fs/xfs/libxfs/xfs_da_btree.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_shared.h"
  22 #include "xfs_format.h"
  23 #include "xfs_log_format.h"
  24 #include "xfs_trans_resv.h"
  25 #include "xfs_bit.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_dir2_priv.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_alloc.h"
  35 #include "xfs_bmap.h"
  36 #include "xfs_attr.h"
  37 #include "xfs_attr_leaf.h"
  38 #include "xfs_error.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_cksum.h"
  41 #include "xfs_buf_item.h"
  42 #include "xfs_log.h"
  43
  44 /*
  45  * xfs_da_btree.c
  46  *
  47  * Routines to implement directories as Btrees of hashed names.
  48  */
  49
  50 /*========================================================================
  51  * Function prototypes for the kernel.
  52  *========================================================================*/
  53
  54 /*
  55  * Routines used for growing the Btree.
  56  */
  57 STATIC int xfs_da3_root_split(xfs_da_state_t *state,
  58                                             xfs_da_state_blk_t *existing_root,
  59                                             xfs_da_state_blk_t *new_child);
  60 STATIC int xfs_da3_node_split(xfs_da_state_t *state,
  61                                             xfs_da_state_blk_t *existing_blk,
  62                                             xfs_da_state_blk_t *split_blk,
  63                                             xfs_da_state_blk_t *blk_to_add,
  64                                             int treelevel,
  65                                             int *result);
  66 STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
  67                                          xfs_da_state_blk_t *node_blk_1,
  68                                          xfs_da_state_blk_t *node_blk_2);
  69 STATIC void xfs_da3_node_add(xfs_da_state_t *state,
  70                                    xfs_da_state_blk_t *old_node_blk,
  71                                    xfs_da_state_blk_t *new_node_blk);
  72
  73 /*
  74  * Routines used for shrinking the Btree.
  75  */
  76 STATIC int xfs_da3_root_join(xfs_da_state_t *state,
  77                                            xfs_da_state_blk_t *root_blk);
  78 STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
  79 STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
  80                                               xfs_da_state_blk_t *drop_blk);
  81 STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
  82                                          xfs_da_state_blk_t *src_node_blk,
  83                                          xfs_da_state_blk_t *dst_node_blk);
  84
  85 /*
  86  * Utility routines.
  87  */
  88 STATIC int      xfs_da3_blk_unlink(xfs_da_state_t *state,
  89                                   xfs_da_state_blk_t *drop_blk,
  90                                   xfs_da_state_blk_t *save_blk);
  91
  92
  93 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
  94
  95 /*
  96  * Allocate a dir-state structure.
  97  * We don't put them on the stack since they're large.
  98  */
  99 xfs_da_state_t *
 100 xfs_da_state_alloc(void)
 101 {
 102         return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
 103 }
 104
 105 /*
 106  * Kill the altpath contents of a da-state structure.
 107  */
 108 STATIC void
 109 xfs_da_state_kill_altpath(xfs_da_state_t *state)
 110 {
 111         int     i;
 112
 113         for (i = 0; i < state->altpath.active; i++)
 114                 state->altpath.blk[i].bp = NULL;
 115         state->altpath.active = 0;
 116 }
 117
 118 /*
 119  * Free a da-state structure.
 120  */
 121 void
 122 xfs_da_state_free(xfs_da_state_t *state)
 123 {
 124         xfs_da_state_kill_altpath(state);
 125 #ifdef DEBUG
 126         memset((char *)state, 0, sizeof(*state));
 127 #endif /* DEBUG */
 128         kmem_zone_free(xfs_da_state_zone, state);
 129 }
 130
 131 static bool
 132 xfs_da3_node_verify(
 133         struct xfs_buf          *bp)
 134 {
 135         struct xfs_mount        *mp = bp->b_target->bt_mount;
 136         struct xfs_da_intnode   *hdr = bp->b_addr;
 137         struct xfs_da3_icnode_hdr ichdr;
 138         const struct xfs_dir_ops *ops;
 139
 140         ops = xfs_dir_get_ops(mp, NULL);
 141
 142         ops->node_hdr_from_disk(&ichdr, hdr);
 143
 144         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 145                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 146
 147                 if (ichdr.magic != XFS_DA3_NODE_MAGIC)
 148                         return false;
 149
 150                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
 151                         return false;
 152                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
 153                         return false;
 154                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
 155                         return false;
 156         } else {
 157                 if (ichdr.magic != XFS_DA_NODE_MAGIC)
 158                         return false;
 159         }
 160         if (ichdr.level == 0)
 161                 return false;
 162         if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
 163                 return false;
 164         if (ichdr.count == 0)
 165                 return false;
 166
 167         /*
 168          * we don't know if the node is for and attribute or directory tree,
 169          * so only fail if the count is outside both bounds
 170          */
 171         if (ichdr.count > mp->m_dir_geo->node_ents &&
 172             ichdr.count > mp->m_attr_geo->node_ents)
 173                 return false;
 174
 175         /* XXX: hash order check? */
 176
 177         return true;
 178 }
 179
 180 static void
 181 xfs_da3_node_write_verify(
 182         struct xfs_buf  *bp)
 183 {
 184         struct xfs_mount        *mp = bp->b_target->bt_mount;
 185         struct xfs_buf_log_item *bip = bp->b_fspriv;
 186         struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 187
 188         if (!xfs_da3_node_verify(bp)) {
 189                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 190                 xfs_verifier_error(bp);
 191                 return;
 192         }
 193
 194         if (!xfs_sb_version_hascrc(&mp->m_sb))
 195                 return;
 196
 197         if (bip)
 198                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 199
 200         xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
 201 }
 202
 203 /*
 204  * leaf/node format detection on trees is sketchy, so a node read can be done on
 205  * leaf level blocks when detection identifies the tree as a node format tree
 206  * incorrectly. In this case, we need to swap the verifier to match the correct
 207  * format of the block being read.
 208  */
 209 static void
 210 xfs_da3_node_read_verify(
 211         struct xfs_buf          *bp)
 212 {
 213         struct xfs_da_blkinfo   *info = bp->b_addr;
 214
 215         switch (be16_to_cpu(info->magic)) {
 216                 case XFS_DA3_NODE_MAGIC:
 217                         if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
 218                                 xfs_buf_ioerror(bp, -EFSBADCRC);
 219                                 break;
 220                         }
 221                         /* fall through */
 222                 case XFS_DA_NODE_MAGIC:
 223                         if (!xfs_da3_node_verify(bp)) {
 224                                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 225                                 break;
 226                         }
 227                         return;
 228                 case XFS_ATTR_LEAF_MAGIC:
 229                 case XFS_ATTR3_LEAF_MAGIC:
 230                         bp->b_ops = &xfs_attr3_leaf_buf_ops;
 231                         bp->b_ops->verify_read(bp);
 232                         return;
 233                 case XFS_DIR2_LEAFN_MAGIC:
 234                 case XFS_DIR3_LEAFN_MAGIC:
 235                         bp->b_ops = &xfs_dir3_leafn_buf_ops;
 236                         bp->b_ops->verify_read(bp);
 237                         return;
 238                 default:
 239                         xfs_buf_ioerror(bp, -EFSCORRUPTED);
 240                         break;
 241         }
 242
 243         /* corrupt block */
 244         xfs_verifier_error(bp);
 245 }
 246
 247 const struct xfs_buf_ops xfs_da3_node_buf_ops = {
 248         .name = "xfs_da3_node",
 249         .verify_read = xfs_da3_node_read_verify,
 250         .verify_write = xfs_da3_node_write_verify,
 251 };
 252
 253 int
 254 xfs_da3_node_read(
 255         struct xfs_trans        *tp,
 256         struct xfs_inode        *dp,
 257         xfs_dablk_t             bno,
 258         xfs_daddr_t             mappedbno,
 259         struct xfs_buf          **bpp,
 260         int                     which_fork)
 261 {
 262         int                     err;
 263
 264         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
 265                                         which_fork, &xfs_da3_node_buf_ops);
 266         if (!err && tp) {
 267                 struct xfs_da_blkinfo   *info = (*bpp)->b_addr;
 268                 int                     type;
 269
 270                 switch (be16_to_cpu(info->magic)) {
 271                 case XFS_DA_NODE_MAGIC:
 272                 case XFS_DA3_NODE_MAGIC:
 273                         type = XFS_BLFT_DA_NODE_BUF;
 274                         break;
 275                 case XFS_ATTR_LEAF_MAGIC:
 276                 case XFS_ATTR3_LEAF_MAGIC:
 277                         type = XFS_BLFT_ATTR_LEAF_BUF;
 278                         break;
 279                 case XFS_DIR2_LEAFN_MAGIC:
 280                 case XFS_DIR3_LEAFN_MAGIC:
 281                         type = XFS_BLFT_DIR_LEAFN_BUF;
 282                         break;
 283                 default:
 284                         type = 0;
 285                         ASSERT(0);
 286                         break;
 287                 }
 288                 xfs_trans_buf_set_type(tp, *bpp, type);
 289         }
 290         return err;
 291 }
 292
 293 /*========================================================================
 294  * Routines used for growing the Btree.
 295  *========================================================================*/
 296
 297 /*
 298  * Create the initial contents of an intermediate node.
 299  */
 300 int
 301 xfs_da3_node_create(
 302         struct xfs_da_args      *args,
 303         xfs_dablk_t             blkno,
 304         int                     level,
 305         struct xfs_buf          **bpp,
 306         int                     whichfork)
 307 {
 308         struct xfs_da_intnode   *node;
 309         struct xfs_trans        *tp = args->trans;
 310         struct xfs_mount        *mp = tp->t_mountp;
 311         struct xfs_da3_icnode_hdr ichdr = {0};
 312         struct xfs_buf          *bp;
 313         int                     error;
 314         struct xfs_inode        *dp = args->dp;
 315
 316         trace_xfs_da_node_create(args);
 317         ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
 318
 319         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
 320         if (error)
 321                 return error;
 322         bp->b_ops = &xfs_da3_node_buf_ops;
 323         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 324         node = bp->b_addr;
 325
 326         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 327                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 328
 329                 memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
 330                 ichdr.magic = XFS_DA3_NODE_MAGIC;
 331                 hdr3->info.blkno = cpu_to_be64(bp->b_bn);
 332                 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
 333                 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
 334         } else {
 335                 ichdr.magic = XFS_DA_NODE_MAGIC;
 336         }
 337         ichdr.level = level;
 338
 339         dp->d_ops->node_hdr_to_disk(node, &ichdr);
 340         xfs_trans_log_buf(tp, bp,
 341                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 342
 343         *bpp = bp;
 344         return 0;
 345 }
 346
 347 /*
 348  * Split a leaf node, rebalance, then possibly split
 349  * intermediate nodes, rebalance, etc.
