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_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_da_btree.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_ialloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_inode_item.h"
  41 #include "xfs_alloc.h"
  42 #include "xfs_btree.h"
  43 #include "xfs_bmap.h"
  44 #include "xfs_attr.h"
  45 #include "xfs_attr_leaf.h"
  46 #include "xfs_dir_leaf.h"
  47 #include "xfs_dir2_data.h"
  48 #include "xfs_dir2_leaf.h"
  49 #include "xfs_dir2_block.h"
  50 #include "xfs_dir2_node.h"
  51 #include "xfs_error.h"
  52
  53 /*
  54  * xfs_da_btree.c
  55  *
  56  * Routines to implement directories as Btrees of hashed names.
  57  */
  58
  59 /*========================================================================
  60  * Function prototypes for the kernel.
  61  *========================================================================*/
  62
  63 /*
  64  * Routines used for growing the Btree.
  65  */
  66 STATIC int xfs_da_root_split(xfs_da_state_t *state,
  67                                             xfs_da_state_blk_t *existing_root,
  68                                             xfs_da_state_blk_t *new_child);
  69 STATIC int xfs_da_node_split(xfs_da_state_t *state,
  70                                             xfs_da_state_blk_t *existing_blk,
  71                                             xfs_da_state_blk_t *split_blk,
  72                                             xfs_da_state_blk_t *blk_to_add,
  73                                             int treelevel,
  74                                             int *result);
  75 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
  76                                          xfs_da_state_blk_t *node_blk_1,
  77                                          xfs_da_state_blk_t *node_blk_2);
  78 STATIC void xfs_da_node_add(xfs_da_state_t *state,
  79                                    xfs_da_state_blk_t *old_node_blk,
  80                                    xfs_da_state_blk_t *new_node_blk);
  81
  82 /*
  83  * Routines used for shrinking the Btree.
  84  */
  85 STATIC int xfs_da_root_join(xfs_da_state_t *state,
  86                                            xfs_da_state_blk_t *root_blk);
  87 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
  88 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
  89                                               xfs_da_state_blk_t *drop_blk);
  90 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
  91                                          xfs_da_state_blk_t *src_node_blk,
  92                                          xfs_da_state_blk_t *dst_node_blk);
  93
  94 /*
  95  * Utility routines.
  96  */
  97 STATIC uint     xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
  98 STATIC int      xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
  99 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
 100 STATIC int      xfs_da_blk_unlink(xfs_da_state_t *state,
 101                                   xfs_da_state_blk_t *drop_blk,
 102                                   xfs_da_state_blk_t *save_blk);
 103 STATIC void     xfs_da_state_kill_altpath(xfs_da_state_t *state);
 104
 105 /*========================================================================
 106  * Routines used for growing the Btree.
 107  *========================================================================*/
 108
 109 /*
 110  * Create the initial contents of an intermediate node.
 111  */
 112 int
 113 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
 114                                  xfs_dabuf_t **bpp, int whichfork)
 115 {
 116         xfs_da_intnode_t *node;
 117         xfs_dabuf_t *bp;
 118         int error;
 119         xfs_trans_t *tp;
 120
 121         tp = args->trans;
 122         error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
 123         if (error)
 124                 return(error);
 125         ASSERT(bp != NULL);
 126         node = bp->data;
 127         node->hdr.info.forw = 0;
 128         node->hdr.info.back = 0;
 129         INT_SET(node->hdr.info.magic, ARCH_CONVERT, XFS_DA_NODE_MAGIC);
 130         node->hdr.info.pad = 0;
 131         node->hdr.count = 0;
 132         INT_SET(node->hdr.level, ARCH_CONVERT, level);
 133
 134         xfs_da_log_buf(tp, bp,
 135                 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 136
 137         *bpp = bp;
 138         return(0);
 139 }
 140
 141 /*
 142  * Split a leaf node, rebalance, then possibly split
 143  * intermediate nodes, rebalance, etc.
 144  */
 145 int                                                     /* error */
 146 xfs_da_split(xfs_da_state_t *state)
 147 {
 148         xfs_da_state_blk_t *oldblk, *newblk, *addblk;
 149         xfs_da_intnode_t *node;
 150         xfs_dabuf_t *bp;
 151         int max, action, error, i;
 152
 153         /*
 154          * Walk back up the tree splitting/inserting/adjusting as necessary.
 155          * If we need to insert and there isn't room, split the node, then
 156          * decide which fragment to insert the new block from below into.
 157          * Note that we may split the root this way, but we need more fixup.
 158          */
 159         max = state->path.active - 1;
 160         ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
 161         ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
 162                state->path.blk[max].magic == XFS_DIRX_LEAF_MAGIC(state->mp));
 163
 164         addblk = &state->path.blk[max];         /* initial dummy value */
 165         for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
 166                 oldblk = &state->path.blk[i];
 167                 newblk = &state->altpath.blk[i];
 168
 169                 /*
 170                  * If a leaf node then
 171                  *     Allocate a new leaf node, then rebalance across them.
 172                  * else if an intermediate node then
 173                  *     We split on the last layer, must we split the node?
 174                  */
 175                 switch (oldblk->magic) {
 176                 case XFS_ATTR_LEAF_MAGIC:
 177                         error = xfs_attr_leaf_split(state, oldblk, newblk);
 178                         if ((error != 0) && (error != ENOSPC)) {
 179                                 return(error);  /* GROT: attr is inconsistent */
 180                         }
 181                         if (!error) {
 182                                 addblk = newblk;
 183                                 break;
 184                         }
 185                         /*
 186                          * Entry wouldn't fit, split the leaf again.
 187                          */
 188                         state->extravalid = 1;
 189                         if (state->inleaf) {
 190                                 state->extraafter = 0;  /* before newblk */
 191                                 error = xfs_attr_leaf_split(state, oldblk,
 192                                                             &state->extrablk);
 193                         } else {
 194                                 state->extraafter = 1;  /* after newblk */
 195                                 error = xfs_attr_leaf_split(state, newblk,
 196                                                             &state->extrablk);
 197                         }
 198                         if (error)
 199                                 return(error);  /* GROT: attr inconsistent */
 200                         addblk = newblk;
 201                         break;
 202                 case XFS_DIR_LEAF_MAGIC:
 203                         ASSERT(XFS_DIR_IS_V1(state->mp));
 204                         error = xfs_dir_leaf_split(state, oldblk, newblk);
 205                         if ((error != 0) && (error != ENOSPC)) {
 206                                 return(error);  /* GROT: dir is inconsistent */
 207                         }
 208                         if (!error) {
 209                                 addblk = newblk;
 210                                 break;
 211                         }
 212                         /*
 213                          * Entry wouldn't fit, split the leaf again.
 214                          */
 215                         state->extravalid = 1;
 216                         if (state->inleaf) {
 217                                 state->extraafter = 0;  /* before newblk */
 218                                 error = xfs_dir_leaf_split(state, oldblk,
 219                                                            &state->extrablk);
 220                                 if (error)
 221                                         return(error);  /* GROT: dir incon. */
 222                                 addblk = newblk;
 223                         } else {
 224                                 state->extraafter = 1;  /* after newblk */
 225                                 error = xfs_dir_leaf_split(state, newblk,
 226                                                            &state->extrablk);
 227                                 if (error)
 228                                         return(error);  /* GROT: dir incon. */
 229                                 addblk = newblk;
 230                         }
 231                         break;
 232                 case XFS_DIR2_LEAFN_MAGIC:
 233                         ASSERT(XFS_DIR_IS_V2(state->mp));
 234                         error = xfs_dir2_leafn_split(state, oldblk, newblk);
 235                         if (error)
 236                                 return error;
 237                         addblk = newblk;
 238                         break;
 239                 case XFS_DA_NODE_MAGIC:
 240                         error = xfs_da_node_split(state, oldblk, newblk, addblk,
 241                                                          max - i, &action);
 242                         xfs_da_buf_done(addblk->bp);
 243                         addblk->bp = NULL;
 244                         if (error)
 245                                 return(error);  /* GROT: dir is inconsistent */
 246                         /*
 247                          * Record the newly split block for the next time thru?
 248                          */
 249                         if (action)
 250                                 addblk = newblk;
 251                         else
 252                                 addblk = NULL;
 253                         break;
 254                 }
 255
 256                 /*
 257                  * Update the btree to show the new hashval for this child.
 258                  */
 259                 xfs_da_fixhashpath(state, &state->path);
 260                 /*
 261                  * If we won't need this block again, it's getting dropped
 262                  * from the active path by the loop control, so we need
 263                  * to mark it done now.
 264                  */
 265                 if (i > 0 || !addblk)
 266                         xfs_da_buf_done(oldblk->bp);
 267         }
 268         if (!addblk)
 269                 return(0);
 270
 271         /*
 272          * Split the root node.
 273          */
 274         ASSERT(state->path.active == 0);
 275         oldblk = &state->path.blk[0];
 276         error = xfs_da_root_split(state, oldblk, addblk);
 277         if (error) {
 278                 xfs_da_buf_done(oldblk->bp);
 279                 xfs_da_buf_done(addblk->bp);
 280                 addblk->bp = NULL;
 281                 return(error);  /* GROT: dir is inconsistent */
 282         }
 283
 284         /*
 285          * Update pointers to the node which used to be block 0 and
 286          * just got bumped because of the addition of a new root node.
 287          * There might be three blocks involved if a double split occurred,
 288          * and the original block 0 could be at any position in the list.
 289          */
 290
 291         node = oldblk->bp->data;
 292         if (node->hdr.info.forw) {
 293                 if (INT_GET(node->hdr.info.forw, ARCH_CONVERT) == addblk->blkno) {
 294                         bp = addblk->bp;
 295                 } else {
 296                         ASSERT(state->extravalid);
 297                         bp = state->extrablk.bp;
 298                 }
 299                 node = bp->data;
 300                 INT_SET(node->hdr.info.back, ARCH_CONVERT, oldblk->blkno);
 301                 xfs_da_log_buf(state->args->trans, bp,
 302                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 303                     sizeof(node->hdr.info)));
 304         }
 305         node = oldblk->bp->data;
 306         if (INT_GET(node->hdr.info.back, ARCH_CONVERT)) {
 307                 if (INT_GET(node->hdr.info.back, ARCH_CONVERT) == addblk->blkno) {
 308                         bp = addblk->bp;
 309                 } else {
 310                         ASSERT(state->extravalid);
 311                         bp = state->extrablk.bp;
 312                 }
 313                 node = bp->data;
 314                 INT_SET(node->hdr.info.forw, ARCH_CONVERT, oldblk->blkno);
 315                 xfs_da_log_buf(state->args->trans, bp,
 316                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 317                     sizeof(node->hdr.info)));
 318         }
 319         xfs_da_buf_done(oldblk->bp);
 320         xfs_da_buf_done(addblk->bp);
 321         addblk->bp = NULL;
 322         return(0);
 323 }
 324
 325 /*
 326  * Split the root.  We have to create a new root and point to the two
 327  * parts (the split old root) that we just created.  Copy block zero to
 328  * the EOF, extending the inode in process.
 329  */
 330 STATIC int                                              /* error */
 331 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
 332                                  xfs_da_state_blk_t *blk2)
 333 {
 334         xfs_da_intnode_t *node, *oldroot;
 335         xfs_da_args_t *args;
 336         xfs_dablk_t blkno;
 337         xfs_dabuf_t *bp;
 338         int error, size;
 339         xfs_inode_t *dp;
 340         xfs_trans_t *tp;
 341         xfs_mount_t *mp;
 342         xfs_dir2_leaf_t *leaf;
 343
 344         /*
 345          * Copy the existing (incorrect) block from the root node position
 346          * to a free space somewhere.
