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_mount.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_dir2.h"
  31 #include "xfs_dir2_format.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_inode_item.h"
  36 #include "xfs_alloc.h"
  37 #include "xfs_bmap.h"
  38 #include "xfs_attr.h"
  39 #include "xfs_attr_leaf.h"
  40 #include "xfs_error.h"
  41 #include "xfs_trace.h"
  42
  43 /*
  44  * xfs_da_btree.c
  45  *
  46  * Routines to implement directories as Btrees of hashed names.
  47  */
  48
  49 /*========================================================================
  50  * Function prototypes for the kernel.
  51  *========================================================================*/
  52
  53 /*
  54  * Routines used for growing the Btree.
  55  */
  56 STATIC int xfs_da_root_split(xfs_da_state_t *state,
  57                                             xfs_da_state_blk_t *existing_root,
  58                                             xfs_da_state_blk_t *new_child);
  59 STATIC int xfs_da_node_split(xfs_da_state_t *state,
  60                                             xfs_da_state_blk_t *existing_blk,
  61                                             xfs_da_state_blk_t *split_blk,
  62                                             xfs_da_state_blk_t *blk_to_add,
  63                                             int treelevel,
  64                                             int *result);
  65 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
  66                                          xfs_da_state_blk_t *node_blk_1,
  67                                          xfs_da_state_blk_t *node_blk_2);
  68 STATIC void xfs_da_node_add(xfs_da_state_t *state,
  69                                    xfs_da_state_blk_t *old_node_blk,
  70                                    xfs_da_state_blk_t *new_node_blk);
  71
  72 /*
  73  * Routines used for shrinking the Btree.
  74  */
  75 STATIC int xfs_da_root_join(xfs_da_state_t *state,
  76                                            xfs_da_state_blk_t *root_blk);
  77 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
  78 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
  79                                               xfs_da_state_blk_t *drop_blk);
  80 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
  81                                          xfs_da_state_blk_t *src_node_blk,
  82                                          xfs_da_state_blk_t *dst_node_blk);
  83
  84 /*
  85  * Utility routines.
  86  */
  87 STATIC uint     xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
  88 STATIC int      xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
  89 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps);
  90 STATIC int      xfs_da_blk_unlink(xfs_da_state_t *state,
  91                                   xfs_da_state_blk_t *drop_blk,
  92                                   xfs_da_state_blk_t *save_blk);
  93 STATIC void     xfs_da_state_kill_altpath(xfs_da_state_t *state);
  94
  95 /*========================================================================
  96  * Routines used for growing the Btree.
  97  *========================================================================*/
  98
  99 /*
 100  * Create the initial contents of an intermediate node.
 101  */
 102 int
 103 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
 104                                  xfs_dabuf_t **bpp, int whichfork)
 105 {
 106         xfs_da_intnode_t *node;
 107         xfs_dabuf_t *bp;
 108         int error;
 109         xfs_trans_t *tp;
 110
 111         tp = args->trans;
 112         error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
 113         if (error)
 114                 return(error);
 115         ASSERT(bp != NULL);
 116         node = bp->data;
 117         node->hdr.info.forw = 0;
 118         node->hdr.info.back = 0;
 119         node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
 120         node->hdr.info.pad = 0;
 121         node->hdr.count = 0;
 122         node->hdr.level = cpu_to_be16(level);
 123
 124         xfs_da_log_buf(tp, bp,
 125                 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 126
 127         *bpp = bp;
 128         return(0);
 129 }
 130
 131 /*
 132  * Split a leaf node, rebalance, then possibly split
 133  * intermediate nodes, rebalance, etc.
 134  */
 135 int                                                     /* error */
 136 xfs_da_split(xfs_da_state_t *state)
 137 {
 138         xfs_da_state_blk_t *oldblk, *newblk, *addblk;
 139         xfs_da_intnode_t *node;
 140         xfs_dabuf_t *bp;
 141         int max, action, error, i;
 142
 143         /*
 144          * Walk back up the tree splitting/inserting/adjusting as necessary.
 145          * If we need to insert and there isn't room, split the node, then
 146          * decide which fragment to insert the new block from below into.
 147          * Note that we may split the root this way, but we need more fixup.
 148          */
 149         max = state->path.active - 1;
 150         ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
 151         ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
 152                state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
 153
 154         addblk = &state->path.blk[max];         /* initial dummy value */
 155         for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
 156                 oldblk = &state->path.blk[i];
 157                 newblk = &state->altpath.blk[i];
 158
 159                 /*
 160                  * If a leaf node then
 161                  *     Allocate a new leaf node, then rebalance across them.
 162                  * else if an intermediate node then
 163                  *     We split on the last layer, must we split the node?
 164                  */
 165                 switch (oldblk->magic) {
 166                 case XFS_ATTR_LEAF_MAGIC:
 167                         error = xfs_attr_leaf_split(state, oldblk, newblk);
 168                         if ((error != 0) && (error != ENOSPC)) {
 169                                 return(error);  /* GROT: attr is inconsistent */
 170                         }
 171                         if (!error) {
 172                                 addblk = newblk;
 173                                 break;
 174                         }
 175                         /*
 176                          * Entry wouldn't fit, split the leaf again.
 177                          */
 178                         state->extravalid = 1;
 179                         if (state->inleaf) {
 180                                 state->extraafter = 0;  /* before newblk */
 181                                 error = xfs_attr_leaf_split(state, oldblk,
 182                                                             &state->extrablk);
 183                         } else {
 184                                 state->extraafter = 1;  /* after newblk */
 185                                 error = xfs_attr_leaf_split(state, newblk,
 186                                                             &state->extrablk);
 187                         }
 188                         if (error)
 189                                 return(error);  /* GROT: attr inconsistent */
 190                         addblk = newblk;
 191                         break;
 192                 case XFS_DIR2_LEAFN_MAGIC:
 193                         error = xfs_dir2_leafn_split(state, oldblk, newblk);
 194                         if (error)
 195                                 return error;
 196                         addblk = newblk;
 197                         break;
 198                 case XFS_DA_NODE_MAGIC:
 199                         error = xfs_da_node_split(state, oldblk, newblk, addblk,
 200                                                          max - i, &action);
 201                         xfs_da_buf_done(addblk->bp);
 202                         addblk->bp = NULL;
 203                         if (error)
 204                                 return(error);  /* GROT: dir is inconsistent */
 205                         /*
 206                          * Record the newly split block for the next time thru?
 207                          */
 208                         if (action)
 209                                 addblk = newblk;
 210                         else
 211                                 addblk = NULL;
 212                         break;
 213                 }
 214
 215                 /*
 216                  * Update the btree to show the new hashval for this child.
 217                  */
 218                 xfs_da_fixhashpath(state, &state->path);
 219                 /*
 220                  * If we won't need this block again, it's getting dropped
 221                  * from the active path by the loop control, so we need
 222                  * to mark it done now.
 223                  */
 224                 if (i > 0 || !addblk)
 225                         xfs_da_buf_done(oldblk->bp);
 226         }
 227         if (!addblk)
 228                 return(0);
 229
 230         /*
 231          * Split the root node.
 232          */
 233         ASSERT(state->path.active == 0);
 234         oldblk = &state->path.blk[0];
 235         error = xfs_da_root_split(state, oldblk, addblk);
 236         if (error) {
 237                 xfs_da_buf_done(oldblk->bp);
 238                 xfs_da_buf_done(addblk->bp);
 239                 addblk->bp = NULL;
 240                 return(error);  /* GROT: dir is inconsistent */
 241         }
 242
 243         /*
 244          * Update pointers to the node which used to be block 0 and
 245          * just got bumped because of the addition of a new root node.
 246          * There might be three blocks involved if a double split occurred,
 247          * and the original block 0 could be at any position in the list.
