs_error.h"
  49
  50 /*
  51  * xfs_da_btree.c
  52  *
  53  * Routines to implement directories as Btrees of hashed names.
  54  */
  55
  56 /*========================================================================
  57  * Function prototypes for the kernel.
  58  *========================================================================*/
  59
  60 /*
  61  * Routines used for growing the Btree.
  62  */
  63 STATIC int xfs_da_root_split(xfs_da_state_t *state,
  64                                             xfs_da_state_blk_t *existing_root,
  65                                             xfs_da_state_blk_t *new_child);
  66 STATIC int xfs_da_node_split(xfs_da_state_t *state,
  67                                             xfs_da_state_blk_t *existing_blk,
  68                                             xfs_da_state_blk_t *split_blk,
  69                                             xfs_da_state_blk_t *blk_to_add,
  70                                             int treelevel,
  71                                             int *result);
  72 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
  73                                          xfs_da_state_blk_t *node_blk_1,
  74                                          xfs_da_state_blk_t *node_blk_2);
  75 STATIC void xfs_da_node_add(xfs_da_state_t *state,
  76                                    xfs_da_state_blk_t *old_node_blk,
  77                                    xfs_da_state_blk_t *new_node_blk);
  78
  79 /*
  80  * Routines used for shrinking the Btree.
  81  */
  82 STATIC int xfs_da_root_join(xfs_da_state_t *state,
  83                                            xfs_da_state_blk_t *root_blk);
  84 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
  85 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
  86                                               xfs_da_state_blk_t *drop_blk);
  87 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
  88                                          xfs_da_state_blk_t *src_node_blk,
  89                                          xfs_da_state_blk_t *dst_node_blk);
  90
  91 /*
  92  * Utility routines.
  93  */
  94 STATIC uint     xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
  95 STATIC int      xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
  96 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
  97 STATIC int      xfs_da_blk_unlink(xfs_da_state_t *state,
  98                                   xfs_da_state_blk_t *drop_blk,
  99                                   xfs_da_state_blk_t *save_blk);
 100 STATIC void     xfs_da_state_kill_altpath(xfs_da_state_t *state);
 101
 102 /*========================================================================
 103  * Routines used for growing the Btree.
 104  *========================================================================*/
 105
 106 /*
 107  * Create the initial contents of an intermediate node.
 108  */
 109 int
 110 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
 111                                  xfs_dabuf_t **bpp, int whichfork)
 112 {
 113         xfs_da_intnode_t *node;
 114         xfs_dabuf_t *bp;
 115         int error;
 116         xfs_trans_t *tp;
 117
 118         tp = args->trans;
 119         error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
 120         if (error)
 121                 return(error);
 122         ASSERT(bp != NULL);
 123         node = bp->data;
 124         node->hdr.info.forw = 0;
 125         node->hdr.info.back = 0;
 126         node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
 127         node->hdr.info.pad = 0;
 128         node->hdr.count = 0;
 129         node->hdr.level = cpu_to_be16(level);
 130
 131         xfs_da_log_buf(tp, bp,
 132                 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 133
 134         *bpp = bp;
 135         return(0);
 136 }
 137
 138 /*
 139  * Split a leaf node, rebalance, then possibly split
 140  * intermediate nodes, rebalance, etc.
 141  */
 142 int                                                     /* error */
 143 xfs_da_split(xfs_da_state_t *state)
 144 {
 145         xfs_da_state_blk_t *oldblk, *newblk, *addblk;
 146         xfs_da_intnode_t *node;
 147         xfs_dabuf_t *bp;
 148         int max, action, error, i;
 149
 150         /*
 151          * Walk back up the tree splitting/inserting/adjusting as necessary.
 152          * If we need to insert and there isn't room, split the node, then
 153          * decide which fragment to insert the new block from below into.
 154          * Note that we may split the root this way, but we need more fixup.
 155          */
 156         max = state->path.active - 1;
 157         ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
 158         ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
 159                state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
 160
 161         addblk = &state->path.blk[max];         /* initial dummy value */
 162         for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
 163                 oldblk = &state->path.blk[i];
 164                 newblk = &state->altpath.blk[i];
 165
 166                 /*
 167                  * If a leaf node then
 168                  *     Allocate a new leaf node, then rebalance across them.
 169                  * else if an intermediate node then
 170                  *     We split on the last layer, must we split the node?
 171                  */
 172                 switch (oldblk->magic) {
 173                 case XFS_ATTR_LEAF_MAGIC:
 174                         error = xfs_attr_leaf_split(state, oldblk, newblk);
 175                         if ((error != 0) && (error != ENOSPC)) {
 176                                 return(error);  /* GROT: attr is inconsistent */
 177                         }
 178                         if (!error) {
 179                                 addblk = newblk;
 180                                 break;
 181                         }
 182                         /*
 183                          * Entry wouldn't fit, split the leaf again.
 184                          */
 185                         state->extravalid = 1;
 186                         if (state->inleaf) {
 187                                 state->extraafter = 0;  /* before newblk */
 188                                 error = xfs_attr_leaf_split(state, oldblk,
 189                                                             &state->extrablk);
 190                         } else {
 191                                 state->extraafter = 1;  /* after newblk */
 192                                 error = xfs_attr_leaf_split(state, newblk,
 193                                                             &state->extrablk);
 194                         }
 195                         if (error)
 196                                 return(error);  /* GROT: attr inconsistent */
 197                         addblk = newblk;
 198                         break;
 199                 case XFS_DIR2_LEAFN_MAGIC:
 200                         error = xfs_dir2_leafn_split(state, oldblk, newblk);
 201                         if (error)
 202                                 return error;
 203                         addblk = newblk;
 204                         break;
 205                 case XFS_DA_NODE_MAGIC:
 206                         error = xfs_da_node_split(state, oldblk, newblk, addblk,
 207                                                          max - i, &action);
 208                         xfs_da_buf_done(addblk->bp);
 209                         addblk->bp = NULL;
 210                         if (error)
 211                                 return(error);  /* GROT: dir is inconsistent */
 212                         /*
 213                          * Record the newly split block for the next time thru?
 214                          */
 215                         if (action)
 216                                 addblk = newblk;
 217                         else
 218                                 addblk = NULL;
 219                         break;
 220                 }
 221
 222                 /*
 223                  * Update the btree to show the new hashval for this child.
 224                  */
 225                 xfs_da_fixhashpath(state, &state->path);
 226                 /*
 227                  * If we won't need this block again, it's getting dropped
 228                  * from the active path by the loop control, so we need
 229                  * to mark it done now.
 230                  */
 231                 if (i > 0 || !addblk)
 232                         xfs_da_buf_done(oldblk->bp);
 233         }
 234         if (!addblk)
 235                 return(0);
 236
 237         /*
 238          * Split the root node.
 239          */
 240         ASSERT(state->path.active == 0);
 241         oldblk = &state->path.blk[0];
 242         error = xfs_da_root_split(state, oldblk, addblk);
 243         if (error) {
 244                 xfs_da_buf_done(oldblk->bp);
 245                 xfs_da_buf_done(addblk->bp);
 246                 addblk->bp = NULL;
 247                 return(error);  /* GROT: dir is inconsistent */
 248         }
 249
 250         /*
 251          * Update pointers to the node which used to be block 0 and
 252          * just got bumped because of the addition of a new root node.
 253          * There might be three blocks involved if a double split occurred,
 254          * and the original block 0 could be at any position in the list.
 255          */
 256
 257         node = oldblk->bp->data;
 258         if (node->hdr.info.forw) {
 259                 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
 260                         bp = addblk->bp;
 261                 } else {
 262                         ASSERT(state->extravalid);
 263                         bp = state->extrablk.bp;
 264                 }
 265                 node = bp->data;
 266                 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
 267                 xfs_da_log_buf(state->args->trans, bp,
 268                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 269                     sizeof(node->hdr.info)));
 270         }
 271         node = oldblk->bp->data;
 272         if (node->hdr.info.back) {
 273                 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
 274                         bp = addblk->bp;
 275                 } else {
 276                         ASSERT(state->extravalid);
 277                         bp = state->extrablk.bp;
 278                 }
 279                 node = bp->data;
 280                 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
 281                 xfs_da_log_buf(state->args->trans, bp,
 282                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 283                     sizeof(node->hdr.info)));
 284         }
 285         xfs_da_buf_done(oldblk->bp);
 286         xfs_da_buf_done(addblk->bp);
 287         addblk->bp = NULL;
 288         return(0);
 289 }
 290
 291 /*
 292  * Split the root.  We have to create a new root and point to the two
 293  * parts (the split old root) that we just created.  Copy block zero to
 294  * the EOF, extending the inode in process.
 295  */
 296 STATIC int                                              /* error */
 297 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
 298                                  xfs_da_state_blk_t *blk2)
 299 {
 300         xfs_da_intnode_t *node, *oldroot;
 301         xfs_da_args_t *args;
 302         xfs_dablk_t blkno;
 303         xfs_dabuf_t *bp;
 304         int error, size;
 305         xfs_inode_t *dp;
 306         xfs_trans_t *tp;
 307         xfs_mount_t *mp;
 308         xfs_dir2_leaf_t *leaf;
 309
 310         /*
 311          * Copy the existing (incorrect) block from the root node position
 312          * to a free space somewhere.
 313          */
 314         args = state->args;
 315         ASSERT(args != NULL);
 316         error = xfs_da_grow_inode(args, &blkno);
 317         if (error)
 318                 return(error);
 319         dp = args->dp;
 320         tp = args->trans;
 321         mp = state->mp;
 322         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 323         if (error)
 324                 return(error);
 325         ASSERT(bp != NULL);
 326         node = bp->data;
 327         oldroot = blk1->bp->data;
 328         if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) {
 329                 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
 330                              (char *)oldroot);
 331         } else {
 332                 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 333                 leaf = (xfs_dir2_leaf_t *)oldroot;
 334                 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
 335                              (char *)leaf);
 336         }
 337         memcpy(node, oldroot, size);
 338         xfs_da_log_buf(tp, bp, 0, size - 1);
 339         xfs_da_buf_done(blk1->bp);
 340         blk1->bp = bp;
 341         blk1->blkno = blkno;
 342
 343         /*
 344          * Set up the new root node.