 347          */
 348         args = state->args;
 349         ASSERT(args != NULL);
 350         error = xfs_da_grow_inode(args, &blkno);
 351         if (error)
 352                 return(error);
 353         dp = args->dp;
 354         tp = args->trans;
 355         mp = state->mp;
 356         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 357         if (error)
 358                 return(error);
 359         ASSERT(bp != NULL);
 360         node = bp->data;
 361         oldroot = blk1->bp->data;
 362         if (INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) {
 363                 size = (int)((char *)&oldroot->btree[INT_GET(oldroot->hdr.count, ARCH_CONVERT)] -
 364                              (char *)oldroot);
 365         } else {
 366                 ASSERT(XFS_DIR_IS_V2(mp));
 367                 ASSERT(INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
 368                 leaf = (xfs_dir2_leaf_t *)oldroot;
 369                 size = (int)((char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] -
 370                              (char *)leaf);
 371         }
 372         memcpy(node, oldroot, size);
 373         xfs_da_log_buf(tp, bp, 0, size - 1);
 374         xfs_da_buf_done(blk1->bp);
 375         blk1->bp = bp;
 376         blk1->blkno = blkno;
 377
 378         /*
 379          * Set up the new root node.
 380          */
 381         error = xfs_da_node_create(args,
 382                 args->whichfork == XFS_DATA_FORK &&
 383                 XFS_DIR_IS_V2(mp) ? mp->m_dirleafblk : 0,
 384                 INT_GET(node->hdr.level, ARCH_CONVERT) + 1, &bp, args->whichfork);
 385         if (error)
 386                 return(error);
 387         node = bp->data;
 388         INT_SET(node->btree[0].hashval, ARCH_CONVERT, blk1->hashval);
 389         INT_SET(node->btree[0].before, ARCH_CONVERT, blk1->blkno);
 390         INT_SET(node->btree[1].hashval, ARCH_CONVERT, blk2->hashval);
 391         INT_SET(node->btree[1].before, ARCH_CONVERT, blk2->blkno);
 392         INT_SET(node->hdr.count, ARCH_CONVERT, 2);
 393
 394 #ifdef DEBUG
 395         if (INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC) {
 396                 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
 397                        blk1->blkno < mp->m_dirfreeblk);
 398                 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
 399                        blk2->blkno < mp->m_dirfreeblk);
 400         }
 401 #endif
 402
 403         /* Header is already logged by xfs_da_node_create */
 404         xfs_da_log_buf(tp, bp,
 405                 XFS_DA_LOGRANGE(node, node->btree,
 406                         sizeof(xfs_da_node_entry_t) * 2));
 407         xfs_da_buf_done(bp);
 408
 409         return(0);
 410 }
 411
 412 /*
 413  * Split the node, rebalance, then add the new entry.
 414  */
 415 STATIC int                                              /* error */
 416 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 417                                  xfs_da_state_blk_t *newblk,
 418                                  xfs_da_state_blk_t *addblk,
 419                                  int treelevel, int *result)
 420 {
 421         xfs_da_intnode_t *node;
 422         xfs_dablk_t blkno;
 423         int newcount, error;
 424         int useextra;
 425
 426         node = oldblk->bp->data;
 427         ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 428
 429         /*
 430          * With V2 the extra block is data or freespace.
 431          */
 432         useextra = state->extravalid && XFS_DIR_IS_V1(state->mp);
 433         newcount = 1 + useextra;
 434         /*
 435          * Do we have to split the node?
 436          */
 437         if ((INT_GET(node->hdr.count, ARCH_CONVERT) + newcount) > state->node_ents) {
 438                 /*
 439                  * Allocate a new node, add to the doubly linked chain of
 440                  * nodes, then move some of our excess entries into it.
 441                  */
 442                 error = xfs_da_grow_inode(state->args, &blkno);
 443                 if (error)
 444                         return(error);  /* GROT: dir is inconsistent */
 445
 446                 error = xfs_da_node_create(state->args, blkno, treelevel,
 447                                            &newblk->bp, state->args->whichfork);
 448                 if (error)
 449                         return(error);  /* GROT: dir is inconsistent */
 450                 newblk->blkno = blkno;
 451                 newblk->magic = XFS_DA_NODE_MAGIC;
 452                 xfs_da_node_rebalance(state, oldblk, newblk);
 453                 error = xfs_da_blk_link(state, oldblk, newblk);
 454                 if (error)
 455                         return(error);
 456                 *result = 1;
 457         } else {
 458                 *result = 0;
 459         }
 460
 461         /*
 462          * Insert the new entry(s) into the correct block
 463          * (updating last hashval in the process).
 464          *
 465          * xfs_da_node_add() inserts BEFORE the given index,
 466          * and as a result of using node_lookup_int() we always
 467          * point to a valid entry (not after one), but a split
 468          * operation always results in a new block whose hashvals
 469          * FOLLOW the current block.
 470          *
 471          * If we had double-split op below us, then add the extra block too.
 472          */
 473         node = oldblk->bp->data;
 474         if (oldblk->index <= INT_GET(node->hdr.count, ARCH_CONVERT)) {
 475                 oldblk->index++;
 476                 xfs_da_node_add(state, oldblk, addblk);
 477                 if (useextra) {
 478                         if (state->extraafter)
 479                                 oldblk->index++;
 480                         xfs_da_node_add(state, oldblk, &state->extrablk);
 481                         state->extravalid = 0;
 482                 }
 483         } else {
 484                 newblk->index++;
 485                 xfs_da_node_add(state, newblk, addblk);
 486                 if (useextra) {
 487                         if (state->extraafter)
 488                                 newblk->index++;
 489                         xfs_da_node_add(state, newblk, &state->extrablk);
 490                         state->extravalid = 0;
 491                 }
 492         }
 493
 494         return(0);
 495 }
 496
 497 /*
 498  * Balance the btree elements between two intermediate nodes,
 499  * usually one full and one empty.
 500  *
 501  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 502  */
 503 STATIC void
 504 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
 505                                      xfs_da_state_blk_t *blk2)
 506 {
 507         xfs_da_intnode_t *node1, *node2, *tmpnode;
 508         xfs_da_node_entry_t *btree_s, *btree_d;
 509         int count, tmp;
 510         xfs_trans_t *tp;
 511
 512         node1 = blk1->bp->data;
 513         node2 = blk2->bp->data;
 514         /*
 515          * Figure out how many entries need to move, and in which direction.
 516          * Swap the nodes around if that makes it simpler.
 517          */
 518         if ((INT_GET(node1->hdr.count, ARCH_CONVERT) > 0) && (INT_GET(node2->hdr.count, ARCH_CONVERT) > 0) &&
 519             ((INT_GET(node2->btree[ 0 ].hashval, ARCH_CONVERT) < INT_GET(node1->btree[ 0 ].hashval, ARCH_CONVERT)) ||
 520              (INT_GET(node2->btree[ INT_GET(node2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
 521               INT_GET(node1->btree[ INT_GET(node1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))) {
 522                 tmpnode = node1;
 523                 node1 = node2;
 524                 node2 = tmpnode;
 525         }
 526         ASSERT(INT_GET(node1->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 527         ASSERT(INT_GET(node2->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 528         count = (INT_GET(node1->hdr.count, ARCH_CONVERT) - INT_GET(node2->hdr.count, ARCH_CONVERT)) / 2;
 529         if (count == 0)
 530                 return;
 531         tp = state->args->trans;
 532         /*
 533          * Two cases: high-to-low and low-to-high.
 534          */
 535         if (count > 0) {
 536                 /*
 537                  * Move elements in node2 up to make a hole.
 538                  */
 539                 if ((tmp = INT_GET(node2->hdr.count, ARCH_CONVERT)) > 0) {
 540                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 541                         btree_s = &node2->btree[0];
 542                         btree_d = &node2->btree[count];
 543                         memmove(btree_d, btree_s, tmp);
 544                 }
 545
 546                 /*
 547                  * Move the req'd B-tree elements from high in node1 to
 548                  * low in node2.
 549                  */
 550                 INT_MOD(node2->hdr.count, ARCH_CONVERT, count);
 551                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 552                 btree_s = &node1->btree[INT_GET(node1->hdr.count, ARCH_CONVERT) - count];
 553                 btree_d = &node2->btree[0];
 554                 memcpy(btree_d, btree_s, tmp);
 555                 INT_MOD(node1->hdr.count, ARCH_CONVERT, -(count));
 556
 557         } else {
 558                 /*
 559                  * Move the req'd B-tree elements from low in node2 to
 560                  * high in node1.
 561                  */
 562                 count = -count;
 563                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 564                 btree_s = &node2->btree[0];
 565                 btree_d = &node1->btree[INT_GET(node1->hdr.count, ARCH_CONVERT)];
 566                 memcpy(btree_d, btree_s, tmp);
 567                 INT_MOD(node1->hdr.count, ARCH_CONVERT, count);
 568                 xfs_da_log_buf(tp, blk1->bp,
 569                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 570
 571                 /*
 572                  * Move elements in node2 down to fill the hole.
 573                  */
 574                 tmp  = INT_GET(node2->hdr.count, ARCH_CONVERT) - count;
 575                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 576                 btree_s = &node2->btree[count];
 577                 btree_d = &node2->btree[0];
 578                 memmove(btree_d, btree_s, tmp);
 579                 INT_MOD(node2->hdr.count, ARCH_CONVERT, -(count));
 580         }
 581
 582         /*
 583          * Log header of node 1 and all current bits of node 2.
 584          */
 585         xfs_da_log_buf(tp, blk1->bp,
 586                 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
 587         xfs_da_log_buf(tp, blk2->bp,
 588                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 589                         sizeof(node2->hdr) +
 590                         sizeof(node2->btree[0]) * INT_GET(node2->hdr.count, ARCH_CONVERT)));
 591
 592         /*
 593          * Record the last hashval from each block for upward propagation.
 594          * (note: don't use the swapped node pointers)
 595          */
 596         node1 = blk1->bp->data;
 597         node2 = blk2->bp->data;
 598         blk1->hashval = INT_GET(node1->btree[ INT_GET(node1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
 599         blk2->hashval = INT_GET(node2->btree[ INT_GET(node2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
 600
 601         /*
 602          * Adjust the expected index for insertion.
 603          */
 604         if (blk1->index >= INT_GET(node1->hdr.count, ARCH_CONVERT)) {
 605                 blk2->index = blk1->index - INT_GET(node1->hdr.count, ARCH_CONVERT);
 606                 blk1->index = INT_GET(node1->hdr.count, ARCH_CONVERT) + 1;      /* make it invalid */
 607         }
 608 }
 609
 610 /*
 611  * Add a new entry to an intermediate node.
 612  */
 613 STATIC void
 614 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 615                                xfs_da_state_blk_t *newblk)
 616 {
 617         xfs_da_intnode_t *node;
 618         xfs_da_node_entry_t *btree;
 619         int tmp;
 620         xfs_mount_t *mp;
 621
 622         node = oldblk->bp->data;
 623         mp = state->mp;
 624         ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 625         ASSERT((oldblk->index >= 0) && (oldblk->index <= INT_GET(node->hdr.count, ARCH_CONVERT)));
 626         ASSERT(newblk->blkno != 0);
 627         if (state->args->whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
 628                 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
 629                        newblk->blkno < mp->m_dirfreeblk);
 630
 631         /*
 632          * We may need to make some room before we insert the new node.
 633          */
 634         tmp = 0;
 635         btree = &node->btree[ oldblk->index ];
 636         if (oldblk->index < INT_GET(node->hdr.count, ARCH_CONVERT)) {
 637                 tmp = (INT_GET(node->hdr.count, ARCH_CONVERT) - oldblk->index) * (uint)sizeof(*btree);
 638                 memmove(btree + 1, btree, tmp);
 639         }
 640         INT_SET(btree->hashval, ARCH_CONVERT, newblk->hashval);
 641         INT_SET(btree->before, ARCH_CONVERT, newblk->blkno);
 642         xfs_da_log_buf(state->args->trans, oldblk->bp,
 643                 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
 644         INT_MOD(node->hdr.count, ARCH_CONVERT, +1);
 645         xfs_da_log_buf(state->args->trans, oldblk->bp,
 646                 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
 647
 648         /*
 649          * Copy the last hash value from the oldblk to propagate upwards.