 248          */
 249
 250         node = oldblk->bp->data;
 251         if (node->hdr.info.forw) {
 252                 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
 253                         bp = addblk->bp;
 254                 } else {
 255                         ASSERT(state->extravalid);
 256                         bp = state->extrablk.bp;
 257                 }
 258                 node = bp->data;
 259                 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
 260                 xfs_da_log_buf(state->args->trans, bp,
 261                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 262                     sizeof(node->hdr.info)));
 263         }
 264         node = oldblk->bp->data;
 265         if (node->hdr.info.back) {
 266                 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
 267                         bp = addblk->bp;
 268                 } else {
 269                         ASSERT(state->extravalid);
 270                         bp = state->extrablk.bp;
 271                 }
 272                 node = bp->data;
 273                 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
 274                 xfs_da_log_buf(state->args->trans, bp,
 275                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 276                     sizeof(node->hdr.info)));
 277         }
 278         xfs_da_buf_done(oldblk->bp);
 279         xfs_da_buf_done(addblk->bp);
 280         addblk->bp = NULL;
 281         return(0);
 282 }
 283
 284 /*
 285  * Split the root.  We have to create a new root and point to the two
 286  * parts (the split old root) that we just created.  Copy block zero to
 287  * the EOF, extending the inode in process.
 288  */
 289 STATIC int                                              /* error */
 290 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
 291                                  xfs_da_state_blk_t *blk2)
 292 {
 293         xfs_da_intnode_t *node, *oldroot;
 294         xfs_da_args_t *args;
 295         xfs_dablk_t blkno;
 296         xfs_dabuf_t *bp;
 297         int error, size;
 298         xfs_inode_t *dp;
 299         xfs_trans_t *tp;
 300         xfs_mount_t *mp;
 301         xfs_dir2_leaf_t *leaf;
 302
 303         /*
 304          * Copy the existing (incorrect) block from the root node position
 305          * to a free space somewhere.
 306          */
 307         args = state->args;
 308         ASSERT(args != NULL);
 309         error = xfs_da_grow_inode(args, &blkno);
 310         if (error)
 311                 return(error);
 312         dp = args->dp;
 313         tp = args->trans;
 314         mp = state->mp;
 315         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 316         if (error)
 317                 return(error);
 318         ASSERT(bp != NULL);
 319         node = bp->data;
 320         oldroot = blk1->bp->data;
 321         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
 322                 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
 323                              (char *)oldroot);
 324         } else {
 325                 ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 326                 leaf = (xfs_dir2_leaf_t *)oldroot;
 327                 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
 328                              (char *)leaf);
 329         }
 330         memcpy(node, oldroot, size);
 331         xfs_da_log_buf(tp, bp, 0, size - 1);
 332         xfs_da_buf_done(blk1->bp);
 333         blk1->bp = bp;
 334         blk1->blkno = blkno;
 335
 336         /*
 337          * Set up the new root node.
 338          */
 339         error = xfs_da_node_create(args,
 340                 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
 341                 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
 342         if (error)
 343                 return(error);
 344         node = bp->data;
 345         node->btree[0].hashval = cpu_to_be32(blk1->hashval);
 346         node->btree[0].before = cpu_to_be32(blk1->blkno);
 347         node->btree[1].hashval = cpu_to_be32(blk2->hashval);
 348         node->btree[1].before = cpu_to_be32(blk2->blkno);
 349         node->hdr.count = cpu_to_be16(2);
 350
 351 #ifdef DEBUG
 352         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
 353                 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
 354                        blk1->blkno < mp->m_dirfreeblk);
 355                 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
 356                        blk2->blkno < mp->m_dirfreeblk);
 357         }
 358 #endif
 359
 360         /* Header is already logged by xfs_da_node_create */
 361         xfs_da_log_buf(tp, bp,
 362                 XFS_DA_LOGRANGE(node, node->btree,
 363                         sizeof(xfs_da_node_entry_t) * 2));
 364         xfs_da_buf_done(bp);
 365
 366         return(0);
 367 }
 368
 369 /*
 370  * Split the node, rebalance, then add the new entry.
 371  */
 372 STATIC int                                              /* error */
 373 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 374                                  xfs_da_state_blk_t *newblk,
 375                                  xfs_da_state_blk_t *addblk,
 376                                  int treelevel, int *result)
 377 {
 378         xfs_da_intnode_t *node;
 379         xfs_dablk_t blkno;
 380         int newcount, error;
 381         int useextra;
 382
 383         node = oldblk->bp->data;
 384         ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 385
 386         /*
 387          * With V2 dirs the extra block is data or freespace.
 388          */
 389         useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
 390         newcount = 1 + useextra;
 391         /*
 392          * Do we have to split the node?
 393          */
 394         if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
 395                 /*
 396                  * Allocate a new node, add to the doubly linked chain of
 397                  * nodes, then move some of our excess entries into it.
 398                  */
 399                 error = xfs_da_grow_inode(state->args, &blkno);
 400                 if (error)
 401                         return(error);  /* GROT: dir is inconsistent */
 402
 403                 error = xfs_da_node_create(state->args, blkno, treelevel,
 404                                            &newblk->bp, state->args->whichfork);
 405                 if (error)
 406                         return(error);  /* GROT: dir is inconsistent */
 407                 newblk->blkno = blkno;
 408                 newblk->magic = XFS_DA_NODE_MAGIC;
 409                 xfs_da_node_rebalance(state, oldblk, newblk);
 410                 error = xfs_da_blk_link(state, oldblk, newblk);
 411                 if (error)
 412                         return(error);
 413                 *result = 1;
 414         } else {
 415                 *result = 0;
 416         }
 417
 418         /*
 419          * Insert the new entry(s) into the correct block
 420          * (updating last hashval in the process).
 421          *
 422          * xfs_da_node_add() inserts BEFORE the given index,
 423          * and as a result of using node_lookup_int() we always
 424          * point to a valid entry (not after one), but a split
 425          * operation always results in a new block whose hashvals
 426          * FOLLOW the current block.
 427          *
 428          * If we had double-split op below us, then add the extra block too.
 429          */
 430         node = oldblk->bp->data;
 431         if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
 432                 oldblk->index++;
 433                 xfs_da_node_add(state, oldblk, addblk);
 434                 if (useextra) {
 435                         if (state->extraafter)
 436                                 oldblk->index++;
 437                         xfs_da_node_add(state, oldblk, &state->extrablk);
 438                         state->extravalid = 0;
 439                 }
 440         } else {
 441                 newblk->index++;
 442                 xfs_da_node_add(state, newblk, addblk);
 443                 if (useextra) {
 444                         if (state->extraafter)
 445                                 newblk->index++;
 446                         xfs_da_node_add(state, newblk, &state->extrablk);
 447                         state->extravalid = 0;
 448                 }
 449         }
 450
 451         return(0);
 452 }
 453
 454 /*
 455  * Balance the btree elements between two intermediate nodes,
 456  * usually one full and one empty.
 457  *
 458  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 459  */
 460 STATIC void
 461 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
 462                                      xfs_da_state_blk_t *blk2)
 463 {
 464         xfs_da_intnode_t *node1, *node2, *tmpnode;
 465         xfs_da_node_entry_t *btree_s, *btree_d;
 466         int count, tmp;
 467         xfs_trans_t *tp;
 468
 469         node1 = blk1->bp->data;
 470         node2 = blk2->bp->data;
 471         /*
 472          * Figure out how many entries need to move, and in which direction.
 473          * Swap the nodes around if that makes it simpler.
 474          */
 475         if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
 476             ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
 477              (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
 478               be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
 479                 tmpnode = node1;
 480                 node1 = node2;
 481                 node2 = tmpnode;
 482         }
 483         ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 484         ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 485         count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
 486         if (count == 0)
 487                 return;
 488         tp = state->args->trans;
 489         /*
 490          * Two cases: high-to-low and low-to-high.
 491          */
 492         if (count > 0) {
 493                 /*
 494                  * Move elements in node2 up to make a hole.
 495                  */
 496                 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
 497                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 498                         btree_s = &node2->btree[0];
 499                         btree_d = &node2->btree[count];
 500                         memmove(btree_d, btree_s, tmp);
 501                 }
 502
 503                 /*
 504                  * Move the req'd B-tree elements from high in node1 to
 505                  * low in node2.