 345          */
 346         error = xfs_da_node_create(args,
 347                 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
 348                 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
 349         if (error)
 350                 return(error);
 351         node = bp->data;
 352         node->btree[0].hashval = cpu_to_be32(blk1->hashval);
 353         node->btree[0].before = cpu_to_be32(blk1->blkno);
 354         node->btree[1].hashval = cpu_to_be32(blk2->hashval);
 355         node->btree[1].before = cpu_to_be32(blk2->blkno);
 356         node->hdr.count = cpu_to_be16(2);
 357
 358 #ifdef DEBUG
 359         if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC) {
 360                 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
 361                        blk1->blkno < mp->m_dirfreeblk);
 362                 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
 363                        blk2->blkno < mp->m_dirfreeblk);
 364         }
 365 #endif
 366
 367         /* Header is already logged by xfs_da_node_create */
 368         xfs_da_log_buf(tp, bp,
 369                 XFS_DA_LOGRANGE(node, node->btree,
 370                         sizeof(xfs_da_node_entry_t) * 2));
 371         xfs_da_buf_done(bp);
 372
 373         return(0);
 374 }
 375
 376 /*
 377  * Split the node, rebalance, then add the new entry.
 378  */
 379 STATIC int                                              /* error */
 380 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 381                                  xfs_da_state_blk_t *newblk,
 382                                  xfs_da_state_blk_t *addblk,
 383                                  int treelevel, int *result)
 384 {
 385         xfs_da_intnode_t *node;
 386         xfs_dablk_t blkno;
 387         int newcount, error;
 388         int useextra;
 389
 390         node = oldblk->bp->data;
 391         ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 392
 393         /*
 394          * With V2 dirs the extra block is data or freespace.
 395          */
 396         useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
 397         newcount = 1 + useextra;
 398         /*
 399          * Do we have to split the node?
 400          */
 401         if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
 402                 /*
 403                  * Allocate a new node, add to the doubly linked chain of
 404                  * nodes, then move some of our excess entries into it.
 405                  */
 406                 error = xfs_da_grow_inode(state->args, &blkno);
 407                 if (error)
 408                         return(error);  /* GROT: dir is inconsistent */
 409
 410                 error = xfs_da_node_create(state->args, blkno, treelevel,
 411                                            &newblk->bp, state->args->whichfork);
 412                 if (error)
 413                         return(error);  /* GROT: dir is inconsistent */
 414                 newblk->blkno = blkno;
 415                 newblk->magic = XFS_DA_NODE_MAGIC;
 416                 xfs_da_node_rebalance(state, oldblk, newblk);
 417                 error = xfs_da_blk_link(state, oldblk, newblk);
 418                 if (error)
 419                         return(error);
 420                 *result = 1;
 421         } else {
 422                 *result = 0;
 423         }
 424
 425         /*
 426          * Insert the new entry(s) into the correct block
 427          * (updating last hashval in the process).
 428          *
 429          * xfs_da_node_add() inserts BEFORE the given index,
 430          * and as a result of using node_lookup_int() we always
 431          * point to a valid entry (not after one), but a split
 432          * operation always results in a new block whose hashvals
 433          * FOLLOW the current block.
 434          *
 435          * If we had double-split op below us, then add the extra block too.
 436          */
 437         node = oldblk->bp->data;
 438         if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
 439                 oldblk->index++;
 440                 xfs_da_node_add(state, oldblk, addblk);
 441                 if (useextra) {
 442                         if (state->extraafter)
 443                                 oldblk->index++;
 444                         xfs_da_node_add(state, oldblk, &state->extrablk);
 445                         state->extravalid = 0;
 446                 }
 447         } else {
 448                 newblk->index++;
 449                 xfs_da_node_add(state, newblk, addblk);
 450                 if (useextra) {
 451                         if (state->extraafter)
 452                                 newblk->index++;
 453                         xfs_da_node_add(state, newblk, &state->extrablk);
 454                         state->extravalid = 0;
 455                 }
 456         }
 457
 458         return(0);
 459 }
 460
 461 /*
 462  * Balance the btree elements between two intermediate nodes,
 463  * usually one full and one empty.
 464  *
 465  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 466  */
 467 STATIC void
 468 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
 469                                      xfs_da_state_blk_t *blk2)
 470 {
 471         xfs_da_intnode_t *node1, *node2, *tmpnode;
 472         xfs_da_node_entry_t *btree_s, *btree_d;
 473         int count, tmp;
 474         xfs_trans_t *tp;
 475
 476         node1 = blk1->bp->data;
 477         node2 = blk2->bp->data;
 478         /*
 479          * Figure out how many entries need to move, and in which direction.
 480          * Swap the nodes around if that makes it simpler.
 481          */
 482         if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
 483             ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
 484              (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
 485               be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
 486                 tmpnode = node1;
 487                 node1 = node2;
 488                 node2 = tmpnode;
 489         }
 490         ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 491         ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 492         count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
 493         if (count == 0)
 494                 return;
 495         tp = state->args->trans;
 496         /*
 497          * Two cases: high-to-low and low-to-high.
 498          */
 499         if (count > 0) {
 500                 /*
 501                  * Move elements in node2 up to make a hole.
 502                  */
 503                 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
 504                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 505                         btree_s = &node2->btree[0];
 506                         btree_d = &node2->btree[count];
 507                         memmove(btree_d, btree_s, tmp);
 508                 }
 509
 510                 /*
 511                  * Move the req'd B-tree elements from high in node1 to
 512                  * low in node2.
 513                  */
 514                 be16_add(&node2->hdr.count, count);
 515                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 516                 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
 517                 btree_d = &node2->btree[0];
 518                 memcpy(btree_d, btree_s, tmp);
 519                 be16_add(&node1->hdr.count, -count);
 520         } else {
 521                 /*
 522                  * Move the req'd B-tree elements from low in node2 to
 523                  * high in node1.
 524                  */
 525                 count = -count;
 526                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 527                 btree_s = &node2->btree[0];
 528                 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
 529                 memcpy(btree_d, btree_s, tmp);
 530                 be16_add(&node1->hdr.count, count);
 531                 xfs_da_log_buf(tp, blk1->bp,
 532                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 533
 534                 /*
 535                  * Move elements in node2 down to fill the hole.
 536                  */
 537                 tmp  = be16_to_cpu(node2->hdr.count) - count;
 538                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 539                 btree_s = &node2->btree[count];
 540                 btree_d = &node2->btree[0];
 541                 memmove(btree_d, btree_s, tmp);
 542                 be16_add(&node2->hdr.count, -count);
 543         }
 544
 545         /*
 546          * Log header of node 1 and all current bits of node 2.
 547          */
 548         xfs_da_log_buf(tp, blk1->bp,
 549                 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
 550         xfs_da_log_buf(tp, blk2->bp,
 551                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 552                         sizeof(node2->hdr) +
 553                         sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
 554
 555         /*
 556          * Record the last hashval from each block for upward propagation.
 557          * (note: don't use the swapped node pointers)
 558          */
 559         node1 = blk1->bp->data;
 560         node2 = blk2->bp->data;
 561         blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
 562         blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
 563
 564         /*
 565          * Adjust the expected index for insertion.
 566          */
 567         if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
 568                 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
 569                 blk1->index = be16_to_cpu(node1->hdr.count) + 1;        /* make it invalid */
 570         }
 571 }
 572
 573 /*
 574  * Add a new entry to an intermediate node.
 575  */
 576 STATIC void
 577 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 578                                xfs_da_state_blk_t *newblk)
 579 {
 580         xfs_da_intnode_t *node;
 581         xfs_da_node_entry_t *btree;
 582         int tmp;
 583         xfs_mount_t *mp;
 584
 585         node = oldblk->bp->data;
 586         mp = state->mp;
 587         ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 588         ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
 589         ASSERT(newblk->blkno != 0);
 590         if (state->args->whichfork == XFS_DATA_FORK)
 591                 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
 592                        newblk->blkno < mp->m_dirfreeblk);
 593
 594         /*
 595          * We may need to make some room before we insert the new node.
 596          */
 597         tmp = 0;
 598         btree = &node->btree[ oldblk->index ];
 599         if (oldblk->index < be16_to_cpu(node->hdr.count)) {
 600                 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
 601                 memmove(btree + 1, btree, tmp);
 602         }
 603         btree->hashval = cpu_to_be32(newblk->hashval);
 604         btree->before = cpu_to_be32(newblk->blkno);
 605         xfs_da_log_buf(state->args->trans, oldblk->bp,
 606                 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
 607         be16_add(&node->hdr.count, 1);
 608         xfs_da_log_buf(state->args->trans, oldblk->bp,
 609                 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 610
 611         /*
 612          * Copy the last hash value from the oldblk to propagate upwards.
 613          */
 614         oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
 615 }
 616
 617 /*========================================================================
 618  * Routines used for shrinking the Btree.
 619  *========================================================================*/
 620
 621 /*
 622  * Deallocate an empty leaf node, remove it from its parent,
 623  * possibly deallocating that block, etc...
 624  */
 625 int
 626 xfs_da_join(xfs_da_state_t *state)
 627 {
 628         xfs_da_state_blk_t *drop_blk, *save_blk;
 629         int action, error;
 630
 631         action = 0;
 632         drop_blk = &state->path.blk[ state->path.active-1 ];
 633         save_blk = &state->altpath.blk[ state->path.active-1 ];
 634         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 635         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 636                drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
 637
 638         /*
 639          * Walk back up the tree joining/deallocating as necessary.