 650          */
 651         oldblk->hashval = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
 652 }
 653
 654 /*========================================================================
 655  * Routines used for shrinking the Btree.
 656  *========================================================================*/
 657
 658 /*
 659  * Deallocate an empty leaf node, remove it from its parent,
 660  * possibly deallocating that block, etc...
 661  */
 662 int
 663 xfs_da_join(xfs_da_state_t *state)
 664 {
 665         xfs_da_state_blk_t *drop_blk, *save_blk;
 666         int action, error;
 667
 668         action = 0;
 669         drop_blk = &state->path.blk[ state->path.active-1 ];
 670         save_blk = &state->altpath.blk[ state->path.active-1 ];
 671         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 672         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 673                drop_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp));
 674
 675         /*
 676          * Walk back up the tree joining/deallocating as necessary.
 677          * When we stop dropping blocks, break out.
 678          */
 679         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 680                  state->path.active--) {
 681                 /*
 682                  * See if we can combine the block with a neighbor.
 683                  *   (action == 0) => no options, just leave
 684                  *   (action == 1) => coalesce, then unlink
 685                  *   (action == 2) => block empty, unlink it
 686                  */
 687                 switch (drop_blk->magic) {
 688                 case XFS_ATTR_LEAF_MAGIC:
 689                         error = xfs_attr_leaf_toosmall(state, &action);
 690                         if (error)
 691                                 return(error);
 692                         if (action == 0)
 693                                 return(0);
 694                         xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
 695                         break;
 696                 case XFS_DIR_LEAF_MAGIC:
 697                         ASSERT(XFS_DIR_IS_V1(state->mp));
 698                         error = xfs_dir_leaf_toosmall(state, &action);
 699                         if (error)
 700                                 return(error);
 701                         if (action == 0)
 702                                 return(0);
 703                         xfs_dir_leaf_unbalance(state, drop_blk, save_blk);
 704                         break;
 705                 case XFS_DIR2_LEAFN_MAGIC:
 706                         ASSERT(XFS_DIR_IS_V2(state->mp));
 707                         error = xfs_dir2_leafn_toosmall(state, &action);
 708                         if (error)
 709                                 return error;
 710                         if (action == 0)
 711                                 return 0;
 712                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 713                         break;
 714                 case XFS_DA_NODE_MAGIC:
 715                         /*
 716                          * Remove the offending node, fixup hashvals,
 717                          * check for a toosmall neighbor.
 718                          */
 719                         xfs_da_node_remove(state, drop_blk);
 720                         xfs_da_fixhashpath(state, &state->path);
 721                         error = xfs_da_node_toosmall(state, &action);
 722                         if (error)
 723                                 return(error);
 724                         if (action == 0)
 725                                 return 0;
 726                         xfs_da_node_unbalance(state, drop_blk, save_blk);
 727                         break;
 728                 }
 729                 xfs_da_fixhashpath(state, &state->altpath);
 730                 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
 731                 xfs_da_state_kill_altpath(state);
 732                 if (error)
 733                         return(error);
 734                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
 735                                                          drop_blk->bp);
 736                 drop_blk->bp = NULL;
 737                 if (error)
 738                         return(error);
 739         }
 740         /*
 741          * We joined all the way to the top.  If it turns out that
 742          * we only have one entry in the root, make the child block
 743          * the new root.
 744          */
 745         xfs_da_node_remove(state, drop_blk);
 746         xfs_da_fixhashpath(state, &state->path);
 747         error = xfs_da_root_join(state, &state->path.blk[0]);
 748         return(error);
 749 }
 750
 751 /*
 752  * We have only one entry in the root.  Copy the only remaining child of
 753  * the old root to block 0 as the new root node.
 754  */
 755 STATIC int
 756 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
 757 {
 758         xfs_da_intnode_t *oldroot;
 759         /* REFERENCED */
 760         xfs_da_blkinfo_t *blkinfo;
 761         xfs_da_args_t *args;
 762         xfs_dablk_t child;
 763         xfs_dabuf_t *bp;
 764         int error;
 765
 766         args = state->args;
 767         ASSERT(args != NULL);
 768         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
 769         oldroot = root_blk->bp->data;
 770         ASSERT(INT_GET(oldroot->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 771         ASSERT(!oldroot->hdr.info.forw);
 772         ASSERT(!oldroot->hdr.info.back);
 773
 774         /*
 775          * If the root has more than one child, then don't do anything.
 776          */
 777         if (INT_GET(oldroot->hdr.count, ARCH_CONVERT) > 1)
 778                 return(0);
 779
 780         /*
 781          * Read in the (only) child block, then copy those bytes into
 782          * the root block's buffer and free the original child block.
 783          */
 784         child = INT_GET(oldroot->btree[ 0 ].before, ARCH_CONVERT);
 785         ASSERT(child != 0);
 786         error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
 787                                              args->whichfork);
 788         if (error)
 789                 return(error);
 790         ASSERT(bp != NULL);
 791         blkinfo = bp->data;
 792         if (INT_GET(oldroot->hdr.level, ARCH_CONVERT) == 1) {
 793                 ASSERT(INT_GET(blkinfo->magic, ARCH_CONVERT) == XFS_DIRX_LEAF_MAGIC(state->mp) ||
 794                        INT_GET(blkinfo->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC);
 795         } else {
 796                 ASSERT(INT_GET(blkinfo->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 797         }
 798         ASSERT(!blkinfo->forw);
 799         ASSERT(!blkinfo->back);
 800         memcpy(root_blk->bp->data, bp->data, state->blocksize);
 801         xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
 802         error = xfs_da_shrink_inode(args, child, bp);
 803         return(error);
 804 }
 805
 806 /*
 807  * Check a node block and its neighbors to see if the block should be
 808  * collapsed into one or the other neighbor.  Always keep the block
 809  * with the smaller block number.
 810  * If the current block is over 50% full, don't try to join it, return 0.
 811  * If the block is empty, fill in the state structure and return 2.
 812  * If it can be collapsed, fill in the state structure and return 1.
 813  * If nothing can be done, return 0.
 814  */
 815 STATIC int
 816 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
 817 {
 818         xfs_da_intnode_t *node;
 819         xfs_da_state_blk_t *blk;
 820         xfs_da_blkinfo_t *info;
 821         int count, forward, error, retval, i;
 822         xfs_dablk_t blkno;
 823         xfs_dabuf_t *bp;
 824
 825         /*
 826          * Check for the degenerate case of the block being over 50% full.
 827          * If so, it's not worth even looking to see if we might be able
 828          * to coalesce with a sibling.
 829          */
 830         blk = &state->path.blk[ state->path.active-1 ];
 831         info = blk->bp->data;
 832         ASSERT(INT_GET(info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 833         node = (xfs_da_intnode_t *)info;
 834         count = INT_GET(node->hdr.count, ARCH_CONVERT);
 835         if (count > (state->node_ents >> 1)) {
 836                 *action = 0;    /* blk over 50%, don't try to join */
 837                 return(0);      /* blk over 50%, don't try to join */
 838         }
 839
 840         /*
 841          * Check for the degenerate case of the block being empty.
 842          * If the block is empty, we'll simply delete it, no need to
 843          * coalesce it with a sibling block.  We choose (aribtrarily)
 844          * to merge with the forward block unless it is NULL.
 845          */
 846         if (count == 0) {
 847                 /*
 848                  * Make altpath point to the block we want to keep and
 849                  * path point to the block we want to drop (this one).
 850                  */
 851                 forward = info->forw;
 852                 memcpy(&state->altpath, &state->path, sizeof(state->path));
 853                 error = xfs_da_path_shift(state, &state->altpath, forward,
 854                                                  0, &retval);
 855                 if (error)
 856                         return(error);
 857                 if (retval) {
 858                         *action = 0;
 859                 } else {
 860                         *action = 2;
 861                 }
 862                 return(0);
 863         }
 864
 865         /*
 866          * Examine each sibling block to see if we can coalesce with
 867          * at least 25% free space to spare.  We need to figure out
 868          * whether to merge with the forward or the backward block.
 869          * We prefer coalescing with the lower numbered sibling so as
 870          * to shrink a directory over time.
 871          */
 872         /* start with smaller blk num */
 873         forward = (INT_GET(info->forw, ARCH_CONVERT)
 874                                 < INT_GET(info->back, ARCH_CONVERT));
 875         for (i = 0; i < 2; forward = !forward, i++) {
 876                 if (forward)
 877                         blkno = INT_GET(info->forw, ARCH_CONVERT);
 878                 else
 879                         blkno = INT_GET(info->back, ARCH_CONVERT);
 880                 if (blkno == 0)
 881                         continue;
 882                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
 883                                         blkno, -1, &bp, state->args->whichfork);
 884                 if (error)
 885                         return(error);
 886                 ASSERT(bp != NULL);
 887
 888                 node = (xfs_da_intnode_t *)info;
 889                 count  = state->node_ents;
 890                 count -= state->node_ents >> 2;
 891                 count -= INT_GET(node->hdr.count, ARCH_CONVERT);
 892                 node = bp->data;
 893                 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 894                 count -= INT_GET(node->hdr.count, ARCH_CONVERT);
 895                 xfs_da_brelse(state->args->trans, bp);
 896                 if (count >= 0)
 897                         break;  /* fits with at least 25% to spare */
 898         }
 899         if (i >= 2) {
 900                 *action = 0;
 901                 return(0);
 902         }
 903
 904         /*
 905          * Make altpath point to the block we want to keep (the lower
 906          * numbered block) and path point to the block we want to drop.
 907          */
 908         memcpy(&state->altpath, &state->path, sizeof(state->path));
 909         if (blkno < blk->blkno) {
 910                 error = xfs_da_path_shift(state, &state->altpath, forward,
 911                                                  0, &retval);
 912                 if (error) {
 913                         return(error);
 914                 }
 915                 if (retval) {
 916                         *action = 0;
 917                         return(0);
 918                 }
 919         } else {
 920                 error = xfs_da_path_shift(state, &state->path, forward,
 921                                                  0, &retval);
 922                 if (error) {
 923                         return(error);
 924                 }
 925                 if (retval) {
 926                         *action = 0;
 927                         return(0);
 928                 }
 929         }
 930         *action = 1;
 931         return(0);
 932 }
 933
 934 /*
 935  * Walk back up the tree adjusting hash values as necessary,
 936  * when we stop making changes, return.
 937  */
 938 void
 939 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
 940 {
 941         xfs_da_state_blk_t *blk;
 942         xfs_da_intnode_t *node;
 943         xfs_da_node_entry_t *btree;
 944         xfs_dahash_t lasthash=0;
 945         int level, count;
 946
 947         level = path->active-1;
 948         blk = &path->blk[ level ];
 949         switch (blk->magic) {
 950         case XFS_ATTR_LEAF_MAGIC:
 951                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
 952                 if (count == 0)
 953                         return;
 954                 break;
 955         case XFS_DIR_LEAF_MAGIC:
 956                 ASSERT(XFS_DIR_IS_V1(state->mp));
 957                 lasthash = xfs_dir_leaf_lasthash(blk->bp, &count);
 958                 if (count == 0)
 959                         return;
 960                 break;
 961         case XFS_DIR2_LEAFN_MAGIC:
 962                 ASSERT(XFS_DIR_IS_V2(state->mp));
 963                 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
 964                 if (count == 0)
 965                         return;
 966                 break;
 967         case XFS_DA_NODE_MAGIC:
 968                 lasthash = xfs_da_node_lasthash(blk->bp, &count);
 969                 if (count == 0)
 970                         return;
 971                 break;
 972         }
 973         for (blk--, level--; level >= 0; blk--, level--) {
 974                 node = blk->bp->data;
 975                 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
 976                 btree = &node->btree[ blk->index ];
 977                 if (INT_GET(btree->hashval, ARCH_CONVERT) == lasthash)
 978                         break;
 979                 blk->hashval = lasthash;
 980                 INT_SET(btree->hashval, ARCH_CONVERT, lasthash);
 981                 xfs_da_log_buf(state->args->trans, blk->bp,
 982                                   XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
 983
 984                 lasthash = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
 985         }
 986 }
 987
 988 /*
 989  * Remove an entry from an intermediate node.