 506                  */
 507                 be16_add_cpu(&node2->hdr.count, count);
 508                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 509                 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
 510                 btree_d = &node2->btree[0];
 511                 memcpy(btree_d, btree_s, tmp);
 512                 be16_add_cpu(&node1->hdr.count, -count);
 513         } else {
 514                 /*
 515                  * Move the req'd B-tree elements from low in node2 to
 516                  * high in node1.
 517                  */
 518                 count = -count;
 519                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 520                 btree_s = &node2->btree[0];
 521                 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
 522                 memcpy(btree_d, btree_s, tmp);
 523                 be16_add_cpu(&node1->hdr.count, count);
 524                 xfs_da_log_buf(tp, blk1->bp,
 525                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 526
 527                 /*
 528                  * Move elements in node2 down to fill the hole.
 529                  */
 530                 tmp  = be16_to_cpu(node2->hdr.count) - count;
 531                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 532                 btree_s = &node2->btree[count];
 533                 btree_d = &node2->btree[0];
 534                 memmove(btree_d, btree_s, tmp);
 535                 be16_add_cpu(&node2->hdr.count, -count);
 536         }
 537
 538         /*
 539          * Log header of node 1 and all current bits of node 2.
 540          */
 541         xfs_da_log_buf(tp, blk1->bp,
 542                 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
 543         xfs_da_log_buf(tp, blk2->bp,
 544                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 545                         sizeof(node2->hdr) +
 546                         sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
 547
 548         /*
 549          * Record the last hashval from each block for upward propagation.
 550          * (note: don't use the swapped node pointers)
 551          */
 552         node1 = blk1->bp->data;
 553         node2 = blk2->bp->data;
 554         blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
 555         blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
 556
 557         /*
 558          * Adjust the expected index for insertion.
 559          */
 560         if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
 561                 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
 562                 blk1->index = be16_to_cpu(node1->hdr.count) + 1;        /* make it invalid */
 563         }
 564 }
 565
 566 /*
 567  * Add a new entry to an intermediate node.
 568  */
 569 STATIC void
 570 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 571                                xfs_da_state_blk_t *newblk)
 572 {
 573         xfs_da_intnode_t *node;
 574         xfs_da_node_entry_t *btree;
 575         int tmp;
 576
 577         node = oldblk->bp->data;
 578         ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 579         ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
 580         ASSERT(newblk->blkno != 0);
 581         if (state->args->whichfork == XFS_DATA_FORK)
 582                 ASSERT(newblk->blkno >= state->mp->m_dirleafblk &&
 583                        newblk->blkno < state->mp->m_dirfreeblk);
 584
 585         /*
 586          * We may need to make some room before we insert the new node.
 587          */
 588         tmp = 0;
 589         btree = &node->btree[ oldblk->index ];
 590         if (oldblk->index < be16_to_cpu(node->hdr.count)) {
 591                 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
 592                 memmove(btree + 1, btree, tmp);
 593         }
 594         btree->hashval = cpu_to_be32(newblk->hashval);
 595         btree->before = cpu_to_be32(newblk->blkno);
 596         xfs_da_log_buf(state->args->trans, oldblk->bp,
 597                 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
 598         be16_add_cpu(&node->hdr.count, 1);
 599         xfs_da_log_buf(state->args->trans, oldblk->bp,
 600                 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 601
 602         /*
 603          * Copy the last hash value from the oldblk to propagate upwards.
 604          */
 605         oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
 606 }
 607
 608 /*========================================================================
 609  * Routines used for shrinking the Btree.
 610  *========================================================================*/
 611
 612 /*
 613  * Deallocate an empty leaf node, remove it from its parent,
 614  * possibly deallocating that block, etc...
 615  */
 616 int
 617 xfs_da_join(xfs_da_state_t *state)
 618 {
 619         xfs_da_state_blk_t *drop_blk, *save_blk;
 620         int action, error;
 621
 622         action = 0;
 623         drop_blk = &state->path.blk[ state->path.active-1 ];
 624         save_blk = &state->altpath.blk[ state->path.active-1 ];
 625         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 626         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 627                drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
 628
 629         /*
 630          * Walk back up the tree joining/deallocating as necessary.
 631          * When we stop dropping blocks, break out.
 632          */
 633         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 634                  state->path.active--) {
 635                 /*
 636                  * See if we can combine the block with a neighbor.
 637                  *   (action == 0) => no options, just leave
 638                  *   (action == 1) => coalesce, then unlink
 639                  *   (action == 2) => block empty, unlink it
 640                  */
 641                 switch (drop_blk->magic) {
 642                 case XFS_ATTR_LEAF_MAGIC:
 643                         error = xfs_attr_leaf_toosmall(state, &action);
 644                         if (error)
 645                                 return(error);
 646                         if (action == 0)
 647                                 return(0);
 648                         xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
 649                         break;
 650                 case XFS_DIR2_LEAFN_MAGIC:
 651                         error = xfs_dir2_leafn_toosmall(state, &action);
 652                         if (error)
 653                                 return error;
 654                         if (action == 0)
 655                                 return 0;
 656                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 657                         break;
 658                 case XFS_DA_NODE_MAGIC:
 659                         /*
 660                          * Remove the offending node, fixup hashvals,
 661                          * check for a toosmall neighbor.
 662                          */
 663                         xfs_da_node_remove(state, drop_blk);
 664                         xfs_da_fixhashpath(state, &state->path);
 665                         error = xfs_da_node_toosmall(state, &action);
 666                         if (error)
 667                                 return(error);
 668                         if (action == 0)
 669                                 return 0;
 670                         xfs_da_node_unbalance(state, drop_blk, save_blk);
 671                         break;
 672                 }
 673                 xfs_da_fixhashpath(state, &state->altpath);
 674                 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
 675                 xfs_da_state_kill_altpath(state);
 676                 if (error)
 677                         return(error);
 678                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
 679                                                          drop_blk->bp);
 680                 drop_blk->bp = NULL;
 681                 if (error)
 682                         return(error);
 683         }
 684         /*
 685          * We joined all the way to the top.  If it turns out that
 686          * we only have one entry in the root, make the child block
 687          * the new root.
 688          */
 689         xfs_da_node_remove(state, drop_blk);
 690         xfs_da_fixhashpath(state, &state->path);
 691         error = xfs_da_root_join(state, &state->path.blk[0]);
 692         return(error);
 693 }
 694
 695 #ifdef  DEBUG
 696 static void
 697 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
 698 {
 699         __be16  magic = blkinfo->magic;
 700
 701         if (level == 1) {
 702                 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
 703                        magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
 704         } else
 705                 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 706         ASSERT(!blkinfo->forw);
 707         ASSERT(!blkinfo->back);
 708 }
 709 #else   /* !DEBUG */
 710 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
 711 #endif  /* !DEBUG */
 712
 713 /*
 714  * We have only one entry in the root.  Copy the only remaining child of
 715  * the old root to block 0 as the new root node.
 716  */
 717 STATIC int
 718 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
 719 {
 720         xfs_da_intnode_t *oldroot;
 721         xfs_da_args_t *args;
 722         xfs_dablk_t child;
 723         xfs_dabuf_t *bp;
 724         int error;
 725
 726         args = state->args;
 727         ASSERT(args != NULL);
 728         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
 729         oldroot = root_blk->bp->data;
 730         ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 731         ASSERT(!oldroot->hdr.info.forw);
 732         ASSERT(!oldroot->hdr.info.back);
 733
 734         /*
 735          * If the root has more than one child, then don't do anything.
 736          */
 737         if (be16_to_cpu(oldroot->hdr.count) > 1)
 738                 return(0);
 739
 740         /*
 741          * Read in the (only) child block, then copy those bytes into
 742          * the root block's buffer and free the original child block.
 743          */
 744         child = be32_to_cpu(oldroot->btree[0].before);
 745         ASSERT(child != 0);
 746         error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
 747                                              args->whichfork);
 748         if (error)
 749                 return(error);
 750         ASSERT(bp != NULL);
 751         xfs_da_blkinfo_onlychild_validate(bp->data,
 752                                         be16_to_cpu(oldroot->hdr.level));
 753
 754         memcpy(root_blk->bp->data, bp->data, state->blocksize);
 755         xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
 756         error = xfs_da_shrink_inode(args, child, bp);
 757         return(error);
 758 }
 759
 760 /*
 761  * Check a node block and its neighbors to see if the block should be
 762  * collapsed into one or the other neighbor.  Always keep the block
 763  * with the smaller block number.