 640          * When we stop dropping blocks, break out.
 641          */
 642         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 643                  state->path.active--) {
 644                 /*
 645                  * See if we can combine the block with a neighbor.
 646                  *   (action == 0) => no options, just leave
 647                  *   (action == 1) => coalesce, then unlink
 648                  *   (action == 2) => block empty, unlink it
 649                  */
 650                 switch (drop_blk->magic) {
 651                 case XFS_ATTR_LEAF_MAGIC:
 652                         error = xfs_attr_leaf_toosmall(state, &action);
 653                         if (error)
 654                                 return(error);
 655                         if (action == 0)
 656                                 return(0);
 657                         xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
 658                         break;
 659                 case XFS_DIR2_LEAFN_MAGIC:
 660                         error = xfs_dir2_leafn_toosmall(state, &action);
 661                         if (error)
 662                                 return error;
 663                         if (action == 0)
 664                                 return 0;
 665                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 666                         break;
 667                 case XFS_DA_NODE_MAGIC:
 668                         /*
 669                          * Remove the offending node, fixup hashvals,
 670                          * check for a toosmall neighbor.
 671                          */
 672                         xfs_da_node_remove(state, drop_blk);
 673                         xfs_da_fixhashpath(state, &state->path);
 674                         error = xfs_da_node_toosmall(state, &action);
 675                         if (error)
 676                                 return(error);
 677                         if (action == 0)
 678                                 return 0;
 679                         xfs_da_node_unbalance(state, drop_blk, save_blk);
 680                         break;
 681                 }
 682                 xfs_da_fixhashpath(state, &state->altpath);
 683                 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
 684                 xfs_da_state_kill_altpath(state);
 685                 if (error)
 686                         return(error);
 687                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
 688                                                          drop_blk->bp);
 689                 drop_blk->bp = NULL;
 690                 if (error)
 691                         return(error);
 692         }
 693         /*
 694          * We joined all the way to the top.  If it turns out that
 695          * we only have one entry in the root, make the child block
 696          * the new root.
 697          */
 698         xfs_da_node_remove(state, drop_blk);
 699         xfs_da_fixhashpath(state, &state->path);
 700         error = xfs_da_root_join(state, &state->path.blk[0]);
 701         return(error);
 702 }
 703
 704 /*
 705  * We have only one entry in the root.  Copy the only remaining child of
 706  * the old root to block 0 as the new root node.
 707  */
 708 STATIC int
 709 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
 710 {
 711         xfs_da_intnode_t *oldroot;
 712         /* REFERENCED */
 713         xfs_da_blkinfo_t *blkinfo;
 714         xfs_da_args_t *args;
 715         xfs_dablk_t child;
 716         xfs_dabuf_t *bp;
 717         int error;
 718
 719         args = state->args;
 720         ASSERT(args != NULL);
 721         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
 722         oldroot = root_blk->bp->data;
 723         ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 724         ASSERT(!oldroot->hdr.info.forw);
 725         ASSERT(!oldroot->hdr.info.back);
 726
 727         /*
 728          * If the root has more than one child, then don't do anything.
 729          */
 730         if (be16_to_cpu(oldroot->hdr.count) > 1)
 731                 return(0);
 732
 733         /*
 734          * Read in the (only) child block, then copy those bytes into
 735          * the root block's buffer and free the original child block.
 736          */
 737         child = be32_to_cpu(oldroot->btree[0].before);
 738         ASSERT(child != 0);
 739         error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
 740                                              args->whichfork);
 741         if (error)
 742                 return(error);
 743         ASSERT(bp != NULL);
 744         blkinfo = bp->data;
 745         if (be16_to_cpu(oldroot->hdr.level) == 1) {
 746                 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIR2_LEAFN_MAGIC ||
 747                        be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC);
 748         } else {
 749                 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC);
 750         }
 751         ASSERT(!blkinfo->forw);
 752         ASSERT(!blkinfo->back);
 753         memcpy(root_blk->bp->data, bp->data, state->blocksize);
 754         xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
 755         error = xfs_da_shrink_inode(args, child, bp);
 756         return(error);
 757 }
 758
 759 /*
 760  * Check a node block and its neighbors to see if the block should be
 761  * collapsed into one or the other neighbor.  Always keep the block
 762  * with the smaller block number.
 763  * If the current block is over 50% full, don't try to join it, return 0.
 764  * If the block is empty, fill in the state structure and return 2.
 765  * If it can be collapsed, fill in the state structure and return 1.
 766  * If nothing can be done, return 0.
 767  */
 768 STATIC int
 769 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
 770 {
 771         xfs_da_intnode_t *node;
 772         xfs_da_state_blk_t *blk;
 773         xfs_da_blkinfo_t *info;
 774         int count, forward, error, retval, i;
 775         xfs_dablk_t blkno;
 776         xfs_dabuf_t *bp;
 777
 778         /*
 779          * Check for the degenerate case of the block being over 50% full.
 780          * If so, it's not worth even looking to see if we might be able
 781          * to coalesce with a sibling.
 782          */
 783         blk = &state->path.blk[ state->path.active-1 ];
 784         info = blk->bp->data;
 785         ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC);
 786         node = (xfs_da_intnode_t *)info;
 787         count = be16_to_cpu(node->hdr.count);
 788         if (count > (state->node_ents >> 1)) {
 789                 *action = 0;    /* blk over 50%, don't try to join */
 790                 return(0);      /* blk over 50%, don't try to join */
 791         }
 792
 793         /*
 794          * Check for the degenerate case of the block being empty.
 795          * If the block is empty, we'll simply delete it, no need to
 796          * coalesce it with a sibling block.  We choose (arbitrarily)
 797          * to merge with the forward block unless it is NULL.
 798          */
 799         if (count == 0) {
 800                 /*
 801                  * Make altpath point to the block we want to keep and
 802                  * path point to the block we want to drop (this one).
 803                  */
 804                 forward = (info->forw != 0);
 805                 memcpy(&state->altpath, &state->path, sizeof(state->path));
 806                 error = xfs_da_path_shift(state, &state->altpath, forward,
 807                                                  0, &retval);
 808                 if (error)
 809                         return(error);
 810                 if (retval) {
 811                         *action = 0;
 812                 } else {
 813                         *action = 2;
 814                 }
 815                 return(0);
 816         }
 817
 818         /*
 819          * Examine each sibling block to see if we can coalesce with
 820          * at least 25% free space to spare.  We need to figure out
 821          * whether to merge with the forward or the backward block.
 822          * We prefer coalescing with the lower numbered sibling so as
 823          * to shrink a directory over time.
 824          */
 825         /* start with smaller blk num */
 826         forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
 827         for (i = 0; i < 2; forward = !forward, i++) {
 828                 if (forward)
 829                         blkno = be32_to_cpu(info->forw);
 830                 else
 831                         blkno = be32_to_cpu(info->back);
 832                 if (blkno == 0)
 833                         continue;
 834                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
 835                                         blkno, -1, &bp, state->args->whichfork);
 836                 if (error)
 837                         return(error);
 838                 ASSERT(bp != NULL);
 839
 840                 node = (xfs_da_intnode_t *)info;
 841                 count  = state->node_ents;
 842                 count -= state->node_ents >> 2;
 843                 count -= be16_to_cpu(node->hdr.count);
 844                 node = bp->data;
 845                 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 846                 count -= be16_to_cpu(node->hdr.count);
 847                 xfs_da_brelse(state->args->trans, bp);
 848                 if (count >= 0)
 849                         break;  /* fits with at least 25% to spare */
 850         }
 851         if (i >= 2) {
 852                 *action = 0;
 853                 return(0);
 854         }
 855
 856         /*
 857          * Make altpath point to the block we want to keep (the lower
 858          * numbered block) and path point to the block we want to drop.
 859          */
 860         memcpy(&state->altpath, &state->path, sizeof(state->path));
 861         if (blkno < blk->blkno) {
 862                 error = xfs_da_path_shift(state, &state->altpath, forward,
 863                                                  0, &retval);
 864                 if (error) {
 865                         return(error);
 866                 }
 867                 if (retval) {
 868                         *action = 0;
 869                         return(0);
 870                 }
 871         } else {
 872                 error = xfs_da_path_shift(state, &state->path, forward,
 873                                                  0, &retval);
 874                 if (error) {
 875                         return(error);
 876                 }
 877                 if (retval) {
 878                         *action = 0;
 879                         return(0);
 880                 }
 881         }
 882         *action = 1;
 883         return(0);
 884 }
 885
 886 /*
 887  * Walk back up the tree adjusting hash values as necessary,
 888  * when we stop making changes, return.
 889  */
 890 void
 891 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
 892 {
 893         xfs_da_state_blk_t *blk;
 894         xfs_da_intnode_t *node;
 895         xfs_da_node_entry_t *btree;
 896         xfs_dahash_t lasthash=0;
 897         int level, count;
 898
 899         level = path->active-1;
 900         blk = &path->blk[ level ];
 901         switch (blk->magic) {
 902         case XFS_ATTR_LEAF_MAGIC:
 903                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
 904                 if (count == 0)
 905                         return;
 906                 break;
 907         case XFS_DIR2_LEAFN_MAGIC:
 908                 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
 909                 if (count == 0)
 910                         return;
 911                 break;
 912         case XFS_DA_NODE_MAGIC:
 913                 lasthash = xfs_da_node_lasthash(blk->bp, &count);
 914                 if (count == 0)
 915                         return;
 916                 break;
 917         }
 918         for (blk--, level--; level >= 0; blk--, level--) {
 919                 node = blk->bp->data;
 920                 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 921                 btree = &node->btree[ blk->index ];
 922                 if (be32_to_cpu(btree->hashval) == lasthash)
 923                         break;
 924                 blk->hashval = lasthash;
 925                 btree->hashval = cpu_to_be32(lasthash);
 926                 xfs_da_log_buf(state->args->trans, blk->bp,
 927                                   XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
 928
 929                 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
 930         }
 931 }
 932
 933 /*
 934  * Remove an entry from an intermediate node.