 350  */
 351 int                                                     /* error */
 352 xfs_da3_split(
 353         struct xfs_da_state     *state)
 354 {
 355         struct xfs_da_state_blk *oldblk;
 356         struct xfs_da_state_blk *newblk;
 357         struct xfs_da_state_blk *addblk;
 358         struct xfs_da_intnode   *node;
 359         struct xfs_buf          *bp;
 360         int                     max;
 361         int                     action = 0;
 362         int                     error;
 363         int                     i;
 364
 365         trace_xfs_da_split(state->args);
 366
 367         /*
 368          * Walk back up the tree splitting/inserting/adjusting as necessary.
 369          * If we need to insert and there isn't room, split the node, then
 370          * decide which fragment to insert the new block from below into.
 371          * Note that we may split the root this way, but we need more fixup.
 372          */
 373         max = state->path.active - 1;
 374         ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
 375         ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
 376                state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
 377
 378         addblk = &state->path.blk[max];         /* initial dummy value */
 379         for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
 380                 oldblk = &state->path.blk[i];
 381                 newblk = &state->altpath.blk[i];
 382
 383                 /*
 384                  * If a leaf node then
 385                  *     Allocate a new leaf node, then rebalance across them.
 386                  * else if an intermediate node then
 387                  *     We split on the last layer, must we split the node?
 388                  */
 389                 switch (oldblk->magic) {
 390                 case XFS_ATTR_LEAF_MAGIC:
 391                         error = xfs_attr3_leaf_split(state, oldblk, newblk);
 392                         if ((error != 0) && (error != -ENOSPC)) {
 393                                 return error;   /* GROT: attr is inconsistent */
 394                         }
 395                         if (!error) {
 396                                 addblk = newblk;
 397                                 break;
 398                         }
 399                         /*
 400                          * Entry wouldn't fit, split the leaf again.
 401                          */
 402                         state->extravalid = 1;
 403                         if (state->inleaf) {
 404                                 state->extraafter = 0;  /* before newblk */
 405                                 trace_xfs_attr_leaf_split_before(state->args);
 406                                 error = xfs_attr3_leaf_split(state, oldblk,
 407                                                             &state->extrablk);
 408                         } else {
 409                                 state->extraafter = 1;  /* after newblk */
 410                                 trace_xfs_attr_leaf_split_after(state->args);
 411                                 error = xfs_attr3_leaf_split(state, newblk,
 412                                                             &state->extrablk);
 413                         }
 414                         if (error)
 415                                 return error;   /* GROT: attr inconsistent */
 416                         addblk = newblk;
 417                         break;
 418                 case XFS_DIR2_LEAFN_MAGIC:
 419                         error = xfs_dir2_leafn_split(state, oldblk, newblk);
 420                         if (error)
 421                                 return error;
 422                         addblk = newblk;
 423                         break;
 424                 case XFS_DA_NODE_MAGIC:
 425                         error = xfs_da3_node_split(state, oldblk, newblk, addblk,
 426                                                          max - i, &action);
 427                         addblk->bp = NULL;
 428                         if (error)
 429                                 return error;   /* GROT: dir is inconsistent */
 430                         /*
 431                          * Record the newly split block for the next time thru?
 432                          */
 433                         if (action)
 434                                 addblk = newblk;
 435                         else
 436                                 addblk = NULL;
 437                         break;
 438                 }
 439
 440                 /*
 441                  * Update the btree to show the new hashval for this child.
 442                  */
 443                 xfs_da3_fixhashpath(state, &state->path);
 444         }
 445         if (!addblk)
 446                 return 0;
 447
 448         /*
 449          * Split the root node.
 450          */
 451         ASSERT(state->path.active == 0);
 452         oldblk = &state->path.blk[0];
 453         error = xfs_da3_root_split(state, oldblk, addblk);
 454         if (error) {
 455                 addblk->bp = NULL;
 456                 return error;   /* GROT: dir is inconsistent */
 457         }
 458
 459         /*
 460          * Update pointers to the node which used to be block 0 and
 461          * just got bumped because of the addition of a new root node.
 462          * There might be three blocks involved if a double split occurred,
 463          * and the original block 0 could be at any position in the list.
 464          *
 465          * Note: the magic numbers and sibling pointers are in the same
 466          * physical place for both v2 and v3 headers (by design). Hence it
 467          * doesn't matter which version of the xfs_da_intnode structure we use
 468          * here as the result will be the same using either structure.
 469          */
 470         node = oldblk->bp->b_addr;
 471         if (node->hdr.info.forw) {
 472                 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
 473                         bp = addblk->bp;
 474                 } else {
 475                         ASSERT(state->extravalid);
 476                         bp = state->extrablk.bp;
 477                 }
 478                 node = bp->b_addr;
 479                 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
 480                 xfs_trans_log_buf(state->args->trans, bp,
 481                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 482                     sizeof(node->hdr.info)));
 483         }
 484         node = oldblk->bp->b_addr;
 485         if (node->hdr.info.back) {
 486                 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
 487                         bp = addblk->bp;
 488                 } else {
 489                         ASSERT(state->extravalid);
 490                         bp = state->extrablk.bp;
 491                 }
 492                 node = bp->b_addr;
 493                 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
 494                 xfs_trans_log_buf(state->args->trans, bp,
 495                     XFS_DA_LOGRANGE(node, &node->hdr.info,
 496                     sizeof(node->hdr.info)));
 497         }
 498         addblk->bp = NULL;
 499         return 0;
 500 }
 501
 502 /*
 503  * Split the root.  We have to create a new root and point to the two
 504  * parts (the split old root) that we just created.  Copy block zero to
 505  * the EOF, extending the inode in process.
 506  */
 507 STATIC int                                              /* error */
 508 xfs_da3_root_split(
 509         struct xfs_da_state     *state,
 510         struct xfs_da_state_blk *blk1,
 511         struct xfs_da_state_blk *blk2)
 512 {
 513         struct xfs_da_intnode   *node;
 514         struct xfs_da_intnode   *oldroot;
 515         struct xfs_da_node_entry *btree;
 516         struct xfs_da3_icnode_hdr nodehdr;
 517         struct xfs_da_args      *args;
 518         struct xfs_buf          *bp;
 519         struct xfs_inode        *dp;
 520         struct xfs_trans        *tp;
 521         struct xfs_dir2_leaf    *leaf;
 522         xfs_dablk_t             blkno;
 523         int                     level;
 524         int                     error;
 525         int                     size;
 526
 527         trace_xfs_da_root_split(state->args);
 528
 529         /*
 530          * Copy the existing (incorrect) block from the root node position
 531          * to a free space somewhere.
 532          */
 533         args = state->args;
 534         error = xfs_da_grow_inode(args, &blkno);
 535         if (error)
 536                 return error;
 537
 538         dp = args->dp;
 539         tp = args->trans;
 540         error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
 541         if (error)
 542                 return error;
 543         node = bp->b_addr;
 544         oldroot = blk1->bp->b_addr;
 545         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
 546             oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
 547                 struct xfs_da3_icnode_hdr icnodehdr;
 548
 549                 dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
 550                 btree = dp->d_ops->node_tree_p(oldroot);
 551                 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
 552                 level = icnodehdr.level;
 553
 554                 /*
 555                  * we are about to copy oldroot to bp, so set up the type
 556                  * of bp while we know exactly what it will be.
 557                  */
 558                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
 559         } else {
 560                 struct xfs_dir3_icleaf_hdr leafhdr;
 561                 struct xfs_dir2_leaf_entry *ents;
 562
 563                 leaf = (xfs_dir2_leaf_t *)oldroot;
 564                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 565                 ents = dp->d_ops->leaf_ents_p(leaf);
 566
 567                 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 568                        leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 569                 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
 570                 level = 0;
 571
 572                 /*
 573                  * we are about to copy oldroot to bp, so set up the type
 574                  * of bp while we know exactly what it will be.
 575                  */
 576                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 577         }
 578
 579         /*
 580          * we can copy most of the information in the node from one block to
 581          * another, but for CRC enabled headers we have to make sure that the
 582          * block specific identifiers are kept intact. We update the buffer
 583          * directly for this.
 584          */
 585         memcpy(node, oldroot, size);
 586         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
 587             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 588                 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
 589
 590                 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
 591         }
 592         xfs_trans_log_buf(tp, bp, 0, size - 1);
 593
 594         bp->b_ops = blk1->bp->b_ops;
 595         xfs_trans_buf_copy_type(bp, blk1->bp);
 596         blk1->bp = bp;
 597         blk1->blkno = blkno;
 598
 599         /*
 600          * Set up the new root node.
 601          */
 602         error = xfs_da3_node_create(args,
 603                 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
 604                 level + 1, &bp, args->whichfork);
 605         if (error)
 606                 return error;
 607
 608         node = bp->b_addr;
 609         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 610         btree = dp->d_ops->node_tree_p(node);
 611         btree[0].hashval = cpu_to_be32(blk1->hashval);
 612         btree[0].before = cpu_to_be32(blk1->blkno);
 613         btree[1].hashval = cpu_to_be32(blk2->hashval);
 614         btree[1].before = cpu_to_be32(blk2->blkno);
 615         nodehdr.count = 2;
 616         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 617
 618 #ifdef DEBUG
 619         if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
 620             oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
 621                 ASSERT(blk1->blkno >= args->geo->leafblk &&
 622                        blk1->blkno < args->geo->freeblk);
 623                 ASSERT(blk2->blkno >= args->geo->leafblk &&
 624                        blk2->blkno < args->geo->freeblk);
 625         }
 626 #endif
 627
 628         /* Header is already logged by xfs_da_node_create */
 629         xfs_trans_log_buf(tp, bp,
 630                 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
 631
 632         return 0;
 633 }
 634
 635 /*
 636  * Split the node, rebalance, then add the new entry.
 637  */
 638 STATIC int                                              /* error */
 639 xfs_da3_node_split(
 640         struct xfs_da_state     *state,
 641         struct xfs_da_state_blk *oldblk,
 642         struct xfs_da_state_blk *newblk,
 643         struct xfs_da_state_blk *addblk,
 644         int                     treelevel,
 645         int                     *result)
 646 {
 647         struct xfs_da_intnode   *node;
 648         struct xfs_da3_icnode_hdr nodehdr;
 649         xfs_dablk_t             blkno;
 650         int                     newcount;
 651         int                     error;
 652         int                     useextra;
 653         struct xfs_inode        *dp = state->args->dp;
 654
 655         trace_xfs_da_node_split(state->args);
 656
 657         node = oldblk->bp->b_addr;
 658         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 659
 660         /*
 661          * With V2 dirs the extra block is data or freespace.
 662          */
 663         useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
 664         newcount = 1 + useextra;
 665         /*
 666          * Do we have to split the node?
 667          */
 668         if (nodehdr.count + newcount > state->args->geo->node_ents) {
 669                 /*
 670                  * Allocate a new node, add to the doubly linked chain of
 671                  * nodes, then move some of our excess entries into it.