 764  * If the current block is over 50% full, don't try to join it, return 0.
 765  * If the block is empty, fill in the state structure and return 2.
 766  * If it can be collapsed, fill in the state structure and return 1.
 767  * If nothing can be done, return 0.
 768  */
 769 STATIC int
 770 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
 771 {
 772         xfs_da_intnode_t *node;
 773         xfs_da_state_blk_t *blk;
 774         xfs_da_blkinfo_t *info;
 775         int count, forward, error, retval, i;
 776         xfs_dablk_t blkno;
 777         xfs_dabuf_t *bp;
 778
 779         /*
 780          * Check for the degenerate case of the block being over 50% full.
 781          * If so, it's not worth even looking to see if we might be able
 782          * to coalesce with a sibling.
 783          */
 784         blk = &state->path.blk[ state->path.active-1 ];
 785         info = blk->bp->data;
 786         ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 787         node = (xfs_da_intnode_t *)info;
 788         count = be16_to_cpu(node->hdr.count);
 789         if (count > (state->node_ents >> 1)) {
 790                 *action = 0;    /* blk over 50%, don't try to join */
 791                 return(0);      /* blk over 50%, don't try to join */
 792         }
 793
 794         /*
 795          * Check for the degenerate case of the block being empty.
 796          * If the block is empty, we'll simply delete it, no need to
 797          * coalesce it with a sibling block.  We choose (arbitrarily)
 798          * to merge with the forward block unless it is NULL.
 799          */
 800         if (count == 0) {
 801                 /*
 802                  * Make altpath point to the block we want to keep and
 803                  * path point to the block we want to drop (this one).
 804                  */
 805                 forward = (info->forw != 0);
 806                 memcpy(&state->altpath, &state->path, sizeof(state->path));
 807                 error = xfs_da_path_shift(state, &state->altpath, forward,
 808                                                  0, &retval);
 809                 if (error)
 810                         return(error);
 811                 if (retval) {
 812                         *action = 0;
 813                 } else {
 814                         *action = 2;
 815                 }
 816                 return(0);
 817         }
 818
 819         /*
 820          * Examine each sibling block to see if we can coalesce with
 821          * at least 25% free space to spare.  We need to figure out
 822          * whether to merge with the forward or the backward block.
 823          * We prefer coalescing with the lower numbered sibling so as
 824          * to shrink a directory over time.
 825          */
 826         /* start with smaller blk num */
 827         forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
 828         for (i = 0; i < 2; forward = !forward, i++) {
 829                 if (forward)
 830                         blkno = be32_to_cpu(info->forw);
 831                 else
 832                         blkno = be32_to_cpu(info->back);
 833                 if (blkno == 0)
 834                         continue;
 835                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
 836                                         blkno, -1, &bp, state->args->whichfork);
 837                 if (error)
 838                         return(error);
 839                 ASSERT(bp != NULL);
 840
 841                 node = (xfs_da_intnode_t *)info;
 842                 count  = state->node_ents;
 843                 count -= state->node_ents >> 2;
 844                 count -= be16_to_cpu(node->hdr.count);
 845                 node = bp->data;
 846                 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 847                 count -= be16_to_cpu(node->hdr.count);
 848                 xfs_da_brelse(state->args->trans, bp);
 849                 if (count >= 0)
 850                         break;  /* fits with at least 25% to spare */
 851         }
 852         if (i >= 2) {
 853                 *action = 0;
 854                 return(0);
 855         }
 856
 857         /*
 858          * Make altpath point to the block we want to keep (the lower
 859          * numbered block) and path point to the block we want to drop.
 860          */
 861         memcpy(&state->altpath, &state->path, sizeof(state->path));
 862         if (blkno < blk->blkno) {
 863                 error = xfs_da_path_shift(state, &state->altpath, forward,
 864                                                  0, &retval);
 865                 if (error) {
 866                         return(error);
 867                 }
 868                 if (retval) {
 869                         *action = 0;
 870                         return(0);
 871                 }
 872         } else {
 873                 error = xfs_da_path_shift(state, &state->path, forward,
 874                                                  0, &retval);
 875                 if (error) {
 876                         return(error);
 877                 }
 878                 if (retval) {
 879                         *action = 0;
 880                         return(0);
 881                 }
 882         }
 883         *action = 1;
 884         return(0);
 885 }
 886
 887 /*
 888  * Walk back up the tree adjusting hash values as necessary,
 889  * when we stop making changes, return.
 890  */
 891 void
 892 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
 893 {
 894         xfs_da_state_blk_t *blk;
 895         xfs_da_intnode_t *node;
 896         xfs_da_node_entry_t *btree;
 897         xfs_dahash_t lasthash=0;
 898         int level, count;
 899
 900         level = path->active-1;
 901         blk = &path->blk[ level ];
 902         switch (blk->magic) {
 903         case XFS_ATTR_LEAF_MAGIC:
 904                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
 905                 if (count == 0)
 906                         return;
 907                 break;
 908         case XFS_DIR2_LEAFN_MAGIC:
 909                 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
 910                 if (count == 0)
 911                         return;
 912                 break;
 913         case XFS_DA_NODE_MAGIC:
 914                 lasthash = xfs_da_node_lasthash(blk->bp, &count);
 915                 if (count == 0)
 916                         return;
 917                 break;
 918         }
 919         for (blk--, level--; level >= 0; blk--, level--) {
 920                 node = blk->bp->data;
 921                 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 922                 btree = &node->btree[ blk->index ];
 923                 if (be32_to_cpu(btree->hashval) == lasthash)
 924                         break;
 925                 blk->hashval = lasthash;
 926                 btree->hashval = cpu_to_be32(lasthash);
 927                 xfs_da_log_buf(state->args->trans, blk->bp,
 928                                   XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
 929
 930                 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
 931         }
 932 }
 933
 934 /*
 935  * Remove an entry from an intermediate node.
 936  */
 937 STATIC void
 938 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
 939 {
 940         xfs_da_intnode_t *node;
 941         xfs_da_node_entry_t *btree;
 942         int tmp;
 943
 944         node = drop_blk->bp->data;
 945         ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
 946         ASSERT(drop_blk->index >= 0);
 947
 948         /*
 949          * Copy over the offending entry, or just zero it out.
 950          */
 951         btree = &node->btree[drop_blk->index];
 952         if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
 953                 tmp  = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
 954                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 955                 memmove(btree, btree + 1, tmp);
 956                 xfs_da_log_buf(state->args->trans, drop_blk->bp,
 957                     XFS_DA_LOGRANGE(node, btree, tmp));
 958                 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
 959         }
 960         memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
 961         xfs_da_log_buf(state->args->trans, drop_blk->bp,
 962             XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
 963         be16_add_cpu(&node->hdr.count, -1);
 964         xfs_da_log_buf(state->args->trans, drop_blk->bp,
 965             XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 966
 967         /*
 968          * Copy the last hash value from the block to propagate upwards.
 969          */
 970         btree--;
 971         drop_blk->hashval = be32_to_cpu(btree->hashval);
 972 }
 973
 974 /*
 975  * Unbalance the btree elements between two intermediate nodes,
 976  * move all Btree elements from one node into another.
 977  */
 978 STATIC void
 979 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
 980                                      xfs_da_state_blk_t *save_blk)
 981 {
 982         xfs_da_intnode_t *drop_node, *save_node;
 983         xfs_da_node_entry_t *btree;
 984         int tmp;
 985         xfs_trans_t *tp;
 986
 987         drop_node = drop_blk->bp->data;
 988         save_node = save_blk->bp->data;
 989         ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 990         ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
 991         tp = state->args->trans;
 992
 993         /*
 994          * If the dying block has lower hashvals, then move all the
 995          * elements in the remaining block up to make a hole.