 935  */
 936 STATIC void
 937 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
 938 {
 939         xfs_da_intnode_t *node;
 940         xfs_da_node_entry_t *btree;
 941         int tmp;
 942
 943         node = drop_blk->bp->data;
 944         ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
 945         ASSERT(drop_blk->index >= 0);
 946
 947         /*
 948          * Copy over the offending entry, or just zero it out.
 949          */
 950         btree = &node->btree[drop_blk->index];
 951         if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
 952                 tmp  = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
 953                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 954                 memmove(btree, btree + 1, tmp);
 955                 xfs_da_log_buf(state->args->trans, drop_blk->bp,
 956                     XFS_DA_LOGRANGE(node, btree, tmp));
 957                 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
 958         }
 959         memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
 960         xfs_da_log_buf(state->args->trans, drop_blk->bp,
 961             XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
 962         be16_add(&node->hdr.count, -1);
 963         xfs_da_log_buf(state->args->trans, drop_blk->bp,
 964             XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 965
 966         /*
 967          * Copy the last hash value from the block to propagate upwards.
 968          */
 969         btree--;
 970         drop_blk->hashval = be32_to_cpu(btree->hashval);
 971 }
 972
 973 /*
 974  * Unbalance the btree elements between two intermediate nodes,
 975  * move all Btree elements from one node into another.
 976  */
 977 STATIC void
 978 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
 979                                      xfs_da_state_blk_t *save_blk)
 980 {
 981         xfs_da_intnode_t *drop_node, *save_node;
 982         xfs_da_node_entry_t *btree;
 983         int tmp;
 984         xfs_trans_t *tp;
 985
 986         drop_node = drop_blk->bp->data;
 987         save_node = save_blk->bp->data;
 988         ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 989         ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
 990         tp = state->args->trans;
 991
 992         /*
 993          * If the dying block has lower hashvals, then move all the
 994          * elements in the remaining block up to make a hole.
 995          */
 996         if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
 997             (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
 998              be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
 999         {
1000                 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1001                 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1002                 memmove(btree, &save_node->btree[0], tmp);
1003                 btree = &save_node->btree[0];
1004                 xfs_da_log_buf(tp, save_blk->bp,
1005                         XFS_DA_LOGRANGE(save_node, btree,
1006                                 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1007                                 sizeof(xfs_da_node_entry_t)));
1008         } else {
1009                 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1010                 xfs_da_log_buf(tp, save_blk->bp,
1011                         XFS_DA_LOGRANGE(save_node, btree,
1012                                 be16_to_cpu(drop_node->hdr.count) *
1013                                 sizeof(xfs_da_node_entry_t)));
1014         }
1015
1016         /*
1017          * Move all the B-tree elements from drop_blk to save_blk.
1018          */
1019         tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1020         memcpy(btree, &drop_node->btree[0], tmp);
1021         be16_add(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1022
1023         xfs_da_log_buf(tp, save_blk->bp,
1024                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1025                         sizeof(save_node->hdr)));
1026
1027         /*
1028          * Save the last hashval in the remaining block for upward propagation.
1029          */
1030         save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1031 }
1032
1033 /*========================================================================
1034  * Routines used for finding things in the Btree.
1035  *========================================================================*/
1036
1037 /*
1038  * Walk down the Btree looking for a particular filename, filling
1039  * in the state structure as we go.
1040  *
1041  * We will set the state structure to point to each of the elements
1042  * in each of the nodes where either the hashval is or should be.
1043  *
1044  * We support duplicate hashval's so for each entry in the current
1045  * node that could contain the desired hashval, descend.  This is a
1046  * pruned depth-first tree search.
1047  */
1048 int                                                     /* error */
1049 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1050 {
1051         xfs_da_state_blk_t *blk;
1052         xfs_da_blkinfo_t *curr;
1053         xfs_da_intnode_t *node;
1054         xfs_da_node_entry_t *btree;
1055         xfs_dablk_t blkno;
1056         int probe, span, max, error, retval;
1057         xfs_dahash_t hashval, btreehashval;
1058         xfs_da_args_t *args;
1059
1060         args = state->args;
1061
1062         /*
1063          * Descend thru the B-tree searching each level for the right
1064          * node to use, until the right hashval is found.
1065          */
1066         blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1067         for (blk = &state->path.blk[0], state->path.active = 1;
1068                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1069                          blk++, state->path.active++) {
1070                 /*
1071                  * Read the next node down in the tree.
1072                  */
1073                 blk->blkno = blkno;
1074                 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1075                                         -1, &blk->bp, args->whichfork);
1076                 if (error) {
1077                         blk->blkno = 0;
1078                         state->path.active--;
1079                         return(error);
1080                 }
1081                 curr = blk->bp->data;
1082                 blk->magic = be16_to_cpu(curr->magic);
1083                 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1084                        blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1085                        blk->magic == XFS_ATTR_LEAF_MAGIC);
1086
1087                 /*
1088                  * Search an intermediate node for a match.
1089                  */
1090                 if (blk->magic == XFS_DA_NODE_MAGIC) {
1091                         node = blk->bp->data;
1092                         max = be16_to_cpu(node->hdr.count);
1093                         btreehashval = node->btree[max-1].hashval;
1094                         blk->hashval = be32_to_cpu(btreehashval);
1095
1096                         /*
1097                          * Binary search.  (note: small blocks will skip loop)
1098                          */
1099                         probe = span = max / 2;
1100                         hashval = args->hashval;
1101                         for (btree = &node->btree[probe]; span > 4;
1102                                    btree = &node->btree[probe]) {
1103                                 span /= 2;
1104                                 btreehashval = be32_to_cpu(btree->hashval);
1105                                 if (btreehashval < hashval)
1106                                         probe += span;
1107                                 else if (btreehashval > hashval)
1108                                         probe -= span;
1109                                 else
1110                                         break;
1111                         }
1112                         ASSERT((probe >= 0) && (probe < max));
1113                         ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
1114
1115                         /*
1116                          * Since we may have duplicate hashval's, find the first
1117                          * matching hashval in the node.
1118                          */
1119                         while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
1120                                 btree--;
1121                                 probe--;
1122                         }
1123                         while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
1124                                 btree++;
1125                                 probe++;
1126                         }
1127
1128                         /*
1129                          * Pick the right block to descend on.
1130                          */
1131                         if (probe == max) {
1132                                 blk->index = max-1;
1133                                 blkno = be32_to_cpu(node->btree[max-1].before);
1134                         } else {
1135                                 blk->index = probe;
1136                                 blkno = be32_to_cpu(btree->before);
1137                         }
1138                 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1139                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1140                         break;
1141                 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1142                         blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1143                         break;
1144                 }
1145         }
1146
1147         /*
1148          * A leaf block that ends in the hashval that we are interested in
1149          * (final hashval == search hashval) means that the next block may
1150          * contain more entries with the same hashval, shift upward to the
1151          * next leaf and keep searching.
1152          */
1153         for (;;) {
1154                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1155                         retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1156                                                         &blk->index, state);
1157                 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1158                         retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1159                         blk->index = args->index;
1160                         args->blkno = blk->blkno;
1161                 } else {
1162                         ASSERT(0);
1163                         return XFS_ERROR(EFSCORRUPTED);
1164                 }
1165                 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1166                     (blk->hashval == args->hashval)) {
1167                         error = xfs_da_path_shift(state, &state->path, 1, 1,
1168                                                          &retval);
1169                         if (error)
1170                                 return(error);
1171                         if (retval == 0) {
1172                                 continue;
1173                         } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1174                                 /* path_shift() gives ENOENT */
1175                                 retval = XFS_ERROR(ENOATTR);
1176                         }
1177                 }
1178                 break;
1179         }
1180         *result = retval;
1181         return(0);
1182 }
1183
1184 /*========================================================================
1185  * Utility routines.
1186  *========================================================================*/
1187
1188 /*
1189  * Link a new block into a doubly linked list of blocks (of whatever type).
1190  */
1191 int                                                     /* error */
1192 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1193                                xfs_da_state_blk_t *new_blk)
1194 {
1195         xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1196         xfs_da_args_t *args;
1197         int before=0, error;
1198         xfs_dabuf_t *bp;
1199
1200         /*
1201          * Set up environment.
1202          */
1203         args = state->args;
1204         ASSERT(args != NULL);
1205         old_info = old_blk->bp->data;
1206         new_info = new_blk->bp->data;
1207         ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1208                old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1209                old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1210         ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1211         ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
1212         ASSERT(old_blk->magic == new_blk->magic);
1213
1214         switch (old_blk->magic) {
1215         case XFS_ATTR_LEAF_MAGIC:
1216                 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1217                 break;
1218         case XFS_DIR2_LEAFN_MAGIC:
1219                 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1220                 break;
1221         case XFS_DA_NODE_MAGIC:
1222                 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1223                 break;
1224         }
1225
1226         /*
1227          * Link blocks in appropriate order.
1228          */
1229         if (before) {
1230                 /*
1231                  * Link new block in before existing block.
1232                  */
1233                 new_info->forw = cpu_to_be32(old_blk->blkno);
1234                 new_info->back = old_info->back;
1235                 if (old_info->back) {
1236                         error = xfs_da_read_buf(args->trans, args->dp,
1237                                                 be32_to_cpu(old_info->back),
1238                                                 -1, &bp, args->whichfork);
1239                         if (error)
1240                                 return(error);
1241                         ASSERT(bp != NULL);
1242                         tmp_info = bp->data;
1243                         ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1244                         ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1245                         tmp_info->forw = cpu_to_be32(new_blk->blkno);
1246                         xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1247                         xfs_da_buf_done(bp);
1248                 }
1249                 old_info->back = cpu_to_be32(new_blk->blkno);
1250         } else {
1251                 /*
1252                  * Link new block in after existing block.