 990  */
 991 STATIC void
 992 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
 993 {
 994         xfs_da_intnode_t *node;
 995         xfs_da_node_entry_t *btree;
 996         int tmp;
 997
 998         node = drop_blk->bp->data;
 999         ASSERT(drop_blk->index < INT_GET(node->hdr.count, ARCH_CONVERT));
1000         ASSERT(drop_blk->index >= 0);
1001
1002         /*
1003          * Copy over the offending entry, or just zero it out.
1004          */
1005         btree = &node->btree[drop_blk->index];
1006         if (drop_blk->index < (INT_GET(node->hdr.count, ARCH_CONVERT)-1)) {
1007                 tmp  = INT_GET(node->hdr.count, ARCH_CONVERT) - drop_blk->index - 1;
1008                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1009                 memmove(btree, btree + 1, tmp);
1010                 xfs_da_log_buf(state->args->trans, drop_blk->bp,
1011                     XFS_DA_LOGRANGE(node, btree, tmp));
1012                 btree = &node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ];
1013         }
1014         memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
1015         xfs_da_log_buf(state->args->trans, drop_blk->bp,
1016             XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
1017         INT_MOD(node->hdr.count, ARCH_CONVERT, -1);
1018         xfs_da_log_buf(state->args->trans, drop_blk->bp,
1019             XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
1020
1021         /*
1022          * Copy the last hash value from the block to propagate upwards.
1023          */
1024         btree--;
1025         drop_blk->hashval = INT_GET(btree->hashval, ARCH_CONVERT);
1026 }
1027
1028 /*
1029  * Unbalance the btree elements between two intermediate nodes,
1030  * move all Btree elements from one node into another.
1031  */
1032 STATIC void
1033 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1034                                      xfs_da_state_blk_t *save_blk)
1035 {
1036         xfs_da_intnode_t *drop_node, *save_node;
1037         xfs_da_node_entry_t *btree;
1038         int tmp;
1039         xfs_trans_t *tp;
1040
1041         drop_node = drop_blk->bp->data;
1042         save_node = save_blk->bp->data;
1043         ASSERT(INT_GET(drop_node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1044         ASSERT(INT_GET(save_node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1045         tp = state->args->trans;
1046
1047         /*
1048          * If the dying block has lower hashvals, then move all the
1049          * elements in the remaining block up to make a hole.
1050          */
1051         if ((INT_GET(drop_node->btree[ 0 ].hashval, ARCH_CONVERT) < INT_GET(save_node->btree[ 0 ].hashval, ARCH_CONVERT)) ||
1052             (INT_GET(drop_node->btree[ INT_GET(drop_node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
1053              INT_GET(save_node->btree[ INT_GET(save_node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))
1054         {
1055                 btree = &save_node->btree[ INT_GET(drop_node->hdr.count, ARCH_CONVERT) ];
1056                 tmp = INT_GET(save_node->hdr.count, ARCH_CONVERT) * (uint)sizeof(xfs_da_node_entry_t);
1057                 memmove(btree, &save_node->btree[0], tmp);
1058                 btree = &save_node->btree[0];
1059                 xfs_da_log_buf(tp, save_blk->bp,
1060                         XFS_DA_LOGRANGE(save_node, btree,
1061                                 (INT_GET(save_node->hdr.count, ARCH_CONVERT) + INT_GET(drop_node->hdr.count, ARCH_CONVERT)) *
1062                                 sizeof(xfs_da_node_entry_t)));
1063         } else {
1064                 btree = &save_node->btree[ INT_GET(save_node->hdr.count, ARCH_CONVERT) ];
1065                 xfs_da_log_buf(tp, save_blk->bp,
1066                         XFS_DA_LOGRANGE(save_node, btree,
1067                                 INT_GET(drop_node->hdr.count, ARCH_CONVERT) *
1068                                 sizeof(xfs_da_node_entry_t)));
1069         }
1070
1071         /*
1072          * Move all the B-tree elements from drop_blk to save_blk.
1073          */
1074         tmp = INT_GET(drop_node->hdr.count, ARCH_CONVERT) * (uint)sizeof(xfs_da_node_entry_t);
1075         memcpy(btree, &drop_node->btree[0], tmp);
1076         INT_MOD(save_node->hdr.count, ARCH_CONVERT, INT_GET(drop_node->hdr.count, ARCH_CONVERT));
1077
1078         xfs_da_log_buf(tp, save_blk->bp,
1079                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1080                         sizeof(save_node->hdr)));
1081
1082         /*
1083          * Save the last hashval in the remaining block for upward propagation.
1084          */
1085         save_blk->hashval = INT_GET(save_node->btree[ INT_GET(save_node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1086 }
1087
1088 /*========================================================================
1089  * Routines used for finding things in the Btree.
1090  *========================================================================*/
1091
1092 /*
1093  * Walk down the Btree looking for a particular filename, filling
1094  * in the state structure as we go.
1095  *
1096  * We will set the state structure to point to each of the elements
1097  * in each of the nodes where either the hashval is or should be.
1098  *
1099  * We support duplicate hashval's so for each entry in the current
1100  * node that could contain the desired hashval, descend.  This is a
1101  * pruned depth-first tree search.
1102  */
1103 int                                                     /* error */
1104 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1105 {
1106         xfs_da_state_blk_t *blk;
1107         xfs_da_blkinfo_t *curr;
1108         xfs_da_intnode_t *node;
1109         xfs_da_node_entry_t *btree;
1110         xfs_dablk_t blkno;
1111         int probe, span, max, error, retval;
1112         xfs_dahash_t hashval;
1113         xfs_da_args_t *args;
1114
1115         args = state->args;
1116
1117         /*
1118          * Descend thru the B-tree searching each level for the right
1119          * node to use, until the right hashval is found.
1120          */
1121         if (args->whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(state->mp))
1122                 blkno = state->mp->m_dirleafblk;
1123         else
1124                 blkno = 0;
1125         for (blk = &state->path.blk[0], state->path.active = 1;
1126                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1127                          blk++, state->path.active++) {
1128                 /*
1129                  * Read the next node down in the tree.
1130                  */
1131                 blk->blkno = blkno;
1132                 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1133                                         -1, &blk->bp, args->whichfork);
1134                 if (error) {
1135                         blk->blkno = 0;
1136                         state->path.active--;
1137                         return(error);
1138                 }
1139                 curr = blk->bp->data;
1140                 ASSERT(INT_GET(curr->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC ||
1141                        INT_GET(curr->magic, ARCH_CONVERT) == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1142                        INT_GET(curr->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC);
1143
1144                 /*
1145                  * Search an intermediate node for a match.
1146                  */
1147                 blk->magic = INT_GET(curr->magic, ARCH_CONVERT);
1148                 if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) {
1149                         node = blk->bp->data;
1150                         blk->hashval = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1151
1152                         /*
1153                          * Binary search.  (note: small blocks will skip loop)
1154                          */
1155                         max = INT_GET(node->hdr.count, ARCH_CONVERT);
1156                         probe = span = max / 2;
1157                         hashval = args->hashval;
1158                         for (btree = &node->btree[probe]; span > 4;
1159                                    btree = &node->btree[probe]) {
1160                                 span /= 2;
1161                                 if (INT_GET(btree->hashval, ARCH_CONVERT) < hashval)
1162                                         probe += span;
1163                                 else if (INT_GET(btree->hashval, ARCH_CONVERT) > hashval)
1164                                         probe -= span;
1165                                 else
1166                                         break;
1167                         }
1168                         ASSERT((probe >= 0) && (probe < max));
1169                         ASSERT((span <= 4) || (INT_GET(btree->hashval, ARCH_CONVERT) == hashval));
1170
1171                         /*
1172                          * Since we may have duplicate hashval's, find the first
1173                          * matching hashval in the node.
1174                          */
1175                         while ((probe > 0) && (INT_GET(btree->hashval, ARCH_CONVERT) >= hashval)) {
1176                                 btree--;
1177                                 probe--;
1178                         }
1179                         while ((probe < max) && (INT_GET(btree->hashval, ARCH_CONVERT) < hashval)) {
1180                                 btree++;
1181                                 probe++;
1182                         }
1183
1184                         /*
1185                          * Pick the right block to descend on.
1186                          */
1187                         if (probe == max) {
1188                                 blk->index = max-1;
1189                                 blkno = INT_GET(node->btree[ max-1 ].before, ARCH_CONVERT);
1190                         } else {
1191                                 blk->index = probe;
1192                                 blkno = INT_GET(btree->before, ARCH_CONVERT);
1193                         }
1194                 }
1195                 else if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC) {
1196                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1197                         break;
1198                 }
1199                 else if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_DIR_LEAF_MAGIC) {
1200                         blk->hashval = xfs_dir_leaf_lasthash(blk->bp, NULL);
1201                         break;
1202                 }
1203                 else if (INT_GET(curr->magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC) {
1204                         blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1205                         break;
1206                 }
1207         }
1208
1209         /*
1210          * A leaf block that ends in the hashval that we are interested in
1211          * (final hashval == search hashval) means that the next block may
1212          * contain more entries with the same hashval, shift upward to the
1213          * next leaf and keep searching.
1214          */
1215         for (;;) {
1216                 if (blk->magic == XFS_DIR_LEAF_MAGIC) {
1217                         ASSERT(XFS_DIR_IS_V1(state->mp));
1218                         retval = xfs_dir_leaf_lookup_int(blk->bp, args,
1219                                                                   &blk->index);
1220                 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1221                         ASSERT(XFS_DIR_IS_V2(state->mp));
1222                         retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1223                                                         &blk->index, state);
1224                 }
1225                 else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1226                         retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1227                         blk->index = args->index;
1228                         args->blkno = blk->blkno;
1229                 }
1230                 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1231                     (blk->hashval == args->hashval)) {
1232                         error = xfs_da_path_shift(state, &state->path, 1, 1,
1233                                                          &retval);
1234                         if (error)
1235                                 return(error);
1236                         if (retval == 0) {
1237                                 continue;
1238                         }
1239                         else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1240                                 /* path_shift() gives ENOENT */
1241                                 retval = XFS_ERROR(ENOATTR);
1242                         }
1243                 }
1244                 break;
1245         }
1246         *result = retval;
1247         return(0);
1248 }
1249
1250 /*========================================================================
1251  * Utility routines.
1252  *========================================================================*/
1253
1254 /*
1255  * Link a new block into a doubly linked list of blocks (of whatever type).
1256  */
1257 int                                                     /* error */
1258 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1259                                xfs_da_state_blk_t *new_blk)
1260 {
1261         xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1262         xfs_da_args_t *args;
1263         int before=0, error;
1264         xfs_dabuf_t *bp;
1265
1266         /*
1267          * Set up environment.
1268          */
1269         args = state->args;
1270         ASSERT(args != NULL);
1271         old_info = old_blk->bp->data;
1272         new_info = new_blk->bp->data;
1273         ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1274                old_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1275                old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1276         ASSERT(old_blk->magic == INT_GET(old_info->magic, ARCH_CONVERT));
1277         ASSERT(new_blk->magic == INT_GET(new_info->magic, ARCH_CONVERT));
1278         ASSERT(old_blk->magic == new_blk->magic);
1279
1280         switch (old_blk->magic) {
1281         case XFS_ATTR_LEAF_MAGIC:
1282                 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1283                 break;
1284         case XFS_DIR_LEAF_MAGIC:
1285                 ASSERT(XFS_DIR_IS_V1(state->mp));
1286                 before = xfs_dir_leaf_order(old_blk->bp, new_blk->bp);
1287                 break;
1288         case XFS_DIR2_LEAFN_MAGIC:
1289                 ASSERT(XFS_DIR_IS_V2(state->mp));
1290                 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1291                 break;
1292         case XFS_DA_NODE_MAGIC:
1293                 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1294                 break;
1295         }
1296
1297         /*
1298          * Link blocks in appropriate order.