 996          */
 997         if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
 998             (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
 999              be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
1000         {
1001                 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1002                 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1003                 memmove(btree, &save_node->btree[0], tmp);
1004                 btree = &save_node->btree[0];
1005                 xfs_da_log_buf(tp, save_blk->bp,
1006                         XFS_DA_LOGRANGE(save_node, btree,
1007                                 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1008                                 sizeof(xfs_da_node_entry_t)));
1009         } else {
1010                 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1011                 xfs_da_log_buf(tp, save_blk->bp,
1012                         XFS_DA_LOGRANGE(save_node, btree,
1013                                 be16_to_cpu(drop_node->hdr.count) *
1014                                 sizeof(xfs_da_node_entry_t)));
1015         }
1016
1017         /*
1018          * Move all the B-tree elements from drop_blk to save_blk.
1019          */
1020         tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1021         memcpy(btree, &drop_node->btree[0], tmp);
1022         be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1023
1024         xfs_da_log_buf(tp, save_blk->bp,
1025                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1026                         sizeof(save_node->hdr)));
1027
1028         /*
1029          * Save the last hashval in the remaining block for upward propagation.
1030          */
1031         save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1032 }
1033
1034 /*========================================================================
1035  * Routines used for finding things in the Btree.
1036  *========================================================================*/
1037
1038 /*
1039  * Walk down the Btree looking for a particular filename, filling
1040  * in the state structure as we go.
1041  *
1042  * We will set the state structure to point to each of the elements
1043  * in each of the nodes where either the hashval is or should be.
1044  *
1045  * We support duplicate hashval's so for each entry in the current
1046  * node that could contain the desired hashval, descend.  This is a
1047  * pruned depth-first tree search.
1048  */
1049 int                                                     /* error */
1050 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1051 {
1052         xfs_da_state_blk_t *blk;
1053         xfs_da_blkinfo_t *curr;
1054         xfs_da_intnode_t *node;
1055         xfs_da_node_entry_t *btree;
1056         xfs_dablk_t blkno;
1057         int probe, span, max, error, retval;
1058         xfs_dahash_t hashval, btreehashval;
1059         xfs_da_args_t *args;
1060
1061         args = state->args;
1062
1063         /*
1064          * Descend thru the B-tree searching each level for the right
1065          * node to use, until the right hashval is found.
1066          */
1067         blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1068         for (blk = &state->path.blk[0], state->path.active = 1;
1069                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1070                          blk++, state->path.active++) {
1071                 /*
1072                  * Read the next node down in the tree.
1073                  */
1074                 blk->blkno = blkno;
1075                 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1076                                         -1, &blk->bp, args->whichfork);
1077                 if (error) {
1078                         blk->blkno = 0;
1079                         state->path.active--;
1080                         return(error);
1081                 }
1082                 curr = blk->bp->data;
1083                 blk->magic = be16_to_cpu(curr->magic);
1084                 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1085                        blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1086                        blk->magic == XFS_ATTR_LEAF_MAGIC);
1087
1088                 /*
1089                  * Search an intermediate node for a match.
1090                  */
1091                 if (blk->magic == XFS_DA_NODE_MAGIC) {
1092                         node = blk->bp->data;
1093                         max = be16_to_cpu(node->hdr.count);
1094                         blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
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(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) &&
1289                node2->hdr.info.magic == cpu_to_be16(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(node->hdr.info.magic == cpu_to_be16(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(node->hdr.info.magic == cpu_to_be16(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->op_flags & XFS_DA_OP_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(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1462                        info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1463                        info->magic == cpu_to_be16(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 __uint8_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 enum xfs_dacmp
1535 xfs_da_compname(
1536         struct xfs_da_args *args,
1537         const unsigned char *name,
1538         int             len)
1539 {
1540         return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1541                                         XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1542 }
1543
1544 static xfs_dahash_t
1545 xfs_default_hashname(
1546         struct xfs_name *name)
1547 {
1548         return xfs_da_hashname(name->name, name->len);
1549 }
1550
1551 const struct xfs_nameops xfs_default_nameops = {
1552         .hashname       = xfs_default_hashname,
1553         .compname       = xfs_da_compname
1554 };
1555
1556 int
1557 xfs_da_grow_inode_int(
1558         struct xfs_da_args      *args,
1559         xfs_fileoff_t           *bno,
1560         int                     count)
1561 {
1562         struct xfs_trans        *tp = args->trans;
1563         struct xfs_inode        *dp = args->dp;
1564         int                     w = args->whichfork;
1565         xfs_drfsbno_t           nblks = dp->i_d.di_nblocks;
1566         struct xfs_bmbt_irec    map, *mapp;
1567         int                     nmap, error, got, i, mapi;
1568
1569         /*
1570          * Find a spot in the file space to put the new block.
1571          */
1572         error = xfs_bmap_first_unused(tp, dp, count, bno, w);
1573         if (error)
1574                 return error;
1575
1576         /*
1577          * Try mapping it in one filesystem block.
1578          */
1579         nmap = 1;
1580         ASSERT(args->firstblock != NULL);
1581         error = xfs_bmapi(tp, dp, *bno, count,
1582                         xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1583                         XFS_BMAPI_CONTIG,
1584                         args->firstblock, args->total, &map, &nmap,
1585                         args->flist);
1586         if (error)
1587                 return error;
1588
1589         ASSERT(nmap <= 1);
1590         if (nmap == 1) {
1591                 mapp = &map;
1592                 mapi = 1;
1593         } else if (nmap == 0 && count > 1) {
1594                 xfs_fileoff_t           b;
1595                 int                     c;
1596
1597                 /*
1598                  * If we didn't get it and the block might work if fragmented,
1599                  * try without the CONTIG flag.  Loop until we get it all.
1600                  */
1601                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1602                 for (b = *bno, mapi = 0; b < *bno + count; ) {
1603                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1604                         c = (int)(*bno + count - b);
1605                         error = xfs_bmapi(tp, dp, b, c,
1606                                         xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|
1607                                         XFS_BMAPI_METADATA,
1608                                         args->firstblock, args->total,
1609                                         &mapp[mapi], &nmap, args->flist);
1610                         if (error)
1611                                 goto out_free_map;
1612                         if (nmap < 1)
1613                                 break;
1614                         mapi += nmap;
1615                         b = mapp[mapi - 1].br_startoff +
1616                             mapp[mapi - 1].br_blockcount;
1617                 }
1618         } else {
1619                 mapi = 0;
1620                 mapp = NULL;
1621         }
1622
1623         /*
1624          * Count the blocks we got, make sure it matches the total.
1625          */
1626         for (i = 0, got = 0; i < mapi; i++)
1627                 got += mapp[i].br_blockcount;
1628         if (got != count || mapp[0].br_startoff != *bno ||
1629             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1630             *bno + count) {
1631                 error = XFS_ERROR(ENOSPC);
1632                 goto out_free_map;
1633         }
1634
1635         /* account for newly allocated blocks in reserved blocks total */
1636         args->total -= dp->i_d.di_nblocks - nblks;
1637
1638 out_free_map:
1639         if (mapp != &map)
1640                 kmem_free(mapp);
1641         return error;
1642 }
1643
1644 /*
1645  * Add a block to the btree ahead of the file.
1646  * Return the new block number to the caller.
1647  */
1648 int
1649 xfs_da_grow_inode(
1650         struct xfs_da_args      *args,
1651         xfs_dablk_t             *new_blkno)
1652 {
1653         xfs_fileoff_t           bno;
1654         int                     count;
1655         int                     error;
1656
1657         if (args->whichfork == XFS_DATA_FORK) {
1658                 bno = args->dp->i_mount->m_dirleafblk;
1659                 count = args->dp->i_mount->m_dirblkfsbs;
1660         } else {
1661                 bno = 0;
1662                 count = 1;
1663         }
1664
1665         error = xfs_da_grow_inode_int(args, &bno, count);
1666         if (!error)
1667                 *new_blkno = (xfs_dablk_t)bno;
1668         return error;
1669 }
1670
1671 /*
1672  * Ick.  We need to always be able to remove a btree block, even
1673  * if there's no space reservation because the filesystem is full.