1253                  */
1254                 new_info->forw = old_info->forw;
1255                 new_info->back = cpu_to_be32(old_blk->blkno);
1256                 if (old_info->forw) {
1257                         error = xfs_da_read_buf(args->trans, args->dp,
1258                                                 be32_to_cpu(old_info->forw),
1259                                                 -1, &bp, args->whichfork);
1260                         if (error)
1261                                 return(error);
1262                         ASSERT(bp != NULL);
1263                         tmp_info = bp->data;
1264                         ASSERT(tmp_info->magic == old_info->magic);
1265                         ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1266                         tmp_info->back = cpu_to_be32(new_blk->blkno);
1267                         xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1268                         xfs_da_buf_done(bp);
1269                 }
1270                 old_info->forw = cpu_to_be32(new_blk->blkno);
1271         }
1272
1273         xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1274         xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1275         return(0);
1276 }
1277
1278 /*
1279  * Compare two intermediate nodes for "order".
1280  */
1281 STATIC int
1282 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1283 {
1284         xfs_da_intnode_t *node1, *node2;
1285
1286         node1 = node1_bp->data;
1287         node2 = node2_bp->data;
1288         ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) &&
1289                (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC));
1290         if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
1291             ((be32_to_cpu(node2->btree[0].hashval) <
1292               be32_to_cpu(node1->btree[0].hashval)) ||
1293              (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1294               be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
1295                 return(1);
1296         }
1297         return(0);
1298 }
1299
1300 /*
1301  * Pick up the last hashvalue from an intermediate node.
1302  */
1303 STATIC uint
1304 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1305 {
1306         xfs_da_intnode_t *node;
1307
1308         node = bp->data;
1309         ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1310         if (count)
1311                 *count = be16_to_cpu(node->hdr.count);
1312         if (!node->hdr.count)
1313                 return(0);
1314         return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1315 }
1316
1317 /*
1318  * Unlink a block from a doubly linked list of blocks.
1319  */
1320 STATIC int                                              /* error */
1321 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1322                                  xfs_da_state_blk_t *save_blk)
1323 {
1324         xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1325         xfs_da_args_t *args;
1326         xfs_dabuf_t *bp;
1327         int error;
1328
1329         /*
1330          * Set up environment.
1331          */
1332         args = state->args;
1333         ASSERT(args != NULL);
1334         save_info = save_blk->bp->data;
1335         drop_info = drop_blk->bp->data;
1336         ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1337                save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1338                save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1339         ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1340         ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
1341         ASSERT(save_blk->magic == drop_blk->magic);
1342         ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1343                (be32_to_cpu(save_info->back) == drop_blk->blkno));
1344         ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1345                (be32_to_cpu(drop_info->back) == save_blk->blkno));
1346
1347         /*
1348          * Unlink the leaf block from the doubly linked chain of leaves.
1349          */
1350         if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1351                 save_info->back = drop_info->back;
1352                 if (drop_info->back) {
1353                         error = xfs_da_read_buf(args->trans, args->dp,
1354                                                 be32_to_cpu(drop_info->back),
1355                                                 -1, &bp, args->whichfork);
1356                         if (error)
1357                                 return(error);
1358                         ASSERT(bp != NULL);
1359                         tmp_info = bp->data;
1360                         ASSERT(tmp_info->magic == save_info->magic);
1361                         ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1362                         tmp_info->forw = cpu_to_be32(save_blk->blkno);
1363                         xfs_da_log_buf(args->trans, bp, 0,
1364                                                     sizeof(*tmp_info) - 1);
1365                         xfs_da_buf_done(bp);
1366                 }
1367         } else {
1368                 save_info->forw = drop_info->forw;
1369                 if (drop_info->forw) {
1370                         error = xfs_da_read_buf(args->trans, args->dp,
1371                                                 be32_to_cpu(drop_info->forw),
1372                                                 -1, &bp, args->whichfork);
1373                         if (error)
1374                                 return(error);
1375                         ASSERT(bp != NULL);
1376                         tmp_info = bp->data;
1377                         ASSERT(tmp_info->magic == save_info->magic);
1378                         ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1379                         tmp_info->back = cpu_to_be32(save_blk->blkno);
1380                         xfs_da_log_buf(args->trans, bp, 0,
1381                                                     sizeof(*tmp_info) - 1);
1382                         xfs_da_buf_done(bp);
1383                 }
1384         }
1385
1386         xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1387         return(0);
1388 }
1389
1390 /*
1391  * Move a path "forward" or "!forward" one block at the current level.
1392  *
1393  * This routine will adjust a "path" to point to the next block
1394  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1395  * Btree, including updating pointers to the intermediate nodes between
1396  * the new bottom and the root.
1397  */
1398 int                                                     /* error */
1399 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1400                                  int forward, int release, int *result)
1401 {
1402         xfs_da_state_blk_t *blk;
1403         xfs_da_blkinfo_t *info;
1404         xfs_da_intnode_t *node;
1405         xfs_da_args_t *args;
1406         xfs_dablk_t blkno=0;
1407         int level, error;
1408
1409         /*
1410          * Roll up the Btree looking for the first block where our
1411          * current index is not at the edge of the block.  Note that
1412          * we skip the bottom layer because we want the sibling block.
1413          */
1414         args = state->args;
1415         ASSERT(args != NULL);
1416         ASSERT(path != NULL);
1417         ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1418         level = (path->active-1) - 1;   /* skip bottom layer in path */
1419         for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1420                 ASSERT(blk->bp != NULL);
1421                 node = blk->bp->data;
1422                 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1423                 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
1424                         blk->index++;
1425                         blkno = be32_to_cpu(node->btree[blk->index].before);
1426                         break;
1427                 } else if (!forward && (blk->index > 0)) {
1428                         blk->index--;
1429                         blkno = be32_to_cpu(node->btree[blk->index].before);
1430                         break;
1431                 }
1432         }
1433         if (level < 0) {
1434                 *result = XFS_ERROR(ENOENT);    /* we're out of our tree */
1435                 ASSERT(args->oknoent);
1436                 return(0);
1437         }
1438
1439         /*
1440          * Roll down the edge of the subtree until we reach the
1441          * same depth we were at originally.
1442          */
1443         for (blk++, level++; level < path->active; blk++, level++) {
1444                 /*
1445                  * Release the old block.
1446                  * (if it's dirty, trans won't actually let go)
1447                  */
1448                 if (release)
1449                         xfs_da_brelse(args->trans, blk->bp);
1450
1451                 /*
1452                  * Read the next child block.
1453                  */
1454                 blk->blkno = blkno;
1455                 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1456                                                      &blk->bp, args->whichfork);
1457                 if (error)
1458                         return(error);
1459                 ASSERT(blk->bp != NULL);
1460                 info = blk->bp->data;
1461                 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC ||
1462                        be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC ||
1463                        be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1464                 blk->magic = be16_to_cpu(info->magic);
1465                 if (blk->magic == XFS_DA_NODE_MAGIC) {
1466                         node = (xfs_da_intnode_t *)info;
1467                         blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1468                         if (forward)
1469                                 blk->index = 0;
1470                         else
1471                                 blk->index = be16_to_cpu(node->hdr.count)-1;
1472                         blkno = be32_to_cpu(node->btree[blk->index].before);
1473                 } else {
1474                         ASSERT(level == path->active-1);
1475                         blk->index = 0;
1476                         switch(blk->magic) {
1477                         case XFS_ATTR_LEAF_MAGIC:
1478                                 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1479                                                                       NULL);
1480                                 break;
1481                         case XFS_DIR2_LEAFN_MAGIC:
1482                                 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1483                                                                        NULL);
1484                                 break;
1485                         default:
1486                                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1487                                        blk->magic == XFS_DIR2_LEAFN_MAGIC);
1488                                 break;
1489                         }
1490                 }
1491         }
1492         *result = 0;
1493         return(0);
1494 }
1495
1496
1497 /*========================================================================
1498  * Utility routines.
1499  *========================================================================*/
1500
1501 /*
1502  * Implement a simple hash on a character string.
1503  * Rotate the hash value by 7 bits, then XOR each character in.
1504  * This is implemented with some source-level loop unrolling.
1505  */
1506 xfs_dahash_t
1507 xfs_da_hashname(const uchar_t *name, int namelen)
1508 {
1509         xfs_dahash_t hash;
1510
1511         /*
1512          * Do four characters at a time as long as we can.
1513          */
1514         for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1515                 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1516                        (name[3] << 0) ^ rol32(hash, 7 * 4);
1517
1518         /*
1519          * Now do the rest of the characters.
1520          */
1521         switch (namelen) {
1522         case 3:
1523                 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1524                        rol32(hash, 7 * 3);
1525         case 2:
1526                 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1527         case 1:
1528                 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1529         default: /* case 0: */
1530                 return hash;
1531         }
1532 }
1533
1534 /*
1535  * Add a block to the btree ahead of the file.
1536  * Return the new block number to the caller.
1537  */
1538 int
1539 xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1540 {
1541         xfs_fileoff_t bno, b;
1542         xfs_bmbt_irec_t map;
1543         xfs_bmbt_irec_t *mapp;
1544         xfs_inode_t *dp;
1545         int nmap, error, w, count, c, got, i, mapi;
1546         xfs_trans_t *tp;
1547         xfs_mount_t *mp;
1548
1549         dp = args->dp;
1550         mp = dp->i_mount;
1551         w = args->whichfork;
1552         tp = args->trans;
1553         /*
1554          * For new directories adjust the file offset and block count.