 672                  */
 673                 error = xfs_da_grow_inode(state->args, &blkno);
 674                 if (error)
 675                         return error;   /* GROT: dir is inconsistent */
 676
 677                 error = xfs_da3_node_create(state->args, blkno, treelevel,
 678                                            &newblk->bp, state->args->whichfork);
 679                 if (error)
 680                         return error;   /* GROT: dir is inconsistent */
 681                 newblk->blkno = blkno;
 682                 newblk->magic = XFS_DA_NODE_MAGIC;
 683                 xfs_da3_node_rebalance(state, oldblk, newblk);
 684                 error = xfs_da3_blk_link(state, oldblk, newblk);
 685                 if (error)
 686                         return error;
 687                 *result = 1;
 688         } else {
 689                 *result = 0;
 690         }
 691
 692         /*
 693          * Insert the new entry(s) into the correct block
 694          * (updating last hashval in the process).
 695          *
 696          * xfs_da3_node_add() inserts BEFORE the given index,
 697          * and as a result of using node_lookup_int() we always
 698          * point to a valid entry (not after one), but a split
 699          * operation always results in a new block whose hashvals
 700          * FOLLOW the current block.
 701          *
 702          * If we had double-split op below us, then add the extra block too.
 703          */
 704         node = oldblk->bp->b_addr;
 705         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 706         if (oldblk->index <= nodehdr.count) {
 707                 oldblk->index++;
 708                 xfs_da3_node_add(state, oldblk, addblk);
 709                 if (useextra) {
 710                         if (state->extraafter)
 711                                 oldblk->index++;
 712                         xfs_da3_node_add(state, oldblk, &state->extrablk);
 713                         state->extravalid = 0;
 714                 }
 715         } else {
 716                 newblk->index++;
 717                 xfs_da3_node_add(state, newblk, addblk);
 718                 if (useextra) {
 719                         if (state->extraafter)
 720                                 newblk->index++;
 721                         xfs_da3_node_add(state, newblk, &state->extrablk);
 722                         state->extravalid = 0;
 723                 }
 724         }
 725
 726         return 0;
 727 }
 728
 729 /*
 730  * Balance the btree elements between two intermediate nodes,
 731  * usually one full and one empty.
 732  *
 733  * NOTE: if blk2 is empty, then it will get the upper half of blk1.
 734  */
 735 STATIC void
 736 xfs_da3_node_rebalance(
 737         struct xfs_da_state     *state,
 738         struct xfs_da_state_blk *blk1,
 739         struct xfs_da_state_blk *blk2)
 740 {
 741         struct xfs_da_intnode   *node1;
 742         struct xfs_da_intnode   *node2;
 743         struct xfs_da_intnode   *tmpnode;
 744         struct xfs_da_node_entry *btree1;
 745         struct xfs_da_node_entry *btree2;
 746         struct xfs_da_node_entry *btree_s;
 747         struct xfs_da_node_entry *btree_d;
 748         struct xfs_da3_icnode_hdr nodehdr1;
 749         struct xfs_da3_icnode_hdr nodehdr2;
 750         struct xfs_trans        *tp;
 751         int                     count;
 752         int                     tmp;
 753         int                     swap = 0;
 754         struct xfs_inode        *dp = state->args->dp;
 755
 756         trace_xfs_da_node_rebalance(state->args);
 757
 758         node1 = blk1->bp->b_addr;
 759         node2 = blk2->bp->b_addr;
 760         dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 761         dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 762         btree1 = dp->d_ops->node_tree_p(node1);
 763         btree2 = dp->d_ops->node_tree_p(node2);
 764
 765         /*
 766          * Figure out how many entries need to move, and in which direction.
 767          * Swap the nodes around if that makes it simpler.
 768          */
 769         if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
 770             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
 771              (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
 772                         be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
 773                 tmpnode = node1;
 774                 node1 = node2;
 775                 node2 = tmpnode;
 776                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 777                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 778                 btree1 = dp->d_ops->node_tree_p(node1);
 779                 btree2 = dp->d_ops->node_tree_p(node2);
 780                 swap = 1;
 781         }
 782
 783         count = (nodehdr1.count - nodehdr2.count) / 2;
 784         if (count == 0)
 785                 return;
 786         tp = state->args->trans;
 787         /*
 788          * Two cases: high-to-low and low-to-high.
 789          */
 790         if (count > 0) {
 791                 /*
 792                  * Move elements in node2 up to make a hole.
 793                  */
 794                 tmp = nodehdr2.count;
 795                 if (tmp > 0) {
 796                         tmp *= (uint)sizeof(xfs_da_node_entry_t);
 797                         btree_s = &btree2[0];
 798                         btree_d = &btree2[count];
 799                         memmove(btree_d, btree_s, tmp);
 800                 }
 801
 802                 /*
 803                  * Move the req'd B-tree elements from high in node1 to
 804                  * low in node2.
 805                  */
 806                 nodehdr2.count += count;
 807                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 808                 btree_s = &btree1[nodehdr1.count - count];
 809                 btree_d = &btree2[0];
 810                 memcpy(btree_d, btree_s, tmp);
 811                 nodehdr1.count -= count;
 812         } else {
 813                 /*
 814                  * Move the req'd B-tree elements from low in node2 to
 815                  * high in node1.
 816                  */
 817                 count = -count;
 818                 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
 819                 btree_s = &btree2[0];
 820                 btree_d = &btree1[nodehdr1.count];
 821                 memcpy(btree_d, btree_s, tmp);
 822                 nodehdr1.count += count;
 823
 824                 xfs_trans_log_buf(tp, blk1->bp,
 825                         XFS_DA_LOGRANGE(node1, btree_d, tmp));
 826
 827                 /*
 828                  * Move elements in node2 down to fill the hole.
 829                  */
 830                 tmp  = nodehdr2.count - count;
 831                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
 832                 btree_s = &btree2[count];
 833                 btree_d = &btree2[0];
 834                 memmove(btree_d, btree_s, tmp);
 835                 nodehdr2.count -= count;
 836         }
 837
 838         /*
 839          * Log header of node 1 and all current bits of node 2.
 840          */
 841         dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
 842         xfs_trans_log_buf(tp, blk1->bp,
 843                 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
 844
 845         dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
 846         xfs_trans_log_buf(tp, blk2->bp,
 847                 XFS_DA_LOGRANGE(node2, &node2->hdr,
 848                                 dp->d_ops->node_hdr_size +
 849                                 (sizeof(btree2[0]) * nodehdr2.count)));
 850
 851         /*
 852          * Record the last hashval from each block for upward propagation.
 853          * (note: don't use the swapped node pointers)
 854          */
 855         if (swap) {
 856                 node1 = blk1->bp->b_addr;
 857                 node2 = blk2->bp->b_addr;
 858                 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
 859                 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
 860                 btree1 = dp->d_ops->node_tree_p(node1);
 861                 btree2 = dp->d_ops->node_tree_p(node2);
 862         }
 863         blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
 864         blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
 865
 866         /*
 867          * Adjust the expected index for insertion.
 868          */
 869         if (blk1->index >= nodehdr1.count) {
 870                 blk2->index = blk1->index - nodehdr1.count;
 871                 blk1->index = nodehdr1.count + 1;       /* make it invalid */
 872         }
 873 }
 874
 875 /*
 876  * Add a new entry to an intermediate node.
 877  */
 878 STATIC void
 879 xfs_da3_node_add(
 880         struct xfs_da_state     *state,
 881         struct xfs_da_state_blk *oldblk,
 882         struct xfs_da_state_blk *newblk)
 883 {
 884         struct xfs_da_intnode   *node;
 885         struct xfs_da3_icnode_hdr nodehdr;
 886         struct xfs_da_node_entry *btree;
 887         int                     tmp;
 888         struct xfs_inode        *dp = state->args->dp;
 889
 890         trace_xfs_da_node_add(state->args);
 891
 892         node = oldblk->bp->b_addr;
 893         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
 894         btree = dp->d_ops->node_tree_p(node);
 895
 896         ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
 897         ASSERT(newblk->blkno != 0);
 898         if (state->args->whichfork == XFS_DATA_FORK)
 899                 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
 900                        newblk->blkno < state->args->geo->freeblk);
 901
 902         /*
 903          * We may need to make some room before we insert the new node.
 904          */
 905         tmp = 0;
 906         if (oldblk->index < nodehdr.count) {
 907                 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
 908                 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
 909         }
 910         btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
 911         btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
 912         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 913                 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
 914                                 tmp + sizeof(*btree)));
 915
 916         nodehdr.count += 1;
 917         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
 918         xfs_trans_log_buf(state->args->trans, oldblk->bp,
 919                 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
 920
 921         /*
 922          * Copy the last hash value from the oldblk to propagate upwards.
 923          */
 924         oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
 925 }
 926
 927 /*========================================================================
 928  * Routines used for shrinking the Btree.
 929  *========================================================================*/
 930
 931 /*
 932  * Deallocate an empty leaf node, remove it from its parent,
 933  * possibly deallocating that block, etc...
 934  */
 935 int
 936 xfs_da3_join(
 937         struct xfs_da_state     *state)
 938 {
 939         struct xfs_da_state_blk *drop_blk;
 940         struct xfs_da_state_blk *save_blk;
 941         int                     action = 0;
 942         int                     error;
 943
 944         trace_xfs_da_join(state->args);
 945
 946         drop_blk = &state->path.blk[ state->path.active-1 ];
 947         save_blk = &state->altpath.blk[ state->path.active-1 ];
 948         ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
 949         ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
 950                drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
 951
 952         /*
 953          * Walk back up the tree joining/deallocating as necessary.
 954          * When we stop dropping blocks, break out.
 955          */
 956         for (  ; state->path.active >= 2; drop_blk--, save_blk--,
 957                  state->path.active--) {
 958                 /*
 959                  * See if we can combine the block with a neighbor.
 960                  *   (action == 0) => no options, just leave
 961                  *   (action == 1) => coalesce, then unlink
 962                  *   (action == 2) => block empty, unlink it
 963                  */
 964                 switch (drop_blk->magic) {
 965                 case XFS_ATTR_LEAF_MAGIC:
 966                         error = xfs_attr3_leaf_toosmall(state, &action);
 967                         if (error)
 968                                 return error;
 969                         if (action == 0)
 970                                 return 0;
 971                         xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
 972                         break;
 973                 case XFS_DIR2_LEAFN_MAGIC:
 974                         error = xfs_dir2_leafn_toosmall(state, &action);
 975                         if (error)
 976                                 return error;
 977                         if (action == 0)
 978                                 return 0;
 979                         xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
 980                         break;
 981                 case XFS_DA_NODE_MAGIC:
 982                         /*
 983                          * Remove the offending node, fixup hashvals,
 984                          * check for a toosmall neighbor.