1674  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1675  * It swaps the target block with the last block in the file.  The
1676  * last block in the file can always be removed since it can't cause
1677  * a bmap btree split to do that.
1678  */
1679 STATIC int
1680 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1681                       xfs_dabuf_t **dead_bufp)
1682 {
1683         xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1684         xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1685         xfs_fileoff_t lastoff;
1686         xfs_inode_t *ip;
1687         xfs_trans_t *tp;
1688         xfs_mount_t *mp;
1689         int error, w, entno, level, dead_level;
1690         xfs_da_blkinfo_t *dead_info, *sib_info;
1691         xfs_da_intnode_t *par_node, *dead_node;
1692         xfs_dir2_leaf_t *dead_leaf2;
1693         xfs_dahash_t dead_hash;
1694
1695         dead_buf = *dead_bufp;
1696         dead_blkno = *dead_blknop;
1697         tp = args->trans;
1698         ip = args->dp;
1699         w = args->whichfork;
1700         ASSERT(w == XFS_DATA_FORK);
1701         mp = ip->i_mount;
1702         lastoff = mp->m_dirfreeblk;
1703         error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1704         if (error)
1705                 return error;
1706         if (unlikely(lastoff == 0)) {
1707                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1708                                  mp);
1709                 return XFS_ERROR(EFSCORRUPTED);
1710         }
1711         /*
1712          * Read the last block in the btree space.
1713          */
1714         last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1715         if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1716                 return error;
1717         /*
1718          * Copy the last block into the dead buffer and log it.
1719          */
1720         memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1721         xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1722         dead_info = dead_buf->data;
1723         /*
1724          * Get values from the moved block.
1725          */
1726         if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
1727                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1728                 dead_level = 0;
1729                 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1730         } else {
1731                 ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1732                 dead_node = (xfs_da_intnode_t *)dead_info;
1733                 dead_level = be16_to_cpu(dead_node->hdr.level);
1734                 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1735         }
1736         sib_buf = par_buf = NULL;
1737         /*
1738          * If the moved block has a left sibling, fix up the pointers.
1739          */
1740         if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1741                 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1742                         goto done;
1743                 sib_info = sib_buf->data;
1744                 if (unlikely(
1745                     be32_to_cpu(sib_info->forw) != last_blkno ||
1746                     sib_info->magic != dead_info->magic)) {
1747                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1748                                          XFS_ERRLEVEL_LOW, mp);
1749                         error = XFS_ERROR(EFSCORRUPTED);
1750                         goto done;
1751                 }
1752                 sib_info->forw = cpu_to_be32(dead_blkno);
1753                 xfs_da_log_buf(tp, sib_buf,
1754                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1755                                         sizeof(sib_info->forw)));
1756                 xfs_da_buf_done(sib_buf);
1757                 sib_buf = NULL;
1758         }
1759         /*
1760          * If the moved block has a right sibling, fix up the pointers.
1761          */
1762         if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1763                 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1764                         goto done;
1765                 sib_info = sib_buf->data;
1766                 if (unlikely(
1767                        be32_to_cpu(sib_info->back) != last_blkno ||
1768                        sib_info->magic != dead_info->magic)) {
1769                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1770                                          XFS_ERRLEVEL_LOW, mp);
1771                         error = XFS_ERROR(EFSCORRUPTED);
1772                         goto done;
1773                 }
1774                 sib_info->back = cpu_to_be32(dead_blkno);
1775                 xfs_da_log_buf(tp, sib_buf,
1776                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1777                                         sizeof(sib_info->back)));
1778                 xfs_da_buf_done(sib_buf);
1779                 sib_buf = NULL;
1780         }
1781         par_blkno = mp->m_dirleafblk;
1782         level = -1;
1783         /*
1784          * Walk down the tree looking for the parent of the moved block.
1785          */
1786         for (;;) {
1787                 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1788                         goto done;
1789                 par_node = par_buf->data;
1790                 if (unlikely(par_node->hdr.info.magic !=
1791                     cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1792                     (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1793                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1794                                          XFS_ERRLEVEL_LOW, mp);
1795                         error = XFS_ERROR(EFSCORRUPTED);
1796                         goto done;
1797                 }
1798                 level = be16_to_cpu(par_node->hdr.level);
1799                 for (entno = 0;
1800                      entno < be16_to_cpu(par_node->hdr.count) &&
1801                      be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1802                      entno++)
1803                         continue;
1804                 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1805                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1806                                          XFS_ERRLEVEL_LOW, mp);
1807                         error = XFS_ERROR(EFSCORRUPTED);
1808                         goto done;
1809                 }
1810                 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1811                 if (level == dead_level + 1)
1812                         break;
1813                 xfs_da_brelse(tp, par_buf);
1814                 par_buf = NULL;
1815         }
1816         /*
1817          * We're in the right parent block.
1818          * Look for the right entry.
1819          */
1820         for (;;) {
1821                 for (;
1822                      entno < be16_to_cpu(par_node->hdr.count) &&
1823                      be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1824                      entno++)
1825                         continue;
1826                 if (entno < be16_to_cpu(par_node->hdr.count))
1827                         break;
1828                 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1829                 xfs_da_brelse(tp, par_buf);
1830                 par_buf = NULL;
1831                 if (unlikely(par_blkno == 0)) {
1832                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1833                                          XFS_ERRLEVEL_LOW, mp);
1834                         error = XFS_ERROR(EFSCORRUPTED);
1835                         goto done;
1836                 }
1837                 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1838                         goto done;
1839                 par_node = par_buf->data;
1840                 if (unlikely(
1841                     be16_to_cpu(par_node->hdr.level) != level ||
1842                     par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) {
1843                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1844                                          XFS_ERRLEVEL_LOW, mp);
1845                         error = XFS_ERROR(EFSCORRUPTED);
1846                         goto done;
1847                 }
1848                 entno = 0;
1849         }
1850         /*
1851          * Update the parent entry pointing to the moved block.
1852          */
1853         par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1854         xfs_da_log_buf(tp, par_buf,
1855                 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1856                                 sizeof(par_node->btree[entno].before)));
1857         xfs_da_buf_done(par_buf);
1858         xfs_da_buf_done(dead_buf);
1859         *dead_blknop = last_blkno;
1860         *dead_bufp = last_buf;
1861         return 0;
1862 done:
1863         if (par_buf)
1864                 xfs_da_brelse(tp, par_buf);
1865         if (sib_buf)
1866                 xfs_da_brelse(tp, sib_buf);
1867         xfs_da_brelse(tp, last_buf);
1868         return error;
1869 }
1870
1871 /*
1872  * Remove a btree block from a directory or attribute.
1873  */
1874 int
1875 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1876                     xfs_dabuf_t *dead_buf)
1877 {
1878         xfs_inode_t *dp;
1879         int done, error, w, count;
1880         xfs_trans_t *tp;
1881         xfs_mount_t *mp;
1882
1883         dp = args->dp;
1884         w = args->whichfork;
1885         tp = args->trans;
1886         mp = dp->i_mount;
1887         if (w == XFS_DATA_FORK)
1888                 count = mp->m_dirblkfsbs;
1889         else
1890                 count = 1;
1891         for (;;) {
1892                 /*
1893                  * Remove extents.  If we get ENOSPC for a dir we have to move
1894                  * the last block to the place we want to kill.
1895                  */
1896                 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1897                                 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1898                                 0, args->firstblock, args->flist,
1899                                 &done)) == ENOSPC) {
1900                         if (w != XFS_DATA_FORK)
1901                                 break;
1902                         if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1903                                         &dead_buf)))
1904                                 break;
1905                 } else {
1906                         break;
1907                 }
1908         }
1909         xfs_da_binval(tp, dead_buf);
1910         return error;
1911 }
1912
1913 /*
1914  * See if the mapping(s) for this btree block are valid, i.e.
1915  * don't contain holes, are logically contiguous, and cover the whole range.