1555          */
1556         if (w == XFS_DATA_FORK) {
1557                 bno = mp->m_dirleafblk;
1558                 count = mp->m_dirblkfsbs;
1559         } else {
1560                 bno = 0;
1561                 count = 1;
1562         }
1563         /*
1564          * Find a spot in the file space to put the new block.
1565          */
1566         if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w)))
1567                 return error;
1568         if (w == XFS_DATA_FORK)
1569                 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1570         /*
1571          * Try mapping it in one filesystem block.
1572          */
1573         nmap = 1;
1574         ASSERT(args->firstblock != NULL);
1575         if ((error = xfs_bmapi(tp, dp, bno, count,
1576                         XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1577                         XFS_BMAPI_CONTIG,
1578                         args->firstblock, args->total, &map, &nmap,
1579                         args->flist, NULL))) {
1580                 return error;
1581         }
1582         ASSERT(nmap <= 1);
1583         if (nmap == 1) {
1584                 mapp = &map;
1585                 mapi = 1;
1586         }
1587         /*
1588          * If we didn't get it and the block might work if fragmented,
1589          * try without the CONTIG flag.  Loop until we get it all.
1590          */
1591         else if (nmap == 0 && count > 1) {
1592                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1593                 for (b = bno, mapi = 0; b < bno + count; ) {
1594                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1595                         c = (int)(bno + count - b);
1596                         if ((error = xfs_bmapi(tp, dp, b, c,
1597                                         XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|
1598                                         XFS_BMAPI_METADATA,
1599                                         args->firstblock, args->total,
1600                                         &mapp[mapi], &nmap, args->flist,
1601                                         NULL))) {
1602                                 kmem_free(mapp, sizeof(*mapp) * count);
1603                                 return error;
1604                         }
1605                         if (nmap < 1)
1606                                 break;
1607                         mapi += nmap;
1608                         b = mapp[mapi - 1].br_startoff +
1609                             mapp[mapi - 1].br_blockcount;
1610                 }
1611         } else {
1612                 mapi = 0;
1613                 mapp = NULL;
1614         }
1615         /*
1616          * Count the blocks we got, make sure it matches the total.
1617          */
1618         for (i = 0, got = 0; i < mapi; i++)
1619                 got += mapp[i].br_blockcount;
1620         if (got != count || mapp[0].br_startoff != bno ||
1621             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1622             bno + count) {
1623                 if (mapp != &map)
1624                         kmem_free(mapp, sizeof(*mapp) * count);
1625                 return XFS_ERROR(ENOSPC);
1626         }
1627         if (mapp != &map)
1628                 kmem_free(mapp, sizeof(*mapp) * count);
1629         *new_blkno = (xfs_dablk_t)bno;
1630         return 0;
1631 }
1632
1633 /*
1634  * Ick.  We need to always be able to remove a btree block, even
1635  * if there's no space reservation because the filesystem is full.
1636  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1637  * It swaps the target block with the last block in the file.  The
1638  * last block in the file can always be removed since it can't cause
1639  * a bmap btree split to do that.
1640  */
1641 STATIC int
1642 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1643                       xfs_dabuf_t **dead_bufp)
1644 {
1645         xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1646         xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1647         xfs_fileoff_t lastoff;
1648         xfs_inode_t *ip;
1649         xfs_trans_t *tp;
1650         xfs_mount_t *mp;
1651         int error, w, entno, level, dead_level;
1652         xfs_da_blkinfo_t *dead_info, *sib_info;
1653         xfs_da_intnode_t *par_node, *dead_node;
1654         xfs_dir2_leaf_t *dead_leaf2;
1655         xfs_dahash_t dead_hash;
1656
1657         dead_buf = *dead_bufp;
1658         dead_blkno = *dead_blknop;
1659         tp = args->trans;
1660         ip = args->dp;
1661         w = args->whichfork;
1662         ASSERT(w == XFS_DATA_FORK);
1663         mp = ip->i_mount;
1664         lastoff = mp->m_dirfreeblk;
1665         error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1666         if (error)
1667                 return error;
1668         if (unlikely(lastoff == 0)) {
1669                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1670                                  mp);
1671                 return XFS_ERROR(EFSCORRUPTED);
1672         }
1673         /*
1674          * Read the last block in the btree space.
1675          */
1676         last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1677         if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1678                 return error;
1679         /*
1680          * Copy the last block into the dead buffer and log it.
1681          */
1682         memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1683         xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1684         dead_info = dead_buf->data;
1685         /*
1686          * Get values from the moved block.
1687          */
1688         if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) {
1689                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1690                 dead_level = 0;
1691                 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1692         } else {
1693                 ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC);
1694                 dead_node = (xfs_da_intnode_t *)dead_info;
1695                 dead_level = be16_to_cpu(dead_node->hdr.level);
1696                 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1697         }
1698         sib_buf = par_buf = NULL;
1699         /*
1700          * If the moved block has a left sibling, fix up the pointers.
1701          */
1702         if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1703                 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1704                         goto done;
1705                 sib_info = sib_buf->data;
1706                 if (unlikely(
1707                     be32_to_cpu(sib_info->forw) != last_blkno ||
1708                     sib_info->magic != dead_info->magic)) {
1709                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1710                                          XFS_ERRLEVEL_LOW, mp);
1711                         error = XFS_ERROR(EFSCORRUPTED);
1712                         goto done;
1713                 }
1714                 sib_info->forw = cpu_to_be32(dead_blkno);
1715                 xfs_da_log_buf(tp, sib_buf,
1716                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1717                                         sizeof(sib_info->forw)));
1718                 xfs_da_buf_done(sib_buf);
1719                 sib_buf = NULL;
1720         }
1721         /*
1722          * If the moved block has a right sibling, fix up the pointers.
1723          */
1724         if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1725                 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1726                         goto done;
1727                 sib_info = sib_buf->data;
1728                 if (unlikely(
1729                        be32_to_cpu(sib_info->back) != last_blkno ||
1730                        sib_info->magic != dead_info->magic)) {
1731                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1732                                          XFS_ERRLEVEL_LOW, mp);
1733                         error = XFS_ERROR(EFSCORRUPTED);
1734                         goto done;
1735                 }
1736                 sib_info->back = cpu_to_be32(dead_blkno);
1737                 xfs_da_log_buf(tp, sib_buf,
1738                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1739                                         sizeof(sib_info->back)));
1740                 xfs_da_buf_done(sib_buf);
1741                 sib_buf = NULL;
1742         }
1743         par_blkno = mp->m_dirleafblk;
1744         level = -1;
1745         /*
1746          * Walk down the tree looking for the parent of the moved block.
1747          */
1748         for (;;) {
1749                 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1750                         goto done;
1751                 par_node = par_buf->data;
1752                 if (unlikely(
1753                     be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC ||
1754                     (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1755                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1756                                          XFS_ERRLEVEL_LOW, mp);
1757                         error = XFS_ERROR(EFSCORRUPTED);
1758                         goto done;
1759                 }
1760                 level = be16_to_cpu(par_node->hdr.level);
1761                 for (entno = 0;
1762                      entno < be16_to_cpu(par_node->hdr.count) &&
1763                      be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1764                      entno++)
1765                         continue;
1766                 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1767                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1768                                          XFS_ERRLEVEL_LOW, mp);
1769                         error = XFS_ERROR(EFSCORRUPTED);
1770                         goto done;
1771                 }
1772                 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1773                 if (level == dead_level + 1)
1774                         break;
1775                 xfs_da_brelse(tp, par_buf);
1776                 par_buf = NULL;
1777         }
1778         /*
1779          * We're in the right parent block.
1780          * Look for the right entry.
1781          */
1782         for (;;) {
1783                 for (;
1784                      entno < be16_to_cpu(par_node->hdr.count) &&
1785                      be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1786                      entno++)
1787                         continue;
1788                 if (entno < be16_to_cpu(par_node->hdr.count))
1789                         break;
1790                 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1791                 xfs_da_brelse(tp, par_buf);
1792                 par_buf = NULL;
1793                 if (unlikely(par_blkno == 0)) {
1794                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1795                                          XFS_ERRLEVEL_LOW, mp);
1796                         error = XFS_ERROR(EFSCORRUPTED);
1797                         goto done;
1798                 }
1799                 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1800                         goto done;
1801                 par_node = par_buf->data;
1802                 if (unlikely(
1803                     be16_to_cpu(par_node->hdr.level) != level ||
1804                     be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) {
1805                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1806                                          XFS_ERRLEVEL_LOW, mp);
1807                         error = XFS_ERROR(EFSCORRUPTED);
1808                         goto done;
1809                 }
1810                 entno = 0;
1811         }
1812         /*
1813          * Update the parent entry pointing to the moved block.
1814          */
1815         par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1816         xfs_da_log_buf(tp, par_buf,
1817                 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1818                                 sizeof(par_node->btree[entno].before)));
1819         xfs_da_buf_done(par_buf);
1820         xfs_da_buf_done(dead_buf);
1821         *dead_blknop = last_blkno;
1822         *dead_bufp = last_buf;
1823         return 0;
1824 done:
1825         if (par_buf)
1826                 xfs_da_brelse(tp, par_buf);
1827         if (sib_buf)
1828                 xfs_da_brelse(tp, sib_buf);
1829         xfs_da_brelse(tp, last_buf);
1830         return error;
1831 }
1832
1833 /*
1834  * Remove a btree block from a directory or attribute.
1835  */
1836 int
1837 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1838                     xfs_dabuf_t *dead_buf)
1839 {
1840         xfs_inode_t *dp;
1841         int done, error, w, count;
1842         xfs_trans_t *tp;
1843         xfs_mount_t *mp;
1844
1845         dp = args->dp;
1846         w = args->whichfork;
1847         tp = args->trans;
1848         mp = dp->i_mount;
1849         if (w == XFS_DATA_FORK)
1850                 count = mp->m_dirblkfsbs;
1851         else
1852                 count = 1;
1853         for (;;) {
1854                 /*
1855                  * Remove extents.  If we get ENOSPC for a dir we have to move
1856                  * the last block to the place we want to kill.