 985                          */
 986                         xfs_da3_node_remove(state, drop_blk);
 987                         xfs_da3_fixhashpath(state, &state->path);
 988                         error = xfs_da3_node_toosmall(state, &action);
 989                         if (error)
 990                                 return error;
 991                         if (action == 0)
 992                                 return 0;
 993                         xfs_da3_node_unbalance(state, drop_blk, save_blk);
 994                         break;
 995                 }
 996                 xfs_da3_fixhashpath(state, &state->altpath);
 997                 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
 998                 xfs_da_state_kill_altpath(state);
 999                 if (error)
1000                         return error;
1001                 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1002                                                          drop_blk->bp);
1003                 drop_blk->bp = NULL;
1004                 if (error)
1005                         return error;
1006         }
1007         /*
1008          * We joined all the way to the top.  If it turns out that
1009          * we only have one entry in the root, make the child block
1010          * the new root.
1011          */
1012         xfs_da3_node_remove(state, drop_blk);
1013         xfs_da3_fixhashpath(state, &state->path);
1014         error = xfs_da3_root_join(state, &state->path.blk[0]);
1015         return error;
1016 }
1017
1018 #ifdef  DEBUG
1019 static void
1020 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1021 {
1022         __be16  magic = blkinfo->magic;
1023
1024         if (level == 1) {
1025                 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1026                        magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1027                        magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1028                        magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1029         } else {
1030                 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1031                        magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1032         }
1033         ASSERT(!blkinfo->forw);
1034         ASSERT(!blkinfo->back);
1035 }
1036 #else   /* !DEBUG */
1037 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1038 #endif  /* !DEBUG */
1039
1040 /*
1041  * We have only one entry in the root.  Copy the only remaining child of
1042  * the old root to block 0 as the new root node.
1043  */
1044 STATIC int
1045 xfs_da3_root_join(
1046         struct xfs_da_state     *state,
1047         struct xfs_da_state_blk *root_blk)
1048 {
1049         struct xfs_da_intnode   *oldroot;
1050         struct xfs_da_args      *args;
1051         xfs_dablk_t             child;
1052         struct xfs_buf          *bp;
1053         struct xfs_da3_icnode_hdr oldroothdr;
1054         struct xfs_da_node_entry *btree;
1055         int                     error;
1056         struct xfs_inode        *dp = state->args->dp;
1057
1058         trace_xfs_da_root_join(state->args);
1059
1060         ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1061
1062         args = state->args;
1063         oldroot = root_blk->bp->b_addr;
1064         dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1065         ASSERT(oldroothdr.forw == 0);
1066         ASSERT(oldroothdr.back == 0);
1067
1068         /*
1069          * If the root has more than one child, then don't do anything.
1070          */
1071         if (oldroothdr.count > 1)
1072                 return 0;
1073
1074         /*
1075          * Read in the (only) child block, then copy those bytes into
1076          * the root block's buffer and free the original child block.
1077          */
1078         btree = dp->d_ops->node_tree_p(oldroot);
1079         child = be32_to_cpu(btree[0].before);
1080         ASSERT(child != 0);
1081         error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1082                                              args->whichfork);
1083         if (error)
1084                 return error;
1085         xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1086
1087         /*
1088          * This could be copying a leaf back into the root block in the case of
1089          * there only being a single leaf block left in the tree. Hence we have
1090          * to update the b_ops pointer as well to match the buffer type change
1091          * that could occur. For dir3 blocks we also need to update the block
1092          * number in the buffer header.
1093          */
1094         memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1095         root_blk->bp->b_ops = bp->b_ops;
1096         xfs_trans_buf_copy_type(root_blk->bp, bp);
1097         if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1098                 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1099                 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1100         }
1101         xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1102                           args->geo->blksize - 1);
1103         error = xfs_da_shrink_inode(args, child, bp);
1104         return error;
1105 }
1106
1107 /*
1108  * Check a node block and its neighbors to see if the block should be
1109  * collapsed into one or the other neighbor.  Always keep the block
1110  * with the smaller block number.
1111  * If the current block is over 50% full, don't try to join it, return 0.
1112  * If the block is empty, fill in the state structure and return 2.
1113  * If it can be collapsed, fill in the state structure and return 1.
1114  * If nothing can be done, return 0.
1115  */
1116 STATIC int
1117 xfs_da3_node_toosmall(
1118         struct xfs_da_state     *state,
1119         int                     *action)
1120 {
1121         struct xfs_da_intnode   *node;
1122         struct xfs_da_state_blk *blk;
1123         struct xfs_da_blkinfo   *info;
1124         xfs_dablk_t             blkno;
1125         struct xfs_buf          *bp;
1126         struct xfs_da3_icnode_hdr nodehdr;
1127         int                     count;
1128         int                     forward;
1129         int                     error;
1130         int                     retval;
1131         int                     i;
1132         struct xfs_inode        *dp = state->args->dp;
1133
1134         trace_xfs_da_node_toosmall(state->args);
1135
1136         /*
1137          * Check for the degenerate case of the block being over 50% full.
1138          * If so, it's not worth even looking to see if we might be able
1139          * to coalesce with a sibling.
1140          */
1141         blk = &state->path.blk[ state->path.active-1 ];
1142         info = blk->bp->b_addr;
1143         node = (xfs_da_intnode_t *)info;
1144         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1145         if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1146                 *action = 0;    /* blk over 50%, don't try to join */
1147                 return 0;       /* blk over 50%, don't try to join */
1148         }
1149
1150         /*
1151          * Check for the degenerate case of the block being empty.
1152          * If the block is empty, we'll simply delete it, no need to
1153          * coalesce it with a sibling block.  We choose (arbitrarily)
1154          * to merge with the forward block unless it is NULL.
1155          */
1156         if (nodehdr.count == 0) {
1157                 /*
1158                  * Make altpath point to the block we want to keep and
1159                  * path point to the block we want to drop (this one).
1160                  */
1161                 forward = (info->forw != 0);
1162                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1163                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1164                                                  0, &retval);
1165                 if (error)
1166                         return error;
1167                 if (retval) {
1168                         *action = 0;
1169                 } else {
1170                         *action = 2;
1171                 }
1172                 return 0;
1173         }
1174
1175         /*
1176          * Examine each sibling block to see if we can coalesce with
1177          * at least 25% free space to spare.  We need to figure out
1178          * whether to merge with the forward or the backward block.
1179          * We prefer coalescing with the lower numbered sibling so as
1180          * to shrink a directory over time.
1181          */
1182         count  = state->args->geo->node_ents;
1183         count -= state->args->geo->node_ents >> 2;
1184         count -= nodehdr.count;
1185
1186         /* start with smaller blk num */
1187         forward = nodehdr.forw < nodehdr.back;
1188         for (i = 0; i < 2; forward = !forward, i++) {
1189                 struct xfs_da3_icnode_hdr thdr;
1190                 if (forward)
1191                         blkno = nodehdr.forw;
1192                 else
1193                         blkno = nodehdr.back;
1194                 if (blkno == 0)
1195                         continue;
1196                 error = xfs_da3_node_read(state->args->trans, dp,
1197                                         blkno, -1, &bp, state->args->whichfork);
1198                 if (error)
1199                         return error;
1200
1201                 node = bp->b_addr;
1202                 dp->d_ops->node_hdr_from_disk(&thdr, node);
1203                 xfs_trans_brelse(state->args->trans, bp);
1204
1205                 if (count - thdr.count >= 0)
1206                         break;  /* fits with at least 25% to spare */
1207         }
1208         if (i >= 2) {
1209                 *action = 0;
1210                 return 0;
1211         }
1212
1213         /*
1214          * Make altpath point to the block we want to keep (the lower
1215          * numbered block) and path point to the block we want to drop.
1216          */
1217         memcpy(&state->altpath, &state->path, sizeof(state->path));
1218         if (blkno < blk->blkno) {
1219                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1220                                                  0, &retval);
1221         } else {
1222                 error = xfs_da3_path_shift(state, &state->path, forward,
1223                                                  0, &retval);
1224         }
1225         if (error)
1226                 return error;
1227         if (retval) {
1228                 *action = 0;
1229                 return 0;
1230         }
1231         *action = 1;
1232         return 0;
1233 }
1234
1235 /*
1236  * Pick up the last hashvalue from an intermediate node.
1237  */
1238 STATIC uint
1239 xfs_da3_node_lasthash(
1240         struct xfs_inode        *dp,
1241         struct xfs_buf          *bp,
1242         int                     *count)
1243 {
1244         struct xfs_da_intnode    *node;
1245         struct xfs_da_node_entry *btree;
1246         struct xfs_da3_icnode_hdr nodehdr;
1247
1248         node = bp->b_addr;
1249         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1250         if (count)
1251                 *count = nodehdr.count;
1252         if (!nodehdr.count)
1253                 return 0;
1254         btree = dp->d_ops->node_tree_p(node);
1255         return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1256 }
1257
1258 /*
1259  * Walk back up the tree adjusting hash values as necessary,
1260  * when we stop making changes, return.
1261  */
1262 void
1263 xfs_da3_fixhashpath(
1264         struct xfs_da_state     *state,
1265         struct xfs_da_state_path *path)
1266 {
1267         struct xfs_da_state_blk *blk;
1268         struct xfs_da_intnode   *node;
1269         struct xfs_da_node_entry *btree;
1270         xfs_dahash_t            lasthash=0;
1271         int                     level;
1272         int                     count;
1273         struct xfs_inode        *dp = state->args->dp;
1274
1275         trace_xfs_da_fixhashpath(state->args);
1276
1277         level = path->active-1;
1278         blk = &path->blk[ level ];
1279         switch (blk->magic) {
1280         case XFS_ATTR_LEAF_MAGIC:
1281                 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1282                 if (count == 0)
1283                         return;
1284                 break;
1285         case XFS_DIR2_LEAFN_MAGIC:
1286                 lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1287                 if (count == 0)
1288                         return;
1289                 break;
1290         case XFS_DA_NODE_MAGIC:
1291                 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1292                 if (count == 0)
1293                         return;
1294                 break;
1295         }
1296         for (blk--, level--; level >= 0; blk--, level--) {
1297                 struct xfs_da3_icnode_hdr nodehdr;
1298
1299                 node = blk->bp->b_addr;
1300                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1301                 btree = dp->d_ops->node_tree_p(node);
1302                 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1303                         break;
1304                 blk->hashval = lasthash;
1305                 btree[blk->index].hashval = cpu_to_be32(lasthash);
1306                 xfs_trans_log_buf(state->args->trans, blk->bp,
1307                                   XFS_DA_LOGRANGE(node, &btree[blk->index],
1308                                                   sizeof(*btree)));
1309
1310                 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1311         }
1312 }
1313
1314 /*
1315  * Remove an entry from an intermediate node.