1916  */
1917 STATIC int
1918 xfs_da_map_covers_blocks(
1919         int             nmap,
1920         xfs_bmbt_irec_t *mapp,
1921         xfs_dablk_t     bno,
1922         int             count)
1923 {
1924         int             i;
1925         xfs_fileoff_t   off;
1926
1927         for (i = 0, off = bno; i < nmap; i++) {
1928                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1929                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
1930                         return 0;
1931                 }
1932                 if (off != mapp[i].br_startoff) {
1933                         return 0;
1934                 }
1935                 off += mapp[i].br_blockcount;
1936         }
1937         return off == bno + count;
1938 }
1939
1940 /*
1941  * Make a dabuf.
1942  * Used for get_buf, read_buf, read_bufr, and reada_buf.
1943  */
1944 STATIC int
1945 xfs_da_do_buf(
1946         xfs_trans_t     *trans,
1947         xfs_inode_t     *dp,
1948         xfs_dablk_t     bno,
1949         xfs_daddr_t     *mappedbnop,
1950         xfs_dabuf_t     **bpp,
1951         int             whichfork,
1952         int             caller)
1953 {
1954         xfs_buf_t       *bp = NULL;
1955         xfs_buf_t       **bplist;
1956         int             error=0;
1957         int             i;
1958         xfs_bmbt_irec_t map;
1959         xfs_bmbt_irec_t *mapp;
1960         xfs_daddr_t     mappedbno;
1961         xfs_mount_t     *mp;
1962         int             nbplist=0;
1963         int             nfsb;
1964         int             nmap;
1965         xfs_dabuf_t     *rbp;
1966
1967         mp = dp->i_mount;
1968         nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
1969         mappedbno = *mappedbnop;
1970         /*
1971          * Caller doesn't have a mapping.  -2 means don't complain
1972          * if we land in a hole.
1973          */
1974         if (mappedbno == -1 || mappedbno == -2) {
1975                 /*
1976                  * Optimize the one-block case.
1977                  */
1978                 if (nfsb == 1) {
1979                         xfs_fsblock_t   fsb;
1980
1981                         if ((error =
1982                             xfs_bmapi_single(trans, dp, whichfork, &fsb,
1983                                     (xfs_fileoff_t)bno))) {
1984                                 return error;
1985                         }
1986                         mapp = &map;
1987                         if (fsb == NULLFSBLOCK) {
1988                                 nmap = 0;
1989                         } else {
1990                                 map.br_startblock = fsb;
1991                                 map.br_startoff = (xfs_fileoff_t)bno;
1992                                 map.br_blockcount = 1;
1993                                 nmap = 1;
1994                         }
1995                 } else {
1996                         mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1997                         nmap = nfsb;
1998                         if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1999                                         nfsb,
2000                                         XFS_BMAPI_METADATA |
2001                                                 xfs_bmapi_aflag(whichfork),
2002                                         NULL, 0, mapp, &nmap, NULL)))
2003                                 goto exit0;
2004                 }
2005         } else {
2006                 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2007                 map.br_startoff = (xfs_fileoff_t)bno;
2008                 map.br_blockcount = nfsb;
2009                 mapp = &map;
2010                 nmap = 1;
2011         }
2012         if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
2013                 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
2014                 if (unlikely(error == EFSCORRUPTED)) {
2015                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2016                                 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2017                                         __func__, (long long)bno,
2018                                         (long long)dp->i_ino);
2019                                 for (i = 0; i < nmap; i++) {
2020                                         xfs_alert(mp,
2021 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2022                                                 i,
2023                                                 (long long)mapp[i].br_startoff,
2024                                                 (long long)mapp[i].br_startblock,
2025                                                 (long long)mapp[i].br_blockcount,
2026                                                 mapp[i].br_state);
2027                                 }
2028                         }
2029                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2030                                          XFS_ERRLEVEL_LOW, mp);
2031                 }
2032                 goto exit0;
2033         }
2034         if (caller != 3 && nmap > 1) {
2035                 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2036                 nbplist = 0;
2037         } else
2038                 bplist = NULL;
2039         /*
2040          * Turn the mapping(s) into buffer(s).
2041          */
2042         for (i = 0; i < nmap; i++) {
2043                 int     nmapped;
2044
2045                 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2046                 if (i == 0)
2047                         *mappedbnop = mappedbno;
2048                 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2049                 switch (caller) {
2050                 case 0:
2051                         bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2052                                 mappedbno, nmapped, 0);
2053                         error = bp ? bp->b_error : XFS_ERROR(EIO);
2054                         break;
2055                 case 1:
2056                 case 2:
2057                         bp = NULL;
2058                         error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2059                                 mappedbno, nmapped, 0, &bp);
2060                         break;
2061                 case 3:
2062                         xfs_buf_readahead(mp->m_ddev_targp, mappedbno, nmapped);
2063                         error = 0;
2064                         bp = NULL;
2065                         break;
2066                 }
2067                 if (error) {
2068                         if (bp)
2069                                 xfs_trans_brelse(trans, bp);
2070                         goto exit1;
2071                 }
2072                 if (!bp)
2073                         continue;
2074                 if (caller == 1) {
2075                         if (whichfork == XFS_ATTR_FORK) {
2076                                 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2077                                                 XFS_ATTR_BTREE_REF);
2078                         } else {
2079                                 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2080                                                 XFS_DIR_BTREE_REF);
2081                         }
2082                 }
2083                 if (bplist) {
2084                         bplist[nbplist++] = bp;
2085                 }
2086         }
2087         /*
2088          * Build a dabuf structure.
2089          */
2090         if (bplist) {
2091                 rbp = xfs_da_buf_make(nbplist, bplist);
2092         } else if (bp)
2093                 rbp = xfs_da_buf_make(1, &bp);
2094         else
2095                 rbp = NULL;
2096         /*
2097          * For read_buf, check the magic number.
2098          */
2099         if (caller == 1) {
2100                 xfs_dir2_data_hdr_t     *hdr = rbp->data;
2101                 xfs_dir2_free_t         *free = rbp->data;
2102                 xfs_da_blkinfo_t        *info = rbp->data;
2103                 uint                    magic, magic1;
2104
2105                 magic = be16_to_cpu(info->magic);
2106                 magic1 = be32_to_cpu(hdr->magic);
2107                 if (unlikely(
2108                     XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2109                                    (magic != XFS_ATTR_LEAF_MAGIC) &&
2110                                    (magic != XFS_DIR2_LEAF1_MAGIC) &&
2111                                    (magic != XFS_DIR2_LEAFN_MAGIC) &&
2112                                    (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2113                                    (magic1 != XFS_DIR2_DATA_MAGIC) &&
2114                                    (free->hdr.magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC)),
2115                                 mp, XFS_ERRTAG_DA_READ_BUF,
2116                                 XFS_RANDOM_DA_READ_BUF))) {
2117                         trace_xfs_da_btree_corrupt(rbp->bps[0], _RET_IP_);
2118                         XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2119                                              XFS_ERRLEVEL_LOW, mp, info);
2120                         error = XFS_ERROR(EFSCORRUPTED);
2121                         xfs_da_brelse(trans, rbp);
2122                         nbplist = 0;
2123                         goto exit1;
2124                 }
2125         }
2126         if (bplist) {
2127                 kmem_free(bplist);
2128         }
2129         if (mapp != &map) {
2130                 kmem_free(mapp);
2131         }
2132         if (bpp)
2133                 *bpp = rbp;
2134         return 0;
2135 exit1:
2136         if (bplist) {
2137                 for (i = 0; i < nbplist; i++)
2138                         xfs_trans_brelse(trans, bplist[i]);
2139                 kmem_free(bplist);
2140         }
2141 exit0:
2142         if (mapp != &map)
2143                 kmem_free(mapp);
2144         if (bpp)
2145                 *bpp = NULL;
2146         return error;
2147 }
2148
2149 /*
2150  * Get a buffer for the dir/attr block.
2151  */
2152 int
2153 xfs_da_get_buf(
2154         xfs_trans_t     *trans,
2155         xfs_inode_t     *dp,
2156         xfs_dablk_t     bno,
2157         xfs_daddr_t             mappedbno,
2158         xfs_dabuf_t     **bpp,
2159         int             whichfork)
2160 {
2161         return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0);
2162 }
2163
2164 /*
2165  * Get a buffer for the dir/attr block, fill in the contents.