1857                  */
1858                 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1859                                 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_METADATA,
1860                                 0, args->firstblock, args->flist, NULL,
1861                                 &done)) == ENOSPC) {
1862                         if (w != XFS_DATA_FORK)
1863                                 break;
1864                         if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1865                                         &dead_buf)))
1866                                 break;
1867                 } else {
1868                         break;
1869                 }
1870         }
1871         xfs_da_binval(tp, dead_buf);
1872         return error;
1873 }
1874
1875 /*
1876  * See if the mapping(s) for this btree block are valid, i.e.
1877  * don't contain holes, are logically contiguous, and cover the whole range.
1878  */
1879 STATIC int
1880 xfs_da_map_covers_blocks(
1881         int             nmap,
1882         xfs_bmbt_irec_t *mapp,
1883         xfs_dablk_t     bno,
1884         int             count)
1885 {
1886         int             i;
1887         xfs_fileoff_t   off;
1888
1889         for (i = 0, off = bno; i < nmap; i++) {
1890                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1891                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
1892                         return 0;
1893                 }
1894                 if (off != mapp[i].br_startoff) {
1895                         return 0;
1896                 }
1897                 off += mapp[i].br_blockcount;
1898         }
1899         return off == bno + count;
1900 }
1901
1902 /*
1903  * Make a dabuf.
1904  * Used for get_buf, read_buf, read_bufr, and reada_buf.
1905  */
1906 STATIC int
1907 xfs_da_do_buf(
1908         xfs_trans_t     *trans,
1909         xfs_inode_t     *dp,
1910         xfs_dablk_t     bno,
1911         xfs_daddr_t     *mappedbnop,
1912         xfs_dabuf_t     **bpp,
1913         int             whichfork,
1914         int             caller,
1915         inst_t          *ra)
1916 {
1917         xfs_buf_t       *bp = NULL;
1918         xfs_buf_t       **bplist;
1919         int             error=0;
1920         int             i;
1921         xfs_bmbt_irec_t map;
1922         xfs_bmbt_irec_t *mapp;
1923         xfs_daddr_t     mappedbno;
1924         xfs_mount_t     *mp;
1925         int             nbplist=0;
1926         int             nfsb;
1927         int             nmap;
1928         xfs_dabuf_t     *rbp;
1929
1930         mp = dp->i_mount;
1931         nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
1932         mappedbno = *mappedbnop;
1933         /*
1934          * Caller doesn't have a mapping.  -2 means don't complain
1935          * if we land in a hole.
1936          */
1937         if (mappedbno == -1 || mappedbno == -2) {
1938                 /*
1939                  * Optimize the one-block case.
1940                  */
1941                 if (nfsb == 1) {
1942                         xfs_fsblock_t   fsb;
1943
1944                         if ((error =
1945                             xfs_bmapi_single(trans, dp, whichfork, &fsb,
1946                                     (xfs_fileoff_t)bno))) {
1947                                 return error;
1948                         }
1949                         mapp = &map;
1950                         if (fsb == NULLFSBLOCK) {
1951                                 nmap = 0;
1952                         } else {
1953                                 map.br_startblock = fsb;
1954                                 map.br_startoff = (xfs_fileoff_t)bno;
1955                                 map.br_blockcount = 1;
1956                                 nmap = 1;
1957                         }
1958                 } else {
1959                         mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1960                         nmap = nfsb;
1961                         if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1962                                         nfsb,
1963                                         XFS_BMAPI_METADATA |
1964                                                 XFS_BMAPI_AFLAG(whichfork),
1965                                         NULL, 0, mapp, &nmap, NULL, NULL)))
1966                                 goto exit0;
1967                 }
1968         } else {
1969                 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1970                 map.br_startoff = (xfs_fileoff_t)bno;
1971                 map.br_blockcount = nfsb;
1972                 mapp = &map;
1973                 nmap = 1;
1974         }
1975         if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
1976                 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
1977                 if (unlikely(error == EFSCORRUPTED)) {
1978                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
1979                                 int     i;
1980                                 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
1981                                         (long long)bno);
1982                                 cmn_err(CE_ALERT, "dir: inode %lld\n",
1983                                         (long long)dp->i_ino);
1984                                 for (i = 0; i < nmap; i++) {
1985                                         cmn_err(CE_ALERT,
1986                                                 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
1987                                                 i,
1988                                                 (long long)mapp[i].br_startoff,
1989                                                 (long long)mapp[i].br_startblock,
1990                                                 (long long)mapp[i].br_blockcount,
1991                                                 mapp[i].br_state);
1992                                 }
1993                         }
1994                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
1995                                          XFS_ERRLEVEL_LOW, mp);
1996                 }
1997                 goto exit0;
1998         }
1999         if (caller != 3 && nmap > 1) {
2000                 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2001                 nbplist = 0;
2002         } else
2003                 bplist = NULL;
2004         /*
2005          * Turn the mapping(s) into buffer(s).
2006          */
2007         for (i = 0; i < nmap; i++) {
2008                 int     nmapped;
2009
2010                 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2011                 if (i == 0)
2012                         *mappedbnop = mappedbno;
2013                 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2014                 switch (caller) {
2015                 case 0:
2016                         bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2017                                 mappedbno, nmapped, 0);
2018                         error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2019                         break;
2020                 case 1:
2021                 case 2:
2022                         bp = NULL;
2023                         error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2024                                 mappedbno, nmapped, 0, &bp);
2025                         break;
2026                 case 3:
2027                         xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2028                         error = 0;
2029                         bp = NULL;
2030                         break;
2031                 }
2032                 if (error) {
2033                         if (bp)
2034                                 xfs_trans_brelse(trans, bp);
2035                         goto exit1;
2036                 }
2037                 if (!bp)
2038                         continue;
2039                 if (caller == 1) {
2040                         if (whichfork == XFS_ATTR_FORK) {
2041                                 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2042                                                 XFS_ATTR_BTREE_REF);
2043                         } else {
2044                                 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2045                                                 XFS_DIR_BTREE_REF);
2046                         }
2047                 }
2048                 if (bplist) {
2049                         bplist[nbplist++] = bp;
2050                 }
2051         }
2052         /*
2053          * Build a dabuf structure.
2054          */
2055         if (bplist) {
2056                 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2057         } else if (bp)
2058                 rbp = xfs_da_buf_make(1, &bp, ra);
2059         else
2060                 rbp = NULL;
2061         /*
2062          * For read_buf, check the magic number.
2063          */
2064         if (caller == 1) {
2065                 xfs_dir2_data_t         *data;
2066                 xfs_dir2_free_t         *free;
2067                 xfs_da_blkinfo_t        *info;
2068                 uint                    magic, magic1;
2069
2070                 info = rbp->data;
2071                 data = rbp->data;
2072                 free = rbp->data;
2073                 magic = be16_to_cpu(info->magic);
2074                 magic1 = be32_to_cpu(data->hdr.magic);
2075                 if (unlikely(
2076                     XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2077                                    (magic != XFS_ATTR_LEAF_MAGIC) &&
2078                                    (magic != XFS_DIR2_LEAF1_MAGIC) &&
2079                                    (magic != XFS_DIR2_LEAFN_MAGIC) &&
2080                                    (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2081                                    (magic1 != XFS_DIR2_DATA_MAGIC) &&
2082                                    (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC),
2083                                 mp, XFS_ERRTAG_DA_READ_BUF,
2084                                 XFS_RANDOM_DA_READ_BUF))) {
2085                         xfs_buftrace("DA READ ERROR", rbp->bps[0]);
2086                         XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2087                                              XFS_ERRLEVEL_LOW, mp, info);
2088                         error = XFS_ERROR(EFSCORRUPTED);
2089                         xfs_da_brelse(trans, rbp);
2090                         nbplist = 0;
2091                         goto exit1;
2092                 }
2093         }
2094         if (bplist) {
2095                 kmem_free(bplist, sizeof(*bplist) * nmap);
2096         }
2097         if (mapp != &map) {
2098                 kmem_free(mapp, sizeof(*mapp) * nfsb);
2099         }
2100         if (bpp)
2101                 *bpp = rbp;
2102         return 0;
2103 exit1:
2104         if (bplist) {
2105                 for (i = 0; i < nbplist; i++)
2106                         xfs_trans_brelse(trans, bplist[i]);
2107                 kmem_free(bplist, sizeof(*bplist) * nmap);
2108         }
2109 exit0:
2110         if (mapp != &map)
2111                 kmem_free(mapp, sizeof(*mapp) * nfsb);
2112         if (bpp)
2113                 *bpp = NULL;
2114         return error;
2115 }
2116
2117 /*
2118  * Get a buffer for the dir/attr block.
2119  */
2120 int
2121 xfs_da_get_buf(
2122         xfs_trans_t     *trans,
2123         xfs_inode_t     *dp,
2124         xfs_dablk_t     bno,
2125         xfs_daddr_t             mappedbno,
2126         xfs_dabuf_t     **bpp,
2127         int             whichfork)
2128 {
2129         return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2130                                                  (inst_t *)__return_address);
2131 }
2132
2133 /*
2134  * Get a buffer for the dir/attr block, fill in the contents.
2135  */
2136 int
2137 xfs_da_read_buf(
2138         xfs_trans_t     *trans,
2139         xfs_inode_t     *dp,
2140         xfs_dablk_t     bno,
2141         xfs_daddr_t             mappedbno,
2142         xfs_dabuf_t     **bpp,
2143         int             whichfork)
2144 {
2145         return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2146                 (inst_t *)__return_address);
2147 }
2148
2149 /*
2150  * Readahead the dir/attr block.