1316  */
1317 STATIC void
1318 xfs_da3_node_remove(
1319         struct xfs_da_state     *state,
1320         struct xfs_da_state_blk *drop_blk)
1321 {
1322         struct xfs_da_intnode   *node;
1323         struct xfs_da3_icnode_hdr nodehdr;
1324         struct xfs_da_node_entry *btree;
1325         int                     index;
1326         int                     tmp;
1327         struct xfs_inode        *dp = state->args->dp;
1328
1329         trace_xfs_da_node_remove(state->args);
1330
1331         node = drop_blk->bp->b_addr;
1332         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1333         ASSERT(drop_blk->index < nodehdr.count);
1334         ASSERT(drop_blk->index >= 0);
1335
1336         /*
1337          * Copy over the offending entry, or just zero it out.
1338          */
1339         index = drop_blk->index;
1340         btree = dp->d_ops->node_tree_p(node);
1341         if (index < nodehdr.count - 1) {
1342                 tmp  = nodehdr.count - index - 1;
1343                 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1344                 memmove(&btree[index], &btree[index + 1], tmp);
1345                 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1346                     XFS_DA_LOGRANGE(node, &btree[index], tmp));
1347                 index = nodehdr.count - 1;
1348         }
1349         memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1350         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1351             XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1352         nodehdr.count -= 1;
1353         dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1354         xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1355             XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1356
1357         /*
1358          * Copy the last hash value from the block to propagate upwards.
1359          */
1360         drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1361 }
1362
1363 /*
1364  * Unbalance the elements between two intermediate nodes,
1365  * move all Btree elements from one node into another.
1366  */
1367 STATIC void
1368 xfs_da3_node_unbalance(
1369         struct xfs_da_state     *state,
1370         struct xfs_da_state_blk *drop_blk,
1371         struct xfs_da_state_blk *save_blk)
1372 {
1373         struct xfs_da_intnode   *drop_node;
1374         struct xfs_da_intnode   *save_node;
1375         struct xfs_da_node_entry *drop_btree;
1376         struct xfs_da_node_entry *save_btree;
1377         struct xfs_da3_icnode_hdr drop_hdr;
1378         struct xfs_da3_icnode_hdr save_hdr;
1379         struct xfs_trans        *tp;
1380         int                     sindex;
1381         int                     tmp;
1382         struct xfs_inode        *dp = state->args->dp;
1383
1384         trace_xfs_da_node_unbalance(state->args);
1385
1386         drop_node = drop_blk->bp->b_addr;
1387         save_node = save_blk->bp->b_addr;
1388         dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1389         dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1390         drop_btree = dp->d_ops->node_tree_p(drop_node);
1391         save_btree = dp->d_ops->node_tree_p(save_node);
1392         tp = state->args->trans;
1393
1394         /*
1395          * If the dying block has lower hashvals, then move all the
1396          * elements in the remaining block up to make a hole.
1397          */
1398         if ((be32_to_cpu(drop_btree[0].hashval) <
1399                         be32_to_cpu(save_btree[0].hashval)) ||
1400             (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1401                         be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1402                 /* XXX: check this - is memmove dst correct? */
1403                 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1404                 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1405
1406                 sindex = 0;
1407                 xfs_trans_log_buf(tp, save_blk->bp,
1408                         XFS_DA_LOGRANGE(save_node, &save_btree[0],
1409                                 (save_hdr.count + drop_hdr.count) *
1410                                                 sizeof(xfs_da_node_entry_t)));
1411         } else {
1412                 sindex = save_hdr.count;
1413                 xfs_trans_log_buf(tp, save_blk->bp,
1414                         XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1415                                 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1416         }
1417
1418         /*
1419          * Move all the B-tree elements from drop_blk to save_blk.
1420          */
1421         tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1422         memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1423         save_hdr.count += drop_hdr.count;
1424
1425         dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1426         xfs_trans_log_buf(tp, save_blk->bp,
1427                 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1428                                 dp->d_ops->node_hdr_size));
1429
1430         /*
1431          * Save the last hashval in the remaining block for upward propagation.
1432          */
1433         save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1434 }
1435
1436 /*========================================================================
1437  * Routines used for finding things in the Btree.
1438  *========================================================================*/
1439
1440 /*
1441  * Walk down the Btree looking for a particular filename, filling
1442  * in the state structure as we go.
1443  *
1444  * We will set the state structure to point to each of the elements
1445  * in each of the nodes where either the hashval is or should be.
1446  *
1447  * We support duplicate hashval's so for each entry in the current
1448  * node that could contain the desired hashval, descend.  This is a
1449  * pruned depth-first tree search.
1450  */
1451 int                                                     /* error */
1452 xfs_da3_node_lookup_int(
1453         struct xfs_da_state     *state,
1454         int                     *result)
1455 {
1456         struct xfs_da_state_blk *blk;
1457         struct xfs_da_blkinfo   *curr;
1458         struct xfs_da_intnode   *node;
1459         struct xfs_da_node_entry *btree;
1460         struct xfs_da3_icnode_hdr nodehdr;
1461         struct xfs_da_args      *args;
1462         xfs_dablk_t             blkno;
1463         xfs_dahash_t            hashval;
1464         xfs_dahash_t            btreehashval;
1465         int                     probe;
1466         int                     span;
1467         int                     max;
1468         int                     error;
1469         int                     retval;
1470         struct xfs_inode        *dp = state->args->dp;
1471
1472         args = state->args;
1473
1474         /*
1475          * Descend thru the B-tree searching each level for the right
1476          * node to use, until the right hashval is found.
1477          */
1478         blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0;
1479         for (blk = &state->path.blk[0], state->path.active = 1;
1480                          state->path.active <= XFS_DA_NODE_MAXDEPTH;
1481                          blk++, state->path.active++) {
1482                 /*
1483                  * Read the next node down in the tree.
1484                  */
1485                 blk->blkno = blkno;
1486                 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1487                                         -1, &blk->bp, args->whichfork);
1488                 if (error) {
1489                         blk->blkno = 0;
1490                         state->path.active--;
1491                         return error;
1492                 }
1493                 curr = blk->bp->b_addr;
1494                 blk->magic = be16_to_cpu(curr->magic);
1495
1496                 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1497                     blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1498                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1499                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1500                         break;
1501                 }
1502
1503                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1504                     blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1505                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1506                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1507                                                                blk->bp, NULL);
1508                         break;
1509                 }
1510
1511                 blk->magic = XFS_DA_NODE_MAGIC;
1512
1513
1514                 /*
1515                  * Search an intermediate node for a match.
1516                  */
1517                 node = blk->bp->b_addr;
1518                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1519                 btree = dp->d_ops->node_tree_p(node);
1520
1521                 max = nodehdr.count;
1522                 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1523
1524                 /*
1525                  * Binary search.  (note: small blocks will skip loop)
1526                  */
1527                 probe = span = max / 2;
1528                 hashval = args->hashval;
1529                 while (span > 4) {
1530                         span /= 2;
1531                         btreehashval = be32_to_cpu(btree[probe].hashval);
1532                         if (btreehashval < hashval)
1533                                 probe += span;
1534                         else if (btreehashval > hashval)
1535                                 probe -= span;
1536                         else
1537                                 break;
1538                 }
1539                 ASSERT((probe >= 0) && (probe < max));
1540                 ASSERT((span <= 4) ||
1541                         (be32_to_cpu(btree[probe].hashval) == hashval));
1542
1543                 /*
1544                  * Since we may have duplicate hashval's, find the first
1545                  * matching hashval in the node.
1546                  */
1547                 while (probe > 0 &&
1548                        be32_to_cpu(btree[probe].hashval) >= hashval) {
1549                         probe--;
1550                 }
1551                 while (probe < max &&
1552                        be32_to_cpu(btree[probe].hashval) < hashval) {
1553                         probe++;
1554                 }
1555
1556                 /*
1557                  * Pick the right block to descend on.
1558                  */
1559                 if (probe == max) {
1560                         blk->index = max - 1;
1561                         blkno = be32_to_cpu(btree[max - 1].before);
1562                 } else {
1563                         blk->index = probe;
1564                         blkno = be32_to_cpu(btree[probe].before);
1565                 }
1566         }
1567
1568         /*
1569          * A leaf block that ends in the hashval that we are interested in
1570          * (final hashval == search hashval) means that the next block may
1571          * contain more entries with the same hashval, shift upward to the
1572          * next leaf and keep searching.
1573          */
1574         for (;;) {
1575                 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1576                         retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1577                                                         &blk->index, state);
1578                 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1579                         retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1580                         blk->index = args->index;
1581                         args->blkno = blk->blkno;
1582                 } else {
1583                         ASSERT(0);
1584                         return -EFSCORRUPTED;
1585                 }
1586                 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1587                     (blk->hashval == args->hashval)) {
1588                         error = xfs_da3_path_shift(state, &state->path, 1, 1,
1589                                                          &retval);
1590                         if (error)
1591                                 return error;
1592                         if (retval == 0) {
1593                                 continue;
1594                         } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1595                                 /* path_shift() gives ENOENT */
1596                                 retval = -ENOATTR;
1597                         }
1598                 }
1599                 break;
1600         }
1601         *result = retval;
1602         return 0;
1603 }
1604
1605 /*========================================================================
1606  * Utility routines.
1607  *========================================================================*/
1608
1609 /*
1610  * Compare two intermediate nodes for "order".
1611  */
1612 STATIC int
1613 xfs_da3_node_order(
1614         struct xfs_inode *dp,
1615         struct xfs_buf  *node1_bp,
1616         struct xfs_buf  *node2_bp)
1617 {
1618         struct xfs_da_intnode   *node1;
1619         struct xfs_da_intnode   *node2;
1620         struct xfs_da_node_entry *btree1;
1621         struct xfs_da_node_entry *btree2;
1622         struct xfs_da3_icnode_hdr node1hdr;
1623         struct xfs_da3_icnode_hdr node2hdr;
1624
1625         node1 = node1_bp->b_addr;
1626         node2 = node2_bp->b_addr;
1627         dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1628         dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1629         btree1 = dp->d_ops->node_tree_p(node1);
1630         btree2 = dp->d_ops->node_tree_p(node2);
1631
1632         if (node1hdr.count > 0 && node2hdr.count > 0 &&
1633             ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1634              (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1635               be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1636                 return 1;
1637         }
1638         return 0;
1639 }
1640
1641 /*
1642  * Link a new block into a doubly linked list of blocks (of whatever type).
1643  */
1644 int                                                     /* error */
1645 xfs_da3_blk_link(
1646         struct xfs_da_state     *state,
1647         struct xfs_da_state_blk *old_blk,
1648         struct xfs_da_state_blk *new_blk)
1649 {
1650         struct xfs_da_blkinfo   *old_info;
1651         struct xfs_da_blkinfo   *new_info;
1652         struct xfs_da_blkinfo   *tmp_info;
1653         struct xfs_da_args      *args;
1654         struct xfs_buf          *bp;
1655         int                     before = 0;
1656         int                     error;
1657         struct xfs_inode        *dp = state->args->dp;
1658
1659         /*
1660          * Set up environment.