2166  */
2167 int
2168 xfs_da_read_buf(
2169         xfs_trans_t     *trans,
2170         xfs_inode_t     *dp,
2171         xfs_dablk_t     bno,
2172         xfs_daddr_t             mappedbno,
2173         xfs_dabuf_t     **bpp,
2174         int             whichfork)
2175 {
2176         return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1);
2177 }
2178
2179 /*
2180  * Readahead the dir/attr block.
2181  */
2182 xfs_daddr_t
2183 xfs_da_reada_buf(
2184         xfs_trans_t     *trans,
2185         xfs_inode_t     *dp,
2186         xfs_dablk_t     bno,
2187         int             whichfork)
2188 {
2189         xfs_daddr_t             rval;
2190
2191         rval = -1;
2192         if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3))
2193                 return -1;
2194         else
2195                 return rval;
2196 }
2197
2198 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2199 kmem_zone_t *xfs_dabuf_zone;            /* dabuf zone */
2200
2201 /*
2202  * Allocate a dir-state structure.
2203  * We don't put them on the stack since they're large.
2204  */
2205 xfs_da_state_t *
2206 xfs_da_state_alloc(void)
2207 {
2208         return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
2209 }
2210
2211 /*
2212  * Kill the altpath contents of a da-state structure.
2213  */
2214 STATIC void
2215 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2216 {
2217         int     i;
2218
2219         for (i = 0; i < state->altpath.active; i++) {
2220                 if (state->altpath.blk[i].bp) {
2221                         if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2222                                 xfs_da_buf_done(state->altpath.blk[i].bp);
2223                         state->altpath.blk[i].bp = NULL;
2224                 }
2225         }
2226         state->altpath.active = 0;
2227 }
2228
2229 /*
2230  * Free a da-state structure.
2231  */
2232 void
2233 xfs_da_state_free(xfs_da_state_t *state)
2234 {
2235         int     i;
2236
2237         xfs_da_state_kill_altpath(state);
2238         for (i = 0; i < state->path.active; i++) {
2239                 if (state->path.blk[i].bp)
2240                         xfs_da_buf_done(state->path.blk[i].bp);
2241         }
2242         if (state->extravalid && state->extrablk.bp)
2243                 xfs_da_buf_done(state->extrablk.bp);
2244 #ifdef DEBUG
2245         memset((char *)state, 0, sizeof(*state));
2246 #endif /* DEBUG */
2247         kmem_zone_free(xfs_da_state_zone, state);
2248 }
2249
2250 /*
2251  * Create a dabuf.
2252  */
2253 /* ARGSUSED */
2254 STATIC xfs_dabuf_t *
2255 xfs_da_buf_make(int nbuf, xfs_buf_t **bps)
2256 {
2257         xfs_buf_t       *bp;
2258         xfs_dabuf_t     *dabuf;
2259         int             i;
2260         int             off;
2261
2262         if (nbuf == 1)
2263                 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
2264         else
2265                 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_NOFS);
2266         dabuf->dirty = 0;
2267         if (nbuf == 1) {
2268                 dabuf->nbuf = 1;
2269                 bp = bps[0];
2270                 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2271                 dabuf->data = bp->b_addr;
2272                 dabuf->bps[0] = bp;
2273         } else {
2274                 dabuf->nbuf = nbuf;
2275                 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2276                         dabuf->bps[i] = bp = bps[i];
2277                         dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2278                 }
2279                 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2280                 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2281                         bp = bps[i];
2282                         memcpy((char *)dabuf->data + off, bp->b_addr,
2283                                 XFS_BUF_COUNT(bp));
2284                 }
2285         }
2286         return dabuf;
2287 }
2288
2289 /*
2290  * Un-dirty a dabuf.
2291  */
2292 STATIC void
2293 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2294 {
2295         xfs_buf_t       *bp;
2296         int             i;
2297         int             off;
2298
2299         if (dabuf->dirty) {
2300                 ASSERT(dabuf->nbuf > 1);
2301                 dabuf->dirty = 0;
2302                 for (i = off = 0; i < dabuf->nbuf;
2303                                 i++, off += XFS_BUF_COUNT(bp)) {
2304                         bp = dabuf->bps[i];
2305                         memcpy(bp->b_addr, dabuf->data + off,
2306                                                 XFS_BUF_COUNT(bp));
2307                 }
2308         }
2309 }
2310
2311 /*
2312  * Release a dabuf.
2313  */
2314 void
2315 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2316 {
2317         ASSERT(dabuf);
2318         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2319         if (dabuf->dirty)
2320                 xfs_da_buf_clean(dabuf);
2321         if (dabuf->nbuf > 1) {
2322                 kmem_free(dabuf->data);
2323                 kmem_free(dabuf);
2324         } else {
2325                 kmem_zone_free(xfs_dabuf_zone, dabuf);
2326         }
2327 }
2328
2329 /*
2330  * Log transaction from a dabuf.
2331  */
2332 void
2333 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2334 {
2335         xfs_buf_t       *bp;
2336         uint            f;
2337         int             i;
2338         uint            l;
2339         int             off;
2340
2341         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2342         if (dabuf->nbuf == 1) {
2343                 ASSERT(dabuf->data == dabuf->bps[0]->b_addr);
2344                 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2345                 return;
2346         }
2347         dabuf->dirty = 1;
2348         ASSERT(first <= last);
2349         for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2350                 bp = dabuf->bps[i];
2351                 f = off;
2352                 l = f + XFS_BUF_COUNT(bp) - 1;
2353                 if (f < first)
2354                         f = first;
2355                 if (l > last)
2356                         l = last;
2357                 if (f <= l)
2358                         xfs_trans_log_buf(tp, bp, f - off, l - off);
2359                 /*
2360                  * B_DONE is set by xfs_trans_log buf.
2361                  * If we don't set it on a new buffer (get not read)
2362                  * then if we don't put anything in the buffer it won't
2363                  * be set, and at commit it it released into the cache,
2364                  * and then a read will fail.
2365                  */
2366                 else if (!(XFS_BUF_ISDONE(bp)))
2367                   XFS_BUF_DONE(bp);
2368         }
2369         ASSERT(last < off);
2370 }
2371
2372 /*
2373  * Release dabuf from a transaction.
2374  * Have to free up the dabuf before the buffers are released,
2375  * since the synchronization on the dabuf is really the lock on the buffer.
2376  */
2377 void
2378 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2379 {
2380         xfs_buf_t       *bp;
2381         xfs_buf_t       **bplist;
2382         int             i;
2383         int             nbuf;
2384
2385         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2386         if ((nbuf = dabuf->nbuf) == 1) {
2387                 bplist = &bp;
2388                 bp = dabuf->bps[0];
2389         } else {
2390                 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2391                 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2392         }
2393         xfs_da_buf_done(dabuf);
2394         for (i = 0; i < nbuf; i++)
2395                 xfs_trans_brelse(tp, bplist[i]);
2396         if (bplist != &bp)
2397                 kmem_free(bplist);
2398 }
2399
2400 /*
2401  * Invalidate dabuf from a transaction.
2402  */
2403 void
2404 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2405 {
2406         xfs_buf_t       *bp;
2407         xfs_buf_t       **bplist;
2408         int             i;
2409         int             nbuf;
2410
2411         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2412         if ((nbuf = dabuf->nbuf) == 1) {
2413                 bplist = &bp;
2414                 bp = dabuf->bps[0];
2415         } else {
2416                 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2417                 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2418         }
2419         xfs_da_buf_done(dabuf);
2420         for (i = 0; i < nbuf; i++)
2421                 xfs_trans_binval(tp, bplist[i]);
2422         if (bplist != &bp)
2423                 kmem_free(bplist);
2424 }
2425
2426 /*
2427  * Get the first daddr from a dabuf.
2428  */
2429 xfs_daddr_t
2430 xfs_da_blkno(xfs_dabuf_t *dabuf)
2431 {
2432         ASSERT(dabuf->nbuf);
2433         ASSERT(dabuf->data);
2434         return XFS_BUF_ADDR(dabuf->bps[0]);
2435 }