2151  */
2152 xfs_daddr_t
2153 xfs_da_reada_buf(
2154         xfs_trans_t     *trans,
2155         xfs_inode_t     *dp,
2156         xfs_dablk_t     bno,
2157         int             whichfork)
2158 {
2159         xfs_daddr_t             rval;
2160
2161         rval = -1;
2162         if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2163                         (inst_t *)__return_address))
2164                 return -1;
2165         else
2166                 return rval;
2167 }
2168
2169 /*
2170  * Calculate the number of bits needed to hold i different values.
2171  */
2172 uint
2173 xfs_da_log2_roundup(uint i)
2174 {
2175         uint rval;
2176
2177         for (rval = 0; rval < NBBY * sizeof(i); rval++) {
2178                 if ((1 << rval) >= i)
2179                         break;
2180         }
2181         return(rval);
2182 }
2183
2184 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2185 kmem_zone_t *xfs_dabuf_zone;            /* dabuf zone */
2186
2187 /*
2188  * Allocate a dir-state structure.
2189  * We don't put them on the stack since they're large.
2190  */
2191 xfs_da_state_t *
2192 xfs_da_state_alloc(void)
2193 {
2194         return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP);
2195 }
2196
2197 /*
2198  * Kill the altpath contents of a da-state structure.
2199  */
2200 STATIC void
2201 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2202 {
2203         int     i;
2204
2205         for (i = 0; i < state->altpath.active; i++) {
2206                 if (state->altpath.blk[i].bp) {
2207                         if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2208                                 xfs_da_buf_done(state->altpath.blk[i].bp);
2209                         state->altpath.blk[i].bp = NULL;
2210                 }
2211         }
2212         state->altpath.active = 0;
2213 }
2214
2215 /*
2216  * Free a da-state structure.
2217  */
2218 void
2219 xfs_da_state_free(xfs_da_state_t *state)
2220 {
2221         int     i;
2222
2223         xfs_da_state_kill_altpath(state);
2224         for (i = 0; i < state->path.active; i++) {
2225                 if (state->path.blk[i].bp)
2226                         xfs_da_buf_done(state->path.blk[i].bp);
2227         }
2228         if (state->extravalid && state->extrablk.bp)
2229                 xfs_da_buf_done(state->extrablk.bp);
2230 #ifdef DEBUG
2231         memset((char *)state, 0, sizeof(*state));
2232 #endif /* DEBUG */
2233         kmem_zone_free(xfs_da_state_zone, state);
2234 }
2235
2236 #ifdef XFS_DABUF_DEBUG
2237 xfs_dabuf_t     *xfs_dabuf_global_list;
2238 lock_t          xfs_dabuf_global_lock;
2239 #endif
2240
2241 /*
2242  * Create a dabuf.
2243  */
2244 /* ARGSUSED */
2245 STATIC xfs_dabuf_t *
2246 xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2247 {
2248         xfs_buf_t       *bp;
2249         xfs_dabuf_t     *dabuf;
2250         int             i;
2251         int             off;
2252
2253         if (nbuf == 1)
2254                 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP);
2255         else
2256                 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP);
2257         dabuf->dirty = 0;
2258 #ifdef XFS_DABUF_DEBUG
2259         dabuf->ra = ra;
2260         dabuf->target = XFS_BUF_TARGET(bps[0]);
2261         dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2262 #endif
2263         if (nbuf == 1) {
2264                 dabuf->nbuf = 1;
2265                 bp = bps[0];
2266                 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2267                 dabuf->data = XFS_BUF_PTR(bp);
2268                 dabuf->bps[0] = bp;
2269         } else {
2270                 dabuf->nbuf = nbuf;
2271                 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2272                         dabuf->bps[i] = bp = bps[i];
2273                         dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2274                 }
2275                 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2276                 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2277                         bp = bps[i];
2278                         memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2279                                 XFS_BUF_COUNT(bp));
2280                 }
2281         }
2282 #ifdef XFS_DABUF_DEBUG
2283         {
2284                 SPLDECL(s);
2285                 xfs_dabuf_t     *p;
2286
2287                 s = mutex_spinlock(&xfs_dabuf_global_lock);
2288                 for (p = xfs_dabuf_global_list; p; p = p->next) {
2289                         ASSERT(p->blkno != dabuf->blkno ||
2290                                p->target != dabuf->target);
2291                 }
2292                 dabuf->prev = NULL;
2293                 if (xfs_dabuf_global_list)
2294                         xfs_dabuf_global_list->prev = dabuf;
2295                 dabuf->next = xfs_dabuf_global_list;
2296                 xfs_dabuf_global_list = dabuf;
2297                 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2298         }
2299 #endif
2300         return dabuf;
2301 }
2302
2303 /*
2304  * Un-dirty a dabuf.
2305  */
2306 STATIC void
2307 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2308 {
2309         xfs_buf_t       *bp;
2310         int             i;
2311         int             off;
2312
2313         if (dabuf->dirty) {
2314                 ASSERT(dabuf->nbuf > 1);
2315                 dabuf->dirty = 0;
2316                 for (i = off = 0; i < dabuf->nbuf;
2317                                 i++, off += XFS_BUF_COUNT(bp)) {
2318                         bp = dabuf->bps[i];
2319                         memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2320                                 XFS_BUF_COUNT(bp));
2321                 }
2322         }
2323 }
2324
2325 /*
2326  * Release a dabuf.
2327  */
2328 void
2329 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2330 {
2331         ASSERT(dabuf);
2332         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2333         if (dabuf->dirty)
2334                 xfs_da_buf_clean(dabuf);
2335         if (dabuf->nbuf > 1)
2336                 kmem_free(dabuf->data, BBTOB(dabuf->bbcount));
2337 #ifdef XFS_DABUF_DEBUG
2338         {
2339                 SPLDECL(s);
2340
2341                 s = mutex_spinlock(&xfs_dabuf_global_lock);
2342                 if (dabuf->prev)
2343                         dabuf->prev->next = dabuf->next;
2344                 else
2345                         xfs_dabuf_global_list = dabuf->next;
2346                 if (dabuf->next)
2347                         dabuf->next->prev = dabuf->prev;
2348                 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2349         }
2350         memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2351 #endif
2352         if (dabuf->nbuf == 1)
2353                 kmem_zone_free(xfs_dabuf_zone, dabuf);
2354         else
2355                 kmem_free(dabuf, XFS_DA_BUF_SIZE(dabuf->nbuf));
2356 }
2357
2358 /*
2359  * Log transaction from a dabuf.
2360  */
2361 void
2362 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2363 {
2364         xfs_buf_t       *bp;
2365         uint            f;
2366         int             i;
2367         uint            l;
2368         int             off;
2369
2370         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2371         if (dabuf->nbuf == 1) {
2372                 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2373                 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2374                 return;
2375         }
2376         dabuf->dirty = 1;
2377         ASSERT(first <= last);
2378         for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2379                 bp = dabuf->bps[i];
2380                 f = off;
2381                 l = f + XFS_BUF_COUNT(bp) - 1;
2382                 if (f < first)
2383                         f = first;
2384                 if (l > last)
2385                         l = last;
2386                 if (f <= l)
2387                         xfs_trans_log_buf(tp, bp, f - off, l - off);
2388                 /*
2389                  * B_DONE is set by xfs_trans_log buf.
2390                  * If we don't set it on a new buffer (get not read)
2391                  * then if we don't put anything in the buffer it won't
2392                  * be set, and at commit it it released into the cache,
2393                  * and then a read will fail.
2394                  */
2395                 else if (!(XFS_BUF_ISDONE(bp)))
2396                   XFS_BUF_DONE(bp);
2397         }
2398         ASSERT(last < off);
2399 }
2400
2401 /*
2402  * Release dabuf from a transaction.
2403  * Have to free up the dabuf before the buffers are released,
2404  * since the synchronization on the dabuf is really the lock on the buffer.
2405  */
2406 void
2407 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2408 {
2409         xfs_buf_t       *bp;
2410         xfs_buf_t       **bplist;
2411         int             i;
2412         int             nbuf;
2413
2414         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2415         if ((nbuf = dabuf->nbuf) == 1) {
2416                 bplist = &bp;
2417                 bp = dabuf->bps[0];
2418         } else {
2419                 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2420                 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2421         }
2422         xfs_da_buf_done(dabuf);
2423         for (i = 0; i < nbuf; i++)
2424                 xfs_trans_brelse(tp, bplist[i]);
2425         if (bplist != &bp)
2426                 kmem_free(bplist, nbuf * sizeof(*bplist));
2427 }
2428
2429 /*
2430  * Invalidate dabuf from a transaction.
2431  */
2432 void
2433 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2434 {
2435         xfs_buf_t       *bp;
2436         xfs_buf_t       **bplist;
2437         int             i;
2438         int             nbuf;
2439
2440         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2441         if ((nbuf = dabuf->nbuf) == 1) {
2442                 bplist = &bp;
2443                 bp = dabuf->bps[0];
2444         } else {
2445                 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2446                 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2447         }
2448         xfs_da_buf_done(dabuf);
2449         for (i = 0; i < nbuf; i++)
2450                 xfs_trans_binval(tp, bplist[i]);
2451         if (bplist != &bp)
2452                 kmem_free(bplist, nbuf * sizeof(*bplist));
2453 }
2454
2455 /*
2456  * Get the first daddr from a dabuf.
2457  */
2458 xfs_daddr_t
2459 xfs_da_blkno(xfs_dabuf_t *dabuf)
2460 {
2461         ASSERT(dabuf->nbuf);
2462         ASSERT(dabuf->data);
2463         return XFS_BUF_ADDR(dabuf->bps[0]);
2464 }