1661          */
1662         args = state->args;
1663         ASSERT(args != NULL);
1664         old_info = old_blk->bp->b_addr;
1665         new_info = new_blk->bp->b_addr;
1666         ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1667                old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1668                old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1669
1670         switch (old_blk->magic) {
1671         case XFS_ATTR_LEAF_MAGIC:
1672                 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1673                 break;
1674         case XFS_DIR2_LEAFN_MAGIC:
1675                 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1676                 break;
1677         case XFS_DA_NODE_MAGIC:
1678                 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1679                 break;
1680         }
1681
1682         /*
1683          * Link blocks in appropriate order.
1684          */
1685         if (before) {
1686                 /*
1687                  * Link new block in before existing block.
1688                  */
1689                 trace_xfs_da_link_before(args);
1690                 new_info->forw = cpu_to_be32(old_blk->blkno);
1691                 new_info->back = old_info->back;
1692                 if (old_info->back) {
1693                         error = xfs_da3_node_read(args->trans, dp,
1694                                                 be32_to_cpu(old_info->back),
1695                                                 -1, &bp, args->whichfork);
1696                         if (error)
1697                                 return error;
1698                         ASSERT(bp != NULL);
1699                         tmp_info = bp->b_addr;
1700                         ASSERT(tmp_info->magic == old_info->magic);
1701                         ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1702                         tmp_info->forw = cpu_to_be32(new_blk->blkno);
1703                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1704                 }
1705                 old_info->back = cpu_to_be32(new_blk->blkno);
1706         } else {
1707                 /*
1708                  * Link new block in after existing block.
1709                  */
1710                 trace_xfs_da_link_after(args);
1711                 new_info->forw = old_info->forw;
1712                 new_info->back = cpu_to_be32(old_blk->blkno);
1713                 if (old_info->forw) {
1714                         error = xfs_da3_node_read(args->trans, dp,
1715                                                 be32_to_cpu(old_info->forw),
1716                                                 -1, &bp, args->whichfork);
1717                         if (error)
1718                                 return error;
1719                         ASSERT(bp != NULL);
1720                         tmp_info = bp->b_addr;
1721                         ASSERT(tmp_info->magic == old_info->magic);
1722                         ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1723                         tmp_info->back = cpu_to_be32(new_blk->blkno);
1724                         xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1725                 }
1726                 old_info->forw = cpu_to_be32(new_blk->blkno);
1727         }
1728
1729         xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1730         xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1731         return 0;
1732 }
1733
1734 /*
1735  * Unlink a block from a doubly linked list of blocks.
1736  */
1737 STATIC int                                              /* error */
1738 xfs_da3_blk_unlink(
1739         struct xfs_da_state     *state,
1740         struct xfs_da_state_blk *drop_blk,
1741         struct xfs_da_state_blk *save_blk)
1742 {
1743         struct xfs_da_blkinfo   *drop_info;
1744         struct xfs_da_blkinfo   *save_info;
1745         struct xfs_da_blkinfo   *tmp_info;
1746         struct xfs_da_args      *args;
1747         struct xfs_buf          *bp;
1748         int                     error;
1749
1750         /*
1751          * Set up environment.
1752          */
1753         args = state->args;
1754         ASSERT(args != NULL);
1755         save_info = save_blk->bp->b_addr;
1756         drop_info = drop_blk->bp->b_addr;
1757         ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1758                save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1759                save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1760         ASSERT(save_blk->magic == drop_blk->magic);
1761         ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1762                (be32_to_cpu(save_info->back) == drop_blk->blkno));
1763         ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1764                (be32_to_cpu(drop_info->back) == save_blk->blkno));
1765
1766         /*
1767          * Unlink the leaf block from the doubly linked chain of leaves.
1768          */
1769         if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1770                 trace_xfs_da_unlink_back(args);
1771                 save_info->back = drop_info->back;
1772                 if (drop_info->back) {
1773                         error = xfs_da3_node_read(args->trans, args->dp,
1774                                                 be32_to_cpu(drop_info->back),
1775                                                 -1, &bp, args->whichfork);
1776                         if (error)
1777                                 return error;
1778                         ASSERT(bp != NULL);
1779                         tmp_info = bp->b_addr;
1780                         ASSERT(tmp_info->magic == save_info->magic);
1781                         ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1782                         tmp_info->forw = cpu_to_be32(save_blk->blkno);
1783                         xfs_trans_log_buf(args->trans, bp, 0,
1784                                                     sizeof(*tmp_info) - 1);
1785                 }
1786         } else {
1787                 trace_xfs_da_unlink_forward(args);
1788                 save_info->forw = drop_info->forw;
1789                 if (drop_info->forw) {
1790                         error = xfs_da3_node_read(args->trans, args->dp,
1791                                                 be32_to_cpu(drop_info->forw),
1792                                                 -1, &bp, args->whichfork);
1793                         if (error)
1794                                 return error;
1795                         ASSERT(bp != NULL);
1796                         tmp_info = bp->b_addr;
1797                         ASSERT(tmp_info->magic == save_info->magic);
1798                         ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1799                         tmp_info->back = cpu_to_be32(save_blk->blkno);
1800                         xfs_trans_log_buf(args->trans, bp, 0,
1801                                                     sizeof(*tmp_info) - 1);
1802                 }
1803         }
1804
1805         xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1806         return 0;
1807 }
1808
1809 /*
1810  * Move a path "forward" or "!forward" one block at the current level.
1811  *
1812  * This routine will adjust a "path" to point to the next block
1813  * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1814  * Btree, including updating pointers to the intermediate nodes between
1815  * the new bottom and the root.
1816  */
1817 int                                                     /* error */
1818 xfs_da3_path_shift(
1819         struct xfs_da_state     *state,
1820         struct xfs_da_state_path *path,
1821         int                     forward,
1822         int                     release,
1823         int                     *result)
1824 {
1825         struct xfs_da_state_blk *blk;
1826         struct xfs_da_blkinfo   *info;
1827         struct xfs_da_intnode   *node;
1828         struct xfs_da_args      *args;
1829         struct xfs_da_node_entry *btree;
1830         struct xfs_da3_icnode_hdr nodehdr;
1831         struct xfs_buf          *bp;
1832         xfs_dablk_t             blkno = 0;
1833         int                     level;
1834         int                     error;
1835         struct xfs_inode        *dp = state->args->dp;
1836
1837         trace_xfs_da_path_shift(state->args);
1838
1839         /*
1840          * Roll up the Btree looking for the first block where our
1841          * current index is not at the edge of the block.  Note that
1842          * we skip the bottom layer because we want the sibling block.
1843          */
1844         args = state->args;
1845         ASSERT(args != NULL);
1846         ASSERT(path != NULL);
1847         ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1848         level = (path->active-1) - 1;   /* skip bottom layer in path */
1849         for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1850                 node = blk->bp->b_addr;
1851                 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1852                 btree = dp->d_ops->node_tree_p(node);
1853
1854                 if (forward && (blk->index < nodehdr.count - 1)) {
1855                         blk->index++;
1856                         blkno = be32_to_cpu(btree[blk->index].before);
1857                         break;
1858                 } else if (!forward && (blk->index > 0)) {
1859                         blk->index--;
1860                         blkno = be32_to_cpu(btree[blk->index].before);
1861                         break;
1862                 }
1863         }
1864         if (level < 0) {
1865                 *result = -ENOENT;      /* we're out of our tree */
1866                 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1867                 return 0;
1868         }
1869
1870         /*
1871          * Roll down the edge of the subtree until we reach the
1872          * same depth we were at originally.
1873          */
1874         for (blk++, level++; level < path->active; blk++, level++) {
1875                 /*
1876                  * Read the next child block into a local buffer.
1877                  */
1878                 error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp,
1879                                           args->whichfork);
1880                 if (error)
1881                         return error;
1882
1883                 /*
1884                  * Release the old block (if it's dirty, the trans doesn't
1885                  * actually let go) and swap the local buffer into the path
1886                  * structure. This ensures failure of the above read doesn't set
1887                  * a NULL buffer in an active slot in the path.
1888                  */
1889                 if (release)
1890                         xfs_trans_brelse(args->trans, blk->bp);
1891                 blk->blkno = blkno;
1892                 blk->bp = bp;
1893
1894                 info = blk->bp->b_addr;
1895                 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1896                        info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1897                        info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1898                        info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1899                        info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1900                        info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1901
1902
1903                 /*
1904                  * Note: we flatten the magic number to a single type so we
1905                  * don't have to compare against crc/non-crc types elsewhere.
1906                  */
1907                 switch (be16_to_cpu(info->magic)) {
1908                 case XFS_DA_NODE_MAGIC:
1909                 case XFS_DA3_NODE_MAGIC:
1910                         blk->magic = XFS_DA_NODE_MAGIC;
1911                         node = (xfs_da_intnode_t *)info;
1912                         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1913                         btree = dp->d_ops->node_tree_p(node);
1914                         blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1915                         if (forward)
1916                                 blk->index = 0;
1917                         else
1918                                 blk->index = nodehdr.count - 1;
1919                         blkno = be32_to_cpu(btree[blk->index].before);
1920                         break;
1921                 case XFS_ATTR_LEAF_MAGIC:
1922                 case XFS_ATTR3_LEAF_MAGIC:
1923                         blk->magic = XFS_ATTR_LEAF_MAGIC;
1924                         ASSERT(level == path->active-1);
1925                         blk->index = 0;
1926                         blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1927                         break;
1928                 case XFS_DIR2_LEAFN_MAGIC:
1929                 case XFS_DIR3_LEAFN_MAGIC:
1930                         blk->magic = XFS_DIR2_LEAFN_MAGIC;
1931                         ASSERT(level == path->active-1);
1932                         blk->index = 0;
1933                         blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1934                                                                blk->bp, NULL);
1935                         break;
1936                 default:
1937                         ASSERT(0);
1938                         break;
1939                 }
1940         }
1941         *result = 0;
1942         return 0;
1943 }
1944
1945
1946 /*========================================================================
1947  * Utility routines.
1948  *========================================================================*/
1949
1950 /*
1951  * Implement a simple hash on a character string.
1952  * Rotate the hash value by 7 bits, then XOR each character in.
1953  * This is implemented with some source-level loop unrolling.
1954  */
1955 xfs_dahash_t
1956 xfs_da_hashname(const __uint8_t *name, int namelen)
1957 {
1958         xfs_dahash_t hash;
1959
1960         /*
1961          * Do four characters at a time as long as we can.
1962          */
1963         for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1964                 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1965                        (name[3] << 0) ^ rol32(hash, 7 * 4);
1966
1967         /*
1968          * Now do the rest of the characters.