1299          */
1300         if (before) {
1301                 /*
1302                  * Link new block in before existing block.
1303                  */
1304                 INT_SET(new_info->forw, ARCH_CONVERT, old_blk->blkno);
1305                 new_info->back = old_info->back; /* INT_: direct copy */
1306                 if (INT_GET(old_info->back, ARCH_CONVERT)) {
1307                         error = xfs_da_read_buf(args->trans, args->dp,
1308                                                 INT_GET(old_info->back,
1309                                                         ARCH_CONVERT), -1, &bp,
1310                                                 args->whichfork);
1311                         if (error)
1312                                 return(error);
1313                         ASSERT(bp != NULL);
1314                         tmp_info = bp->data;
1315                         ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT) == INT_GET(old_info->magic, ARCH_CONVERT));
1316                         ASSERT(INT_GET(tmp_info->forw, ARCH_CONVERT) == old_blk->blkno);
1317                         INT_SET(tmp_info->forw, ARCH_CONVERT, new_blk->blkno);
1318                         xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1319                         xfs_da_buf_done(bp);
1320                 }
1321                 INT_SET(old_info->back, ARCH_CONVERT, new_blk->blkno);
1322         } else {
1323                 /*
1324                  * Link new block in after existing block.
1325                  */
1326                 new_info->forw = old_info->forw; /* INT_: direct copy */
1327                 INT_SET(new_info->back, ARCH_CONVERT, old_blk->blkno);
1328                 if (INT_GET(old_info->forw, ARCH_CONVERT)) {
1329                         error = xfs_da_read_buf(args->trans, args->dp,
1330                                                 INT_GET(old_info->forw, ARCH_CONVERT), -1, &bp,
1331                                                 args->whichfork);
1332                         if (error)
1333                                 return(error);
1334                         ASSERT(bp != NULL);
1335                         tmp_info = bp->data;
1336                         ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT)
1337                                     == INT_GET(old_info->magic, ARCH_CONVERT));
1338                         ASSERT(INT_GET(tmp_info->back, ARCH_CONVERT)
1339                                     == old_blk->blkno);
1340                         INT_SET(tmp_info->back, ARCH_CONVERT, new_blk->blkno);
1341                         xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1342                         xfs_da_buf_done(bp);
1343                 }
1344                 INT_SET(old_info->forw, ARCH_CONVERT, new_blk->blkno);
1345         }
1346
1347         xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1348         xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1349         return(0);
1350 }
1351
1352 /*
1353  * Compare two intermediate nodes for "order".
1354  */
1355 STATIC int
1356 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1357 {
1358         xfs_da_intnode_t *node1, *node2;
1359
1360         node1 = node1_bp->data;
1361         node2 = node2_bp->data;
1362         ASSERT((INT_GET(node1->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) &&
1363                (INT_GET(node2->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC));
1364         if ((INT_GET(node1->hdr.count, ARCH_CONVERT) > 0) && (INT_GET(node2->hdr.count, ARCH_CONVERT) > 0) &&
1365             ((INT_GET(node2->btree[ 0 ].hashval, ARCH_CONVERT) <
1366               INT_GET(node1->btree[ 0 ].hashval, ARCH_CONVERT)) ||
1367              (INT_GET(node2->btree[ INT_GET(node2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
1368               INT_GET(node1->btree[ INT_GET(node1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))) {
1369                 return(1);
1370         }
1371         return(0);
1372 }
1373
1374 /*
1375  * Pick up the last hashvalue from an intermediate node.
1376  */
1377 STATIC uint
1378 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1379 {
1380         xfs_da_intnode_t *node;
1381
1382         node = bp->data;
1383         ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1384         if (count)
1385                 *count = INT_GET(node->hdr.count, ARCH_CONVERT);
1386         if (!node->hdr.count)
1387                 return(0);
1388         return(INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT));
1389 }
1390
1391 /*
1392  * Unlink a block from a doubly linked list of blocks.
1393  */
1394 STATIC int                                              /* error */
1395 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1396                                  xfs_da_state_blk_t *save_blk)
1397 {
1398         xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1399         xfs_da_args_t *args;
1400         xfs_dabuf_t *bp;
1401         int error;
1402
1403         /*
1404          * Set up environment.
1405          */
1406         args = state->args;
1407         ASSERT(args != NULL);
1408         save_info = save_blk->bp->data;
1409         drop_info = drop_blk->bp->data;
1410         ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1411                save_blk->magic == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1412                save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1413         ASSERT(save_blk->magic == INT_GET(save_info->magic, ARCH_CONVERT));
1414         ASSERT(drop_blk->magic == INT_GET(drop_info->magic, ARCH_CONVERT));
1415         ASSERT(save_blk->magic == drop_blk->magic);
1416         ASSERT((INT_GET(save_info->forw, ARCH_CONVERT) == drop_blk->blkno) ||
1417                (INT_GET(save_info->back, ARCH_CONVERT) == drop_blk->blkno));
1418         ASSERT((INT_GET(drop_info->forw, ARCH_CONVERT) == save_blk->blkno) ||
1419                (INT_GET(drop_info->back, ARCH_CONVERT) == save_blk->blkno));
1420
1421         /*
1422          * Unlink the leaf block from the doubly linked chain of leaves.
1423          */
1424         if (INT_GET(save_info->back, ARCH_CONVERT) == drop_blk->blkno) {
1425                 save_info->back = drop_info->back; /* INT_: direct copy */
1426                 if (INT_GET(drop_info->back, ARCH_CONVERT)) {
1427                         error = xfs_da_read_buf(args->trans, args->dp,
1428                                                 INT_GET(drop_info->back,
1429                                                         ARCH_CONVERT), -1, &bp,
1430                                                 args->whichfork);
1431                         if (error)
1432                                 return(error);
1433                         ASSERT(bp != NULL);
1434                         tmp_info = bp->data;
1435                         ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT) == INT_GET(save_info->magic, ARCH_CONVERT));
1436                         ASSERT(INT_GET(tmp_info->forw, ARCH_CONVERT) == drop_blk->blkno);
1437                         INT_SET(tmp_info->forw, ARCH_CONVERT, save_blk->blkno);
1438                         xfs_da_log_buf(args->trans, bp, 0,
1439                                                     sizeof(*tmp_info) - 1);
1440                         xfs_da_buf_done(bp);
1441                 }
1442         } else {
1443                 save_info->forw = drop_info->forw; /* INT_: direct copy */
1444                 if (INT_GET(drop_info->forw, ARCH_CONVERT)) {
1445                         error = xfs_da_read_buf(args->trans, args->dp,
1446                                                 INT_GET(drop_info->forw, ARCH_CONVERT), -1, &bp,
1447                                                 args->whichfork);
1448                         if (error)
1449                                 return(error);
1450                         ASSERT(bp != NULL);
1451                         tmp_info = bp->data;
1452                         ASSERT(INT_GET(tmp_info->magic, ARCH_CONVERT)
1453                                     == INT_GET(save_info->magic, ARCH_CONVERT));
1454                         ASSERT(INT_GET(tmp_info->back, ARCH_CONVERT)
1455                                     == drop_blk->blkno);
1456                         INT_SET(tmp_info->back, ARCH_CONVERT, save_blk->blkno);
1457                         xfs_da_log_buf(args->trans, bp, 0,
1458                                                     sizeof(*tmp_info) - 1);
1459                         xfs_da_buf_done(bp);
1460                 }
1461         }
1462
1463         xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1464         return(0);
1465 }
1466
1467 /*
1468  * Move a path "forward" or "!forward" one block at the current level.
1469  *
1470  * This routine will adjust a "path" to point to the next block
1471  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1472  * Btree, including updating pointers to the intermediate nodes between
1473  * the new bottom and the root.
1474  */
1475 int                                                     /* error */
1476 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1477                                  int forward, int release, int *result)
1478 {
1479         xfs_da_state_blk_t *blk;
1480         xfs_da_blkinfo_t *info;
1481         xfs_da_intnode_t *node;
1482         xfs_da_args_t *args;
1483         xfs_dablk_t blkno=0;
1484         int level, error;
1485
1486         /*
1487          * Roll up the Btree looking for the first block where our
1488          * current index is not at the edge of the block.  Note that
1489          * we skip the bottom layer because we want the sibling block.
1490          */
1491         args = state->args;
1492         ASSERT(args != NULL);
1493         ASSERT(path != NULL);
1494         ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1495         level = (path->active-1) - 1;   /* skip bottom layer in path */
1496         for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1497                 ASSERT(blk->bp != NULL);
1498                 node = blk->bp->data;
1499                 ASSERT(INT_GET(node->hdr.info.magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1500                 if (forward && (blk->index < INT_GET(node->hdr.count, ARCH_CONVERT)-1)) {
1501                         blk->index++;
1502                         blkno = INT_GET(node->btree[ blk->index ].before, ARCH_CONVERT);
1503                         break;
1504                 } else if (!forward && (blk->index > 0)) {
1505                         blk->index--;
1506                         blkno = INT_GET(node->btree[ blk->index ].before, ARCH_CONVERT);
1507                         break;
1508                 }
1509         }
1510         if (level < 0) {
1511                 *result = XFS_ERROR(ENOENT);    /* we're out of our tree */
1512                 ASSERT(args->oknoent);
1513                 return(0);
1514         }
1515
1516         /*
1517          * Roll down the edge of the subtree until we reach the
1518          * same depth we were at originally.
1519          */
1520         for (blk++, level++; level < path->active; blk++, level++) {
1521                 /*
1522                  * Release the old block.
1523                  * (if it's dirty, trans won't actually let go)
1524                  */
1525                 if (release)
1526                         xfs_da_brelse(args->trans, blk->bp);
1527
1528                 /*
1529                  * Read the next child block.
1530                  */
1531                 blk->blkno = blkno;
1532                 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1533                                                      &blk->bp, args->whichfork);
1534                 if (error)
1535                         return(error);
1536                 ASSERT(blk->bp != NULL);
1537                 info = blk->bp->data;
1538                 ASSERT(INT_GET(info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC ||
1539                        INT_GET(info->magic, ARCH_CONVERT) == XFS_DIRX_LEAF_MAGIC(state->mp) ||
1540                        INT_GET(info->magic, ARCH_CONVERT) == XFS_ATTR_LEAF_MAGIC);
1541                 blk->magic = INT_GET(info->magic, ARCH_CONVERT);
1542                 if (INT_GET(info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC) {
1543                         node = (xfs_da_intnode_t *)info;
1544                         blk->hashval = INT_GET(node->btree[ INT_GET(node->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1545                         if (forward)
1546                                 blk->index = 0;
1547                         else
1548                                 blk->index = INT_GET(node->hdr.count, ARCH_CONVERT)-1;
1549                         blkno = INT_GET(node->btree[ blk->index ].before, ARCH_CONVERT);
1550                 } else {
1551                         ASSERT(level == path->active-1);
1552                         blk->index = 0;
1553                         switch(blk->magic) {
1554                         case XFS_ATTR_LEAF_MAGIC:
1555                                 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1556                                                                       NULL);
1557                                 break;
1558                         case XFS_DIR_LEAF_MAGIC:
1559                                 ASSERT(XFS_DIR_IS_V1(state->mp));
1560                                 blk->hashval = xfs_dir_leaf_lasthash(blk->bp,
1561                                                                      NULL);
1562                                 break;
1563                         case XFS_DIR2_LEAFN_MAGIC:
1564                                 ASSERT(XFS_DIR_IS_V2(state->mp));
1565                                 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1566                                                                        NULL);
1567                                 break;
1568                         default:
1569                                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1570                                        blk->magic ==
1571                                        XFS_DIRX_LEAF_MAGIC(state->mp));
1572                                 break;
1573                         }
1574                 }
1575         }
1576         *result = 0;
1577         return(0);
1578 }
1579
1580
1581 /*========================================================================
1582  * Utility routines.
1583  *========================================================================*/
1584
1585 /*
1586  * Implement a simple hash on a character string.
1587  * Rotate the hash value by 7 bits, then XOR each character in.
1588  * This is implemented with some source-level loop unrolling.
1589  */
1590 xfs_dahash_t
1591 xfs_da_hashname(const uchar_t *name, int namelen)
1592 {
1593         xfs_dahash_t hash;
1594
1595 #ifdef SLOWVERSION
1596         /*
1597          * This is the old one-byte-at-a-time version.
1598          */
1599         for (hash = 0; namelen > 0; namelen--)
1600                 hash = *name++ ^ rol32(hash, 7);
1601
1602         return(hash);
1603 #else
1604         /*
1605          * Do four characters at a time as long as we can.
1606          */
1607         for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1608                 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1609                        (name[3] << 0) ^ rol32(hash, 7 * 4);
1610
1611         /*
1612          * Now do the rest of the characters.
1613          */
1614         switch (namelen) {
1615         case 3:
1616                 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1617                        rol32(hash, 7 * 3);
1618         case 2:
1619                 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1620         case 1:
1621                 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1622         case 0:
1623                 return hash;
1624         }
1625         /* NOTREACHED */
1626 #endif
1627         return 0; /* keep gcc happy */
1628 }
1629
1630 /*
1631  * Add a block to the btree ahead of the file.
1632  * Return the new block number to the caller.
1633  */
1634 int
1635 xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1636 {
1637         xfs_fileoff_t bno, b;
1638         xfs_bmbt_irec_t map;
1639         xfs_bmbt_irec_t *mapp;
1640         xfs_inode_t *dp;
1641         int nmap, error, w, count, c, got, i, mapi;
1642         xfs_fsize_t size;
1643         xfs_trans_t *tp;
1644         xfs_mount_t *mp;
1645
1646         dp = args->dp;
1647         mp = dp->i_mount;
1648         w = args->whichfork;
1649         tp = args->trans;
1650         /*
1651          * For new directories adjust the file offset and block count.
1652          */
1653         if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp)) {
1654                 bno = mp->m_dirleafblk;
1655                 count = mp->m_dirblkfsbs;
1656         } else {
1657                 bno = 0;
1658                 count = 1;
1659         }
1660         /*
1661          * Find a spot in the file space to put the new block.
1662          */
1663         if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w))) {
1664                 return error;
1665         }
1666         if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
1667                 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1668         /*
1669          * Try mapping it in one filesystem block.
1670          */
1671         nmap = 1;
1672         ASSERT(args->firstblock != NULL);
1673         if ((error = xfs_bmapi(tp, dp, bno, count,
1674                         XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1675                         XFS_BMAPI_CONTIG,
1676                         args->firstblock, args->total, &map, &nmap,
1677                         args->flist))) {
1678                 return error;
1679         }
1680         ASSERT(nmap <= 1);
1681         if (nmap == 1) {
1682                 mapp = &map;
1683                 mapi = 1;
1684         }
1685         /*
1686          * If we didn't get it and the block might work if fragmented,
1687          * try without the CONTIG flag.  Loop until we get it all.
1688          */
1689         else if (nmap == 0 && count > 1) {
1690                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1691                 for (b = bno, mapi = 0; b < bno + count; ) {
1692                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1693                         c = (int)(bno + count - b);
1694                         if ((error = xfs_bmapi(tp, dp, b, c,
1695                                         XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|
1696                                         XFS_BMAPI_METADATA,
1697                                         args->firstblock, args->total,
1698                                         &mapp[mapi], &nmap, args->flist))) {
1699                                 kmem_free(mapp, sizeof(*mapp) * count);
1700                                 return error;
1701                         }
1702                         if (nmap < 1)
1703                                 break;
1704                         mapi += nmap;
1705                         b = mapp[mapi - 1].br_startoff +
1706                             mapp[mapi - 1].br_blockcount;
1707                 }
1708         } else {
1709                 mapi = 0;
1710                 mapp = NULL;
1711         }
1712         /*
1713          * Count the blocks we got, make sure it matches the total.
1714          */
1715         for (i = 0, got = 0; i < mapi; i++)
1716                 got += mapp[i].br_blockcount;
1717         if (got != count || mapp[0].br_startoff != bno ||
1718             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1719             bno + count) {
1720                 if (mapp != &map)
1721                         kmem_free(mapp, sizeof(*mapp) * count);
1722                 return XFS_ERROR(ENOSPC);
1723         }
1724         if (mapp != &map)
1725                 kmem_free(mapp, sizeof(*mapp) * count);
1726         *new_blkno = (xfs_dablk_t)bno;
1727         /*
1728          * For version 1 directories, adjust the file size if it changed.
1729          */
1730         if (w == XFS_DATA_FORK && XFS_DIR_IS_V1(mp)) {
1731                 ASSERT(mapi == 1);
1732                 if ((error = xfs_bmap_last_offset(tp, dp, &bno, w)))
1733                         return error;
1734                 size = XFS_FSB_TO_B(mp, bno);
1735                 if (size != dp->i_d.di_size) {
1736                         dp->i_d.di_size = size;
1737                         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1738                 }
1739         }
1740         return 0;
1741 }
1742
1743 /*
1744  * Ick.  We need to always be able to remove a btree block, even
1745  * if there's no space reservation because the filesystem is full.
1746  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1747  * It swaps the target block with the last block in the file.  The
1748  * last block in the file can always be removed since it can't cause
1749  * a bmap btree split to do that.
1750  */
1751 STATIC int
1752 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1753                       xfs_dabuf_t **dead_bufp)
1754 {
1755         xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1756         xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1757         xfs_fileoff_t lastoff;
1758         xfs_inode_t *ip;
1759         xfs_trans_t *tp;
1760         xfs_mount_t *mp;
1761         int error, w, entno, level, dead_level;
1762         xfs_da_blkinfo_t *dead_info, *sib_info;
1763         xfs_da_intnode_t *par_node, *dead_node;
1764         xfs_dir_leafblock_t *dead_leaf;
1765         xfs_dir2_leaf_t *dead_leaf2;
1766         xfs_dahash_t dead_hash;
1767
1768         dead_buf = *dead_bufp;
1769         dead_blkno = *dead_blknop;
1770         tp = args->trans;
1771         ip = args->dp;
1772         w = args->whichfork;
1773         ASSERT(w == XFS_DATA_FORK);
1774         mp = ip->i_mount;
1775         if (XFS_DIR_IS_V2(mp)) {
1776                 lastoff = mp->m_dirfreeblk;
1777                 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1778         } else
1779                 error = xfs_bmap_last_offset(tp, ip, &lastoff, w);
1780         if (error)
1781                 return error;
1782         if (unlikely(lastoff == 0)) {
1783                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1784                                  mp);
1785                 return XFS_ERROR(EFSCORRUPTED);
1786         }
1787         /*
1788          * Read the last block in the btree space.
1789          */
1790         last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1791         if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1792                 return error;
1793         /*
1794          * Copy the last block into the dead buffer and log it.
1795          */
1796         memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1797         xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1798         dead_info = dead_buf->data;
1799         /*
1800          * Get values from the moved block.
1801          */
1802         if (INT_GET(dead_info->magic, ARCH_CONVERT) == XFS_DIR_LEAF_MAGIC) {
1803                 ASSERT(XFS_DIR_IS_V1(mp));
1804                 dead_leaf = (xfs_dir_leafblock_t *)dead_info;
1805                 dead_level = 0;
1806                 dead_hash =
1807                         INT_GET(dead_leaf->entries[INT_GET(dead_leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1808         } else if (INT_GET(dead_info->magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC) {
1809                 ASSERT(XFS_DIR_IS_V2(mp));
1810                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1811                 dead_level = 0;
1812                 dead_hash = INT_GET(dead_leaf2->ents[INT_GET(dead_leaf2->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1813         } else {
1814                 ASSERT(INT_GET(dead_info->magic, ARCH_CONVERT) == XFS_DA_NODE_MAGIC);
1815                 dead_node = (xfs_da_intnode_t *)dead_info;
1816                 dead_level = INT_GET(dead_node->hdr.level, ARCH_CONVERT);
1817                 dead_hash = INT_GET(dead_node->btree[INT_GET(dead_node->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT);
1818         }
1819         sib_buf = par_buf = NULL;
1820         /*
1821          * If the moved block has a left sibling, fix up the pointers.
1822          */
1823         if ((sib_blkno = INT_GET(dead_info->back, ARCH_CONVERT))) {
1824                 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1825                         goto done;
1826                 sib_info = sib_buf->data;
1827                 if (unlikely(
1828                     INT_GET(sib_info->forw, ARCH_CONVERT) != last_blkno ||
1829                     INT_GET(sib_info->magic, ARCH_CONVERT) != INT_GET(dead_info->magic, ARCH_CONVERT))) {
1830                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1831                                          XFS_ERRLEVEL_LOW, mp);
1832                         error = XFS_ERROR(EFSCORRUPTED);
1833                         goto done;
1834                 }
1835                 INT_SET(sib_info->forw, ARCH_CONVERT, dead_blkno);
1836                 xfs_da_log_buf(tp, sib_buf,
1837                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1838                                         sizeof(sib_info->forw)));
1839                 xfs_da_buf_done(sib_buf);
1840                 sib_buf = NULL;
1841         }
1842         /*
1843          * If the moved block has a right sibling, fix up the pointers.
1844          */
1845         if ((sib_blkno = INT_GET(dead_info->forw, ARCH_CONVERT))) {
1846                 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1847                         goto done;
1848                 sib_info = sib_buf->data;
1849                 if (unlikely(
1850                        INT_GET(sib_info->back, ARCH_CONVERT) != last_blkno
1851                     || INT_GET(sib_info->magic, ARCH_CONVERT)
1852                                 != INT_GET(dead_info->magic, ARCH_CONVERT))) {
1853                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1854                                          XFS_ERRLEVEL_LOW, mp);
1855                         error = XFS_ERROR(EFSCORRUPTED);
1856                         goto done;
1857                 }
1858                 INT_SET(sib_info->back, ARCH_CONVERT, dead_blkno);
1859                 xfs_da_log_buf(tp, sib_buf,
1860                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1861                                         sizeof(sib_info->back)));
1862                 xfs_da_buf_done(sib_buf);
1863                 sib_buf = NULL;
1864         }
1865         par_blkno = XFS_DIR_IS_V1(mp) ? 0 : mp->m_dirleafblk;
1866         level = -1;
1867         /*
1868          * Walk down the tree looking for the parent of the moved block.
1869          */
1870         for (;;) {
1871                 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1872                         goto done;
1873                 par_node = par_buf->data;
1874                 if (unlikely(
1875                     INT_GET(par_node->hdr.info.magic, ARCH_CONVERT) != XFS_DA_NODE_MAGIC ||
1876                     (level >= 0 && level != INT_GET(par_node->hdr.level, ARCH_CONVERT) + 1))) {
1877                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1878                                          XFS_ERRLEVEL_LOW, mp);
1879                         error = XFS_ERROR(EFSCORRUPTED);
1880                         goto done;
1881                 }
1882                 level = INT_GET(par_node->hdr.level, ARCH_CONVERT);
1883                 for (entno = 0;
1884                      entno < INT_GET(par_node->hdr.count, ARCH_CONVERT) &&
1885                      INT_GET(par_node->btree[entno].hashval, ARCH_CONVERT) < dead_hash;
1886                      entno++)
1887                         continue;
1888                 if (unlikely(entno == INT_GET(par_node->hdr.count, ARCH_CONVERT))) {
1889                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1890                                          XFS_ERRLEVEL_LOW, mp);
1891                         error = XFS_ERROR(EFSCORRUPTED);
1892                         goto done;
1893                 }
1894                 par_blkno = INT_GET(par_node->btree[entno].before, ARCH_CONVERT);
1895                 if (level == dead_level + 1)
1896                         break;
1897                 xfs_da_brelse(tp, par_buf);
1898                 par_buf = NULL;
1899         }
1900         /*
1901          * We're in the right parent block.