1969          */
1970         switch (namelen) {
1971         case 3:
1972                 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1973                        rol32(hash, 7 * 3);
1974         case 2:
1975                 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1976         case 1:
1977                 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1978         default: /* case 0: */
1979                 return hash;
1980         }
1981 }
1982
1983 enum xfs_dacmp
1984 xfs_da_compname(
1985         struct xfs_da_args *args,
1986         const unsigned char *name,
1987         int             len)
1988 {
1989         return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1990                                         XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1991 }
1992
1993 static xfs_dahash_t
1994 xfs_default_hashname(
1995         struct xfs_name *name)
1996 {
1997         return xfs_da_hashname(name->name, name->len);
1998 }
1999
2000 const struct xfs_nameops xfs_default_nameops = {
2001         .hashname       = xfs_default_hashname,
2002         .compname       = xfs_da_compname
2003 };
2004
2005 int
2006 xfs_da_grow_inode_int(
2007         struct xfs_da_args      *args,
2008         xfs_fileoff_t           *bno,
2009         int                     count)
2010 {
2011         struct xfs_trans        *tp = args->trans;
2012         struct xfs_inode        *dp = args->dp;
2013         int                     w = args->whichfork;
2014         xfs_rfsblock_t          nblks = dp->i_d.di_nblocks;
2015         struct xfs_bmbt_irec    map, *mapp;
2016         int                     nmap, error, got, i, mapi;
2017
2018         /*
2019          * Find a spot in the file space to put the new block.
2020          */
2021         error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2022         if (error)
2023                 return error;
2024
2025         /*
2026          * Try mapping it in one filesystem block.
2027          */
2028         nmap = 1;
2029         ASSERT(args->firstblock != NULL);
2030         error = xfs_bmapi_write(tp, dp, *bno, count,
2031                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2032                         args->firstblock, args->total, &map, &nmap,
2033                         args->flist);
2034         if (error)
2035                 return error;
2036
2037         ASSERT(nmap <= 1);
2038         if (nmap == 1) {
2039                 mapp = &map;
2040                 mapi = 1;
2041         } else if (nmap == 0 && count > 1) {
2042                 xfs_fileoff_t           b;
2043                 int                     c;
2044
2045                 /*
2046                  * If we didn't get it and the block might work if fragmented,
2047                  * try without the CONTIG flag.  Loop until we get it all.
2048                  */
2049                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2050                 for (b = *bno, mapi = 0; b < *bno + count; ) {
2051                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2052                         c = (int)(*bno + count - b);
2053                         error = xfs_bmapi_write(tp, dp, b, c,
2054                                         xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2055                                         args->firstblock, args->total,
2056                                         &mapp[mapi], &nmap, args->flist);
2057                         if (error)
2058                                 goto out_free_map;
2059                         if (nmap < 1)
2060                                 break;
2061                         mapi += nmap;
2062                         b = mapp[mapi - 1].br_startoff +
2063                             mapp[mapi - 1].br_blockcount;
2064                 }
2065         } else {
2066                 mapi = 0;
2067                 mapp = NULL;
2068         }
2069
2070         /*
2071          * Count the blocks we got, make sure it matches the total.
2072          */
2073         for (i = 0, got = 0; i < mapi; i++)
2074                 got += mapp[i].br_blockcount;
2075         if (got != count || mapp[0].br_startoff != *bno ||
2076             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2077             *bno + count) {
2078                 error = -ENOSPC;
2079                 goto out_free_map;
2080         }
2081
2082         /* account for newly allocated blocks in reserved blocks total */
2083         args->total -= dp->i_d.di_nblocks - nblks;
2084
2085 out_free_map:
2086         if (mapp != &map)
2087                 kmem_free(mapp);
2088         return error;
2089 }
2090
2091 /*
2092  * Add a block to the btree ahead of the file.
2093  * Return the new block number to the caller.
2094  */
2095 int
2096 xfs_da_grow_inode(
2097         struct xfs_da_args      *args,
2098         xfs_dablk_t             *new_blkno)
2099 {
2100         xfs_fileoff_t           bno;
2101         int                     error;
2102
2103         trace_xfs_da_grow_inode(args);
2104
2105         bno = args->geo->leafblk;
2106         error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2107         if (!error)
2108                 *new_blkno = (xfs_dablk_t)bno;
2109         return error;
2110 }
2111
2112 /*
2113  * Ick.  We need to always be able to remove a btree block, even
2114  * if there's no space reservation because the filesystem is full.
2115  * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2116  * It swaps the target block with the last block in the file.  The
2117  * last block in the file can always be removed since it can't cause
2118  * a bmap btree split to do that.
2119  */
2120 STATIC int
2121 xfs_da3_swap_lastblock(
2122         struct xfs_da_args      *args,
2123         xfs_dablk_t             *dead_blknop,
2124         struct xfs_buf          **dead_bufp)
2125 {
2126         struct xfs_da_blkinfo   *dead_info;
2127         struct xfs_da_blkinfo   *sib_info;
2128         struct xfs_da_intnode   *par_node;
2129         struct xfs_da_intnode   *dead_node;
2130         struct xfs_dir2_leaf    *dead_leaf2;
2131         struct xfs_da_node_entry *btree;
2132         struct xfs_da3_icnode_hdr par_hdr;
2133         struct xfs_inode        *dp;
2134         struct xfs_trans        *tp;
2135         struct xfs_mount        *mp;
2136         struct xfs_buf          *dead_buf;
2137         struct xfs_buf          *last_buf;
2138         struct xfs_buf          *sib_buf;
2139         struct xfs_buf          *par_buf;
2140         xfs_dahash_t            dead_hash;
2141         xfs_fileoff_t           lastoff;
2142         xfs_dablk_t             dead_blkno;
2143         xfs_dablk_t             last_blkno;
2144         xfs_dablk_t             sib_blkno;
2145         xfs_dablk_t             par_blkno;
2146         int                     error;
2147         int                     w;
2148         int                     entno;
2149         int                     level;
2150         int                     dead_level;
2151
2152         trace_xfs_da_swap_lastblock(args);
2153
2154         dead_buf = *dead_bufp;
2155         dead_blkno = *dead_blknop;
2156         tp = args->trans;
2157         dp = args->dp;
2158         w = args->whichfork;
2159         ASSERT(w == XFS_DATA_FORK);
2160         mp = dp->i_mount;
2161         lastoff = args->geo->freeblk;
2162         error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2163         if (error)
2164                 return error;
2165         if (unlikely(lastoff == 0)) {
2166                 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2167                                  mp);
2168                 return -EFSCORRUPTED;
2169         }
2170         /*
2171          * Read the last block in the btree space.
2172          */
2173         last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2174         error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2175         if (error)
2176                 return error;
2177         /*
2178          * Copy the last block into the dead buffer and log it.
2179          */
2180         memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2181         xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2182         dead_info = dead_buf->b_addr;
2183         /*
2184          * Get values from the moved block.
2185          */
2186         if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2187             dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2188                 struct xfs_dir3_icleaf_hdr leafhdr;
2189                 struct xfs_dir2_leaf_entry *ents;
2190
2191                 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2192                 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2193                 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2194                 dead_level = 0;
2195                 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2196         } else {
2197                 struct xfs_da3_icnode_hdr deadhdr;
2198
2199                 dead_node = (xfs_da_intnode_t *)dead_info;
2200                 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2201                 btree = dp->d_ops->node_tree_p(dead_node);
2202                 dead_level = deadhdr.level;
2203                 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2204         }
2205         sib_buf = par_buf = NULL;
2206         /*
2207          * If the moved block has a left sibling, fix up the pointers.
2208          */
2209         if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2210                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2211                 if (error)
2212                         goto done;
2213                 sib_info = sib_buf->b_addr;
2214                 if (unlikely(
2215                     be32_to_cpu(sib_info->forw) != last_blkno ||
2216                     sib_info->magic != dead_info->magic)) {
2217                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2218                                          XFS_ERRLEVEL_LOW, mp);
2219                         error = -EFSCORRUPTED;
2220                         goto done;
2221                 }
2222                 sib_info->forw = cpu_to_be32(dead_blkno);
2223                 xfs_trans_log_buf(tp, sib_buf,
2224                         XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2225                                         sizeof(sib_info->forw)));
2226                 sib_buf = NULL;
2227         }
2228         /*
2229          * If the moved block has a right sibling, fix up the pointers.
2230          */
2231         if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2232                 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2233                 if (error)
2234                         goto done;
2235                 sib_info = sib_buf->b_addr;
2236                 if (unlikely(
2237                        be32_to_cpu(sib_info->back) != last_blkno ||
2238                        sib_info->magic != dead_info->magic)) {
2239                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2240                                          XFS_ERRLEVEL_LOW, mp);
2241                         error = -EFSCORRUPTED;
2242                         goto done;
2243                 }
2244                 sib_info->back = cpu_to_be32(dead_blkno);
2245                 xfs_trans_log_buf(tp, sib_buf,
2246                         XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2247                                         sizeof(sib_info->back)));
2248                 sib_buf = NULL;
2249         }
2250         par_blkno = args->geo->leafblk;
2251         level = -1;
2252         /*
2253          * Walk down the tree looking for the parent of the moved block.
2254          */
2255         for (;;) {
2256                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2257                 if (error)
2258                         goto done;
2259                 par_node = par_buf->b_addr;
2260                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2261                 if (level >= 0 && level != par_hdr.level + 1) {
2262                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2263                                          XFS_ERRLEVEL_LOW, mp);
2264                         error = -EFSCORRUPTED;
2265                         goto done;
2266                 }
2267                 level = par_hdr.level;
2268                 btree = dp->d_ops->node_tree_p(par_node);
2269                 for (entno = 0;
2270                      entno < par_hdr.count &&
2271                      be32_to_cpu(btree[entno].hashval) < dead_hash;
2272                      entno++)
2273                         continue;
2274                 if (entno == par_hdr.count) {
2275                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2276                                          XFS_ERRLEVEL_LOW, mp);
2277                         error = -EFSCORRUPTED;
2278                         goto done;
2279                 }
2280                 par_blkno = be32_to_cpu(btree[entno].before);
2281                 if (level == dead_level + 1)
2282                         break;
2283                 xfs_trans_brelse(tp, par_buf);
2284                 par_buf = NULL;
2285         }
2286         /*
2287          * We're in the right parent block.
2288          * Look for the right entry.
2289          */
2290         for (;;) {
2291                 for (;
2292                      entno < par_hdr.count &&
2293                      be32_to_cpu(btree[entno].before) != last_blkno;
2294                      entno++)
2295                         continue;
2296                 if (entno < par_hdr.count)
2297                         break;
2298                 par_blkno = par_hdr.forw;
2299                 xfs_trans_brelse(tp, par_buf);
2300                 par_buf = NULL;
2301                 if (unlikely(par_blkno == 0)) {
2302                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2303                                          XFS_ERRLEVEL_LOW, mp);
2304                         error = -EFSCORRUPTED;
2305                         goto done;
2306                 }
2307                 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2308                 if (error)
2309                         goto done;
2310                 par_node = par_buf->b_addr;
2311                 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2312                 if (par_hdr.level != level) {
2313                         XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2314                                          XFS_ERRLEVEL_LOW, mp);
2315                         error = -EFSCORRUPTED;
2316                         goto done;
2317                 }
2318                 btree = dp->d_ops->node_tree_p(par_node);
2319                 entno = 0;
2320         }
2321         /*
2322          * Update the parent entry pointing to the moved block.