1902          * Look for the right entry.
1903          */
1904         for (;;) {
1905                 for (;
1906                      entno < INT_GET(par_node->hdr.count, ARCH_CONVERT) &&
1907                      INT_GET(par_node->btree[entno].before, ARCH_CONVERT) != last_blkno;
1908                      entno++)
1909                         continue;
1910                 if (entno < INT_GET(par_node->hdr.count, ARCH_CONVERT))
1911                         break;
1912                 par_blkno = INT_GET(par_node->hdr.info.forw, ARCH_CONVERT);
1913                 xfs_da_brelse(tp, par_buf);
1914                 par_buf = NULL;
1915                 if (unlikely(par_blkno == 0)) {
1916                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1917                                          XFS_ERRLEVEL_LOW, mp);
1918                         error = XFS_ERROR(EFSCORRUPTED);
1919                         goto done;
1920                 }
1921                 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1922                         goto done;
1923                 par_node = par_buf->data;
1924                 if (unlikely(
1925                     INT_GET(par_node->hdr.level, ARCH_CONVERT) != level ||
1926                     INT_GET(par_node->hdr.info.magic, ARCH_CONVERT) != XFS_DA_NODE_MAGIC)) {
1927                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1928                                          XFS_ERRLEVEL_LOW, mp);
1929                         error = XFS_ERROR(EFSCORRUPTED);
1930                         goto done;
1931                 }
1932                 entno = 0;
1933         }
1934         /*
1935          * Update the parent entry pointing to the moved block.
1936          */
1937         INT_SET(par_node->btree[entno].before, ARCH_CONVERT, dead_blkno);
1938         xfs_da_log_buf(tp, par_buf,
1939                 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1940                                 sizeof(par_node->btree[entno].before)));
1941         xfs_da_buf_done(par_buf);
1942         xfs_da_buf_done(dead_buf);
1943         *dead_blknop = last_blkno;
1944         *dead_bufp = last_buf;
1945         return 0;
1946 done:
1947         if (par_buf)
1948                 xfs_da_brelse(tp, par_buf);
1949         if (sib_buf)
1950                 xfs_da_brelse(tp, sib_buf);
1951         xfs_da_brelse(tp, last_buf);
1952         return error;
1953 }
1954
1955 /*
1956  * Remove a btree block from a directory or attribute.
1957  */
1958 int
1959 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1960                     xfs_dabuf_t *dead_buf)
1961 {
1962         xfs_inode_t *dp;
1963         int done, error, w, count;
1964         xfs_fileoff_t bno;
1965         xfs_fsize_t size;
1966         xfs_trans_t *tp;
1967         xfs_mount_t *mp;
1968
1969         dp = args->dp;
1970         w = args->whichfork;
1971         tp = args->trans;
1972         mp = dp->i_mount;
1973         if (w == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
1974                 count = mp->m_dirblkfsbs;
1975         else
1976                 count = 1;
1977         for (;;) {
1978                 /*
1979                  * Remove extents.  If we get ENOSPC for a dir we have to move
1980                  * the last block to the place we want to kill.
1981                  */
1982                 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1983                                 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_METADATA,
1984                                 0, args->firstblock, args->flist,
1985                                 &done)) == ENOSPC) {
1986                         if (w != XFS_DATA_FORK)
1987                                 goto done;
1988                         if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1989                                         &dead_buf)))
1990                                 goto done;
1991                 } else if (error)
1992                         goto done;
1993                 else
1994                         break;
1995         }
1996         ASSERT(done);
1997         xfs_da_binval(tp, dead_buf);
1998         /*
1999          * Adjust the directory size for version 1.
2000          */
2001         if (w == XFS_DATA_FORK && XFS_DIR_IS_V1(mp)) {
2002                 if ((error = xfs_bmap_last_offset(tp, dp, &bno, w)))
2003                         return error;
2004                 size = XFS_FSB_TO_B(dp->i_mount, bno);
2005                 if (size != dp->i_d.di_size) {
2006                         dp->i_d.di_size = size;
2007                         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2008                 }
2009         }
2010         return 0;
2011 done:
2012         xfs_da_binval(tp, dead_buf);
2013         return error;
2014 }
2015
2016 /*
2017  * See if the mapping(s) for this btree block are valid, i.e.
2018  * don't contain holes, are logically contiguous, and cover the whole range.
2019  */
2020 STATIC int
2021 xfs_da_map_covers_blocks(
2022         int             nmap,
2023         xfs_bmbt_irec_t *mapp,
2024         xfs_dablk_t     bno,
2025         int             count)
2026 {
2027         int             i;
2028         xfs_fileoff_t   off;
2029
2030         for (i = 0, off = bno; i < nmap; i++) {
2031                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2032                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
2033                         return 0;
2034                 }
2035                 if (off != mapp[i].br_startoff) {
2036                         return 0;
2037                 }
2038                 off += mapp[i].br_blockcount;
2039         }
2040         return off == bno + count;
2041 }
2042
2043 /*
2044  * Make a dabuf.
2045  * Used for get_buf, read_buf, read_bufr, and reada_buf.
2046  */
2047 STATIC int
2048 xfs_da_do_buf(
2049         xfs_trans_t     *trans,
2050         xfs_inode_t     *dp,
2051         xfs_dablk_t     bno,
2052         xfs_daddr_t     *mappedbnop,
2053         xfs_dabuf_t     **bpp,
2054         int             whichfork,
2055         int             caller,
2056         inst_t          *ra)
2057 {
2058         xfs_buf_t       *bp = NULL;
2059         xfs_buf_t       **bplist;
2060         int             error=0;
2061         int             i;
2062         xfs_bmbt_irec_t map;
2063         xfs_bmbt_irec_t *mapp;
2064         xfs_daddr_t     mappedbno;
2065         xfs_mount_t     *mp;
2066         int             nbplist=0;
2067         int             nfsb;
2068         int             nmap;
2069         xfs_dabuf_t     *rbp;
2070
2071         mp = dp->i_mount;
2072         if (whichfork == XFS_DATA_FORK && XFS_DIR_IS_V2(mp))
2073                 nfsb = mp->m_dirblkfsbs;
2074         else
2075                 nfsb = 1;
2076         mappedbno = *mappedbnop;
2077         /*
2078          * Caller doesn't have a mapping.  -2 means don't complain
2079          * if we land in a hole.
2080          */
2081         if (mappedbno == -1 || mappedbno == -2) {
2082                 /*
2083                  * Optimize the one-block case.
2084                  */
2085                 if (nfsb == 1) {
2086                         xfs_fsblock_t   fsb;
2087
2088                         if ((error =
2089                             xfs_bmapi_single(trans, dp, whichfork, &fsb,
2090                                     (xfs_fileoff_t)bno))) {
2091                                 return error;
2092                         }
2093                         mapp = &map;
2094                         if (fsb == NULLFSBLOCK) {
2095                                 nmap = 0;
2096                         } else {
2097                                 map.br_startblock = fsb;
2098                                 map.br_startoff = (xfs_fileoff_t)bno;
2099                                 map.br_blockcount = 1;
2100                                 nmap = 1;
2101                         }
2102                 } else {
2103                         mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
2104                         nmap = nfsb;
2105                         if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
2106                                         nfsb,
2107                                         XFS_BMAPI_METADATA |
2108                                                 XFS_BMAPI_AFLAG(whichfork),
2109                                         NULL, 0, mapp, &nmap, NULL)))
2110                                 goto exit0;
2111                 }
2112         } else {
2113                 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2114                 map.br_startoff = (xfs_fileoff_t)bno;
2115                 map.br_blockcount = nfsb;
2116                 mapp = &map;
2117                 nmap = 1;
2118         }
2119         if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
2120                 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
2121                 if (unlikely(error == EFSCORRUPTED)) {
2122                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2123                                 int     i;
2124                                 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
2125                                         (long long)bno);
2126                                 cmn_err(CE_ALERT, "dir: inode %lld\n",
2127                                         (long long)dp->i_ino);
2128                                 for (i = 0; i < nmap; i++) {
2129                                         cmn_err(CE_ALERT,
2130                                                 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
2131                                                 i,
2132                                                 (long long)mapp[i].br_startoff,
2133                                                 (long long)mapp[i].br_startblock,
2134                                                 (long long)mapp[i].br_blockcount,
2135                                                 mapp[i].br_state);
2136                                 }
2137                         }
2138                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2139                                          XFS_ERRLEVEL_LOW, mp);
2140                 }
2141                 goto exit0;
2142         }
2143         if (caller != 3 && nmap > 1) {
2144                 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2145                 nbplist = 0;
2146         } else
2147                 bplist = NULL;
2148         /*
2149          * Turn the mapping(s) into buffer(s).
2150          */
2151         for (i = 0; i < nmap; i++) {
2152                 int     nmapped;
2153
2154                 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2155                 if (i == 0)
2156                         *mappedbnop = mappedbno;
2157                 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2158                 switch (caller) {
2159                 case 0:
2160                         bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2161                                 mappedbno, nmapped, 0);
2162                         error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2163                         break;
2164                 case 1:
2165                 case 2:
2166                         bp = NULL;
2167                         error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2168                                 mappedbno, nmapped, 0, &bp);
2169                         break;
2170                 case 3:
2171                         xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2172                         error = 0;
2173                         bp = NULL;
2174                         break;
2175                 }
2176                 if (error) {
2177                         if (bp)
2178                                 xfs_trans_brelse(trans, bp);
2179                         goto exit1;
2180                 }
2181                 if (!bp)
2182                         continue;
2183                 if (caller == 1) {
2184                         if (whichfork == XFS_ATTR_FORK) {
2185                                 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2186                                                 XFS_ATTR_BTREE_REF);
2187                         } else {
2188                                 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2189                                                 XFS_DIR_BTREE_REF);
2190                         }
2191                 }
2192                 if (bplist) {
2193                         bplist[nbplist++] = bp;
2194                 }
2195         }
2196         /*
2197          * Build a dabuf structure.
2198          */
2199         if (bplist) {
2200                 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2201         } else if (bp)
2202                 rbp = xfs_da_buf_make(1, &bp, ra);
2203         else
2204                 rbp = NULL;
2205         /*
2206          * For read_buf, check the magic number.
2207          */
2208         if (caller == 1) {
2209                 xfs_dir2_data_t         *data;
2210                 xfs_dir2_free_t         *free;
2211                 xfs_da_blkinfo_t        *info;
2212                 uint                    magic, magic1;
2213
2214                 info = rbp->data;
2215                 data = rbp->data;
2216                 free = rbp->data;
2217                 magic = INT_GET(info->magic, ARCH_CONVERT);
2218                 magic1 = INT_GET(data->hdr.magic, ARCH_CONVERT);
2219                 if (unlikely(
2220                     XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2221                                    (magic != XFS_DIR_LEAF_MAGIC) &&
2222                                    (magic != XFS_ATTR_LEAF_MAGIC) &&
2223                                    (magic != XFS_DIR2_LEAF1_MAGIC) &&
2224                                    (magic != XFS_DIR2_LEAFN_MAGIC) &&
2225                                    (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2226                                    (magic1 != XFS_DIR2_DATA_MAGIC) &&
2227                                    (INT_GET(free->hdr.magic, ARCH_CONVERT) != XFS_DIR2_FREE_MAGIC),
2228                                 mp, XFS_ERRTAG_DA_READ_BUF,
2229                                 XFS_RANDOM_DA_READ_BUF))) {
2230                         xfs_buftrace("DA READ ERROR", rbp->bps[0]);
2231                         XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2232                                              XFS_ERRLEVEL_LOW, mp, info);
2233                         error = XFS_ERROR(EFSCORRUPTED);
2234                         xfs_da_brelse(trans, rbp);
2235                         nbplist = 0;
2236                         goto exit1;
2237                 }
2238         }
2239         if (bplist) {
2240                 kmem_free(bplist, sizeof(*bplist) * nmap);
2241         }
2242         if (mapp != &map) {
2243                 kmem_free(mapp, sizeof(*mapp) * nfsb);
2244         }
2245         if (bpp)
2246                 *bpp = rbp;
2247         return 0;
2248 exit1:
2249         if (bplist) {
2250                 for (i = 0; i < nbplist; i++)
2251                         xfs_trans_brelse(trans, bplist[i]);
2252                 kmem_free(bplist, sizeof(*bplist) * nmap);
2253         }
2254 exit0:
2255         if (mapp != &map)
2256                 kmem_free(mapp, sizeof(*mapp) * nfsb);
2257         if (bpp)
2258                 *bpp = NULL;
2259         return error;
2260 }
2261
2262 /*
2263  * Get a buffer for the dir/attr block.
2264  */
2265 int
2266 xfs_da_get_buf(
2267         xfs_trans_t     *trans,
2268         xfs_inode_t     *dp,
2269         xfs_dablk_t     bno,
2270         xfs_daddr_t             mappedbno,
2271         xfs_dabuf_t     **bpp,
2272         int             whichfork)
2273 {
2274         return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2275                                                  (inst_t *)__return_address);
2276 }
2277
2278 /*
2279  * Get a buffer for the dir/attr block, fill in the contents.
2280  */
2281 int
2282 xfs_da_read_buf(
2283         xfs_trans_t     *trans,
2284         xfs_inode_t     *dp,
2285         xfs_dablk_t     bno,
2286         xfs_daddr_t             mappedbno,
2287         xfs_dabuf_t     **bpp,
2288         int             whichfork)
2289 {
2290         return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2291                 (inst_t *)__return_address);
2292 }
2293
2294 /*
2295  * Readahead the dir/attr block.
2296  */
2297 xfs_daddr_t
2298 xfs_da_reada_buf(
2299         xfs_trans_t     *trans,
2300         xfs_inode_t     *dp,
2301         xfs_dablk_t     bno,
2302         int             whichfork)
2303 {
2304         xfs_daddr_t             rval;
2305
2306         rval = -1;
2307         if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2308                         (inst_t *)__return_address))
2309                 return -1;
2310         else
2311                 return rval;
2312 }
2313
2314 /*
2315  * Calculate the number of bits needed to hold i different values.
2316  */
2317 uint
2318 xfs_da_log2_roundup(uint i)
2319 {
2320         uint rval;
2321
2322         for (rval = 0; rval < NBBY * sizeof(i); rval++) {
2323                 if ((1 << rval) >= i)
2324                         break;
2325         }
2326         return(rval);
2327 }
2328
2329 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2330 kmem_zone_t *xfs_dabuf_zone;            /* dabuf zone */
2331
2332 /*
2333  * Allocate a dir-state structure.
2334  * We don't put them on the stack since they're large.
2335  */
2336 xfs_da_state_t *
2337 xfs_da_state_alloc(void)
2338 {
2339         return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP);
2340 }
2341
2342 /*
2343  * Kill the altpath contents of a da-state structure.
2344  */
2345 STATIC void
2346 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2347 {
2348         int     i;
2349
2350         for (i = 0; i < state->altpath.active; i++) {
2351                 if (state->altpath.blk[i].bp) {
2352                         if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2353                                 xfs_da_buf_done(state->altpath.blk[i].bp);
2354                         state->altpath.blk[i].bp = NULL;
2355                 }
2356         }
2357         state->altpath.active = 0;
2358 }
2359
2360 /*
2361  * Free a da-state structure.
2362  */
2363 void
2364 xfs_da_state_free(xfs_da_state_t *state)
2365 {
2366         int     i;
2367
2368         xfs_da_state_kill_altpath(state);
2369         for (i = 0; i < state->path.active; i++) {
2370                 if (state->path.blk[i].bp)
2371                         xfs_da_buf_done(state->path.blk[i].bp);
2372         }
2373         if (state->extravalid && state->extrablk.bp)
2374                 xfs_da_buf_done(state->extrablk.bp);
2375 #ifdef DEBUG
2376         memset((char *)state, 0, sizeof(*state));
2377 #endif /* DEBUG */
2378         kmem_zone_free(xfs_da_state_zone, state);
2379 }
2380
2381 #ifdef XFS_DABUF_DEBUG
2382 xfs_dabuf_t     *xfs_dabuf_global_list;
2383 lock_t          xfs_dabuf_global_lock;
2384 #endif
2385
2386 /*
2387  * Create a dabuf.
2388  */
2389 /* ARGSUSED */
2390 STATIC xfs_dabuf_t *
2391 xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2392 {
2393         xfs_buf_t       *bp;
2394         xfs_dabuf_t     *dabuf;
2395         int             i;
2396         int             off;
2397
2398         if (nbuf == 1)
2399                 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP);
2400         else
2401                 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP);
2402         dabuf->dirty = 0;
2403 #ifdef XFS_DABUF_DEBUG
2404         dabuf->ra = ra;
2405         dabuf->target = XFS_BUF_TARGET(bps[0]);
2406         dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2407 #endif
2408         if (nbuf == 1) {
2409                 dabuf->nbuf = 1;
2410                 bp = bps[0];
2411                 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2412                 dabuf->data = XFS_BUF_PTR(bp);
2413                 dabuf->bps[0] = bp;
2414         } else {
2415                 dabuf->nbuf = nbuf;
2416                 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2417                         dabuf->bps[i] = bp = bps[i];
2418                         dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2419                 }
2420                 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2421                 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2422                         bp = bps[i];
2423                         memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2424                                 XFS_BUF_COUNT(bp));
2425                 }
2426         }
2427 #ifdef XFS_DABUF_DEBUG
2428         {
2429                 SPLDECL(s);
2430                 xfs_dabuf_t     *p;
2431
2432                 s = mutex_spinlock(&xfs_dabuf_global_lock);
2433                 for (p = xfs_dabuf_global_list; p; p = p->next) {
2434                         ASSERT(p->blkno != dabuf->blkno ||
2435                                p->target != dabuf->target);
2436                 }
2437                 dabuf->prev = NULL;
2438                 if (xfs_dabuf_global_list)
2439                         xfs_dabuf_global_list->prev = dabuf;
2440                 dabuf->next = xfs_dabuf_global_list;
2441                 xfs_dabuf_global_list = dabuf;
2442                 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2443         }
2444 #endif
2445         return dabuf;
2446 }
2447
2448 /*
2449  * Un-dirty a dabuf.
2450  */
2451 STATIC void
2452 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2453 {
2454         xfs_buf_t       *bp;
2455         int             i;
2456         int             off;
2457
2458         if (dabuf->dirty) {
2459                 ASSERT(dabuf->nbuf > 1);
2460                 dabuf->dirty = 0;
2461                 for (i = off = 0; i < dabuf->nbuf;
2462                                 i++, off += XFS_BUF_COUNT(bp)) {
2463                         bp = dabuf->bps[i];
2464                         memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2465                                 XFS_BUF_COUNT(bp));
2466                 }
2467         }
2468 }
2469
2470 /*
2471  * Release a dabuf.
2472  */
2473 void
2474 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2475 {
2476         ASSERT(dabuf);
2477         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2478         if (dabuf->dirty)
2479                 xfs_da_buf_clean(dabuf);
2480         if (dabuf->nbuf > 1)
2481                 kmem_free(dabuf->data, BBTOB(dabuf->bbcount));
2482 #ifdef XFS_DABUF_DEBUG
2483         {
2484                 SPLDECL(s);
2485
2486                 s = mutex_spinlock(&xfs_dabuf_global_lock);
2487                 if (dabuf->prev)
2488                         dabuf->prev->next = dabuf->next;
2489                 else
2490                         xfs_dabuf_global_list = dabuf->next;
2491                 if (dabuf->next)
2492                         dabuf->next->prev = dabuf->prev;
2493                 mutex_spinunlock(&xfs_dabuf_global_lock, s);
2494         }
2495         memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2496 #endif
2497         if (dabuf->nbuf == 1)
2498                 kmem_zone_free(xfs_dabuf_zone, dabuf);
2499         else
2500                 kmem_free(dabuf, XFS_DA_BUF_SIZE(dabuf->nbuf));
2501 }
2502
2503 /*
2504  * Log transaction from a dabuf.
2505  */
2506 void
2507 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2508 {
2509         xfs_buf_t       *bp;
2510         uint            f;
2511         int             i;
2512         uint            l;
2513         int             off;
2514
2515         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2516         if (dabuf->nbuf == 1) {
2517                 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2518                 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2519                 return;
2520         }
2521         dabuf->dirty = 1;
2522         ASSERT(first <= last);
2523         for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2524                 bp = dabuf->bps[i];
2525                 f = off;
2526                 l = f + XFS_BUF_COUNT(bp) - 1;
2527                 if (f < first)
2528                         f = first;
2529                 if (l > last)
2530                         l = last;
2531                 if (f <= l)
2532                         xfs_trans_log_buf(tp, bp, f - off, l - off);
2533                 /*
2534                  * B_DONE is set by xfs_trans_log buf.
2535                  * If we don't set it on a new buffer (get not read)
2536                  * then if we don't put anything in the buffer it won't
2537                  * be set, and at commit it it released into the cache,
2538                  * and then a read will fail.
2539                  */
2540                 else if (!(XFS_BUF_ISDONE(bp)))
2541                   XFS_BUF_DONE(bp);
2542         }
2543         ASSERT(last < off);
2544 }
2545
2546 /*
2547  * Release dabuf from a transaction.
2548  * Have to free up the dabuf before the buffers are released,
2549  * since the synchronization on the dabuf is really the lock on the buffer.
2550  */
2551 void
2552 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2553 {
2554         xfs_buf_t       *bp;
2555         xfs_buf_t       **bplist;
2556         int             i;
2557         int             nbuf;
2558
2559         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2560         if ((nbuf = dabuf->nbuf) == 1) {
2561                 bplist = &bp;
2562                 bp = dabuf->bps[0];
2563         } else {
2564                 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2565                 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2566         }
2567         xfs_da_buf_done(dabuf);
2568         for (i = 0; i < nbuf; i++)
2569                 xfs_trans_brelse(tp, bplist[i]);
2570         if (bplist != &bp)
2571                 kmem_free(bplist, nbuf * sizeof(*bplist));
2572 }
2573
2574 /*
2575  * Invalidate dabuf from a transaction.
2576  */
2577 void
2578 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2579 {
2580         xfs_buf_t       *bp;
2581         xfs_buf_t       **bplist;
2582         int             i;
2583         int             nbuf;
2584
2585         ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2586         if ((nbuf = dabuf->nbuf) == 1) {
2587                 bplist = &bp;
2588                 bp = dabuf->bps[0];
2589         } else {
2590                 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2591                 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2592         }
2593         xfs_da_buf_done(dabuf);
2594         for (i = 0; i < nbuf; i++)
2595                 xfs_trans_binval(tp, bplist[i]);
2596         if (bplist != &bp)
2597                 kmem_free(bplist, nbuf * sizeof(*bplist));
2598 }
2599
2600 /*
2601  * Get the first daddr from a dabuf.
2602  */
2603 xfs_daddr_t
2604 xfs_da_blkno(xfs_dabuf_t *dabuf)
2605 {
2606         ASSERT(dabuf->nbuf);
2607         ASSERT(dabuf->data);
2608         return XFS_BUF_ADDR(dabuf->bps[0]);
2609 }