2323          */
2324         btree[entno].before = cpu_to_be32(dead_blkno);
2325         xfs_trans_log_buf(tp, par_buf,
2326                 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2327                                 sizeof(btree[entno].before)));
2328         *dead_blknop = last_blkno;
2329         *dead_bufp = last_buf;
2330         return 0;
2331 done:
2332         if (par_buf)
2333                 xfs_trans_brelse(tp, par_buf);
2334         if (sib_buf)
2335                 xfs_trans_brelse(tp, sib_buf);
2336         xfs_trans_brelse(tp, last_buf);
2337         return error;
2338 }
2339
2340 /*
2341  * Remove a btree block from a directory or attribute.
2342  */
2343 int
2344 xfs_da_shrink_inode(
2345         xfs_da_args_t   *args,
2346         xfs_dablk_t     dead_blkno,
2347         struct xfs_buf  *dead_buf)
2348 {
2349         xfs_inode_t *dp;
2350         int done, error, w, count;
2351         xfs_trans_t *tp;
2352
2353         trace_xfs_da_shrink_inode(args);
2354
2355         dp = args->dp;
2356         w = args->whichfork;
2357         tp = args->trans;
2358         count = args->geo->fsbcount;
2359         for (;;) {
2360                 /*
2361                  * Remove extents.  If we get ENOSPC for a dir we have to move
2362                  * the last block to the place we want to kill.
2363                  */
2364                 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2365                                     xfs_bmapi_aflag(w), 0, args->firstblock,
2366                                     args->flist, &done);
2367                 if (error == -ENOSPC) {
2368                         if (w != XFS_DATA_FORK)
2369                                 break;
2370                         error = xfs_da3_swap_lastblock(args, &dead_blkno,
2371                                                       &dead_buf);
2372                         if (error)
2373                                 break;
2374                 } else {
2375                         break;
2376                 }
2377         }
2378         xfs_trans_binval(tp, dead_buf);
2379         return error;
2380 }
2381
2382 /*
2383  * See if the mapping(s) for this btree block are valid, i.e.
2384  * don't contain holes, are logically contiguous, and cover the whole range.
2385  */
2386 STATIC int
2387 xfs_da_map_covers_blocks(
2388         int             nmap,
2389         xfs_bmbt_irec_t *mapp,
2390         xfs_dablk_t     bno,
2391         int             count)
2392 {
2393         int             i;
2394         xfs_fileoff_t   off;
2395
2396         for (i = 0, off = bno; i < nmap; i++) {
2397                 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2398                     mapp[i].br_startblock == DELAYSTARTBLOCK) {
2399                         return 0;
2400                 }
2401                 if (off != mapp[i].br_startoff) {
2402                         return 0;
2403                 }
2404                 off += mapp[i].br_blockcount;
2405         }
2406         return off == bno + count;
2407 }
2408
2409 /*
2410  * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2411  *
2412  * For the single map case, it is assumed that the caller has provided a pointer
2413  * to a valid xfs_buf_map.  For the multiple map case, this function will
2414  * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2415  * map pointer with the allocated map.
2416  */
2417 static int
2418 xfs_buf_map_from_irec(
2419         struct xfs_mount        *mp,
2420         struct xfs_buf_map      **mapp,
2421         int                     *nmaps,
2422         struct xfs_bmbt_irec    *irecs,
2423         int                     nirecs)
2424 {
2425         struct xfs_buf_map      *map;
2426         int                     i;
2427
2428         ASSERT(*nmaps == 1);
2429         ASSERT(nirecs >= 1);
2430
2431         if (nirecs > 1) {
2432                 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2433                                   KM_SLEEP | KM_NOFS);
2434                 if (!map)
2435                         return -ENOMEM;
2436                 *mapp = map;
2437         }
2438
2439         *nmaps = nirecs;
2440         map = *mapp;
2441         for (i = 0; i < *nmaps; i++) {
2442                 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2443                        irecs[i].br_startblock != HOLESTARTBLOCK);
2444                 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2445                 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2446         }
2447         return 0;
2448 }
2449
2450 /*
2451  * Map the block we are given ready for reading. There are three possible return
2452  * values:
2453  *      -1 - will be returned if we land in a hole and mappedbno == -2 so the
2454  *           caller knows not to execute a subsequent read.
2455  *       0 - if we mapped the block successfully
2456  *      >0 - positive error number if there was an error.
2457  */
2458 static int
2459 xfs_dabuf_map(
2460         struct xfs_inode        *dp,
2461         xfs_dablk_t             bno,
2462         xfs_daddr_t             mappedbno,
2463         int                     whichfork,
2464         struct xfs_buf_map      **map,
2465         int                     *nmaps)
2466 {
2467         struct xfs_mount        *mp = dp->i_mount;
2468         int                     nfsb;
2469         int                     error = 0;
2470         struct xfs_bmbt_irec    irec;
2471         struct xfs_bmbt_irec    *irecs = &irec;
2472         int                     nirecs;
2473
2474         ASSERT(map && *map);
2475         ASSERT(*nmaps == 1);
2476
2477         if (whichfork == XFS_DATA_FORK)
2478                 nfsb = mp->m_dir_geo->fsbcount;
2479         else
2480                 nfsb = mp->m_attr_geo->fsbcount;
2481
2482         /*
2483          * Caller doesn't have a mapping.  -2 means don't complain
2484          * if we land in a hole.
2485          */
2486         if (mappedbno == -1 || mappedbno == -2) {
2487                 /*
2488                  * Optimize the one-block case.
2489                  */
2490                 if (nfsb != 1)
2491                         irecs = kmem_zalloc(sizeof(irec) * nfsb,
2492                                             KM_SLEEP | KM_NOFS);
2493
2494                 nirecs = nfsb;
2495                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2496                                        &nirecs, xfs_bmapi_aflag(whichfork));
2497                 if (error)
2498                         goto out;
2499         } else {
2500                 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2501                 irecs->br_startoff = (xfs_fileoff_t)bno;
2502                 irecs->br_blockcount = nfsb;
2503                 irecs->br_state = 0;
2504                 nirecs = 1;
2505         }
2506
2507         if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2508                 error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2509                 if (unlikely(error == -EFSCORRUPTED)) {
2510                         if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2511                                 int i;
2512                                 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2513                                         __func__, (long long)bno,
2514                                         (long long)dp->i_ino);
2515                                 for (i = 0; i < *nmaps; i++) {
2516                                         xfs_alert(mp,
2517 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2518                                                 i,
2519                                                 (long long)irecs[i].br_startoff,
2520                                                 (long long)irecs[i].br_startblock,
2521                                                 (long long)irecs[i].br_blockcount,
2522                                                 irecs[i].br_state);
2523                                 }
2524                         }
2525                         XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2526                                          XFS_ERRLEVEL_LOW, mp);
2527                 }
2528                 goto out;
2529         }
2530         error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2531 out:
2532         if (irecs != &irec)
2533                 kmem_free(irecs);
2534         return error;
2535 }
2536
2537 /*
2538  * Get a buffer for the dir/attr block.
2539  */
2540 int
2541 xfs_da_get_buf(
2542         struct xfs_trans        *trans,
2543         struct xfs_inode        *dp,
2544         xfs_dablk_t             bno,
2545         xfs_daddr_t             mappedbno,
2546         struct xfs_buf          **bpp,
2547         int                     whichfork)
2548 {
2549         struct xfs_buf          *bp;
2550         struct xfs_buf_map      map;
2551         struct xfs_buf_map      *mapp;
2552         int                     nmap;
2553         int                     error;
2554
2555         *bpp = NULL;
2556         mapp = &map;
2557         nmap = 1;
2558         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2559                                 &mapp, &nmap);
2560         if (error) {
2561                 /* mapping a hole is not an error, but we don't continue */
2562                 if (error == -1)
2563                         error = 0;
2564                 goto out_free;
2565         }
2566
2567         bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2568                                     mapp, nmap, 0);
2569         error = bp ? bp->b_error : -EIO;
2570         if (error) {
2571                 if (bp)
2572                         xfs_trans_brelse(trans, bp);
2573                 goto out_free;
2574         }
2575
2576         *bpp = bp;
2577
2578 out_free:
2579         if (mapp != &map)
2580                 kmem_free(mapp);
2581
2582         return error;
2583 }
2584
2585 /*
2586  * Get a buffer for the dir/attr block, fill in the contents.
2587  */
2588 int
2589 xfs_da_read_buf(
2590         struct xfs_trans        *trans,
2591         struct xfs_inode        *dp,
2592         xfs_dablk_t             bno,
2593         xfs_daddr_t             mappedbno,
2594         struct xfs_buf          **bpp,
2595         int                     whichfork,
2596         const struct xfs_buf_ops *ops)
2597 {
2598         struct xfs_buf          *bp;
2599         struct xfs_buf_map      map;
2600         struct xfs_buf_map      *mapp;
2601         int                     nmap;
2602         int                     error;
2603
2604         *bpp = NULL;
2605         mapp = &map;
2606         nmap = 1;
2607         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2608                                 &mapp, &nmap);
2609         if (error) {
2610                 /* mapping a hole is not an error, but we don't continue */
2611                 if (error == -1)
2612                         error = 0;
2613                 goto out_free;
2614         }
2615
2616         error = xfs_trans_read_buf_map(dp->i_mount, trans,
2617                                         dp->i_mount->m_ddev_targp,
2618                                         mapp, nmap, 0, &bp, ops);
2619         if (error)
2620                 goto out_free;
2621
2622         if (whichfork == XFS_ATTR_FORK)
2623                 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2624         else
2625                 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2626         *bpp = bp;
2627 out_free:
2628         if (mapp != &map)
2629                 kmem_free(mapp);
2630
2631         return error;
2632 }
2633
2634 /*
2635  * Readahead the dir/attr block.
2636  */
2637 xfs_daddr_t
2638 xfs_da_reada_buf(
2639         struct xfs_inode        *dp,
2640         xfs_dablk_t             bno,
2641         xfs_daddr_t             mappedbno,
2642         int                     whichfork,
2643         const struct xfs_buf_ops *ops)
2644 {
2645         struct xfs_buf_map      map;
2646         struct xfs_buf_map      *mapp;
2647         int                     nmap;
2648         int                     error;
2649
2650         mapp = &map;
2651         nmap = 1;
2652         error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2653                                 &mapp, &nmap);
2654         if (error) {
2655                 /* mapping a hole is not an error, but we don't continue */
2656                 if (error == -1)
2657                         error = 0;
2658                 goto out_free;
2659         }
2660
2661         mappedbno = mapp[0].bm_bn;
2662         xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2663
2664 out_free:
2665         if (mapp != &map)
2666                 kmem_free(mapp);
2667
2668         if (error)
2669                 return -1;
2670         return mappedbno;
2671 }